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_old, 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_old *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(*ff)) {
 722                 rc = 0;
 723                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 724                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 725                 /*
 726                  * XXX: use 0 as the index for compatibility, the caller will
 727                  * handle index related issues when necessary.
 728                  */
 729                 if (!rc)
 730                         ostid_to_fid(fid, ostid, 0);
 731         } else if (rc == sizeof(struct filter_fid)) {
 732                 rc = 1;
 733         } else if (rc >= 0) {
 734                 rc = -EINVAL;
 735         }
 736
 737         return rc;
 738 }
 739
 740 static int osd_lma_self_repair(struct osd_thread_info *info,
 741                                struct osd_device *osd, struct inode *inode,
 742                                const struct lu_fid *fid, __u32 compat)
 743 {
 744         handle_t *jh;
 745         int rc;
 746
 747         LASSERT(current->journal_info == NULL);
 748
 749         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 750                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 751         if (IS_ERR(jh)) {
 752                 rc = PTR_ERR(jh);
 753                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 754                       osd_name(osd), rc);
 755                 return rc;
 756         }
 757
 758         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 759         if (rc != 0)
 760                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 761                       osd_name(osd), rc);
 762         ldiskfs_journal_stop(jh);
 763         return rc;
 764 }
 765
 766 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 767 {
 768         struct osd_thread_info *info = osd_oti_get(env);
 769         struct osd_device *osd = osd_obj2dev(obj);
 770         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 771         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 772         struct inode *inode = obj->oo_inode;
 773         struct dentry *dentry = &info->oti_obj_dentry;
 774         struct lu_fid *fid = NULL;
 775         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 776         int rc;
 777
 778         ENTRY;
 779
 780         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 781                              (void *)loa, sizeof(*loa));
 782         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 783                 fid = &lma->lma_self_fid;
 784                 rc = osd_get_idif(info, inode, dentry, fid);
 785                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 786                         /*
 787                          * For the given OST-object, if it has neither LMA nor
 788                          * FID in XATTR_NAME_FID, then the given FID (which is
 789                          * contained in the @obj, from client RPC for locating
 790                          * the OST-object) is trusted. We use it to generate
 791                          * the LMA.
 792                          */
 793                         osd_lma_self_repair(info, osd, inode, rfid,
 794                                             LMAC_FID_ON_OST);
 795                         RETURN(0);
 796                 }
 797         }
 798
 799         if (rc < 0)
 800                 RETURN(rc);
 801
 802         if (rc > 0) {
 803                 rc = 0;
 804                 lustre_lma_swab(lma);
 805                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 806                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 807                               S_ISREG(inode->i_mode)))) {
 808                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 809                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 810                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 811                               PFID(rfid), inode->i_ino);
 812                         rc = -EOPNOTSUPP;
 813                 } else {
 814                         fid = &lma->lma_self_fid;
 815                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 816                             osd->od_is_ost)
 817                                 obj->oo_pfid_in_lma = 1;
 818                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 819                             !osd->od_is_ost)
 820                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 821                 }
 822         }
 823
 824         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 825                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 826                         struct ost_id   *oi   = &info->oti_ostid;
 827                         struct lu_fid   *fid1 = &info->oti_fid3;
 828                         __u32            idx  = fid_idif_ost_idx(rfid);
 829
 830                         /*
 831                          * For old IDIF, the OST index is not part of the IDIF,
 832                          * Means that different OSTs may have the same IDIFs.
 833                          * Under such case, we need to make some compatible
 834                          * check to make sure to trigger OI scrub properly.
 835                          */
 836                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 837                                 /* Given @rfid is new, LMA is old. */
 838                                 fid_to_ostid(fid, oi);
 839                                 ostid_to_fid(fid1, oi, idx);
 840                                 if (lu_fid_eq(fid1, rfid)) {
 841                                         if (osd->od_index_in_idif)
 842                                                 osd_lma_self_repair(info, osd,
 843                                                         inode, rfid,
 844                                                         LMAC_FID_ON_OST);
 845                                         RETURN(0);
 846                                 }
 847                         }
 848                 }
 849
 850                 rc = -EREMCHG;
 851         }
 852
 853         RETURN(rc);
 854 }
 855
 856 struct osd_check_lmv_buf {
 857 #ifdef HAVE_DIR_CONTEXT
 858         /* please keep it as first member */
 859         struct dir_context ctx;
 860 #endif
 861         struct osd_thread_info *oclb_info;
 862         struct osd_device *oclb_dev;
 863         struct osd_idmap_cache *oclb_oic;
 864         int oclb_items;
 865         bool oclb_found;
 866 };
 867
 868 /**
 869  * It is called internally by ->readdir() to filter out the
 870  * local slave object's FID of the striped directory.
 871  *
 872  * \retval      1 found the local slave's FID
 873  * \retval      0 continue to check next item
 874  * \retval      -ve for failure
 875  */
 876 #ifdef HAVE_FILLDIR_USE_CTX
 877 static int osd_stripe_dir_filldir(struct dir_context *buf,
 878 #else
 879 static int osd_stripe_dir_filldir(void *buf,
 880 #endif
 881                                   const char *name, int namelen,
 882                                   loff_t offset, __u64 ino, unsigned int d_type)
 883 {
 884         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 885         struct osd_thread_info *oti = oclb->oclb_info;
 886         struct lu_fid *fid = &oti->oti_fid3;
 887         struct osd_inode_id *id = &oti->oti_id3;
 888         struct osd_device *dev = oclb->oclb_dev;
 889         struct osd_idmap_cache *oic = oclb->oclb_oic;
 890         struct inode *inode;
 891
 892         oclb->oclb_items++;
 893
 894         if (name[0] == '.')
 895                 return 0;
 896
 897         fid_zero(fid);
 898         sscanf(name + 1, SFID, RFID(fid));
 899         if (!fid_is_sane(fid))
 900                 return 0;
 901
 902         if (osd_remote_fid(oti->oti_env, dev, fid))
 903                 return 0;
 904
 905         osd_id_gen(id, ino, OSD_OII_NOGEN);
 906         inode = osd_iget(oti, dev, id);
 907         if (IS_ERR(inode))
 908                 return PTR_ERR(inode);
 909
 910         iput(inode);
 911         osd_add_oi_cache(oti, dev, id, fid);
 912         oic->oic_fid = *fid;
 913         oic->oic_lid = *id;
 914         oic->oic_dev = dev;
 915         osd_oii_insert(dev, oic, true);
 916         oclb->oclb_found = true;
 917
 918         return 1;
 919 }
 920
 921 /*
 922  * When lookup item under striped directory, we need to locate the master
 923  * MDT-object of the striped directory firstly, then the client will send
 924  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 925  * and the item's name. If the system is restored from MDT file level backup,
 926  * then before the OI scrub completely built the OI files, the OI mappings of
 927  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 928  * not a problem for the master MDT-object. Because when locate the master
 929  * MDT-object, we will do name based lookup (for the striped directory itself)
 930  * firstly, during such process we can setup the correct OI mapping for the
 931  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 932  * the client will not trigger name based lookup on the MDT to locate the slave
 933  * MDT-object before locating item under the striped directory, then when
 934  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 935  * is invalid and does not know what the right OI mapping is, then the MDT has
 936  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 937  * the OI file, related OI mapping is unknown yet, please try again later. And
 938  * then client will re-try the RPC again and again until related OI mapping has
 939  * been updated. That is quite inefficient.
 940  *
 941  * To resolve above trouble, we will handle it as the following two cases:
 942  *
 943  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 944  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 945  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 946  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 947  *    directly. Please check osd_fid_lookup().
 948  *
 949  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 950  *    Under such case, during lookup the master MDT-object, we will lookup the
 951  *    slave MDT-object via readdir against the master MDT-object, because the
 952  *    slave MDT-objects information are stored as sub-directories with the name
 953  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 954  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 955  *    the correct OI mapping for the slave MDT-object.
 956  */
 957 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 958                          struct inode *inode, struct osd_idmap_cache *oic)
 959 {
 960         struct lu_buf *buf = &oti->oti_big_buf;
 961         struct dentry *dentry = &oti->oti_obj_dentry;
 962         struct file *filp = &oti->oti_file;
 963         const struct file_operations *fops;
 964         struct lmv_mds_md_v1 *lmv1;
 965         struct osd_check_lmv_buf oclb = {
 966 #ifdef HAVE_DIR_CONTEXT
 967                 .ctx.actor = osd_stripe_dir_filldir,
 968 #endif
 969                 .oclb_info = oti,
 970                 .oclb_dev = dev,
 971                 .oclb_oic = oic,
 972                 .oclb_found = false,
 973         };
 974         int rc = 0;
 975
 976         ENTRY;
 977
 978 again:
 979         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 980                              buf->lb_len);
 981         if (rc == -ERANGE) {
 982                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 983                 if (rc > 0) {
 984                         lu_buf_realloc(buf, rc);
 985                         if (buf->lb_buf == NULL)
 986                                 GOTO(out, rc = -ENOMEM);
 987
 988                         goto again;
 989                 }
 990         }
 991
 992         if (unlikely(rc == 0 || rc == -ENODATA))
 993                 GOTO(out, rc = 0);
 994
 995         if (rc < 0)
 996                 GOTO(out, rc);
 997
 998         if (unlikely(buf->lb_buf == NULL)) {
 999                 lu_buf_realloc(buf, rc);
1000                 if (buf->lb_buf == NULL)
1001                         GOTO(out, rc = -ENOMEM);
1002
1003                 goto again;
1004         }
1005
1006         lmv1 = buf->lb_buf;
1007         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1008                 GOTO(out, rc = 0);
1009
1010         fops = inode->i_fop;
1011         dentry->d_inode = inode;
1012         dentry->d_sb = inode->i_sb;
1013         filp->f_pos = 0;
1014         filp->f_path.dentry = dentry;
1015         filp->f_mode = FMODE_64BITHASH;
1016         filp->f_mapping = inode->i_mapping;
1017         filp->f_op = fops;
1018         filp->private_data = NULL;
1019         set_file_inode(filp, inode);
1020
1021         do {
1022                 oclb.oclb_items = 0;
1023 #ifdef HAVE_DIR_CONTEXT
1024                 oclb.ctx.pos = filp->f_pos;
1025 #ifdef HAVE_ITERATE_SHARED
1026                 rc = fops->iterate_shared(filp, &oclb.ctx);
1027 #else
1028                 rc = fops->iterate(filp, &oclb.ctx);
1029 #endif
1030                 filp->f_pos = oclb.ctx.pos;
1031 #else
1032                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1033 #endif
1034         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1035                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1036         fops->release(inode, filp);
1037
1038 out:
1039         if (rc < 0)
1040                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1041                        DFID": rc = %d\n", osd_ino2name(inode),
1042                        inode->i_ino, inode->i_generation,
1043                        PFID(&oic->oic_fid), rc);
1044         else
1045                 rc = 0;
1046
1047         RETURN(rc);
1048 }
1049
1050 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1051                           const struct lu_fid *fid,
1052                           const struct lu_object_conf *conf)
1053 {
1054         struct osd_thread_info *info;
1055         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1056         struct osd_device *dev;
1057         struct osd_idmap_cache *oic;
1058         struct osd_inode_id *id;
1059         struct inode *inode = NULL;
1060         struct lustre_scrub *scrub;
1061         struct scrub_file *sf;
1062         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1063         __u32 saved_ino;
1064         __u32 saved_gen;
1065         int result = 0;
1066         int rc1 = 0;
1067         bool remote = false;
1068         bool trusted = true;
1069         bool updated = false;
1070         bool checked = false;
1071
1072         ENTRY;
1073
1074         LINVRNT(osd_invariant(obj));
1075         LASSERT(obj->oo_inode == NULL);
1076         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1077
1078         dev = osd_dev(ldev);
1079         scrub = &dev->od_scrub.os_scrub;
1080         sf = &scrub->os_file;
1081         info = osd_oti_get(env);
1082         LASSERT(info);
1083         oic = &info->oti_cache;
1084
1085         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1086                 RETURN(-ENOENT);
1087
1088         /*
1089          * For the object is created as locking anchor, or for the object to
1090          * be created on disk. No need to osd_oi_lookup() at here because FID
1091          * shouldn't never be re-used, if it's really a duplicate FID from
1092          * unexpected reason, we should be able to detect it later by calling
1093          * do_create->osd_oi_insert().
1094          */
1095         if (conf && conf->loc_flags & LOC_F_NEW)
1096                 GOTO(out, result = 0);
1097
1098         /* Search order: 1. per-thread cache. */
1099         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1100                 id = &oic->oic_lid;
1101                 goto iget;
1102         }
1103
1104         id = &info->oti_id;
1105         if (!list_empty(&scrub->os_inconsistent_items)) {
1106                 /* Search order: 2. OI scrub pending list. */
1107                 result = osd_oii_lookup(dev, fid, id);
1108                 if (!result)
1109                         goto iget;
1110         }
1111
1112         /*
1113          * The OI mapping in the OI file can be updated by the OI scrub
1114          * when we locate the inode via FID. So it may be not trustable.
1115          */
1116         trusted = false;
1117
1118         /* Search order: 3. OI files. */
1119         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1120         if (result == -ENOENT) {
1121                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1122                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1123                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1124                                       sf->sf_oi_bitmap))
1125                         GOTO(out, result = 0);
1126
1127                 goto trigger;
1128         }
1129
1130         if (result)
1131                 GOTO(out, result);
1132
1133 iget:
1134         obj->oo_inode = NULL;
1135         /* for later passes through checks, not true on first pass */
1136         if (!IS_ERR_OR_NULL(inode))
1137                 iput(inode);
1138
1139         inode = osd_iget_check(info, dev, fid, id, trusted);
1140         if (!IS_ERR(inode)) {
1141                 obj->oo_inode = inode;
1142                 result = 0;
1143                 if (remote)
1144                         goto trigger;
1145
1146                 goto check_lma;
1147         }
1148
1149         result = PTR_ERR(inode);
1150         if (result == -ENOENT || result == -ESTALE)
1151                 GOTO(out, result = 0);
1152
1153         if (result != -EREMCHG)
1154                 GOTO(out, result);
1155
1156 trigger:
1157         /*
1158          * We still have chance to get the valid inode: for the
1159          * object which is referenced by remote name entry, the
1160          * object on the local MDT will be linked under the dir
1161          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1162          *
1163          * We do not know whether the object for the given FID
1164          * is referenced by some remote name entry or not, and
1165          * especially for DNE II, a multiple-linked object may
1166          * have many name entries reside on many MDTs.
1167          *
1168          * To simplify the operation, OSD will not distinguish
1169          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1170          * only happened for the RPC from other MDT during the
1171          * OI scrub, or for the client side RPC with FID only,
1172          * such as FID to path, or from old connected client.
1173          */
1174         if (!remote) {
1175                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1176                 if (!rc1) {
1177                         remote = true;
1178                         trusted = true;
1179                         flags |= SS_AUTO_PARTIAL;
1180                         flags &= ~SS_AUTO_FULL;
1181                         goto iget;
1182                 }
1183         }
1184
1185         if (thread_is_running(&scrub->os_thread)) {
1186                 if (scrub->os_partial_scan && !scrub->os_in_join)
1187                         goto join;
1188
1189                 if (IS_ERR_OR_NULL(inode) || result)
1190                         GOTO(out, result = -EINPROGRESS);
1191
1192                 LASSERT(remote);
1193                 LASSERT(obj->oo_inode == inode);
1194
1195                 osd_add_oi_cache(info, dev, id, fid);
1196                 osd_oii_insert(dev, oic, true);
1197                 goto found;
1198         }
1199
1200         if (dev->od_auto_scrub_interval == AS_NEVER) {
1201                 if (!remote)
1202                         GOTO(out, result = -EREMCHG);
1203
1204                 LASSERT(!result);
1205                 LASSERT(obj->oo_inode == inode);
1206
1207                 osd_add_oi_cache(info, dev, id, fid);
1208                 goto found;
1209         }
1210
1211 join:
1212         rc1 = osd_scrub_start(env, dev, flags);
1213         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1214                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1215         if (rc1 && rc1 != -EALREADY)
1216                 GOTO(out, result = -EREMCHG);
1217
1218         if (IS_ERR_OR_NULL(inode) || result)
1219                 GOTO(out, result = -EINPROGRESS);
1220
1221         LASSERT(remote);
1222         LASSERT(obj->oo_inode == inode);
1223
1224         osd_add_oi_cache(info, dev, id, fid);
1225         osd_oii_insert(dev, oic, true);
1226         goto found;
1227
1228 check_lma:
1229         checked = true;
1230         if (unlikely(obj->oo_header))
1231                 goto found;
1232
1233         result = osd_check_lma(env, obj);
1234         if (!result)
1235                 goto found;
1236
1237         LASSERTF(id->oii_ino == inode->i_ino &&
1238                  id->oii_gen == inode->i_generation,
1239                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1240                  PFID(fid), id->oii_ino, id->oii_gen,
1241                  inode->i_ino, inode->i_generation);
1242
1243         saved_ino = inode->i_ino;
1244         saved_gen = inode->i_generation;
1245
1246         if (unlikely(result == -ENODATA)) {
1247                 /*
1248                  * If the OI scrub updated the OI mapping by race, it
1249                  * must be valid. Trust the inode that has no LMA EA.
1250                  */
1251                 if (updated)
1252                         goto found;
1253
1254                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1255                 if (!result) {
1256                         /*
1257                          * The OI mapping is still there, the inode is still
1258                          * valid. It is just becaues the inode has no LMA EA.
1259                          */
1260                         if (saved_ino == id->oii_ino &&
1261                             saved_gen == id->oii_gen)
1262                                 goto found;
1263
1264                         /*
1265                          * It is the OI scrub updated the OI mapping by race.
1266                          * The new OI mapping must be valid.
1267                          */
1268                         trusted = true;
1269                         updated = true;
1270                         goto iget;
1271                 }
1272
1273                 /*
1274                  * "result == -ENOENT" means that the OI mappinghas been
1275                  * removed by race, so the inode belongs to other object.
1276                  *
1277                  * Others error can be returned  directly.
1278                  */
1279                 if (result == -ENOENT) {
1280                         LASSERT(trusted);
1281
1282                         obj->oo_inode = NULL;
1283                         result = 0;
1284                 }
1285         }
1286
1287         if (result != -EREMCHG)
1288                 GOTO(out, result);
1289
1290         LASSERT(!updated);
1291
1292         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1293         /*
1294          * "result == -ENOENT" means the cached OI mapping has been removed
1295          * from the OI file by race, above inode belongs to other object.
1296          */
1297         if (result == -ENOENT) {
1298                 LASSERT(trusted);
1299
1300                 obj->oo_inode = NULL;
1301                 GOTO(out, result = 0);
1302         }
1303
1304         if (result)
1305                 GOTO(out, result);
1306
1307         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1308                 result = -EREMCHG;
1309                 goto trigger;
1310         }
1311
1312         /*
1313          * It is the OI scrub updated the OI mapping by race.
1314          * The new OI mapping must be valid.
1315          */
1316         trusted = true;
1317         updated = true;
1318         goto iget;
1319
1320 found:
1321         if (!checked) {
1322                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1323                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1324
1325                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1326                 if (!result) {
1327                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1328                             dev->od_is_ost)
1329                                 obj->oo_pfid_in_lma = 1;
1330                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1331                             !dev->od_is_ost)
1332                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1333                 } else if (result != -ENODATA) {
1334                         GOTO(out, result);
1335                 }
1336         }
1337
1338         obj->oo_compat_dot_created = 1;
1339         obj->oo_compat_dotdot_created = 1;
1340
1341         if (S_ISDIR(inode->i_mode) &&
1342             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1343                 osd_check_lmv(info, dev, inode, oic);
1344
1345         result = osd_attach_jinode(inode);
1346         if (result)
1347                 GOTO(out, result);
1348
1349         if (!ldiskfs_pdo)
1350                 GOTO(out, result = 0);
1351
1352         LASSERT(!obj->oo_hl_head);
1353         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1354
1355         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1356
1357 out:
1358         if (result || !obj->oo_inode) {
1359                 if (!IS_ERR_OR_NULL(inode))
1360                         iput(inode);
1361
1362                 obj->oo_inode = NULL;
1363                 if (trusted)
1364                         fid_zero(&oic->oic_fid);
1365         }
1366
1367         LINVRNT(osd_invariant(obj));
1368         return result;
1369 }
1370
1371 /*
1372  * Concurrency: shouldn't matter.
1373  */
1374 static void osd_object_init0(struct osd_object *obj)
1375 {
1376         LASSERT(obj->oo_inode != NULL);
1377         obj->oo_dt.do_body_ops = &osd_body_ops;
1378         obj->oo_dt.do_lu.lo_header->loh_attr |=
1379                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1380 }
1381
1382 /*
1383  * Concurrency: no concurrent access is possible that early in object
1384  * life-cycle.
1385  */
1386 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1387                            const struct lu_object_conf *conf)
1388 {
1389         struct osd_object *obj = osd_obj(l);
1390         int result;
1391
1392         LINVRNT(osd_invariant(obj));
1393
1394         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1395                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1396                 l->lo_header->loh_attr |= LOHA_EXISTS;
1397                 return 0;
1398         }
1399
1400         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1401         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1402         if (result == 0 && obj->oo_inode != NULL) {
1403                 struct osd_thread_info *oti = osd_oti_get(env);
1404                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1405
1406                 osd_object_init0(obj);
1407                 if (unlikely(obj->oo_header))
1408                         return 0;
1409
1410                 result = osd_get_lma(oti, obj->oo_inode,
1411                                      &oti->oti_obj_dentry, loa);
1412                 if (!result) {
1413                         /*
1414                          * Convert LMAI flags to lustre LMA flags
1415                          * and cache it to oo_lma_flags
1416                          */
1417                         obj->oo_lma_flags =
1418                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1419                 } else if (result == -ENODATA) {
1420                         result = 0;
1421                 }
1422         }
1423
1424         LINVRNT(osd_invariant(obj));
1425         return result;
1426 }
1427
1428 /*
1429  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1430  * The left part is for value, binary mode.
1431  */
1432 struct osd_xattr_entry {
1433         struct list_head        oxe_list;
1434         size_t                  oxe_len;
1435         size_t                  oxe_namelen;
1436         bool                    oxe_exist;
1437         struct rcu_head         oxe_rcu;
1438         char                    oxe_buf[0];
1439 };
1440
1441 static int osd_oxc_get(struct osd_object *obj, const char *name,
1442                        struct lu_buf *buf)
1443 {
1444         struct osd_xattr_entry *tmp;
1445         struct osd_xattr_entry *oxe = NULL;
1446         size_t namelen = strlen(name);
1447         int rc;
1448
1449         ENTRY;
1450
1451         rcu_read_lock();
1452         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1453                 if (namelen == tmp->oxe_namelen &&
1454                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1455                         oxe = tmp;
1456                         break;
1457                 }
1458         }
1459
1460         if (oxe == NULL)
1461                 GOTO(out, rc = -ENOENT);
1462
1463         if (!oxe->oxe_exist)
1464                 GOTO(out, rc = -ENODATA);
1465
1466         /* vallen */
1467         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1468         LASSERT(rc > 0);
1469
1470         if (buf->lb_buf == NULL)
1471                 GOTO(out, rc);
1472
1473         if (buf->lb_len < rc)
1474                 GOTO(out, rc = -ERANGE);
1475
1476         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1477         EXIT;
1478 out:
1479         rcu_read_unlock();
1480
1481         return rc;
1482 }
1483
1484 static void osd_oxc_free(struct rcu_head *head)
1485 {
1486         struct osd_xattr_entry *oxe;
1487
1488         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1489         OBD_FREE(oxe, oxe->oxe_len);
1490 }
1491
1492 static void osd_oxc_add(struct osd_object *obj, const char *name,
1493                         const char *buf, int buflen)
1494 {
1495         struct osd_xattr_entry *oxe;
1496         struct osd_xattr_entry *old = NULL;
1497         struct osd_xattr_entry *tmp;
1498         size_t namelen = strlen(name);
1499         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1500
1501         OBD_ALLOC(oxe, len);
1502         if (oxe == NULL)
1503                 return;
1504
1505         INIT_LIST_HEAD(&oxe->oxe_list);
1506         oxe->oxe_len = len;
1507         oxe->oxe_namelen = namelen;
1508         memcpy(oxe->oxe_buf, name, namelen);
1509         if (buflen > 0) {
1510                 LASSERT(buf != NULL);
1511                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1512                 oxe->oxe_exist = true;
1513         } else {
1514                 oxe->oxe_exist = false;
1515         }
1516
1517         /* this should be rarely called, just remove old and add new */
1518         spin_lock(&obj->oo_guard);
1519         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1520                 if (namelen == tmp->oxe_namelen &&
1521                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1522                         old = tmp;
1523                         break;
1524                 }
1525         }
1526         if (old != NULL) {
1527                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1528                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1529         } else {
1530                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1531         }
1532         spin_unlock(&obj->oo_guard);
1533 }
1534
1535 static void osd_oxc_del(struct osd_object *obj, const char *name)
1536 {
1537         struct osd_xattr_entry *oxe;
1538         size_t namelen = strlen(name);
1539
1540         spin_lock(&obj->oo_guard);
1541         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1542                 if (namelen == oxe->oxe_namelen &&
1543                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1544                         list_del_rcu(&oxe->oxe_list);
1545                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1546                         break;
1547                 }
1548         }
1549         spin_unlock(&obj->oo_guard);
1550 }
1551
1552 static void osd_oxc_fini(struct osd_object *obj)
1553 {
1554         struct osd_xattr_entry *oxe, *next;
1555
1556         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1557                 list_del(&oxe->oxe_list);
1558                 OBD_FREE(oxe, oxe->oxe_len);
1559         }
1560 }
1561
1562 /*
1563  * Concurrency: no concurrent access is possible that late in object
1564  * life-cycle.
1565  */
1566 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1567 {
1568         struct osd_object *obj = osd_obj(l);
1569         struct lu_object_header *h = obj->oo_header;
1570
1571         LINVRNT(osd_invariant(obj));
1572
1573         osd_oxc_fini(obj);
1574         dt_object_fini(&obj->oo_dt);
1575         if (obj->oo_hl_head != NULL)
1576                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1577         OBD_FREE_PTR(obj);
1578         if (unlikely(h)) {
1579                 lu_object_header_fini(h);
1580                 OBD_FREE_PTR(h);
1581         }
1582 }
1583
1584 /*
1585  * Concurrency: no concurrent access is possible that late in object
1586  * life-cycle.
1587  */
1588 static void osd_index_fini(struct osd_object *o)
1589 {
1590         struct iam_container *bag;
1591
1592         if (o->oo_dir != NULL) {
1593                 bag = &o->oo_dir->od_container;
1594                 if (o->oo_inode != NULL) {
1595                         if (bag->ic_object == o->oo_inode)
1596                                 iam_container_fini(bag);
1597                 }
1598                 OBD_FREE_PTR(o->oo_dir);
1599                 o->oo_dir = NULL;
1600         }
1601 }
1602
1603 /*
1604  * Concurrency: no concurrent access is possible that late in object
1605  * life-cycle (for all existing callers, that is. New callers have to provide
1606  * their own locking.)
1607  */
1608 static int osd_inode_unlinked(const struct inode *inode)
1609 {
1610         return inode->i_nlink == 0;
1611 }
1612
1613 enum {
1614         OSD_TXN_OI_DELETE_CREDITS    = 20,
1615         OSD_TXN_INODE_DELETE_CREDITS = 20
1616 };
1617
1618 /*
1619  * Journal
1620  */
1621
1622 #if OSD_THANDLE_STATS
1623 /**
1624  * Set time when the handle is allocated
1625  */
1626 static void osd_th_alloced(struct osd_thandle *oth)
1627 {
1628         oth->oth_alloced = ktime_get();
1629 }
1630
1631 /**
1632  * Set time when the handle started
1633  */
1634 static void osd_th_started(struct osd_thandle *oth)
1635 {
1636         oth->oth_started = ktime_get();
1637 }
1638
1639 /**
1640  * Check whether the we deal with this handle for too long.
1641  */
1642 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1643                                 ktime_t alloced, ktime_t started,
1644                                 ktime_t closed)
1645 {
1646         ktime_t now = ktime_get();
1647
1648         LASSERT(dev != NULL);
1649
1650         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1651                             ktime_us_delta(started, alloced));
1652         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1653                             ktime_us_delta(closed, started));
1654         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1655                             ktime_us_delta(now, closed));
1656
1657         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1658                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1659                                 oth, now, alloced, started, closed);
1660                 libcfs_debug_dumpstack(NULL);
1661         }
1662 }
1663
1664 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1665 {                                                                       \
1666         ktime_t __closed = ktime_get();                                 \
1667         ktime_t __alloced = oth->oth_alloced;                           \
1668         ktime_t __started = oth->oth_started;                           \
1669                                                                         \
1670         expr;                                                           \
1671         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1672 }
1673
1674 #else /* OSD_THANDLE_STATS */
1675
1676 #define osd_th_alloced(h)                  do {} while(0)
1677 #define osd_th_started(h)                  do {} while(0)
1678 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1679
1680 #endif /* OSD_THANDLE_STATS */
1681
1682 /*
1683  * Concurrency: doesn't access mutable data.
1684  */
1685 static int osd_param_is_not_sane(const struct osd_device *dev,
1686                                  const struct thandle *th)
1687 {
1688         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1689
1690         return oh->ot_credits > osd_transaction_size(dev);
1691 }
1692
1693 /*
1694  * Concurrency: shouldn't matter.
1695  */
1696 static void osd_trans_commit_cb(struct super_block *sb,
1697                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1698 {
1699         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1700         struct thandle *th = &oh->ot_super;
1701         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1702         struct dt_txn_commit_cb *dcb, *tmp;
1703
1704         LASSERT(oh->ot_handle == NULL);
1705
1706         if (error)
1707                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1708
1709         dt_txn_hook_commit(th);
1710
1711         /* call per-transaction callbacks if any */
1712         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1713                                  dcb_linkage) {
1714                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1715                          "commit callback entry: magic=%x name='%s'\n",
1716                          dcb->dcb_magic, dcb->dcb_name);
1717                 list_del_init(&dcb->dcb_linkage);
1718                 dcb->dcb_func(NULL, th, dcb, error);
1719         }
1720
1721         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1722         lu_device_put(lud);
1723         th->th_dev = NULL;
1724
1725         OBD_FREE_PTR(oh);
1726 }
1727
1728 #ifndef HAVE_SB_START_WRITE
1729 # define sb_start_write(sb) do {} while (0)
1730 # define sb_end_write(sb) do {} while (0)
1731 #endif
1732
1733 static struct thandle *osd_trans_create(const struct lu_env *env,
1734                                         struct dt_device *d)
1735 {
1736         struct osd_thread_info *oti = osd_oti_get(env);
1737         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1738         struct osd_thandle *oh;
1739         struct thandle *th;
1740
1741         ENTRY;
1742
1743         if (d->dd_rdonly) {
1744                 CERROR("%s: someone try to start transaction under "
1745                        "readonly mode, should be disabled.\n",
1746                        osd_name(osd_dt_dev(d)));
1747                 dump_stack();
1748                 RETURN(ERR_PTR(-EROFS));
1749         }
1750
1751         /* on pending IO in this thread should left from prev. request */
1752         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1753
1754         sb_start_write(osd_sb(osd_dt_dev(d)));
1755
1756         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1757         if (!oh) {
1758                 sb_end_write(osd_sb(osd_dt_dev(d)));
1759                 RETURN(ERR_PTR(-ENOMEM));
1760         }
1761
1762         oh->ot_quota_trans = &oti->oti_quota_trans;
1763         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1764         th = &oh->ot_super;
1765         th->th_dev = d;
1766         th->th_result = 0;
1767         oh->ot_credits = 0;
1768         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1769         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1770         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1771         osd_th_alloced(oh);
1772
1773         memset(oti->oti_declare_ops, 0,
1774                sizeof(oti->oti_declare_ops));
1775         memset(oti->oti_declare_ops_cred, 0,
1776                sizeof(oti->oti_declare_ops_cred));
1777         memset(oti->oti_declare_ops_used, 0,
1778                sizeof(oti->oti_declare_ops_used));
1779
1780         oti->oti_ins_cache_depth++;
1781
1782         RETURN(th);
1783 }
1784
1785 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1786 {
1787         struct osd_thread_info *oti = osd_oti_get(env);
1788         struct osd_thandle *oh;
1789
1790         oh = container_of0(th, struct osd_thandle, ot_super);
1791         LASSERT(oh != NULL);
1792
1793         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1794               oti->oti_declare_ops[OSD_OT_CREATE],
1795               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1796               oti->oti_declare_ops_used[OSD_OT_CREATE],
1797               oti->oti_declare_ops[OSD_OT_DESTROY],
1798               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1799               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1800         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1801               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1802               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1803               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1805               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1806               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1807         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1808               oti->oti_declare_ops[OSD_OT_WRITE],
1809               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1810               oti->oti_declare_ops_used[OSD_OT_WRITE],
1811               oti->oti_declare_ops[OSD_OT_PUNCH],
1812               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1813               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1814               oti->oti_declare_ops[OSD_OT_QUOTA],
1815               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1816               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1817         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_INSERT],
1819               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1820               oti->oti_declare_ops_used[OSD_OT_INSERT],
1821               oti->oti_declare_ops[OSD_OT_DELETE],
1822               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1823               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1824         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1825               oti->oti_declare_ops[OSD_OT_REF_ADD],
1826               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1827               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops[OSD_OT_REF_DEL],
1829               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1830               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1831 }
1832
1833 /*
1834  * Concurrency: shouldn't matter.
1835  */
1836 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1837                            struct thandle *th)
1838 {
1839         struct osd_thread_info *oti = osd_oti_get(env);
1840         struct osd_device *dev = osd_dt_dev(d);
1841         handle_t *jh;
1842         struct osd_thandle *oh;
1843         int rc;
1844
1845         ENTRY;
1846
1847         LASSERT(current->journal_info == NULL);
1848
1849         oh = container_of0(th, struct osd_thandle, ot_super);
1850         LASSERT(oh != NULL);
1851         LASSERT(oh->ot_handle == NULL);
1852
1853         rc = dt_txn_hook_start(env, d, th);
1854         if (rc != 0)
1855                 GOTO(out, rc);
1856
1857         if (unlikely(osd_param_is_not_sane(dev, th))) {
1858                 static unsigned long last_printed;
1859                 static int last_credits;
1860
1861                 /*
1862                  * don't make noise on a tiny testing systems
1863                  * actual credits misuse will be caught anyway
1864                  */
1865                 if (last_credits != oh->ot_credits &&
1866                     time_after(jiffies, last_printed +
1867                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1868                     osd_transaction_size(dev) > 512) {
1869                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1870                               oh->ot_credits, osd_transaction_size(dev));
1871                         osd_trans_dump_creds(env, th);
1872                         libcfs_debug_dumpstack(NULL);
1873                         last_credits = oh->ot_credits;
1874                         last_printed = jiffies;
1875                 }
1876                 /*
1877                  * XXX Limit the credits to 'max_transaction_buffers', and
1878                  *     let the underlying filesystem to catch the error if
1879                  *     we really need so many credits.
1880                  *
1881                  *     This should be removed when we can calculate the
1882                  *     credits precisely.
1883                  */
1884                 oh->ot_credits = osd_transaction_size(dev);
1885         } else if (ldiskfs_track_declares_assert != 0) {
1886                 /*
1887                  * reserve few credits to prevent an assertion in JBD
1888                  * our debugging mechanism will be able to detected
1889                  * overuse. this can help to debug single-update
1890                  * transactions
1891                  */
1892                 oh->ot_credits += 10;
1893                 if (unlikely(osd_param_is_not_sane(dev, th)))
1894                         oh->ot_credits = osd_transaction_size(dev);
1895         }
1896
1897         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1898                 GOTO(out, rc = -EIO);
1899
1900         /*
1901          * XXX temporary stuff. Some abstraction layer should
1902          * be used.
1903          */
1904         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1905         osd_th_started(oh);
1906         if (!IS_ERR(jh)) {
1907                 oh->ot_handle = jh;
1908                 LASSERT(oti->oti_txns == 0);
1909
1910                 lu_device_get(&d->dd_lu_dev);
1911                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1912                               "osd-tx", th);
1913                 oti->oti_txns++;
1914                 rc = 0;
1915         } else {
1916                 rc = PTR_ERR(jh);
1917         }
1918 out:
1919         RETURN(rc);
1920 }
1921
1922 static int osd_seq_exists(const struct lu_env *env,
1923                           struct osd_device *osd, u64 seq)
1924 {
1925         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1926         struct seq_server_site *ss = osd_seq_site(osd);
1927         int rc;
1928
1929         ENTRY;
1930
1931         LASSERT(ss != NULL);
1932         LASSERT(ss->ss_server_fld != NULL);
1933
1934         rc = osd_fld_lookup(env, osd, seq, range);
1935         if (rc != 0) {
1936                 if (rc != -ENOENT)
1937                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1938                                osd_name(osd), seq, rc);
1939                 RETURN(0);
1940         }
1941
1942         RETURN(ss->ss_node_id == range->lsr_index);
1943 }
1944
1945 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1946 {
1947         struct dt_txn_commit_cb *dcb;
1948         struct dt_txn_commit_cb *tmp;
1949
1950         /* call per-transaction stop callbacks if any */
1951         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1952                                  dcb_linkage) {
1953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1954                          "commit callback entry: magic=%x name='%s'\n",
1955                          dcb->dcb_magic, dcb->dcb_name);
1956                 list_del_init(&dcb->dcb_linkage);
1957                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1958         }
1959 }
1960
1961 /*
1962  * Concurrency: shouldn't matter.
1963  */
1964 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1965                           struct thandle *th)
1966 {
1967         struct osd_thread_info *oti = osd_oti_get(env);
1968         struct osd_thandle *oh;
1969         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1970         struct osd_device *osd = osd_dt_dev(th->th_dev);
1971         struct qsd_instance *qsd = osd->od_quota_slave;
1972         struct lquota_trans *qtrans;
1973         struct list_head truncates = LIST_HEAD_INIT(truncates);
1974         int rc = 0, remove_agents = 0;
1975
1976         ENTRY;
1977
1978         oh = container_of0(th, struct osd_thandle, ot_super);
1979
1980         remove_agents = oh->ot_remove_agents;
1981
1982         qtrans = oh->ot_quota_trans;
1983         oh->ot_quota_trans = NULL;
1984
1985         /* move locks to local list, stop tx, execute truncates */
1986         list_splice(&oh->ot_trunc_locks, &truncates);
1987
1988         if (oh->ot_handle != NULL) {
1989                 int rc2;
1990
1991                 handle_t *hdl = oh->ot_handle;
1992
1993                 /*
1994                  * add commit callback
1995                  * notice we don't do this in osd_trans_start()
1996                  * as underlying transaction can change during truncate
1997                  */
1998                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1999                                              &oh->ot_jcb);
2000
2001                 LASSERT(oti->oti_txns == 1);
2002                 oti->oti_txns--;
2003
2004                 rc = dt_txn_hook_stop(env, th);
2005                 if (rc != 0)
2006                         CERROR("%s: failed in transaction hook: rc = %d\n",
2007                                osd_name(osd), rc);
2008
2009                 osd_trans_stop_cb(oh, rc);
2010                 /* hook functions might modify th_sync */
2011                 hdl->h_sync = th->th_sync;
2012
2013                 oh->ot_handle = NULL;
2014                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2015                 if (rc2 != 0)
2016                         CERROR("%s: failed to stop transaction: rc = %d\n",
2017                                osd_name(osd), rc2);
2018                 if (!rc)
2019                         rc = rc2;
2020
2021                 osd_process_truncates(&truncates);
2022         } else {
2023                 osd_trans_stop_cb(oh, th->th_result);
2024                 OBD_FREE_PTR(oh);
2025         }
2026
2027         osd_trunc_unlock_all(&truncates);
2028
2029         /* inform the quota slave device that the transaction is stopping */
2030         qsd_op_end(env, qsd, qtrans);
2031
2032         /*
2033          * as we want IO to journal and data IO be concurrent, we don't block
2034          * awaiting data IO completion in osd_do_bio(), instead we wait here
2035          * once transaction is submitted to the journal. all reqular requests
2036          * don't do direct IO (except read/write), thus this wait_event becomes
2037          * no-op for them.
2038          *
2039          * IMPORTANT: we have to wait till any IO submited by the thread is
2040          * completed otherwise iobuf may be corrupted by different request
2041          */
2042         wait_event(iobuf->dr_wait,
2043                        atomic_read(&iobuf->dr_numreqs) == 0);
2044         osd_fini_iobuf(osd, iobuf);
2045         if (!rc)
2046                 rc = iobuf->dr_error;
2047
2048         if (unlikely(remove_agents != 0))
2049                 osd_process_scheduled_agent_removals(env, osd);
2050
2051         oti->oti_ins_cache_depth--;
2052         /* reset OI cache for safety */
2053         if (oti->oti_ins_cache_depth == 0)
2054                 oti->oti_ins_cache_used = 0;
2055
2056         sb_end_write(osd_sb(osd));
2057
2058         RETURN(rc);
2059 }
2060
2061 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2062 {
2063         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2064                                                ot_super);
2065
2066         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2067         LASSERT(&dcb->dcb_func != NULL);
2068         if (dcb->dcb_flags & DCB_TRANS_STOP)
2069                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2070         else
2071                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2072
2073         return 0;
2074 }
2075
2076 /*
2077  * Called just before object is freed. Releases all resources except for
2078  * object itself (that is released by osd_object_free()).
2079  *
2080  * Concurrency: no concurrent access is possible that late in object
2081  * life-cycle.
2082  */
2083 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2084 {
2085         struct osd_object *obj = osd_obj(l);
2086         struct inode *inode = obj->oo_inode;
2087
2088         LINVRNT(osd_invariant(obj));
2089
2090         /*
2091          * If object is unlinked remove fid->ino mapping from object index.
2092          */
2093
2094         osd_index_fini(obj);
2095         if (inode != NULL) {
2096                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
2097                 qid_t                    uid = i_uid_read(inode);
2098                 qid_t                    gid = i_gid_read(inode);
2099
2100                 obj->oo_inode = NULL;
2101                 iput(inode);
2102                 if (!obj->oo_header && qsd) {
2103                         struct osd_thread_info *info = osd_oti_get(env);
2104                         struct lquota_id_info *qi = &info->oti_qi;
2105
2106                         /* Release granted quota to master if necessary */
2107                         qi->lqi_id.qid_uid = uid;
2108                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2109
2110                         qi->lqi_id.qid_uid = gid;
2111                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2112
2113                         qi->lqi_id.qid_uid = i_projid_read(inode);
2114                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2115                 }
2116         }
2117 }
2118
2119 /*
2120  * Concurrency: ->loo_object_release() is called under site spin-lock.
2121  */
2122 static void osd_object_release(const struct lu_env *env,
2123                                struct lu_object *l)
2124 {
2125         struct osd_object *o = osd_obj(l);
2126
2127         /*
2128          * nobody should be releasing a non-destroyed object with nlink=0
2129          * the API allows this, but ldiskfs doesn't like and then report
2130          * this inode as deleted
2131          */
2132         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2133                   o->oo_inode->i_nlink == 0));
2134 }
2135
2136 /*
2137  * Concurrency: shouldn't matter.
2138  */
2139 static int osd_object_print(const struct lu_env *env, void *cookie,
2140                             lu_printer_t p, const struct lu_object *l)
2141 {
2142         struct osd_object *o = osd_obj(l);
2143         struct iam_descr *d;
2144
2145         if (o->oo_dir != NULL)
2146                 d = o->oo_dir->od_container.ic_descr;
2147         else
2148                 d = NULL;
2149         return (*p)(env, cookie,
2150                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2151                     o, o->oo_inode,
2152                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2153                     o->oo_inode ? o->oo_inode->i_generation : 0,
2154                     d ? d->id_ops->id_name : "plain");
2155 }
2156
2157 /*
2158  * Concurrency: shouldn't matter.
2159  */
2160 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2161                 struct obd_statfs *sfs)
2162 {
2163         struct osd_device *osd = osd_dt_dev(d);
2164         struct super_block *sb = osd_sb(osd);
2165         struct kstatfs *ksfs;
2166         __u64 reserved;
2167         int result = 0;
2168
2169         if (unlikely(osd->od_mnt == NULL))
2170                 return -EINPROGRESS;
2171
2172         /* osd_lproc.c call this without env, allocate ksfs for that case */
2173         if (unlikely(env == NULL)) {
2174                 OBD_ALLOC_PTR(ksfs);
2175                 if (ksfs == NULL)
2176                         return -ENOMEM;
2177         } else {
2178                 ksfs = &osd_oti_get(env)->oti_ksfs;
2179         }
2180
2181         result = sb->s_op->statfs(sb->s_root, ksfs);
2182         if (result)
2183                 goto out;
2184
2185         statfs_pack(sfs, ksfs);
2186         if (unlikely(sb->s_flags & MS_RDONLY))
2187                 sfs->os_state |= OS_STATE_READONLY;
2188         if (ldiskfs_has_feature_extents(sb))
2189                 sfs->os_maxbytes = sb->s_maxbytes;
2190         else
2191                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2192
2193         /*
2194          * Reserve some space so to avoid fragmenting the filesystem too much.
2195          * Fragmentation not only impacts performance, but can also increase
2196          * metadata overhead significantly, causing grant calculation to be
2197          * wrong.
2198          *
2199          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2200          */
2201         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2202         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2203         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2204                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2205
2206         sfs->os_blocks -= reserved;
2207         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2208         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2209
2210 out:
2211         if (unlikely(env == NULL))
2212                 OBD_FREE_PTR(ksfs);
2213         return result;
2214 }
2215
2216 /**
2217  * Estimate space needed for file creations. We assume the largest filename
2218  * which is 2^64 - 1, hence a filename of 20 chars.
2219  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2220  */
2221 #ifdef __LDISKFS_DIR_REC_LEN
2222 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2223 #else
2224 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2225 #endif
2226
2227 /*
2228  * Concurrency: doesn't access mutable data.
2229  */
2230 static void osd_conf_get(const struct lu_env *env,
2231                          const struct dt_device *dev,
2232                          struct dt_device_param *param)
2233 {
2234         struct osd_device *d = osd_dt_dev(dev);
2235         struct super_block *sb = osd_sb(d);
2236         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2237         const char *name;
2238         int ea_overhead;
2239
2240         /*
2241          * XXX should be taken from not-yet-existing fs abstraction layer.
2242          */
2243         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2244         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2245         param->ddp_symlink_max  = sb->s_blocksize;
2246         param->ddp_mount_type   = LDD_MT_LDISKFS;
2247         if (ldiskfs_has_feature_extents(sb))
2248                 param->ddp_maxbytes = sb->s_maxbytes;
2249         else
2250                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2251         /*
2252          * inode are statically allocated, so per-inode space consumption
2253          * is the space consumed by the directory entry
2254          */
2255         param->ddp_inodespace     = PER_OBJ_USAGE;
2256         /*
2257          * EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2258          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2259          * maximum length to not under count the actual number of extents
2260          * needed for writing a file.
2261          */
2262         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2263         /* worst-case extent insertion metadata overhead */
2264         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2265         param->ddp_mntopts = 0;
2266         if (test_opt(sb, XATTR_USER))
2267                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2268         if (test_opt(sb, POSIX_ACL))
2269                 param->ddp_mntopts |= MNTOPT_ACL;
2270
2271         /*
2272          * LOD might calculate the max stripe count based on max_ea_size,
2273          * so we need take account in the overhead as well,
2274          * xattr_header + magic + xattr_entry_head
2275          */
2276         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2277                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2278
2279 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2280         if (ldiskfs_has_feature_ea_inode(sb))
2281                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2282                                                                 ea_overhead;
2283         else
2284 #endif
2285                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2286
2287         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2288                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2289
2290         /*
2291          * Preferred RPC size for efficient disk IO.  4MB shows good
2292          * all-around performance for ldiskfs, but use bigalloc chunk size
2293          * by default if larger.
2294          */
2295 #if defined(LDISKFS_CLUSTER_SIZE)
2296         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2297                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2298         else
2299 #endif
2300                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2301
2302         param->ddp_t10_cksum_type = 0;
2303         if (bi) {
2304                 unsigned short interval = blk_integrity_interval(bi);
2305                 name = blk_integrity_name(bi);
2306                 /*
2307                  * Expected values:
2308                  * T10-DIF-TYPE1-CRC
2309                  * T10-DIF-TYPE3-CRC
2310                  * T10-DIF-TYPE1-IP
2311                  * T10-DIF-TYPE3-IP
2312                  */
2313                 if (strncmp(name, "T10-DIF-TYPE",
2314                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2315                         /* also skip "1/3-" at end */
2316                         const int type_off = sizeof("T10-DIF-TYPE.");
2317                         char type_number = name[type_off - 2];
2318
2319                         if (interval != 512 && interval != 4096) {
2320                                 CERROR("%s: unsupported T10PI sector size %u\n",
2321                                        d->od_svname, interval);
2322                         } else if (type_number != '1' && type_number != '3') {
2323                                 CERROR("%s: unsupported T10PI type %s\n",
2324                                        d->od_svname, name);
2325                         } else if (strcmp(name + type_off, "CRC") == 0) {
2326                                 d->od_t10_type = type_number == '1' ?
2327                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2328                                 param->ddp_t10_cksum_type = interval == 512 ?
2329                                         OBD_CKSUM_T10CRC512 :
2330                                         OBD_CKSUM_T10CRC4K;
2331                         } else if (strcmp(name + type_off, "IP") == 0) {
2332                                 d->od_t10_type = type_number == '1' ?
2333                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2334                                 param->ddp_t10_cksum_type = interval == 512 ?
2335                                         OBD_CKSUM_T10IP512 :
2336                                         OBD_CKSUM_T10IP4K;
2337                         } else {
2338                                 CERROR("%s: unsupported checksum type of "
2339                                        "T10PI type '%s'",
2340                                        d->od_svname, name);
2341                         }
2342
2343                 } else {
2344                         CERROR("%s: unsupported T10PI type '%s'",
2345                                d->od_svname, name);
2346                 }
2347         }
2348 }
2349
2350 /*
2351  * Concurrency: shouldn't matter.
2352  */
2353 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2354 {
2355         int rc;
2356         struct super_block *s = osd_sb(osd_dt_dev(d));
2357         ENTRY;
2358
2359         down_read(&s->s_umount);
2360         rc = s->s_op->sync_fs(s, 1);
2361         up_read(&s->s_umount);
2362
2363         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2364                rc);
2365
2366         return rc;
2367 }
2368
2369 /**
2370  * Start commit for OSD device.
2371  *
2372  * An implementation of dt_commit_async method for OSD device.
2373  * Asychronously starts underlayng fs sync and thereby a transaction
2374  * commit.
2375  *
2376  * \param env environment
2377  * \param d dt device
2378  *
2379  * \see dt_device_operations
2380  */
2381 static int osd_commit_async(const struct lu_env *env,
2382                             struct dt_device *d)
2383 {
2384         struct super_block *s = osd_sb(osd_dt_dev(d));
2385         int rc;
2386
2387         ENTRY;
2388
2389         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2390         down_read(&s->s_umount);
2391         rc = s->s_op->sync_fs(s, 0);
2392         up_read(&s->s_umount);
2393
2394         RETURN(rc);
2395 }
2396
2397 /* Our own copy of the set readonly functions if present, or NU if not. */
2398 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2399 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2400 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2401
2402 /*
2403  * Concurrency: shouldn't matter.
2404  */
2405 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2406 {
2407         struct super_block *sb = osd_sb(osd_dt_dev(d));
2408         struct block_device *dev = sb->s_bdev;
2409         int rc = -EOPNOTSUPP;
2410
2411         ENTRY;
2412
2413         if (priv_dev_set_rdonly) {
2414                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2415
2416                 rc = 0;
2417                 CERROR("*** setting %s read-only ***\n",
2418                        osd_dt_dev(d)->od_svname);
2419
2420                 if (sb->s_op->freeze_fs) {
2421                         rc = sb->s_op->freeze_fs(sb);
2422                         if (rc)
2423                                 goto out;
2424                 }
2425
2426                 if (jdev && (jdev != dev)) {
2427                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2428                                (long)jdev);
2429                         priv_dev_set_rdonly(jdev);
2430                 }
2431                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2432                 priv_dev_set_rdonly(dev);
2433
2434                 if (sb->s_op->unfreeze_fs)
2435                         sb->s_op->unfreeze_fs(sb);
2436         }
2437
2438 out:
2439         if (rc)
2440                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2441                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2442
2443         RETURN(rc);
2444 }
2445
2446 /**
2447  * Note: we do not count into QUOTA here.
2448  * If we mount with --data_journal we may need more.
2449  */
2450 const int osd_dto_credits_noquota[DTO_NR] = {
2451         /**
2452          * Insert.
2453          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2454          * SINGLEDATA_TRANS_BLOCKS(8)
2455          * XXX Note: maybe iam need more, since iam have more level than
2456          *           EXT3 htree.
2457          */
2458         [DTO_INDEX_INSERT]  = 16,
2459         /**
2460          * Delete
2461          * just modify a single entry, probably merge few within a block
2462          */
2463         [DTO_INDEX_DELETE]  = 1,
2464         /**
2465          * Used for OI scrub
2466          */
2467         [DTO_INDEX_UPDATE]  = 16,
2468         /**
2469          * 4(inode, inode bits, groups, GDT)
2470          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2471          */
2472         [DTO_OBJECT_CREATE] = 4,
2473         /**
2474          * 4(inode, inode bits, groups, GDT)
2475          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2476          */
2477         [DTO_OBJECT_DELETE] = 4,
2478         /**
2479          * Attr set credits (inode)
2480          */
2481         [DTO_ATTR_SET_BASE] = 1,
2482         /**
2483          * Xattr set. The same as xattr of EXT3.
2484          * DATA_TRANS_BLOCKS(14)
2485          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2486          * are also counted in. Do not know why?
2487          */
2488         [DTO_XATTR_SET]     = 14,
2489         /**
2490          * credits for inode change during write.
2491          */
2492         [DTO_WRITE_BASE]    = 3,
2493         /**
2494          * credits for single block write.
2495          */
2496         [DTO_WRITE_BLOCK]   = 14,
2497         /**
2498          * Attr set credits for chown.
2499          * This is extra credits for setattr, and it is null without quota
2500          */
2501         [DTO_ATTR_SET_CHOWN] = 0
2502 };
2503
2504 static const struct dt_device_operations osd_dt_ops = {
2505         .dt_root_get       = osd_root_get,
2506         .dt_statfs         = osd_statfs,
2507         .dt_trans_create   = osd_trans_create,
2508         .dt_trans_start    = osd_trans_start,
2509         .dt_trans_stop     = osd_trans_stop,
2510         .dt_trans_cb_add   = osd_trans_cb_add,
2511         .dt_conf_get       = osd_conf_get,
2512         .dt_sync           = osd_sync,
2513         .dt_ro             = osd_ro,
2514         .dt_commit_async   = osd_commit_async,
2515 };
2516
2517 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2518                           unsigned int role)
2519 {
2520         struct osd_object *obj = osd_dt_obj(dt);
2521         struct osd_thread_info *oti = osd_oti_get(env);
2522
2523         LINVRNT(osd_invariant(obj));
2524
2525         LASSERT(obj->oo_owner != env);
2526         down_read_nested(&obj->oo_sem, role);
2527
2528         LASSERT(obj->oo_owner == NULL);
2529         oti->oti_r_locks++;
2530 }
2531
2532 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2533                            unsigned int role)
2534 {
2535         struct osd_object *obj = osd_dt_obj(dt);
2536         struct osd_thread_info *oti = osd_oti_get(env);
2537
2538         LINVRNT(osd_invariant(obj));
2539
2540         LASSERT(obj->oo_owner != env);
2541         down_write_nested(&obj->oo_sem, role);
2542
2543         LASSERT(obj->oo_owner == NULL);
2544         obj->oo_owner = env;
2545         oti->oti_w_locks++;
2546 }
2547
2548 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2549 {
2550         struct osd_object *obj = osd_dt_obj(dt);
2551         struct osd_thread_info *oti = osd_oti_get(env);
2552
2553         LINVRNT(osd_invariant(obj));
2554
2555         LASSERT(oti->oti_r_locks > 0);
2556         oti->oti_r_locks--;
2557         up_read(&obj->oo_sem);
2558 }
2559
2560 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2561 {
2562         struct osd_object *obj = osd_dt_obj(dt);
2563         struct osd_thread_info *oti = osd_oti_get(env);
2564
2565         LINVRNT(osd_invariant(obj));
2566
2567         LASSERT(obj->oo_owner == env);
2568         LASSERT(oti->oti_w_locks > 0);
2569         oti->oti_w_locks--;
2570         obj->oo_owner = NULL;
2571         up_write(&obj->oo_sem);
2572 }
2573
2574 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2575 {
2576         struct osd_object *obj = osd_dt_obj(dt);
2577
2578         LINVRNT(osd_invariant(obj));
2579
2580         return obj->oo_owner == env;
2581 }
2582
2583 static struct timespec *osd_inode_time(const struct lu_env *env,
2584                                        struct inode *inode, __u64 seconds)
2585 {
2586         struct osd_thread_info *oti = osd_oti_get(env);
2587         struct timespec *t = &oti->oti_time;
2588
2589         t->tv_sec = seconds;
2590         t->tv_nsec = 0;
2591         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2592         return t;
2593 }
2594
2595 static void osd_inode_getattr(const struct lu_env *env,
2596                               struct inode *inode, struct lu_attr *attr)
2597 {
2598         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2599                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2600                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2601                            LA_BLKSIZE | LA_TYPE;
2602
2603         attr->la_atime   = LTIME_S(inode->i_atime);
2604         attr->la_mtime   = LTIME_S(inode->i_mtime);
2605         attr->la_ctime   = LTIME_S(inode->i_ctime);
2606         attr->la_mode    = inode->i_mode;
2607         attr->la_size    = i_size_read(inode);
2608         attr->la_blocks  = inode->i_blocks;
2609         attr->la_uid     = i_uid_read(inode);
2610         attr->la_gid     = i_gid_read(inode);
2611         attr->la_projid  = i_projid_read(inode);
2612         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2613         attr->la_nlink   = inode->i_nlink;
2614         attr->la_rdev    = inode->i_rdev;
2615         attr->la_blksize = 1 << inode->i_blkbits;
2616         attr->la_blkbits = inode->i_blkbits;
2617         /*
2618          * Ext4 did not transfer inherit flags from raw inode
2619          * to inode flags, and ext4 internally test raw inode
2620          * @i_flags directly. Instead of patching ext4, we do it here.
2621          */
2622         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2623                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2624 }
2625
2626 static int osd_attr_get(const struct lu_env *env,
2627                         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_flags |= LUSTRE_ORPHAN_FL;
2644         spin_unlock(&obj->oo_guard);
2645
2646         return 0;
2647 }
2648
2649 static int osd_declare_attr_qid(const struct lu_env *env,
2650                                 struct osd_object *obj,
2651                                 struct osd_thandle *oh, long long bspace,
2652                                 qid_t old_id, qid_t new_id, bool enforce,
2653                                 unsigned int type, bool ignore_edquot)
2654 {
2655         int rc;
2656         struct osd_thread_info *info = osd_oti_get(env);
2657         struct lquota_id_info  *qi = &info->oti_qi;
2658
2659         qi->lqi_type = type;
2660         /* inode accounting */
2661         qi->lqi_is_blk = false;
2662
2663         /* one more inode for the new id ... */
2664         qi->lqi_id.qid_uid = new_id;
2665         qi->lqi_space      = 1;
2666         /* Reserve credits for the new id */
2667         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2668         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2669                 rc = 0;
2670         if (rc)
2671                 RETURN(rc);
2672
2673         /* and one less inode for the current id */
2674         qi->lqi_id.qid_uid = old_id;
2675         qi->lqi_space = -1;
2676         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2677         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2678                 rc = 0;
2679         if (rc)
2680                 RETURN(rc);
2681
2682         /* block accounting */
2683         qi->lqi_is_blk = true;
2684
2685         /* more blocks for the new id ... */
2686         qi->lqi_id.qid_uid = new_id;
2687         qi->lqi_space      = bspace;
2688         /*
2689          * Credits for the new uid has been reserved, re-use "obj"
2690          * to save credit reservation.
2691          */
2692         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2693         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2694                 rc = 0;
2695         if (rc)
2696                 RETURN(rc);
2697
2698         /* and finally less blocks for the current uid */
2699         qi->lqi_id.qid_uid = old_id;
2700         qi->lqi_space      = -bspace;
2701         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2702         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2703                 rc = 0;
2704
2705         RETURN(rc);
2706 }
2707
2708 static int osd_declare_attr_set(const struct lu_env *env,
2709                                 struct dt_object *dt,
2710                                 const struct lu_attr *attr,
2711                                 struct thandle *handle)
2712 {
2713         struct osd_thandle *oh;
2714         struct osd_object *obj;
2715         qid_t uid;
2716         qid_t gid;
2717         long long bspace;
2718         int rc = 0;
2719         bool enforce;
2720
2721         ENTRY;
2722
2723         LASSERT(dt != NULL);
2724         LASSERT(handle != NULL);
2725
2726         obj = osd_dt_obj(dt);
2727         LASSERT(osd_invariant(obj));
2728
2729         oh = container_of0(handle, struct osd_thandle, ot_super);
2730         LASSERT(oh->ot_handle == NULL);
2731
2732         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2733                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2734
2735         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2736                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2737
2738         if (attr == NULL || obj->oo_inode == NULL)
2739                 RETURN(rc);
2740
2741         bspace   = obj->oo_inode->i_blocks << 9;
2742         bspace   = toqb(bspace);
2743
2744         /*
2745          * Changing ownership is always preformed by super user, it should not
2746          * fail with EDQUOT unless required explicitly.
2747          *
2748          * We still need to call the osd_declare_qid() to calculate the journal
2749          * credits for updating quota accounting files and to trigger quota
2750          * space adjustment once the operation is completed.
2751          */
2752         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2753                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2754
2755                 if (!ignore_edquot)
2756                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2757                                "(the quota space is %lld)\n",
2758                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2759                                attr->la_gid, bspace);
2760
2761                 /* USERQUOTA */
2762                 uid = i_uid_read(obj->oo_inode);
2763                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2764                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2765                                           attr->la_uid, enforce, USRQUOTA,
2766                                           true);
2767                 if (rc)
2768                         RETURN(rc);
2769
2770                 gid = i_gid_read(obj->oo_inode);
2771                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2772                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2773                                           i_gid_read(obj->oo_inode),
2774                                           attr->la_gid, enforce, GRPQUOTA,
2775                                           ignore_edquot);
2776                 if (rc)
2777                         RETURN(rc);
2778
2779         }
2780 #ifdef HAVE_PROJECT_QUOTA
2781         if (attr->la_valid & LA_PROJID) {
2782                 __u32 projid = i_projid_read(obj->oo_inode);
2783
2784                 enforce = (attr->la_valid & LA_PROJID) &&
2785                                         (attr->la_projid != projid);
2786                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2787                                           (qid_t)projid, (qid_t)attr->la_projid,
2788                                           enforce, PRJQUOTA, true);
2789                 if (rc)
2790                         RETURN(rc);
2791         }
2792 #endif
2793         RETURN(rc);
2794 }
2795
2796 static int osd_inode_setattr(const struct lu_env *env,
2797                              struct inode *inode, const struct lu_attr *attr)
2798 {
2799         __u64 bits = attr->la_valid;
2800
2801         /* Only allow set size for regular file */
2802         if (!S_ISREG(inode->i_mode))
2803                 bits &= ~(LA_SIZE | LA_BLOCKS);
2804
2805         if (bits == 0)
2806                 return 0;
2807
2808         if (bits & LA_ATIME)
2809                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2810         if (bits & LA_CTIME)
2811                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2812         if (bits & LA_MTIME)
2813                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2814         if (bits & LA_SIZE) {
2815                 spin_lock(&inode->i_lock);
2816                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2817                 i_size_write(inode, attr->la_size);
2818                 spin_unlock(&inode->i_lock);
2819         }
2820
2821         /*
2822          * OSD should not change "i_blocks" which is used by quota.
2823          * "i_blocks" should be changed by ldiskfs only.
2824          */
2825         if (bits & LA_MODE)
2826                 inode->i_mode = (inode->i_mode & S_IFMT) |
2827                                 (attr->la_mode & ~S_IFMT);
2828         if (bits & LA_UID)
2829                 i_uid_write(inode, attr->la_uid);
2830         if (bits & LA_GID)
2831                 i_gid_write(inode, attr->la_gid);
2832         if (bits & LA_PROJID)
2833                 i_projid_write(inode, attr->la_projid);
2834         if (bits & LA_NLINK)
2835                 set_nlink(inode, attr->la_nlink);
2836         if (bits & LA_RDEV)
2837                 inode->i_rdev = attr->la_rdev;
2838
2839         if (bits & LA_FLAGS) {
2840                 /* always keep S_NOCMTIME */
2841                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2842                                  S_NOCMTIME;
2843                 /*
2844                  * Ext4 did not transfer inherit flags from
2845                  * @inode->i_flags to raw inode i_flags when writing
2846                  * flags, we do it explictly here.
2847                  */
2848                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2849                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2850                 else
2851                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2852         }
2853         return 0;
2854 }
2855
2856 #ifdef HAVE_PROJECT_QUOTA
2857 static int osd_transfer_project(struct inode *inode, __u32 projid)
2858 {
2859         struct super_block *sb = inode->i_sb;
2860         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2861         int err;
2862         kprojid_t kprojid;
2863         struct ldiskfs_iloc iloc;
2864         struct ldiskfs_inode *raw_inode;
2865         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2866
2867         if (!ldiskfs_has_feature_project(sb)) {
2868                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2869                         == LDISKFS_DEF_PROJID);
2870                 if (projid != LDISKFS_DEF_PROJID)
2871                         return -EOPNOTSUPP;
2872                 else
2873                         return 0;
2874         }
2875
2876         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2877                 return -EOPNOTSUPP;
2878
2879         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2880         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2881                 return 0;
2882
2883         err = ldiskfs_get_inode_loc(inode, &iloc);
2884         if (err)
2885                 return err;
2886
2887         raw_inode = ldiskfs_raw_inode(&iloc);
2888         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2889                 err = -EOVERFLOW;
2890                 brelse(iloc.bh);
2891                 return err;
2892         }
2893         brelse(iloc.bh);
2894
2895         dquot_initialize(inode);
2896         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2897         if (transfer_to[PRJQUOTA]) {
2898                 err = __dquot_transfer(inode, transfer_to);
2899                 dqput(transfer_to[PRJQUOTA]);
2900                 if (err)
2901                         return err;
2902         }
2903
2904         return err;
2905 }
2906 #endif
2907
2908 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2909 {
2910         int rc;
2911
2912         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2913             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2914                 struct iattr iattr;
2915
2916                 ll_vfs_dq_init(inode);
2917                 iattr.ia_valid = 0;
2918                 if (attr->la_valid & LA_UID)
2919                         iattr.ia_valid |= ATTR_UID;
2920                 if (attr->la_valid & LA_GID)
2921                         iattr.ia_valid |= ATTR_GID;
2922                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2923                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2924
2925                 rc = ll_vfs_dq_transfer(inode, &iattr);
2926                 if (rc) {
2927                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2928                                "enforcement enabled on the ldiskfs "
2929                                "filesystem?\n", inode->i_sb->s_id, rc);
2930                         return rc;
2931                 }
2932         }
2933
2934         /* Handle project id transfer here properly */
2935         if (attr->la_valid & LA_PROJID &&
2936             attr->la_projid != i_projid_read(inode)) {
2937 #ifdef HAVE_PROJECT_QUOTA
2938                 rc = osd_transfer_project(inode, attr->la_projid);
2939 #else
2940                 rc = -ENOTSUPP;
2941 #endif
2942                 if (rc) {
2943                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2944                                "enforcement enabled on the ldiskfs "
2945                                "filesystem?\n", inode->i_sb->s_id, rc);
2946                         return rc;
2947                 }
2948         }
2949         return 0;
2950 }
2951
2952 static int osd_attr_set(const struct lu_env *env,
2953                         struct dt_object *dt,
2954                         const struct lu_attr *attr,
2955                         struct thandle *handle)
2956 {
2957         struct osd_object *obj = osd_dt_obj(dt);
2958         struct inode *inode;
2959         int rc;
2960
2961         if (!dt_object_exists(dt))
2962                 return -ENOENT;
2963
2964         LASSERT(handle != NULL);
2965         LASSERT(!dt_object_remote(dt));
2966         LASSERT(osd_invariant(obj));
2967
2968         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2969
2970         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2971             !osd_obj2dev(obj)->od_is_ost) {
2972                 struct osd_thread_info *oti = osd_oti_get(env);
2973                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2974                 struct lu_fid *fid1 = &oti->oti_fid;
2975                 struct osd_inode_id *id = &oti->oti_id;
2976                 struct iam_path_descr *ipd;
2977                 struct iam_container *bag;
2978                 struct osd_thandle *oh;
2979                 int rc;
2980
2981                 fid_cpu_to_be(fid1, fid0);
2982                 memset(id, 1, sizeof(*id));
2983                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2984                                   fid0)->oi_dir.od_container;
2985                 ipd = osd_idx_ipd_get(env, bag);
2986                 if (unlikely(ipd == NULL))
2987                         RETURN(-ENOMEM);
2988
2989                 oh = container_of0(handle, struct osd_thandle, ot_super);
2990                 rc = iam_update(oh->ot_handle, bag,
2991                                 (const struct iam_key *)fid1,
2992                                 (const struct iam_rec *)id, ipd);
2993                 osd_ipd_put(env, bag, ipd);
2994                 return(rc > 0 ? 0 : rc);
2995         }
2996
2997         inode = obj->oo_inode;
2998
2999         rc = osd_quota_transfer(inode, attr);
3000         if (rc)
3001                 return rc;
3002
3003         spin_lock(&obj->oo_guard);
3004         rc = osd_inode_setattr(env, inode, attr);
3005         spin_unlock(&obj->oo_guard);
3006         if (rc != 0)
3007                 GOTO(out, rc);
3008
3009         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3010
3011         if (!(attr->la_valid & LA_FLAGS))
3012                 GOTO(out, rc);
3013
3014         /* Let's check if there are extra flags need to be set into LMA */
3015         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3016                 struct osd_thread_info *info = osd_oti_get(env);
3017                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3018
3019                 LASSERT(!obj->oo_pfid_in_lma);
3020
3021                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3022                                  &info->oti_ost_attrs);
3023                 if (rc)
3024                         GOTO(out, rc);
3025
3026                 lma->lma_incompat |=
3027                         lustre_to_lma_flags(attr->la_flags);
3028                 lustre_lma_swab(lma);
3029                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3030                                      lma, sizeof(*lma), XATTR_REPLACE);
3031                 if (rc != 0) {
3032                         struct osd_device *osd = osd_obj2dev(obj);
3033
3034                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3035                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3036                               lma->lma_incompat, rc);
3037                 } else {
3038                         obj->oo_lma_flags =
3039                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3040                 }
3041                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3042         }
3043 out:
3044         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3045
3046         return rc;
3047 }
3048
3049 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3050                                            struct osd_object *obj,
3051                                            const char *name, const int namelen)
3052 {
3053         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3054 }
3055
3056 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3057                       umode_t mode, struct dt_allocation_hint *hint,
3058                       struct thandle *th)
3059 {
3060         int result;
3061         struct osd_device *osd = osd_obj2dev(obj);
3062         struct osd_thandle *oth;
3063         struct dt_object *parent = NULL;
3064         struct inode *inode;
3065
3066         LINVRNT(osd_invariant(obj));
3067         LASSERT(obj->oo_inode == NULL);
3068         LASSERT(obj->oo_hl_head == NULL);
3069
3070         if (S_ISDIR(mode) && ldiskfs_pdo) {
3071                 obj->oo_hl_head =
3072                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3073                 if (obj->oo_hl_head == NULL)
3074                         return -ENOMEM;
3075         }
3076
3077         oth = container_of(th, struct osd_thandle, ot_super);
3078         LASSERT(oth->ot_handle->h_transaction != NULL);
3079
3080         if (hint != NULL && hint->dah_parent != NULL &&
3081             !dt_object_remote(hint->dah_parent))
3082                 parent = hint->dah_parent;
3083
3084         inode = ldiskfs_create_inode(oth->ot_handle,
3085                                      parent ? osd_dt_obj(parent)->oo_inode :
3086                                               osd_sb(osd)->s_root->d_inode,
3087                                      mode);
3088         if (!IS_ERR(inode)) {
3089                 /* Do not update file c/mtime in ldiskfs. */
3090                 inode->i_flags |= S_NOCMTIME;
3091
3092                 /*
3093                  * For new created object, it must be consistent,
3094                  * and it is unnecessary to scrub against it.
3095                  */
3096                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3097
3098                 obj->oo_inode = inode;
3099                 result = 0;
3100         } else {
3101                 if (obj->oo_hl_head != NULL) {
3102                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3103                         obj->oo_hl_head = NULL;
3104                 }
3105                 result = PTR_ERR(inode);
3106         }
3107         LINVRNT(osd_invariant(obj));
3108         return result;
3109 }
3110
3111 enum {
3112         OSD_NAME_LEN = 255
3113 };
3114
3115 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3116                      struct lu_attr *attr,
3117                      struct dt_allocation_hint *hint,
3118                      struct dt_object_format *dof,
3119                      struct thandle *th)
3120 {
3121         int result;
3122         struct osd_thandle *oth;
3123         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3124
3125         LASSERT(S_ISDIR(attr->la_mode));
3126
3127         oth = container_of(th, struct osd_thandle, ot_super);
3128         LASSERT(oth->ot_handle->h_transaction != NULL);
3129         result = osd_mkfile(info, obj, mode, hint, th);
3130
3131         return result;
3132 }
3133
3134 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3135                         struct lu_attr *attr,
3136                         struct dt_allocation_hint *hint,
3137                         struct dt_object_format *dof,
3138                         struct thandle *th)
3139 {
3140         int result;
3141         struct osd_thandle *oth;
3142         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3143
3144         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3145
3146         LASSERT(S_ISREG(attr->la_mode));
3147
3148         oth = container_of(th, struct osd_thandle, ot_super);
3149         LASSERT(oth->ot_handle->h_transaction != NULL);
3150
3151         result = osd_mkfile(info, obj, mode, hint, th);
3152         if (result == 0) {
3153                 LASSERT(obj->oo_inode != NULL);
3154                 if (feat->dif_flags & DT_IND_VARKEY)
3155                         result = iam_lvar_create(obj->oo_inode,
3156                                                  feat->dif_keysize_max,
3157                                                  feat->dif_ptrsize,
3158                                                  feat->dif_recsize_max,
3159                                                  oth->ot_handle);
3160                 else
3161                         result = iam_lfix_create(obj->oo_inode,
3162                                                  feat->dif_keysize_max,
3163                                                  feat->dif_ptrsize,
3164                                                  feat->dif_recsize_max,
3165                                                  oth->ot_handle);
3166         }
3167         return result;
3168 }
3169
3170 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3171                      struct lu_attr *attr,
3172                      struct dt_allocation_hint *hint,
3173                      struct dt_object_format *dof,
3174                      struct thandle *th)
3175 {
3176         LASSERT(S_ISREG(attr->la_mode));
3177         return osd_mkfile(info, obj, (attr->la_mode &
3178                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3179 }
3180
3181 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3182                      struct lu_attr *attr,
3183                      struct dt_allocation_hint *hint,
3184                      struct dt_object_format *dof,
3185                      struct thandle *th)
3186 {
3187         LASSERT(S_ISLNK(attr->la_mode));
3188         return osd_mkfile(info, obj, (attr->la_mode &
3189                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3190 }
3191
3192 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3193                      struct lu_attr *attr,
3194                      struct dt_allocation_hint *hint,
3195                      struct dt_object_format *dof,
3196                      struct thandle *th)
3197 {
3198         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3199         int result;
3200
3201         LINVRNT(osd_invariant(obj));
3202         LASSERT(obj->oo_inode == NULL);
3203         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3204                 S_ISFIFO(mode) || S_ISSOCK(mode));
3205
3206         result = osd_mkfile(info, obj, mode, hint, th);
3207         if (result == 0) {
3208                 LASSERT(obj->oo_inode != NULL);
3209                 /*
3210                  * This inode should be marked dirty for i_rdev.  Currently
3211                  * that is done in the osd_attr_init().
3212                  */
3213                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3214                                    attr->la_rdev);
3215         }
3216         LINVRNT(osd_invariant(obj));
3217         return result;
3218 }
3219
3220 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3221                               struct lu_attr *,
3222                               struct dt_allocation_hint *hint,
3223                               struct dt_object_format *dof,
3224                               struct thandle *);
3225
3226 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3227 {
3228         osd_obj_type_f result;
3229
3230         switch (type) {
3231         case DFT_DIR:
3232                 result = osd_mkdir;
3233                 break;
3234         case DFT_REGULAR:
3235                 result = osd_mkreg;
3236                 break;
3237         case DFT_SYM:
3238                 result = osd_mksym;
3239                 break;
3240         case DFT_NODE:
3241                 result = osd_mknod;
3242                 break;
3243         case DFT_INDEX:
3244                 result = osd_mk_index;
3245                 break;
3246
3247         default:
3248                 LBUG();
3249                 break;
3250         }
3251         return result;
3252 }
3253
3254
3255 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3256                         struct dt_object *parent, struct dt_object *child,
3257                         umode_t child_mode)
3258 {
3259         LASSERT(ah);
3260
3261         ah->dah_parent = parent;
3262         ah->dah_mode = child_mode;
3263
3264         if (parent != NULL && !dt_object_remote(parent)) {
3265                 /* will help to find FID->ino at dt_insert("..") */
3266                 struct osd_object *pobj = osd_dt_obj(parent);
3267
3268                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3269         }
3270 }
3271
3272 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3273                           struct lu_attr *attr, struct dt_object_format *dof)
3274 {
3275         struct inode *inode = obj->oo_inode;
3276         __u64 valid = attr->la_valid;
3277         int result;
3278
3279         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3280
3281         if (dof->dof_type != DFT_NODE)
3282                 attr->la_valid &= ~LA_RDEV;
3283         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3284                 attr->la_valid &= ~LA_ATIME;
3285         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3286                 attr->la_valid &= ~LA_CTIME;
3287         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3288                 attr->la_valid &= ~LA_MTIME;
3289
3290         result = osd_quota_transfer(inode, attr);
3291         if (result)
3292                 return;
3293
3294         if (attr->la_valid != 0) {
3295                 result = osd_inode_setattr(info->oti_env, inode, attr);
3296                 /*
3297                  * The osd_inode_setattr() should always succeed here.  The
3298                  * only error that could be returned is EDQUOT when we are
3299                  * trying to change the UID or GID of the inode. However, this
3300                  * should not happen since quota enforcement is no longer
3301                  * enabled on ldiskfs (lquota takes care of it).
3302                  */
3303                 LASSERTF(result == 0, "%d\n", result);
3304                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3305         }
3306
3307         attr->la_valid = valid;
3308 }
3309
3310 /**
3311  * Helper function for osd_create()
3312  *
3313  * \retval 0, on success
3314  */
3315 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3316                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3317                         struct dt_object_format *dof, struct thandle *th)
3318 {
3319         int result;
3320         __u32 umask;
3321
3322         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3323
3324         /* we drop umask so that permissions we pass are not affected */
3325         umask = current->fs->umask;
3326         current->fs->umask = 0;
3327
3328         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3329                                                   th);
3330         if (likely(obj->oo_inode != NULL)) {
3331                 LASSERT(obj->oo_inode->i_state & I_NEW);
3332
3333                 /*
3334                  * Unlock the inode before attr initialization to avoid
3335                  * unnecessary dqget operations. LU-6378
3336                  */
3337                 unlock_new_inode(obj->oo_inode);
3338         }
3339
3340         if (likely(result == 0)) {
3341                 osd_attr_init(info, obj, attr, dof);
3342                 osd_object_init0(obj);
3343         }
3344
3345         /* restore previous umask value */
3346         current->fs->umask = umask;
3347
3348         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3349
3350         return result;
3351 }
3352
3353 /**
3354  * Helper function for osd_create()
3355  *
3356  * \retval 0, on success
3357  */
3358 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3359                            const struct lu_fid *fid, struct thandle *th)
3360 {
3361         struct osd_thread_info *info = osd_oti_get(env);
3362         struct osd_inode_id    *id   = &info->oti_id;
3363         struct osd_device      *osd  = osd_obj2dev(obj);
3364         struct osd_thandle     *oh;
3365         int rc;
3366
3367         LASSERT(obj->oo_inode != NULL);
3368
3369         oh = container_of0(th, struct osd_thandle, ot_super);
3370         LASSERT(oh->ot_handle);
3371         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3372
3373         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3374         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3375                            OI_CHECK_FLD, NULL);
3376         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3377
3378         return rc;
3379 }
3380
3381 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3382                    u64 seq, struct lu_seq_range *range)
3383 {
3384         struct seq_server_site *ss = osd_seq_site(osd);
3385
3386         if (fid_seq_is_idif(seq)) {
3387                 fld_range_set_ost(range);
3388                 range->lsr_index = idif_ost_idx(seq);
3389                 return 0;
3390         }
3391
3392         if (!fid_seq_in_fldb(seq)) {
3393                 fld_range_set_mdt(range);
3394                 if (ss != NULL)
3395                         /*
3396                          * FIXME: If ss is NULL, it suppose not get lsr_index
3397                          * at all
3398                          */
3399                         range->lsr_index = ss->ss_node_id;
3400                 return 0;
3401         }
3402
3403         LASSERT(ss != NULL);
3404         fld_range_set_any(range);
3405         /* OSD will only do local fld lookup */
3406         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3407 }
3408
3409 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3410                               struct lu_attr *attr,
3411                               struct dt_allocation_hint *hint,
3412                               struct dt_object_format *dof,
3413                               struct thandle *handle)
3414 {
3415         struct osd_thandle *oh;
3416         int rc;
3417
3418         ENTRY;
3419
3420         LASSERT(handle != NULL);
3421
3422         oh = container_of0(handle, struct osd_thandle, ot_super);
3423         LASSERT(oh->ot_handle == NULL);
3424
3425         /*
3426          * EA object consumes more credits than regular object: osd_mk_index
3427          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3428          * leaf_node, could involves the block, block bitmap, groups, GDT
3429          * change for each block, so add 4 * 2 credits in that case.
3430          */
3431         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3432                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3433                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3434         /*
3435          * Reuse idle OI block may cause additional one OI block
3436          * to be changed.
3437          */
3438         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3439                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3440
3441         /* will help to find FID->ino mapping at dt_insert() */
3442         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3443                                    osd_dt_obj(dt));
3444         if (rc != 0)
3445                 RETURN(rc);
3446
3447         if (!attr)
3448                 RETURN(0);
3449
3450         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3451                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3452                                    NULL, OSD_QID_INODE);
3453         if (rc != 0)
3454                 RETURN(rc);
3455
3456         RETURN(rc);
3457 }
3458
3459 /**
3460  * Called to destroy on-disk representation of the object
3461  *
3462  * Concurrency: must be locked
3463  */
3464 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3465                                struct thandle *th)
3466 {
3467         struct osd_object *obj = osd_dt_obj(dt);
3468         struct inode *inode = obj->oo_inode;
3469         struct osd_thandle *oh;
3470         int rc;
3471
3472         ENTRY;
3473
3474         if (inode == NULL)
3475                 RETURN(-ENOENT);
3476
3477         oh = container_of0(th, struct osd_thandle, ot_super);
3478         LASSERT(oh->ot_handle == NULL);
3479
3480         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3481                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3482
3483         /* For removing agent entry */
3484         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3485                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3486
3487         /*
3488          * Recycle idle OI leaf may cause additional three OI blocks
3489          * to be changed.
3490          */
3491         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3492                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3493                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3494         /* one less inode */
3495         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3496                                    i_projid_read(inode), -1, oh, obj, NULL,
3497                                    OSD_QID_INODE);
3498         if (rc)
3499                 RETURN(rc);
3500         /* data to be truncated */
3501         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3502                                    i_projid_read(inode), 0, oh, obj, NULL,
3503                                    OSD_QID_BLK);
3504         if (rc)
3505                 RETURN(rc);
3506
3507         /*
3508          * will help to find FID->ino when this object is being
3509          * added to PENDING
3510          */
3511         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3512
3513         RETURN(rc);
3514 }
3515
3516 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3517                        struct thandle *th)
3518 {
3519         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3520         struct osd_object *obj = osd_dt_obj(dt);
3521         struct inode *inode = obj->oo_inode;
3522         struct osd_device *osd = osd_obj2dev(obj);
3523         struct osd_thandle *oh;
3524         int result;
3525
3526         ENTRY;
3527
3528         oh = container_of0(th, struct osd_thandle, ot_super);
3529         LASSERT(oh->ot_handle);
3530         LASSERT(inode);
3531         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3532
3533         if (unlikely(fid_is_acct(fid)))
3534                 RETURN(-EPERM);
3535
3536         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3537                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3538                 if (result != 0)
3539                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3540                                osd_name(osd), PFID(fid), result);
3541         }
3542
3543         if (S_ISDIR(inode->i_mode)) {
3544                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3545                         inode->i_nlink == 2);
3546
3547                 spin_lock(&obj->oo_guard);
3548                 clear_nlink(inode);
3549                 spin_unlock(&obj->oo_guard);
3550                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3551         }
3552
3553         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3554
3555         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3556
3557         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3558                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3559                                        oh->ot_handle, OI_CHECK_FLD);
3560
3561         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3562         /* XXX: add to ext3 orphan list */
3563         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3564
3565         /* not needed in the cache anymore */
3566         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3567         obj->oo_destroyed = 1;
3568
3569         RETURN(0);
3570 }
3571
3572 /**
3573  * Put the fid into lustre_mdt_attrs, and then place the structure
3574  * inode's ea. This fid should not be altered during the life time
3575  * of the inode.
3576  *
3577  * \retval +ve, on success
3578  * \retval -ve, on error
3579  *
3580  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3581  */
3582 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3583                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3584 {
3585         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3586         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3587         int rc;
3588
3589         ENTRY;
3590
3591         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3592                 RETURN(0);
3593
3594         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3595                 rc = -ENOMEM;
3596
3597         lustre_loa_init(loa, fid, compat, incompat);
3598         lustre_loa_swab(loa, false);
3599
3600         /*
3601          * For the OST device with 256 bytes inode size by default,
3602          * the PFID EA will be stored together with LMA EA to avoid
3603          * performance trouble. Otherwise the PFID EA can be stored
3604          * independently. LU-8998
3605          */
3606         if ((compat & LMAC_FID_ON_OST) &&
3607             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3608                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3609                                      sizeof(*loa), XATTR_CREATE);
3610         else
3611                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3612                                      sizeof(*lma), XATTR_CREATE);
3613         /*
3614          * LMA may already exist, but we need to check that all the
3615          * desired compat/incompat flags have been added.
3616          */
3617         if (unlikely(rc == -EEXIST)) {
3618                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3619                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3620                 if (rc < 0)
3621                         RETURN(rc);
3622
3623                 if (rc < sizeof(*lma))
3624                         RETURN(-EINVAL);
3625
3626                 lustre_loa_swab(loa, true);
3627                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3628                     ((compat == 0 && incompat == 0) ||
3629                      (!(~lma->lma_compat & compat) &&
3630                       !(~lma->lma_incompat & incompat))))
3631                         RETURN(0);
3632
3633                 lma->lma_self_fid = *fid;
3634                 lma->lma_compat |= compat;
3635                 lma->lma_incompat |= incompat;
3636                 if (rc == sizeof(*lma)) {
3637                         lustre_lma_swab(lma);
3638                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3639                                              sizeof(*lma), XATTR_REPLACE);
3640                 } else {
3641                         lustre_loa_swab(loa, false);
3642                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3643                                              sizeof(*loa), XATTR_REPLACE);
3644                 }
3645         }
3646
3647         RETURN(rc);
3648 }
3649
3650 /**
3651  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3652  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3653  * To have compatilibility with 1.8 ldiskfs driver we need to have
3654  * magic number at start of fid data.
3655  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3656  * its inmemory API.
3657  */
3658 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3659                                          const struct lu_fid *fid)
3660 {
3661         if (!fid_is_namespace_visible(fid) ||
3662             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3663                 param->edp_magic = 0;
3664                 return;
3665         }
3666
3667         param->edp_magic = LDISKFS_LUFID_MAGIC;
3668         param->edp_len =  sizeof(struct lu_fid) + 1;
3669         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3670 }
3671
3672 /**
3673  * Try to read the fid from inode ea into dt_rec.
3674  *
3675  * \param fid object fid.
3676  *
3677  * \retval 0 on success
3678  */
3679 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3680                           __u32 ino, struct lu_fid *fid,
3681                           struct osd_inode_id *id)
3682 {
3683         struct osd_thread_info *info  = osd_oti_get(env);
3684         struct inode *inode;
3685
3686         ENTRY;
3687
3688         osd_id_gen(id, ino, OSD_OII_NOGEN);
3689         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3690         if (IS_ERR(inode))
3691                 RETURN(PTR_ERR(inode));
3692
3693         iput(inode);
3694         RETURN(0);
3695 }
3696
3697 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3698                                         struct inode *dir,
3699                                         struct inode *parent_dir,
3700                                         const struct lu_fid *dot_fid,
3701                                         const struct lu_fid *dot_dot_fid,
3702                                         struct osd_thandle *oth)
3703 {
3704         struct ldiskfs_dentry_param *dot_ldp;
3705         struct ldiskfs_dentry_param *dot_dot_ldp;
3706         __u32 saved_nlink = dir->i_nlink;
3707         int rc;
3708
3709         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3710         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3711
3712         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3713         dot_ldp->edp_magic = 0;
3714
3715         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3716                                     dir, dot_ldp, dot_dot_ldp);
3717         /*
3718          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3719          * the subseqent ref_add() will increase the dir->i_nlink
3720          * as 3. That is incorrect for new created directory.
3721          *
3722          * It looks like hack, because we want to make the OSD API
3723          * to be order-independent for new created directory object
3724          * between dt_insert(..) and ref_add() operations.
3725          *
3726          * Here, we only restore the in-RAM dir-inode's nlink attr,
3727          * becuase if the nlink attr is not 2, then there will be
3728          * ref_add() called following the dt_insert(..), such call
3729          * will make both the in-RAM and on-disk dir-inode's nlink
3730          * attr to be set as 2. LU-7447
3731          */
3732         set_nlink(dir, saved_nlink);
3733         return rc;
3734 }
3735
3736 /**
3737  * Create an local agent inode for remote entry
3738  */
3739 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3740                                                   struct osd_device *osd,
3741                                                   struct osd_object *pobj,
3742                                                   const struct lu_fid *fid,
3743                                                   __u32 type,
3744                                                   struct thandle *th)
3745 {
3746         struct osd_thread_info *info = osd_oti_get(env);
3747         struct inode *local;
3748         struct osd_thandle *oh;
3749         int rc;
3750
3751         ENTRY;
3752
3753         LASSERT(th);
3754         oh = container_of(th, struct osd_thandle, ot_super);
3755         LASSERT(oh->ot_handle->h_transaction != NULL);
3756
3757         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3758         if (IS_ERR(local)) {
3759                 CERROR("%s: create local error %d\n", osd_name(osd),
3760                        (int)PTR_ERR(local));
3761                 RETURN(local);
3762         }
3763
3764         /*
3765          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3766          * correct gid on remote file, not agent here
3767          */
3768         local->i_gid = current_fsgid();
3769         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3770
3771         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3772          * That gives e2fsck something to look at and be happy, and allows
3773          * debugging if we need to determine where this symlink came from.
3774          */
3775         if (S_ISLNK(type)) {
3776                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3777                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3778                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3779
3780                 i_size_write(local, rc);
3781                 LDISKFS_I(local)->i_disksize = rc;
3782         }
3783         unlock_new_inode(local);
3784
3785         /* Agent inode should not have project ID */
3786 #ifdef  HAVE_PROJECT_QUOTA
3787         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3788             i_projid_read(pobj->oo_inode) != 0) {
3789                 rc = osd_transfer_project(local, 0);
3790                 if (rc) {
3791                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3792                                "quota enforcement enabled on the ldiskfs "
3793                                "filesystem?\n", local->i_sb->s_id, rc);
3794                         RETURN(ERR_PTR(rc));
3795                 }
3796         }
3797 #endif
3798         /* Set special LMA flag for local agent inode */
3799         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3800         if (rc != 0) {
3801                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3802                        osd_name(osd), PFID(fid), rc);
3803                 RETURN(ERR_PTR(rc));
3804         }
3805
3806         if (!S_ISDIR(type))
3807                 RETURN(local);
3808
3809         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3810                                          lu_object_fid(&pobj->oo_dt.do_lu),
3811                                          fid, oh);
3812         if (rc != 0) {
3813                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3814                         osd_name(osd), PFID(fid), rc);
3815                 RETURN(ERR_PTR(rc));
3816         }
3817
3818         RETURN(local);
3819 }
3820
3821 /**
3822  * when direntry is deleted, we have to take care of possible agent inode
3823  * referenced by that. unfortunately we can't do this at that point:
3824  * iget() within a running transaction leads to deadlock and we better do
3825  * not call that every delete declaration to save performance. so we put
3826  * a potention agent inode on a list and process that once the transaction
3827  * is over. Notice it's not any worse in terms of real orphans as regular
3828  * object destroy doesn't put inodes on the on-disk orphan list. this should
3829  * be addressed separately
3830  */
3831 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3832                                             struct osd_thandle *oh,
3833                                             __u32 ino)
3834 {
3835         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3836         struct osd_obj_orphan *oor;
3837
3838         OBD_ALLOC_PTR(oor);
3839         if (oor == NULL)
3840                 return -ENOMEM;
3841
3842         oor->oor_ino = ino;
3843         oor->oor_env = (struct lu_env *)env;
3844         spin_lock(&osd->od_osfs_lock);
3845         list_add(&oor->oor_list, &osd->od_orphan_list);
3846         spin_unlock(&osd->od_osfs_lock);
3847
3848         oh->ot_remove_agents = 1;
3849
3850         return 0;
3851
3852 }
3853
3854 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3855                                                 struct osd_device *osd)
3856 {
3857         struct osd_thread_info *info = osd_oti_get(env);
3858         struct osd_obj_orphan *oor, *tmp;
3859         struct osd_inode_id id;
3860         struct list_head list;
3861         struct inode *inode;
3862         struct lu_fid fid;
3863         handle_t *jh;
3864         __u32 ino;
3865
3866         INIT_LIST_HEAD(&list);
3867
3868         spin_lock(&osd->od_osfs_lock);
3869         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3870                 if (oor->oor_env == env) {
3871                         list_del(&oor->oor_list);
3872                         list_add(&oor->oor_list, &list);
3873                 }
3874         }
3875         spin_unlock(&osd->od_osfs_lock);
3876
3877         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3878
3879                 ino = oor->oor_ino;
3880
3881                 list_del(&oor->oor_list);
3882                 OBD_FREE_PTR(oor);
3883
3884                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3885                 inode = osd_iget_fid(info, osd, &id, &fid);
3886                 if (IS_ERR(inode))
3887                         continue;
3888
3889                 if (!osd_remote_fid(env, osd, &fid)) {
3890                         iput(inode);
3891                         continue;
3892                 }
3893
3894                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3895                 clear_nlink(inode);
3896                 mark_inode_dirty(inode);
3897                 ldiskfs_journal_stop(jh);
3898                 iput(inode);
3899         }
3900
3901         return 0;
3902 }
3903
3904 /**
3905  * OSD layer object create function for OST objects (b=11826).
3906  *
3907  * The FID is inserted into inode xattr here.
3908  *
3909  * \retval   0, on success
3910  * \retval -ve, on error
3911  */
3912 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3913                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3914                       struct dt_object_format *dof, struct thandle *th)
3915 {
3916         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3917         struct osd_object *obj = osd_dt_obj(dt);
3918         struct osd_thread_info *info = osd_oti_get(env);
3919         int result, on_ost = 0;
3920
3921         ENTRY;
3922
3923         if (dt_object_exists(dt))
3924                 RETURN(-EEXIST);
3925
3926         LINVRNT(osd_invariant(obj));
3927         LASSERT(!dt_object_remote(dt));
3928         LASSERT(osd_is_write_locked(env, obj));
3929         LASSERT(th != NULL);
3930
3931         if (unlikely(fid_is_acct(fid)))
3932                 /*
3933                  * Quota files can't be created from the kernel any more,
3934                  * 'tune2fs -O quota' will take care of creating them
3935                  */
3936                 RETURN(-EPERM);
3937
3938         result = __osd_create(info, obj, attr, hint, dof, th);
3939         if (result == 0) {
3940                 if (fid_is_idif(fid) &&
3941                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3942                         struct lu_fid *tfid = &info->oti_fid;
3943                         struct ost_id *oi   = &info->oti_ostid;
3944
3945                         fid_to_ostid(fid, oi);
3946                         ostid_to_fid(tfid, oi, 0);
3947                         on_ost = 1;
3948                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3949                                                 LMAC_FID_ON_OST, 0);
3950                 } else {
3951                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3952                                                fid, OI_CHECK_FLD);
3953                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3954                                                 on_ost ? LMAC_FID_ON_OST : 0,
3955                                                 0);
3956                 }
3957                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3958                         obj->oo_dt.do_body_ops = &osd_body_ops;
3959         }
3960
3961         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3962                 result = __osd_oi_insert(env, obj, fid, th);
3963
3964         /*
3965          * a small optimization - dt_insert() isn't usually applied
3966          * to OST objects, so we don't need to cache OI mapping for
3967          * OST objects
3968          */
3969         if (result == 0 && on_ost == 0) {
3970                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3971
3972                 result = osd_idc_find_and_init(env, osd, obj);
3973                 LASSERT(result == 0);
3974         }
3975
3976         LASSERT(ergo(result == 0,
3977                      dt_object_exists(dt) && !dt_object_remote(dt)));
3978         LINVRNT(osd_invariant(obj));
3979         RETURN(result);
3980 }
3981
3982 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3983                                struct thandle *handle)
3984 {
3985         struct osd_thandle *oh;
3986
3987         /* it's possible that object doesn't exist yet */
3988         LASSERT(handle != NULL);
3989
3990         oh = container_of0(handle, struct osd_thandle, ot_super);
3991         LASSERT(oh->ot_handle == NULL);
3992
3993         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3994                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3995
3996         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3997
3998         return 0;
3999 }
4000
4001 /*
4002  * Concurrency: @dt is write locked.
4003  */
4004 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4005                        struct thandle *th)
4006 {
4007         struct osd_object *obj = osd_dt_obj(dt);
4008         struct inode *inode = obj->oo_inode;
4009         struct osd_thandle *oh;
4010         int rc = 0;
4011
4012         if (!dt_object_exists(dt) || obj->oo_destroyed)
4013                 return -ENOENT;
4014
4015         LINVRNT(osd_invariant(obj));
4016         LASSERT(!dt_object_remote(dt));
4017         LASSERT(osd_is_write_locked(env, obj));
4018         LASSERT(th != NULL);
4019
4020         oh = container_of0(th, struct osd_thandle, ot_super);
4021         LASSERT(oh->ot_handle != NULL);
4022
4023         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4024
4025         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4026                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4027         /*
4028          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4029          * (65000) subdirectories by storing "1" in i_nlink if the link count
4030          * would otherwise overflow. Directory tranversal tools understand
4031          * that (st_nlink == 1) indicates that the filesystem dose not track
4032          * hard links count on the directory, and will not abort subdirectory
4033          * scanning early once (st_nlink - 2) subdirs have been found.
4034          *
4035          * This also has to properly handle the case of inodes with nlink == 0
4036          * in case they are being linked into the PENDING directory
4037          */
4038         spin_lock(&obj->oo_guard);
4039         if (unlikely(inode->i_nlink == 0))
4040                 /* inc_nlink from 0 may cause WARN_ON */
4041                 set_nlink(inode, 1);
4042         else {
4043                 ldiskfs_inc_count(oh->ot_handle, inode);
4044                 if (!S_ISDIR(inode->i_mode))
4045                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4046         }
4047         spin_unlock(&obj->oo_guard);
4048
4049         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4050         LINVRNT(osd_invariant(obj));
4051
4052         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4053
4054         return rc;
4055 }
4056
4057 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4058                                struct thandle *handle)
4059 {
4060         struct osd_thandle *oh;
4061
4062         if (!dt_object_exists(dt))
4063                 return -ENOENT;
4064
4065         LASSERT(!dt_object_remote(dt));
4066         LASSERT(handle != NULL);
4067
4068         oh = container_of0(handle, struct osd_thandle, ot_super);
4069         LASSERT(oh->ot_handle == NULL);
4070
4071         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4072                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4073
4074         return 0;
4075 }
4076
4077 /*
4078  * Concurrency: @dt is write locked.
4079  */
4080 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4081                        struct thandle *th)
4082 {
4083         struct osd_object *obj = osd_dt_obj(dt);
4084         struct inode *inode = obj->oo_inode;
4085         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4086         struct osd_thandle *oh;
4087
4088         if (!dt_object_exists(dt))
4089                 return -ENOENT;
4090
4091         LINVRNT(osd_invariant(obj));
4092         LASSERT(!dt_object_remote(dt));
4093         LASSERT(osd_is_write_locked(env, obj));
4094         LASSERT(th != NULL);
4095
4096         oh = container_of0(th, struct osd_thandle, ot_super);
4097         LASSERT(oh->ot_handle != NULL);
4098
4099         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4100
4101         spin_lock(&obj->oo_guard);
4102         /*
4103          * That can be result of upgrade from old Lustre version and
4104          * applied only to local files.  Just skip this ref_del call.
4105          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4106          */
4107         if (inode->i_nlink == 0) {
4108                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4109                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4110                              ", maybe an upgraded file? (LU-3915)\n",
4111                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4112                 spin_unlock(&obj->oo_guard);
4113                 return 0;
4114         }
4115
4116         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4117                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4118
4119         ldiskfs_dec_count(oh->ot_handle, inode);
4120         spin_unlock(&obj->oo_guard);
4121
4122         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4123         LINVRNT(osd_invariant(obj));
4124
4125         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4126
4127         return 0;
4128 }
4129
4130 /*
4131  * Concurrency: @dt is read locked.
4132  */
4133 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4134                          struct lu_buf *buf, const char *name)
4135 {
4136         struct osd_object      *obj    = osd_dt_obj(dt);
4137         struct inode           *inode  = obj->oo_inode;
4138         struct osd_thread_info *info   = osd_oti_get(env);
4139         struct dentry          *dentry = &info->oti_obj_dentry;
4140         bool cache_xattr = false;
4141         int rc;
4142
4143         LASSERT(buf);
4144
4145         /* version get is not real XATTR but uses xattr API */
4146         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4147                 dt_obj_version_t *ver = buf->lb_buf;
4148
4149                 /*
4150                  * for version we are just using xattr API but change inode
4151                  * field instead
4152                  */
4153                 if (buf->lb_len == 0)
4154                         return sizeof(dt_obj_version_t);
4155
4156                 if (buf->lb_len < sizeof(dt_obj_version_t))
4157                         return -ERANGE;
4158
4159                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4160                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4161
4162                 *ver = LDISKFS_I(inode)->i_fs_version;
4163
4164                 return sizeof(dt_obj_version_t);
4165         }
4166
4167         if (!dt_object_exists(dt))
4168                 return -ENOENT;
4169
4170         LASSERT(!dt_object_remote(dt));
4171         LASSERT(inode->i_op != NULL);
4172 #ifdef HAVE_IOP_XATTR
4173         LASSERT(inode->i_op->getxattr != NULL);
4174 #endif
4175
4176         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4177             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4178                 cache_xattr = true;
4179
4180         if (cache_xattr) {
4181                 rc = osd_oxc_get(obj, name, buf);
4182                 if (rc != -ENOENT)
4183                         return rc;
4184         }
4185
4186         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4187                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4188                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4189                 struct filter_fid *ff;
4190                 struct ost_layout *ol;
4191
4192                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4193                 if (rc)
4194                         return rc;
4195
4196                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4197
4198                 rc = sizeof(*ff);
4199                 if (buf->lb_len == 0 || !buf->lb_buf)
4200                         return rc;
4201
4202                 if (buf->lb_len < rc)
4203                         return -ERANGE;
4204
4205                 ff = buf->lb_buf;
4206                 ol = &ff->ff_layout;
4207                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4208                                                   PFID_STRIPE_IDX_BITS);
4209                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4210                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4211                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4212                 if (lma->lma_compat & LMAC_COMP_INFO) {
4213                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4214                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4215                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4216                 } else {
4217                         ol->ol_comp_start = 0;
4218                         ol->ol_comp_end = 0;
4219                         ol->ol_comp_id = 0;
4220                 }
4221         } else {
4222                 rc = __osd_xattr_get(inode, dentry, name,
4223                                      buf->lb_buf, buf->lb_len);
4224         }
4225
4226         if (cache_xattr) {
4227                 if (rc == -ENOENT || rc == -ENODATA)
4228                         osd_oxc_add(obj, name, NULL, 0);
4229                 else if (rc > 0 && buf->lb_buf != NULL)
4230                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4231         }
4232
4233         return rc;
4234 }
4235
4236 static int osd_declare_xattr_set(const struct lu_env *env,
4237                                  struct dt_object *dt,
4238                                  const struct lu_buf *buf, const char *name,
4239                                  int fl, struct thandle *handle)
4240 {
4241         struct osd_thandle *oh;
4242         int credits = 0;
4243         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4244
4245         LASSERT(handle != NULL);
4246
4247         oh = container_of0(handle, struct osd_thandle, ot_super);
4248         LASSERT(oh->ot_handle == NULL);
4249
4250         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4251                 /*
4252                  * For non-upgrading case, the LMA is set first and
4253                  * usually fit inode. But for upgrade case, the LMA
4254                  * may be in another separated EA block.
4255                  */
4256                 if (dt_object_exists(dt)) {
4257                         if (fl == LU_XATTR_REPLACE)
4258                                 credits = 1;
4259                         else
4260                                 goto upgrade;
4261                 }
4262         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4263                 credits = 1;
4264         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4265                 /* We may need to delete the old PFID EA. */
4266                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4267                 if (fl == LU_XATTR_REPLACE)
4268                         credits += 1;
4269                 else
4270                         goto upgrade;
4271         } else {
4272                 /*
4273                  * If some name entry resides on remote MDT, then will create
4274                  * agent entry under remote parent. On the other hand, if the
4275                  * remote entry will be removed, then related agent entry may
4276                  * need to be removed from the remote parent. So there may be
4277                  * kinds of cases, let's declare enough credits. The credits
4278                  * for create agent entry is enough for remove case.
4279                  */
4280                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4281                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4282                         if (dt_object_exists(dt))
4283                                 credits += 1; /* For updating LMA */
4284                 }
4285
4286 upgrade:
4287                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4288
4289                 if (buf != NULL) {
4290                         ssize_t buflen;
4291
4292                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4293                                 /*
4294                                  * learn xattr size from osd_xattr_get if
4295                                  * attribute has not been read yet
4296                                  */
4297                                 buflen = __osd_xattr_get(
4298                                     osd_dt_obj(dt)->oo_inode,
4299                                     &osd_oti_get(env)->oti_obj_dentry,
4300                                     name, NULL, 0);
4301                                 if (buflen < 0)
4302                                         buflen = 0;
4303                         } else {
4304                                 buflen = buf->lb_len;
4305                         }
4306
4307                         if (buflen > sb->s_blocksize) {
4308                                 credits += osd_calc_bkmap_credits(
4309                                     sb, NULL, 0, -1,
4310                                     (buflen + sb->s_blocksize - 1) >>
4311                                     sb->s_blocksize_bits);
4312                         }
4313                 }
4314                 /*
4315                  * xattr set may involve inode quota change, reserve credits for
4316                  * dquot_initialize()
4317                  */
4318                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4319         }
4320
4321         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4322
4323         return 0;
4324 }
4325
4326 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4327                               const struct lu_buf *buf, int fl,
4328                               struct thandle *handle)
4329 {
4330         struct osd_thread_info *info = osd_oti_get(env);
4331         struct dentry *dentry = &info->oti_obj_dentry;
4332         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4333         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4334         struct inode *inode = obj->oo_inode;
4335         struct filter_fid *ff = buf->lb_buf;
4336         struct ost_layout *ol = &ff->ff_layout;
4337         int flags = XATTR_REPLACE;
4338         int rc;
4339
4340         ENTRY;
4341
4342         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4343                 RETURN(-EINVAL);
4344
4345         rc = osd_get_lma(info, inode, dentry, loa);
4346         if (rc == -ENODATA) {
4347                 /* Usually for upgarding from old device */
4348                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4349                                 LMAC_FID_ON_OST, 0);
4350                 flags = XATTR_CREATE;
4351         } else if (rc) {
4352                 RETURN(rc);
4353         }
4354
4355         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4356                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4357                         RETURN(-EEXIST);
4358
4359                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4360                         /* Separate PFID EA from LMA */
4361                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4362                         lustre_lma_swab(lma);
4363                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4364                                              sizeof(*lma), XATTR_REPLACE);
4365                         if (!rc) {
4366                                 obj->oo_pfid_in_lma = 0;
4367                                 rc = LU_XATTR_CREATE;
4368                         }
4369
4370                         RETURN(rc);
4371                 }
4372         } else {
4373                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4374                         RETURN(fl);
4375
4376                 /*
4377                  * Old client does not send stripe information,
4378                  * then store the PFID EA on disk separatedly.
4379                  */
4380                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4381                              ol->ol_stripe_size == 0))
4382                         RETURN(fl);
4383
4384                 /* Remove old PFID EA entry firstly. */
4385                 ll_vfs_dq_init(inode);
4386                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4387                 if (rc == -ENODATA) {
4388                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4389                                 RETURN(rc);
4390                 } else if (rc) {
4391                         RETURN(rc);
4392                 }
4393         }
4394
4395         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4396         if (likely(ol->ol_stripe_size != 0)) {
4397                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4398                                              PFID_STRIPE_IDX_BITS;
4399                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4400                 lma->lma_compat |= LMAC_STRIPE_INFO;
4401                 if (ol->ol_comp_id != 0) {
4402                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4403                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4404                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4405                         lma->lma_compat |= LMAC_COMP_INFO;
4406                 }
4407         }
4408
4409         lustre_loa_swab(loa, false);
4410
4411         /* Store the PFID EA inside LMA. */
4412         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4413                              flags);
4414         if (!rc)
4415                 obj->oo_pfid_in_lma = 1;
4416
4417         RETURN(rc);
4418 }
4419
4420 /*
4421  * In DNE environment, the object (in spite of regular file or directory)
4422  * and its name entry may reside on different MDTs. Under such case, we will
4423  * create an agent entry on the MDT where the object resides. The agent entry
4424  * references the object locally, that makes the object to be visible to the
4425  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4426  * such as 'tar' can handle the object properly.
4427  *
4428  * We handle the agent entry during set linkEA that is the common interface
4429  * for both regular file and directroy, can handle kinds of cases, such as
4430  * create/link/unlink/rename, and so on.
4431  *
4432  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4433  *
4434  * XXX: There are two known issues:
4435  * 1. For one object, we will create at most one agent entry even if there
4436  *    may be more than one cross-MDTs hard links on the object. So the local
4437  *    e2fsck may claim that the object's nlink is larger than the name entries
4438  *    that reference such inode. And in further, the e2fsck will fix the nlink
4439  *    attribute to match the local references. Then it will cause the object's
4440  *    nlink attribute to be inconsistent with the global references. it is bad
4441  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4442  *    other hand, the global namespace LFSCK can repair the object's attribute
4443  *    according to the linkEA.
4444  * 2. There may be too many hard links on the object as to its linkEA overflow,
4445  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4446  *    case happened, then at this point, we do not know whether there are some
4447  *    cross-MDTs reference. But there are local references, it guarantees that
4448  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4449  */
4450 static int osd_xattr_handle_linkea(const struct lu_env *env,
4451                                    struct osd_device *osd,
4452                                    struct osd_object *obj,
4453                                    const struct lu_buf *buf,
4454                                    struct thandle *handle)
4455 {
4456         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4457         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4458         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4459         struct lu_name tmpname;
4460         struct osd_thandle *oh;
4461         int rc;
4462         bool remote = false;
4463
4464         ENTRY;
4465
4466         oh = container_of0(handle, struct osd_thandle, ot_super);
4467         LASSERT(oh->ot_handle != NULL);
4468
4469         rc = linkea_init_with_rec(&ldata);
4470         if (!rc) {
4471                 linkea_first_entry(&ldata);
4472                 while (ldata.ld_lee != NULL && !remote) {
4473                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4474                                             &tmpname, tfid);
4475                         if (osd_remote_fid(env, osd, tfid) > 0)
4476                                 remote = true;
4477                         else
4478                                 linkea_next_entry(&ldata);
4479                 }
4480         } else if (rc == -ENODATA) {
4481                 rc = 0;
4482         } else {
4483                 RETURN(rc);
4484         }
4485
4486         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4487                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4488                 if (rc)
4489                         CERROR("%s: failed to remove agent entry for "DFID
4490                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4491         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4492                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4493                 if (rc)
4494                         CERROR("%s: failed to create agent entry for "DFID
4495                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4496         }
4497
4498         RETURN(rc);
4499 }
4500
4501 /*
4502  * Concurrency: @dt is write locked.
4503  */
4504 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4505                          const struct lu_buf *buf, const char *name, int fl,
4506                          struct thandle *handle)
4507 {
4508         struct osd_object *obj = osd_dt_obj(dt);
4509         struct osd_device *osd = osd_obj2dev(obj);
4510         struct inode *inode = obj->oo_inode;
4511         struct osd_thread_info *info = osd_oti_get(env);
4512         int fs_flags = 0;
4513         int len;
4514         int rc;
4515
4516         ENTRY;
4517
4518         LASSERT(handle);
4519         LASSERT(buf);
4520
4521         /* version set is not real XATTR */
4522         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4523                 dt_obj_version_t *version = buf->lb_buf;
4524
4525                 /*
4526                  * for version we are just using xattr API but change inode
4527                  * field instead
4528                  */
4529                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4530
4531                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4532                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4533
4534                 LDISKFS_I(inode)->i_fs_version = *version;
4535                 /*
4536                  * Version is set after all inode operations are finished,
4537                  * so we should mark it dirty here
4538                  */
4539                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4540
4541                 RETURN(0);
4542         }
4543
4544         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4545                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4546
4547         len = buf->lb_len;
4548         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4549
4550         /*
4551          * For the OST device with 256 bytes inode size by default,
4552          * the PFID EA will be stored together with LMA EA to avoid
4553          * performance trouble. Otherwise the PFID EA can be stored
4554          * independently. LU-8998
4555          */
4556         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4557             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4558                 LASSERT(buf->lb_buf);
4559
4560                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4561                 if (fl <= 0)
4562                         RETURN(fl);
4563         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4564                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4565                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4566
4567                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4568                 if (rc)
4569                         RETURN(rc);
4570
4571                 lma->lma_incompat |= LMAI_STRIPED;
4572                 lustre_lma_swab(lma);
4573                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4574                                      sizeof(*lma), XATTR_REPLACE);
4575                 if (rc != 0)
4576                         RETURN(rc);
4577         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4578                 LASSERT(!osd->od_is_ost);
4579
4580                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4581                 if (rc)
4582                         RETURN(rc);
4583         }
4584
4585         if (fl & LU_XATTR_REPLACE)
4586                 fs_flags |= XATTR_REPLACE;
4587
4588         if (fl & LU_XATTR_CREATE)
4589                 fs_flags |= XATTR_CREATE;
4590
4591         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4592         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4593
4594         if (rc == 0 &&
4595             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4596              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4597                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4598
4599         return rc;
4600 }
4601
4602 /*
4603  * Concurrency: @dt is read locked.
4604  */
4605 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4606                           const struct lu_buf *buf)
4607 {
4608         struct osd_object *obj = osd_dt_obj(dt);
4609         struct inode *inode = obj->oo_inode;
4610         struct osd_thread_info *info = osd_oti_get(env);
4611         struct dentry *dentry = &info->oti_obj_dentry;
4612
4613         if (!dt_object_exists(dt))
4614                 return -ENOENT;
4615
4616         LASSERT(!dt_object_remote(dt));
4617         LASSERT(inode->i_op != NULL);
4618         LASSERT(inode->i_op->listxattr != NULL);
4619
4620         dentry->d_inode = inode;
4621         dentry->d_sb = inode->i_sb;
4622         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4623 }
4624
4625 static int osd_declare_xattr_del(const struct lu_env *env,
4626                                  struct dt_object *dt, const char *name,
4627                                  struct thandle *handle)
4628 {
4629         struct osd_thandle *oh;
4630         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4631
4632         LASSERT(!dt_object_remote(dt));
4633         LASSERT(handle != NULL);
4634
4635         oh = container_of0(handle, struct osd_thandle, ot_super);
4636         LASSERT(oh->ot_handle == NULL);
4637
4638         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4639                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4640         /*
4641          * xattr del may involve inode quota change, reserve credits for
4642          * dquot_initialize()
4643          */
4644         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4645
4646         return 0;
4647 }
4648
4649 /*
4650  * Concurrency: @dt is write locked.
4651  */
4652 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4653                          const char *name, struct thandle *handle)
4654 {
4655         struct osd_object *obj = osd_dt_obj(dt);
4656         struct inode *inode = obj->oo_inode;
4657         struct osd_thread_info *info = osd_oti_get(env);
4658         struct dentry *dentry = &info->oti_obj_dentry;
4659         int rc;
4660
4661         if (!dt_object_exists(dt))
4662                 return -ENOENT;
4663
4664         LASSERT(!dt_object_remote(dt));
4665         LASSERT(inode->i_op != NULL);
4666         LASSERT(handle != NULL);
4667 #ifdef HAVE_IOP_XATTR
4668         LASSERT(inode->i_op->removexattr != NULL);
4669 #endif
4670
4671         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4672
4673         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4674                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4675
4676                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4677                                  &info->oti_ost_attrs);
4678                 if (!rc) {
4679                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4680
4681                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4682                         lustre_lma_swab(lma);
4683                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4684                                              sizeof(*lma), XATTR_REPLACE);
4685                         if (!rc)
4686                                 obj->oo_pfid_in_lma = 0;
4687                 }
4688         } else {
4689                 ll_vfs_dq_init(inode);
4690                 dentry->d_inode = inode;
4691                 dentry->d_sb = inode->i_sb;
4692                 rc = osd_removexattr(dentry, inode, name);
4693         }
4694
4695         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4696
4697         if (rc == 0 &&
4698             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4699              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4700                 osd_oxc_del(obj, name);
4701
4702         return rc;
4703 }
4704
4705 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4706                            __u64 start, __u64 end)
4707 {
4708         struct osd_object *obj = osd_dt_obj(dt);
4709         struct inode *inode = obj->oo_inode;
4710         struct osd_thread_info *info = osd_oti_get(env);
4711         struct dentry *dentry = &info->oti_obj_dentry;
4712         struct file *file = &info->oti_file;
4713         int rc;
4714
4715         ENTRY;
4716
4717         dentry->d_inode = inode;
4718         dentry->d_sb = inode->i_sb;
4719         file->f_path.dentry = dentry;
4720         file->f_mapping = inode->i_mapping;
4721         file->f_op = inode->i_fop;
4722         set_file_inode(file, inode);
4723
4724         rc = ll_vfs_fsync_range(file, start, end, 0);
4725
4726         RETURN(rc);
4727 }
4728
4729 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4730 {
4731         return 0;
4732 }
4733
4734 /*
4735  * Index operations.
4736  */
4737
4738 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4739                                const struct dt_index_features *feat)
4740 {
4741         struct iam_descr *descr;
4742
4743         if (osd_object_is_root(o))
4744                 return feat == &dt_directory_features;
4745
4746         LASSERT(o->oo_dir != NULL);
4747
4748         descr = o->oo_dir->od_container.ic_descr;
4749         if (feat == &dt_directory_features) {
4750                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4751                         return 1;
4752                 else
4753                         return 0;
4754         } else {
4755                 return feat->dif_keysize_min <= descr->id_key_size &&
4756                        descr->id_key_size <= feat->dif_keysize_max &&
4757                        feat->dif_recsize_min <= descr->id_rec_size &&
4758                        descr->id_rec_size <= feat->dif_recsize_max &&
4759                        !(feat->dif_flags & (DT_IND_VARKEY |
4760                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4761                        ergo(feat->dif_flags & DT_IND_UPDATE,
4762                             1 /* XXX check that object (and fs) is writable */);
4763         }
4764 }
4765
4766 static int osd_iam_container_init(const struct lu_env *env,
4767                                   struct osd_object *obj,
4768                                   struct osd_directory *dir)
4769 {
4770         struct iam_container *bag = &dir->od_container;
4771         int result;
4772
4773         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4774         if (result != 0)
4775                 return result;
4776
4777         result = iam_container_setup(bag);
4778         if (result == 0)
4779                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4780         else
4781                 iam_container_fini(bag);
4782
4783         return result;
4784 }
4785
4786
4787 /*
4788  * Concurrency: no external locking is necessary.
4789  */
4790 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4791                          const struct dt_index_features *feat)
4792 {
4793         int result;
4794         int skip_iam = 0;
4795         struct osd_object *obj = osd_dt_obj(dt);
4796
4797         LINVRNT(osd_invariant(obj));
4798
4799         if (osd_object_is_root(obj)) {
4800                 dt->do_index_ops = &osd_index_ea_ops;
4801                 result = 0;
4802         } else if (feat == &dt_directory_features) {
4803                 dt->do_index_ops = &osd_index_ea_ops;
4804                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4805                         result = 0;
4806                 else
4807                         result = -ENOTDIR;
4808                 skip_iam = 1;
4809         } else if (unlikely(feat == &dt_otable_features)) {
4810                 dt->do_index_ops = &osd_otable_ops;
4811                 return 0;
4812         } else if (unlikely(feat == &dt_acct_features)) {
4813                 dt->do_index_ops = &osd_acct_index_ops;
4814                 result = 0;
4815                 skip_iam = 1;
4816         } else if (!osd_has_index(obj)) {
4817                 struct osd_directory *dir;
4818                 struct osd_device *osd = osd_obj2dev(obj);
4819                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4820
4821                 OBD_ALLOC_PTR(dir);
4822                 if (dir) {
4823
4824                         spin_lock(&obj->oo_guard);
4825                         if (obj->oo_dir == NULL)
4826                                 obj->oo_dir = dir;
4827                         else
4828                                 /*
4829                                  * Concurrent thread allocated container data.
4830                                  */
4831                                 OBD_FREE_PTR(dir);
4832                         spin_unlock(&obj->oo_guard);
4833                         /*
4834                          * Now, that we have container data, serialize its
4835                          * initialization.
4836                          */
4837                         down_write(&obj->oo_ext_idx_sem);
4838                         /*
4839                          * recheck under lock.
4840                          */
4841
4842                         if (osd_has_index(obj)) {
4843                                 result = 0;
4844                                 goto unlock;
4845                         }
4846
4847                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4848                         if (result || feat == &dt_lfsck_namespace_features ||
4849                             feat == &dt_lfsck_layout_orphan_features ||
4850                             feat == &dt_lfsck_layout_dangling_features)
4851                                 goto unlock;
4852
4853                         result = osd_index_register(osd, fid,
4854                                                     feat->dif_keysize_max,
4855                                                     feat->dif_recsize_max);
4856                         if (result < 0)
4857                                 CWARN("%s: failed to register index "
4858                                       DFID": rc = %d\n",
4859                                       osd_name(osd), PFID(fid), result);
4860                         else if (result > 0)
4861                                 result = 0;
4862                         else
4863                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4864                                        " (%d/%d) registered\n",
4865                                        osd_name(osd), PFID(fid),
4866                                        (int)feat->dif_keysize_max,
4867                                        (int)feat->dif_recsize_max);
4868
4869 unlock:
4870                         up_write(&obj->oo_ext_idx_sem);
4871                 } else {
4872                         result = -ENOMEM;
4873                 }
4874         } else {
4875                 result = 0;
4876         }
4877
4878         if (result == 0 && skip_iam == 0) {
4879                 if (!osd_iam_index_probe(env, obj, feat))
4880                         result = -ENOTDIR;
4881         }
4882         LINVRNT(osd_invariant(obj));
4883
4884         return result;
4885 }
4886
4887 static int osd_otable_it_attr_get(const struct lu_env *env,
4888                                  struct dt_object *dt,
4889                                  struct lu_attr *attr)
4890 {
4891         attr->la_valid = 0;
4892         return 0;
4893 }
4894
4895 static const struct dt_object_operations osd_obj_ops = {
4896         .do_read_lock           = osd_read_lock,
4897         .do_write_lock          = osd_write_lock,
4898         .do_read_unlock         = osd_read_unlock,
4899         .do_write_unlock        = osd_write_unlock,
4900         .do_write_locked        = osd_write_locked,
4901         .do_attr_get            = osd_attr_get,
4902         .do_declare_attr_set    = osd_declare_attr_set,
4903         .do_attr_set            = osd_attr_set,
4904         .do_ah_init             = osd_ah_init,
4905         .do_declare_create      = osd_declare_create,
4906         .do_create              = osd_create,
4907         .do_declare_destroy     = osd_declare_destroy,
4908         .do_destroy             = osd_destroy,
4909         .do_index_try           = osd_index_try,
4910         .do_declare_ref_add     = osd_declare_ref_add,
4911         .do_ref_add             = osd_ref_add,
4912         .do_declare_ref_del     = osd_declare_ref_del,
4913         .do_ref_del             = osd_ref_del,
4914         .do_xattr_get           = osd_xattr_get,
4915         .do_declare_xattr_set   = osd_declare_xattr_set,
4916         .do_xattr_set           = osd_xattr_set,
4917         .do_declare_xattr_del   = osd_declare_xattr_del,
4918         .do_xattr_del           = osd_xattr_del,
4919         .do_xattr_list          = osd_xattr_list,
4920         .do_object_sync         = osd_object_sync,
4921         .do_invalidate          = osd_invalidate,
4922 };
4923
4924 static const struct dt_object_operations osd_obj_otable_it_ops = {
4925         .do_attr_get    = osd_otable_it_attr_get,
4926         .do_index_try   = osd_index_try,
4927 };
4928
4929 static int osd_index_declare_iam_delete(const struct lu_env *env,
4930                                         struct dt_object *dt,
4931                                         const struct dt_key *key,
4932                                         struct thandle *handle)
4933 {
4934         struct osd_thandle *oh;
4935
4936         oh = container_of0(handle, struct osd_thandle, ot_super);
4937         LASSERT(oh->ot_handle == NULL);
4938
4939         /* Recycle  may cause additional three blocks to be changed. */
4940         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4941                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4942
4943         return 0;
4944 }
4945
4946 /**
4947  *      delete a (key, value) pair from index \a dt specified by \a key
4948  *
4949  *      \param  dt      osd index object
4950  *      \param  key     key for index
4951  *      \param  rec     record reference
4952  *      \param  handle  transaction handler
4953  *
4954  *      \retval  0  success
4955  *      \retval -ve   failure
4956  */
4957 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4958                                 const struct dt_key *key,
4959                                 struct thandle *handle)
4960 {
4961         struct osd_thread_info *oti = osd_oti_get(env);
4962         struct osd_object *obj = osd_dt_obj(dt);
4963         struct osd_thandle *oh;
4964         struct iam_path_descr *ipd;
4965         struct iam_container *bag = &obj->oo_dir->od_container;
4966         int rc;
4967
4968         ENTRY;
4969
4970         if (!dt_object_exists(dt))
4971                 RETURN(-ENOENT);
4972
4973         LINVRNT(osd_invariant(obj));
4974         LASSERT(!dt_object_remote(dt));
4975         LASSERT(bag->ic_object == obj->oo_inode);
4976         LASSERT(handle != NULL);
4977
4978         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4979
4980         ipd = osd_idx_ipd_get(env, bag);
4981         if (unlikely(ipd == NULL))
4982                 RETURN(-ENOMEM);
4983
4984         oh = container_of0(handle, struct osd_thandle, ot_super);
4985         LASSERT(oh->ot_handle != NULL);
4986         LASSERT(oh->ot_handle->h_transaction != NULL);
4987
4988         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4989                 /* swab quota uid/gid provided by caller */
4990                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4991                 key = (const struct dt_key *)&oti->oti_quota_id;
4992         }
4993
4994         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4995         osd_ipd_put(env, bag, ipd);
4996         LINVRNT(osd_invariant(obj));
4997         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4998         RETURN(rc);
4999 }
5000
5001 static int osd_index_declare_ea_delete(const struct lu_env *env,
5002                                        struct dt_object *dt,
5003                                        const struct dt_key *key,
5004                                        struct thandle *handle)
5005 {
5006         struct osd_thandle *oh;
5007         struct inode *inode;
5008         int rc, credits;
5009
5010         ENTRY;
5011
5012         LASSERT(!dt_object_remote(dt));
5013         LASSERT(handle != NULL);
5014
5015         oh = container_of0(handle, struct osd_thandle, ot_super);
5016         LASSERT(oh->ot_handle == NULL);
5017
5018         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5019         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5020
5021         inode = osd_dt_obj(dt)->oo_inode;
5022         if (inode == NULL)
5023                 RETURN(-ENOENT);
5024
5025         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5026                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5027                                    NULL, OSD_QID_BLK);
5028         RETURN(rc);
5029 }
5030
5031 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5032                                           struct dt_rec *fid)
5033 {
5034         struct osd_fid_pack *rec;
5035         int rc = -ENODATA;
5036
5037         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5038                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5039                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5040                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5041                         rc = -EINVAL;
5042         }
5043         return rc;
5044 }
5045
5046 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5047                           const struct lu_fid *fid)
5048 {
5049         struct seq_server_site *ss = osd_seq_site(osd);
5050
5051         ENTRY;
5052
5053         /* FID seqs not in FLDB, must be local seq */
5054         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5055                 RETURN(0);
5056
5057         /*
5058          * If FLD is not being initialized yet, it only happens during the
5059          * initialization, likely during mgs initialization, and we assume
5060          * this is local FID.
5061          */
5062         if (ss == NULL || ss->ss_server_fld == NULL)
5063                 RETURN(0);
5064
5065         /* Only check the local FLDB here */
5066         if (osd_seq_exists(env, osd, fid_seq(fid)))
5067                 RETURN(0);
5068
5069         RETURN(1);
5070 }
5071
5072 static void osd_take_care_of_agent(const struct lu_env *env,
5073                                    struct osd_device *osd,
5074                                    struct osd_thandle *oh,
5075                                    struct ldiskfs_dir_entry_2 *de)
5076 {
5077         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5078         struct osd_idmap_cache *idc;
5079         int rc, schedule = 0;
5080
5081         LASSERT(de != NULL);
5082
5083         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5084         if (likely(rc == 0)) {
5085                 idc = osd_idc_find_or_init(env, osd, fid);
5086                 if (IS_ERR(idc) || idc->oic_remote)
5087                         schedule = 1;
5088         } else if (rc == -ENODATA) {
5089                 /*
5090                  * can't get FID, postpone to the end of the
5091                  * transaction when iget() is safe
5092                  */
5093                 schedule = 1;
5094         } else {
5095                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5096         }
5097         if (schedule)
5098                 osd_schedule_agent_inode_removal(env, oh,
5099                                                  le32_to_cpu(de->inode));
5100 }
5101
5102 /**
5103  * Index delete function for interoperability mode (b11826).
5104  * It will remove the directory entry added by osd_index_ea_insert().
5105  * This entry is needed to maintain name->fid mapping.
5106  *
5107  * \param key,  key i.e. file entry to be deleted
5108  *
5109  * \retval   0, on success
5110  * \retval -ve, on error
5111  */
5112 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5113                                const struct dt_key *key, struct thandle *handle)
5114 {
5115         struct osd_object *obj = osd_dt_obj(dt);
5116         struct inode *dir = obj->oo_inode;
5117         struct dentry *dentry;
5118         struct osd_thandle *oh;
5119         struct ldiskfs_dir_entry_2 *de = NULL;
5120         struct buffer_head *bh;
5121         struct htree_lock *hlock = NULL;
5122         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5123         int rc;
5124
5125         ENTRY;
5126
5127         if (!dt_object_exists(dt))
5128                 RETURN(-ENOENT);
5129
5130         LINVRNT(osd_invariant(obj));
5131         LASSERT(!dt_object_remote(dt));
5132         LASSERT(handle != NULL);
5133
5134         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5135
5136         oh = container_of(handle, struct osd_thandle, ot_super);
5137         LASSERT(oh->ot_handle != NULL);
5138         LASSERT(oh->ot_handle->h_transaction != NULL);
5139
5140         ll_vfs_dq_init(dir);
5141         dentry = osd_child_dentry_get(env, obj,
5142                                       (char *)key, strlen((char *)key));
5143
5144         if (obj->oo_hl_head != NULL) {
5145                 hlock = osd_oti_get(env)->oti_hlock;
5146                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5147                                    dir, LDISKFS_HLOCK_DEL);
5148         } else {
5149                 down_write(&obj->oo_ext_idx_sem);
5150         }
5151
5152         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5153         if (!IS_ERR(bh)) {
5154                 /*
5155                  * If this is not the ".." entry, it might be a remote DNE
5156                  * entry and  we need to check if the FID is for a remote
5157                  * MDT.  If the FID is  not in the directory entry (e.g.
5158                  * upgraded 1.8 filesystem without dirdata enabled) then
5159                  * we need to get the FID from the LMA. For a remote directory
5160                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5161                  * case.
5162                  *
5163                  * Delete the entry before the agent inode in order to
5164                  * simplify error handling.  At worst an error after deleting
5165                  * the entry first might leak the agent inode afterward. The
5166                  * reverse would need filesystem abort in case of error deleting
5167                  * the entry after the agent had been removed, or leave a
5168                  * dangling entry pointing at a random inode.
5169                  */
5170                 if (strcmp((char *)key, dotdot) != 0)
5171                         osd_take_care_of_agent(env, osd, oh, de);
5172                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5173                 brelse(bh);
5174         } else {
5175                 rc = PTR_ERR(bh);
5176         }
5177         if (hlock != NULL)
5178                 ldiskfs_htree_unlock(hlock);
5179         else
5180                 up_write(&obj->oo_ext_idx_sem);
5181
5182         GOTO(out, rc);
5183 out:
5184         LASSERT(osd_invariant(obj));
5185         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5186         RETURN(rc);
5187 }
5188
5189 /**
5190  *      Lookup index for \a key and copy record to \a rec.
5191  *
5192  *      \param  dt      osd index object
5193  *      \param  key     key for index
5194  *      \param  rec     record reference
5195  *
5196  *      \retval  +ve  success : exact mach
5197  *      \retval  0    return record with key not greater than \a key
5198  *      \retval -ve   failure
5199  */
5200 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5201                                 struct dt_rec *rec, const struct dt_key *key)
5202 {
5203         struct osd_object *obj = osd_dt_obj(dt);
5204         struct iam_path_descr *ipd;
5205         struct iam_container *bag = &obj->oo_dir->od_container;
5206         struct osd_thread_info *oti = osd_oti_get(env);
5207         struct iam_iterator *it = &oti->oti_idx_it;
5208         struct iam_rec *iam_rec;
5209         int rc;
5210
5211         ENTRY;
5212
5213         if (!dt_object_exists(dt))
5214                 RETURN(-ENOENT);
5215
5216         LASSERT(osd_invariant(obj));
5217         LASSERT(!dt_object_remote(dt));
5218         LASSERT(bag->ic_object == obj->oo_inode);
5219
5220         ipd = osd_idx_ipd_get(env, bag);
5221         if (IS_ERR(ipd))
5222                 RETURN(-ENOMEM);
5223
5224         /* got ipd now we can start iterator. */
5225         iam_it_init(it, bag, 0, ipd);
5226
5227         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5228                 /* swab quota uid/gid provided by caller */
5229                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5230                 key = (const struct dt_key *)&oti->oti_quota_id;
5231         }
5232
5233         rc = iam_it_get(it, (struct iam_key *)key);
5234         if (rc >= 0) {
5235                 if (S_ISDIR(obj->oo_inode->i_mode))
5236                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5237                 else
5238                         iam_rec = (struct iam_rec *)rec;
5239
5240                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5241
5242                 if (S_ISDIR(obj->oo_inode->i_mode))
5243                         osd_fid_unpack((struct lu_fid *)rec,
5244                                        (struct osd_fid_pack *)iam_rec);
5245                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5246                         osd_quota_unpack(obj, rec);
5247         }
5248
5249         iam_it_put(it);
5250         iam_it_fini(it);
5251         osd_ipd_put(env, bag, ipd);
5252
5253         LINVRNT(osd_invariant(obj));
5254
5255         RETURN(rc);
5256 }
5257
5258 static int osd_index_declare_iam_insert(const struct lu_env *env,
5259                                         struct dt_object *dt,
5260                                         const struct dt_rec *rec,
5261                                         const struct dt_key *key,
5262                                         struct thandle *handle)
5263 {
5264         struct osd_thandle *oh;
5265
5266         LASSERT(handle != NULL);
5267
5268         oh = container_of0(handle, struct osd_thandle, ot_super);
5269         LASSERT(oh->ot_handle == NULL);
5270
5271         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5272                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5273
5274         return 0;
5275 }
5276
5277 /**
5278  *      Inserts (key, value) pair in \a dt index object.
5279  *
5280  *      \param  dt      osd index object
5281  *      \param  key     key for index
5282  *      \param  rec     record reference
5283  *      \param  th      transaction handler
5284  *
5285  *      \retval  0  success
5286  *      \retval -ve failure
5287  */
5288 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5289                                 const struct dt_rec *rec,
5290                                 const struct dt_key *key, struct thandle *th)
5291 {
5292         struct osd_object *obj = osd_dt_obj(dt);
5293         struct iam_path_descr *ipd;
5294         struct osd_thandle *oh;
5295         struct iam_container *bag;
5296         struct osd_thread_info *oti = osd_oti_get(env);
5297         struct iam_rec *iam_rec;
5298         int rc;
5299
5300         ENTRY;
5301
5302         if (!dt_object_exists(dt))
5303                 RETURN(-ENOENT);
5304
5305         LINVRNT(osd_invariant(obj));
5306         LASSERT(!dt_object_remote(dt));
5307
5308         bag = &obj->oo_dir->od_container;
5309         LASSERT(bag->ic_object == obj->oo_inode);
5310         LASSERT(th != NULL);
5311
5312         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5313
5314         ipd = osd_idx_ipd_get(env, bag);
5315         if (unlikely(ipd == NULL))
5316                 RETURN(-ENOMEM);
5317
5318         oh = container_of0(th, struct osd_thandle, ot_super);
5319         LASSERT(oh->ot_handle != NULL);
5320         LASSERT(oh->ot_handle->h_transaction != NULL);
5321         if (S_ISDIR(obj->oo_inode->i_mode)) {
5322                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5323                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5324                              &oti->oti_fid);
5325         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5326                 /* pack quota uid/gid */
5327                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5328                 key = (const struct dt_key *)&oti->oti_quota_id;
5329                 /* pack quota record */
5330                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5331                 iam_rec = (struct iam_rec *)rec;
5332         } else {
5333                 iam_rec = (struct iam_rec *)rec;
5334         }
5335
5336         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5337                         iam_rec, ipd);
5338         osd_ipd_put(env, bag, ipd);
5339         LINVRNT(osd_invariant(obj));
5340         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5341         RETURN(rc);
5342 }
5343
5344 /**
5345  * Calls ldiskfs_add_entry() to add directory entry
5346  * into the directory. This is required for
5347  * interoperability mode (b11826)
5348  *
5349  * \retval   0, on success
5350  * \retval -ve, on error
5351  */
5352 static int __osd_ea_add_rec(struct osd_thread_info *info,
5353                             struct osd_object *pobj, struct inode  *cinode,
5354                             const char *name, const struct lu_fid *fid,
5355                             struct htree_lock *hlock, struct thandle *th)
5356 {
5357         struct ldiskfs_dentry_param *ldp;
5358         struct dentry *child;
5359         struct osd_thandle *oth;
5360         int rc;
5361
5362         oth = container_of(th, struct osd_thandle, ot_super);
5363         LASSERT(oth->ot_handle != NULL);
5364         LASSERT(oth->ot_handle->h_transaction != NULL);
5365         LASSERT(pobj->oo_inode);
5366
5367         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5368         if (unlikely(pobj->oo_inode ==
5369                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5370                 ldp->edp_magic = 0;
5371         else
5372                 osd_get_ldiskfs_dirent_param(ldp, fid);
5373         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5374         child->d_fsdata = (void *)ldp;
5375         ll_vfs_dq_init(pobj->oo_inode);
5376         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5377                                    child, cinode, hlock);
5378         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5379                 struct ldiskfs_dir_entry_2      *de;
5380                 struct buffer_head              *bh;
5381                 int                              rc1;
5382
5383                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5384                                             NULL, hlock);
5385                 if (!IS_ERR(bh)) {
5386                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5387                                                                bh);
5388                         if (rc1 == 0) {
5389                                 if (S_ISDIR(cinode->i_mode))
5390                                         de->file_type = LDISKFS_DIRENT_LUFID |
5391                                                         LDISKFS_FT_REG_FILE;
5392                                 else
5393                                         de->file_type = LDISKFS_DIRENT_LUFID |
5394                                                         LDISKFS_FT_DIR;
5395                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5396                                                               NULL, bh);
5397                         }
5398                         brelse(bh);
5399                 }
5400         }
5401
5402         RETURN(rc);
5403 }
5404
5405 /**
5406  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5407  * into the directory.Also sets flags into osd object to
5408  * indicate dot and dotdot are created. This is required for
5409  * interoperability mode (b11826)
5410  *
5411  * \param dir   directory for dot and dotdot fixup.
5412  * \param obj   child object for linking
5413  *
5414  * \retval   0, on success
5415  * \retval -ve, on error
5416  */
5417 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5418                               struct osd_object *dir,
5419                               struct inode *parent_dir, const char *name,
5420                               const struct lu_fid *dot_fid,
5421                               const struct lu_fid *dot_dot_fid,
5422                               struct thandle *th)
5423 {
5424         struct inode *inode = dir->oo_inode;
5425         struct osd_thandle *oth;
5426         int result = 0;
5427
5428         oth = container_of(th, struct osd_thandle, ot_super);
5429         LASSERT(oth->ot_handle->h_transaction != NULL);
5430         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5431
5432         if (strcmp(name, dot) == 0) {
5433                 if (dir->oo_compat_dot_created) {
5434                         result = -EEXIST;
5435                 } else {
5436                         LASSERT(inode->i_ino == parent_dir->i_ino);
5437                         dir->oo_compat_dot_created = 1;
5438                         result = 0;
5439                 }
5440         } else if (strcmp(name, dotdot) == 0) {
5441                 if (!dir->oo_compat_dot_created)
5442                         return -EINVAL;
5443                 /* in case of rename, dotdot is already created */
5444                 if (dir->oo_compat_dotdot_created) {
5445                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5446                                                 dot_dot_fid, NULL, th);
5447                 }
5448
5449                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5450                         struct lu_fid tfid = *dot_dot_fid;
5451
5452                         tfid.f_oid--;
5453                         result = osd_add_dot_dotdot_internal(info,
5454                                         dir->oo_inode, parent_dir, dot_fid,
5455                                         &tfid, oth);
5456                 } else {
5457                         result = osd_add_dot_dotdot_internal(info,
5458                                         dir->oo_inode, parent_dir, dot_fid,
5459                                         dot_dot_fid, oth);
5460                 }
5461
5462                 if (result == 0)
5463                         dir->oo_compat_dotdot_created = 1;
5464         }
5465
5466         return result;
5467 }
5468
5469
5470 /**
5471  * It will call the appropriate osd_add* function and return the
5472  * value, return by respective functions.
5473  */
5474 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5475                           struct inode *cinode, const char *name,
5476                           const struct lu_fid *fid, struct thandle *th)
5477 {
5478         struct osd_thread_info *info = osd_oti_get(env);
5479         struct htree_lock *hlock;
5480         int rc;
5481
5482         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5483
5484         if (name[0] == '.' && (name[1] == '\0' ||
5485                                (name[1] == '.' && name[2] == '\0'))) {
5486                 if (hlock != NULL) {
5487                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5488                                            pobj->oo_inode, 0);
5489                 } else {
5490                         down_write(&pobj->oo_ext_idx_sem);
5491                 }
5492
5493                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5494                                         lu_object_fid(&pobj->oo_dt.do_lu),
5495                                         fid, th);
5496         } else {
5497                 if (hlock != NULL) {
5498                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5499                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5500                 } else {
5501                         down_write(&pobj->oo_ext_idx_sem);
5502                 }
5503
5504                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5505                         struct lu_fid *tfid = &info->oti_fid;
5506
5507                         *tfid = *fid;
5508                         tfid->f_ver = ~0;
5509                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5510                                               tfid, hlock, th);
5511                 } else {
5512                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5513                                               hlock, th);
5514                 }
5515         }
5516         if (hlock != NULL)
5517                 ldiskfs_htree_unlock(hlock);
5518         else
5519                 up_write(&pobj->oo_ext_idx_sem);
5520
5521         return rc;
5522 }
5523
5524 static int
5525 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5526                       struct osd_idmap_cache *oic)
5527 {
5528         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5529         struct lu_fid *fid = &oic->oic_fid;
5530         struct osd_inode_id *id = &oic->oic_lid;
5531         struct inode *inode = NULL;
5532         int once = 0;
5533         bool insert;
5534         int rc;
5535
5536         ENTRY;
5537
5538         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5539                 RETURN(0);
5540
5541         if (thread_is_running(&scrub->os_thread) &&
5542             scrub->os_pos_current > id->oii_ino)
5543                 RETURN(0);
5544
5545         if (dev->od_auto_scrub_interval == AS_NEVER ||
5546             ktime_get_real_seconds() <
5547             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5548                 RETURN(0);
5549
5550 again:
5551         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5552         if (rc == -ENOENT) {
5553                 __u32 gen = id->oii_gen;
5554
5555                 insert = true;
5556                 if (inode != NULL)
5557                         goto trigger;
5558
5559                 inode = osd_iget(oti, dev, id);
5560                 /* The inode has been removed (by race maybe). */
5561                 if (IS_ERR(inode)) {
5562                         rc = PTR_ERR(inode);
5563
5564                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5565                 }
5566
5567                 /* The OI mapping is lost. */
5568                 if (gen != OSD_OII_NOGEN)
5569                         goto trigger;
5570
5571                 /*
5572                  * The inode may has been reused by others, we do not know,
5573                  * leave it to be handled by subsequent osd_fid_lookup().
5574                  */
5575                 GOTO(out, rc = 0);
5576         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5577                 GOTO(out, rc);
5578         }
5579
5580         insert = false;
5581
5582 trigger:
5583         if (thread_is_running(&scrub->os_thread)) {
5584                 if (inode == NULL) {
5585                         inode = osd_iget(oti, dev, id);
5586                         /* The inode has been removed (by race maybe). */
5587                         if (IS_ERR(inode)) {
5588                                 rc = PTR_ERR(inode);
5589
5590                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5591                         }
5592                 }
5593
5594                 rc = osd_oii_insert(dev, oic, insert);
5595                 /*
5596                  * There is race condition between osd_oi_lookup and OI scrub.
5597                  * The OI scrub finished just after osd_oi_lookup() failure.
5598                  * Under such case, it is unnecessary to trigger OI scrub again,
5599                  * but try to call osd_oi_lookup() again.
5600                  */
5601                 if (unlikely(rc == -EAGAIN))
5602                         goto again;
5603
5604                 if (!S_ISDIR(inode->i_mode))
5605                         rc = 0;
5606                 else
5607                         rc = osd_check_lmv(oti, dev, inode, oic);
5608
5609                 GOTO(out, rc);
5610         }
5611
5612         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5613                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5614                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5615                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5616                        "%s: trigger partial OI scrub for RPC inconsistency "
5617                        "checking FID "DFID": rc = %d\n",
5618                        osd_dev2name(dev), PFID(fid), rc);
5619                 if (rc == 0 || rc == -EALREADY)
5620                         goto again;
5621         }
5622
5623         GOTO(out, rc);
5624
5625 out:
5626         if (inode)
5627                 iput(inode);
5628
5629         RETURN(rc);
5630 }
5631
5632 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5633                                struct osd_device *dev,
5634                                struct osd_idmap_cache *oic,
5635                                struct lu_fid *fid, __u32 ino)
5636 {
5637         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5638         struct inode *inode;
5639         int rc;
5640
5641         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5642         inode = osd_iget(oti, dev, &oic->oic_lid);
5643         if (IS_ERR(inode)) {
5644                 fid_zero(&oic->oic_fid);
5645                 return PTR_ERR(inode);
5646         }
5647
5648         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5649         iput(inode);
5650         if (rc != 0)
5651                 fid_zero(&oic->oic_fid);
5652         else
5653                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5654         return rc;
5655 }
5656
5657 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5658                       struct osd_inode_id *id, const struct lu_fid *fid)
5659 {
5660         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5661                id->oii_ino, id->oii_gen, info);
5662         info->oti_cache.oic_lid = *id;
5663         info->oti_cache.oic_fid = *fid;
5664         info->oti_cache.oic_dev = osd;
5665 }
5666
5667 /**
5668  * Get parent FID from the linkEA.
5669  *
5670  * For a directory which parent resides on remote MDT, to satisfy the
5671  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5672  * the other hand, to make the lookup(..) on the directory can return
5673  * the real parent FID, we append the real parent FID after its ".."
5674  * name entry in the /REMOTE_PARENT_DIR.
5675  *
5676  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5677  * backup. So after the restore, we cannot get the right parent FID from
5678  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5679  * we have stored the real parent FID in the directory object's linkEA,
5680  * we can parse the linkEA for the real parent FID.
5681  *
5682  * \param[in] env       pointer to the thread context
5683  * \param[in] obj       pointer to the object to be handled
5684  * \param[out]fid       pointer to the buffer to hold the parent FID
5685  *
5686  * \retval              0 for getting the real parent FID successfully
5687  * \retval              negative error number on failure
5688  */
5689 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5690                                     struct osd_object *obj,
5691                                     struct lu_fid *fid)
5692 {
5693         struct osd_thread_info *oti = osd_oti_get(env);
5694         struct lu_buf *buf = &oti->oti_big_buf;
5695         struct dentry *dentry = &oti->oti_obj_dentry;
5696         struct inode *inode = obj->oo_inode;
5697         struct linkea_data ldata = { NULL };
5698         int rc;
5699
5700         ENTRY;
5701
5702         fid_zero(fid);
5703         if (!S_ISDIR(inode->i_mode))
5704                 RETURN(-EIO);
5705
5706 again:
5707         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5708                              buf->lb_buf, buf->lb_len);
5709         if (rc == -ERANGE) {
5710                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5711                                      NULL, 0);
5712                 if (rc > 0) {
5713                         lu_buf_realloc(buf, rc);
5714                         if (buf->lb_buf == NULL)
5715                                 RETURN(-ENOMEM);
5716
5717                         goto again;
5718                 }
5719         }
5720
5721         if (unlikely(rc == 0))
5722                 RETURN(-ENODATA);
5723
5724         if (rc < 0)
5725                 RETURN(rc);
5726
5727         if (unlikely(buf->lb_buf == NULL)) {
5728                 lu_buf_realloc(buf, rc);
5729                 if (buf->lb_buf == NULL)
5730                         RETURN(-ENOMEM);
5731
5732                 goto again;
5733         }
5734
5735         ldata.ld_buf = buf;
5736         rc = linkea_init_with_rec(&ldata);
5737         if (!rc) {
5738                 linkea_first_entry(&ldata);
5739                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5740         }
5741
5742         RETURN(rc);
5743 }
5744
5745 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5746                                     struct inode *inode,
5747                                     const struct lu_fid *pfid,
5748                                     const char *name, const int namelen)
5749 {
5750         struct osd_thread_info *oti = osd_oti_get(env);
5751         struct lu_buf *buf = &oti->oti_big_buf;
5752         struct dentry *dentry = &oti->oti_obj_dentry;
5753         struct linkea_data ldata = { NULL };
5754         struct lu_name cname = { .ln_name = name,
5755                                  .ln_namelen = namelen };
5756         int rc;
5757
5758         ENTRY;
5759
5760 again:
5761         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5762                              buf->lb_buf, buf->lb_len);
5763         if (rc == -ERANGE)
5764                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5765
5766         if (rc < 0)
5767                 RETURN(rc);
5768
5769         if (unlikely(rc == 0))
5770                 RETURN(-ENODATA);
5771
5772         if (buf->lb_len < rc) {
5773                 lu_buf_realloc(buf, rc);
5774                 if (buf->lb_buf == NULL)
5775                         RETURN(-ENOMEM);
5776
5777                 goto again;
5778         }
5779
5780         ldata.ld_buf = buf;
5781         rc = linkea_init_with_rec(&ldata);
5782         if (!rc)
5783                 rc = linkea_links_find(&ldata, &cname, pfid);
5784
5785         RETURN(rc);
5786 }
5787
5788 /**
5789  * Calls ->lookup() to find dentry. From dentry get inode and
5790  * read inode's ea to get fid. This is required for  interoperability
5791  * mode (b11826)
5792  *
5793  * \retval   0, on success
5794  * \retval -ve, on error
5795  */
5796 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5797                              struct dt_rec *rec, const struct dt_key *key)
5798 {
5799         struct inode *dir = obj->oo_inode;
5800         struct dentry *dentry;
5801         struct ldiskfs_dir_entry_2 *de;
5802         struct buffer_head *bh;
5803         struct lu_fid *fid = (struct lu_fid *)rec;
5804         struct htree_lock *hlock = NULL;
5805         int ino;
5806         int rc;
5807
5808         ENTRY;
5809
5810         LASSERT(dir->i_op != NULL);
5811         LASSERT(dir->i_op->lookup != NULL);
5812
5813         dentry = osd_child_dentry_get(env, obj,
5814                                       (char *)key, strlen((char *)key));
5815
5816         if (obj->oo_hl_head != NULL) {
5817                 hlock = osd_oti_get(env)->oti_hlock;
5818                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5819                                    dir, LDISKFS_HLOCK_LOOKUP);
5820         } else {
5821                 down_read(&obj->oo_ext_idx_sem);
5822         }
5823
5824         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5825         if (!IS_ERR(bh)) {
5826                 struct osd_thread_info *oti = osd_oti_get(env);
5827                 struct osd_inode_id *id = &oti->oti_id;
5828                 struct osd_idmap_cache *oic = &oti->oti_cache;
5829                 struct osd_device *dev = osd_obj2dev(obj);
5830
5831                 ino = le32_to_cpu(de->inode);
5832                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5833                         brelse(bh);
5834                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5835                         GOTO(out, rc);
5836                 }
5837
5838                 rc = osd_get_fid_from_dentry(de, rec);
5839
5840                 /* done with de, release bh */
5841                 brelse(bh);
5842                 if (rc != 0) {
5843                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5844                                 const char *name = (const char *)key;
5845
5846                                 /*
5847                                  * If the parent is on remote MDT, and there
5848                                  * is no FID-in-dirent, then we have to get
5849                                  * the parent FID from the linkEA.
5850                                  */
5851                                 if (likely(strlen(name) == 2 &&
5852                                            name[0] == '.' && name[1] == '.'))
5853                                         rc = osd_get_pfid_from_linkea(env, obj,
5854                                                                       fid);
5855                         } else {
5856                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5857                         }
5858                 } else {
5859                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5860                 }
5861
5862                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5863                         fid_zero(&oic->oic_fid);
5864
5865                         GOTO(out, rc);
5866                 }
5867
5868                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5869                 rc = osd_consistency_check(oti, dev, oic);
5870                 if (rc == -ENOENT)
5871                         fid_zero(&oic->oic_fid);
5872                 else
5873                         /* Other error should not affect lookup result. */
5874                         rc = 0;
5875         } else {
5876                 rc = PTR_ERR(bh);
5877         }
5878
5879         GOTO(out, rc);
5880
5881 out:
5882         if (hlock != NULL)
5883                 ldiskfs_htree_unlock(hlock);
5884         else
5885                 up_read(&obj->oo_ext_idx_sem);
5886         return rc;
5887 }
5888
5889 static int osd_index_declare_ea_insert(const struct lu_env *env,
5890                                        struct dt_object *dt,
5891                                        const struct dt_rec *rec,
5892                                        const struct dt_key *key,
5893                                        struct thandle *handle)
5894 {
5895         struct osd_thandle *oh;
5896         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5897         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5898         const struct lu_fid *fid = rec1->rec_fid;
5899         int credits, rc = 0;
5900         struct osd_idmap_cache *idc;
5901
5902         ENTRY;
5903
5904         LASSERT(!dt_object_remote(dt));
5905         LASSERT(handle != NULL);
5906         LASSERT(fid != NULL);
5907         LASSERT(rec1->rec_type != 0);
5908
5909         oh = container_of0(handle, struct osd_thandle, ot_super);
5910         LASSERT(oh->ot_handle == NULL);
5911
5912         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5913
5914         /*
5915          * we can't call iget() while a transactions is running
5916          * (this can lead to a deadlock), but we need to know
5917          * inum and object type. so we find this information at
5918          * declaration and cache in per-thread info
5919          */
5920         idc = osd_idc_find_or_init(env, osd, fid);
5921         if (IS_ERR(idc))
5922                 RETURN(PTR_ERR(idc));
5923         if (idc->oic_remote) {
5924                 /*
5925                  * a reference to remote inode is represented by an
5926                  * agent inode which we have to create
5927                  */
5928                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5929                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5930         }
5931
5932         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5933
5934         if (osd_dt_obj(dt)->oo_inode != NULL) {
5935                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5936
5937                 /*
5938                  * We ignore block quota on meta pool (MDTs), so needn't
5939                  * calculate how many blocks will be consumed by this index
5940                  * insert
5941                  */
5942                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5943                                            i_gid_read(inode),
5944                                            i_projid_read(inode), 0,
5945                                            oh, osd_dt_obj(dt), NULL,
5946                                            OSD_QID_BLK);
5947                 if (rc)
5948                         RETURN(rc);
5949
5950 #ifdef HAVE_PROJECT_QUOTA
5951                 /*
5952                  * Reserve credits for local agent inode to transfer
5953                  * to 0, quota enforcement is ignored in this case.
5954                  */
5955                 if (idc->oic_remote &&
5956                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5957                     i_projid_read(inode) != 0)
5958                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5959                                                   0, i_projid_read(inode),
5960                                                   0, false, PRJQUOTA, true);
5961 #endif
5962         }
5963
5964         RETURN(rc);
5965 }
5966
5967 /**
5968  * Index add function for interoperability mode (b11826).
5969  * It will add the directory entry.This entry is needed to
5970  * maintain name->fid mapping.
5971  *
5972  * \param key it is key i.e. file entry to be inserted
5973  * \param rec it is value of given key i.e. fid
5974  *
5975  * \retval   0, on success
5976  * \retval -ve, on error
5977  */
5978 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5979                                const struct dt_rec *rec,
5980                                const struct dt_key *key, struct thandle *th)
5981 {
5982         struct osd_object *obj = osd_dt_obj(dt);
5983         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5984         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5985         const struct lu_fid *fid = rec1->rec_fid;
5986         const char *name = (const char *)key;
5987         struct osd_thread_info *oti = osd_oti_get(env);
5988         struct inode *child_inode = NULL;
5989         struct osd_idmap_cache *idc;
5990         int rc;
5991
5992         ENTRY;
5993
5994         if (!dt_object_exists(dt))
5995                 RETURN(-ENOENT);
5996
5997         LASSERT(osd_invariant(obj));
5998         LASSERT(!dt_object_remote(dt));
5999         LASSERT(th != NULL);
6000
6001         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6002
6003         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6004
6005         idc = osd_idc_find(env, osd, fid);
6006         if (unlikely(idc == NULL)) {
6007                 idc = osd_idc_find_or_init(env, osd, fid);
6008                 if (IS_ERR(idc)) {
6009                         /*
6010                          * this dt_insert() wasn't declared properly, so
6011                          * FID is missing in OI cache. we better do not
6012                          * lookup FID in FLDB/OI and don't risk to deadlock,
6013                          * but in some special cases (lfsck testing, etc)
6014                          * it's much simpler than fixing a caller.
6015                          *
6016                          * normally this error should be placed after the first
6017                          * find, but migrate may attach source stripes to
6018                          * target, which doesn't create stripes.
6019                          */
6020                         CERROR("%s: "DFID" wasn't declared for insert\n",
6021                                osd_name(osd), PFID(fid));
6022                         dump_stack();
6023                         RETURN(PTR_ERR(idc));
6024                 }
6025         }
6026
6027         if (idc->oic_remote) {
6028                 /* Insert remote entry */
6029                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6030                         child_inode =
6031                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6032                 } else {
6033                         child_inode = osd_create_local_agent_inode(env, osd,
6034                                         obj, fid, rec1->rec_type & S_IFMT, th);
6035                         if (IS_ERR(child_inode))
6036                                 RETURN(PTR_ERR(child_inode));
6037                 }
6038         } else {
6039                 /* Insert local entry */
6040                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6041                         /* for a reason OI cache wasn't filled properly */
6042                         CERROR("%s: OIC for "DFID" isn't filled\n",
6043                                osd_name(osd), PFID(fid));
6044                         RETURN(-EINVAL);
6045                 }
6046                 child_inode = oti->oti_inode;
6047                 if (unlikely(child_inode == NULL)) {
6048                         struct ldiskfs_inode_info *lii;
6049
6050                         OBD_ALLOC_PTR(lii);
6051                         if (lii == NULL)
6052                                 RETURN(-ENOMEM);
6053                         child_inode = oti->oti_inode = &lii->vfs_inode;
6054                 }
6055                 child_inode->i_sb = osd_sb(osd);
6056                 child_inode->i_ino = idc->oic_lid.oii_ino;
6057                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6058         }
6059
6060         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6061
6062         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6063                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6064
6065         if (child_inode && child_inode != oti->oti_inode)
6066                 iput(child_inode);
6067         LASSERT(osd_invariant(obj));
6068         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6069         RETURN(rc);
6070 }
6071
6072 /**
6073  *  Initialize osd Iterator for given osd index object.
6074  *
6075  *  \param  dt      osd index object
6076  */
6077
6078 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6079                                      struct dt_object *dt,
6080                                      __u32 unused)
6081 {
6082         struct osd_it_iam *it;
6083         struct osd_object *obj = osd_dt_obj(dt);
6084         struct lu_object *lo = &dt->do_lu;
6085         struct iam_path_descr *ipd;
6086         struct iam_container *bag = &obj->oo_dir->od_container;
6087
6088         if (!dt_object_exists(dt))
6089                 return ERR_PTR(-ENOENT);
6090
6091         OBD_ALLOC_PTR(it);
6092         if (it == NULL)
6093                 return ERR_PTR(-ENOMEM);
6094
6095         ipd = osd_it_ipd_get(env, bag);
6096         if (likely(ipd != NULL)) {
6097                 it->oi_obj = obj;
6098                 it->oi_ipd = ipd;
6099                 lu_object_get(lo);
6100                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6101                 return (struct dt_it *)it;
6102         } else {
6103                 OBD_FREE_PTR(it);
6104                 return ERR_PTR(-ENOMEM);
6105         }
6106 }
6107
6108 /**
6109  * free given Iterator.
6110  */
6111 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6112 {
6113         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6114         struct osd_object *obj = it->oi_obj;
6115
6116         iam_it_fini(&it->oi_it);
6117         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6118         osd_object_put(env, obj);
6119         OBD_FREE_PTR(it);
6120 }
6121
6122 /**
6123  *  Move Iterator to record specified by \a key
6124  *
6125  *  \param  di      osd iterator
6126  *  \param  key     key for index
6127  *
6128  *  \retval +ve  di points to record with least key not larger than key
6129  *  \retval  0   di points to exact matched key
6130  *  \retval -ve  failure
6131  */
6132
6133 static int osd_it_iam_get(const struct lu_env *env,
6134                           struct dt_it *di, const struct dt_key *key)
6135 {
6136         struct osd_thread_info *oti = osd_oti_get(env);
6137         struct osd_it_iam *it = (struct osd_it_iam *)di;
6138
6139         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6140                 /* swab quota uid/gid */
6141                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6142                 key = (struct dt_key *)&oti->oti_quota_id;
6143         }
6144
6145         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6146 }
6147
6148 /**
6149  *  Release Iterator
6150  *
6151  *  \param  di      osd iterator
6152  */
6153 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6154 {
6155         struct osd_it_iam *it = (struct osd_it_iam *)di;
6156
6157         iam_it_put(&it->oi_it);
6158 }
6159
6160 /**
6161  *  Move iterator by one record
6162  *
6163  *  \param  di      osd iterator
6164  *
6165  *  \retval +1   end of container reached
6166  *  \retval  0   success
6167  *  \retval -ve  failure
6168  */
6169
6170 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6171 {
6172         struct osd_it_iam *it = (struct osd_it_iam *)di;
6173
6174         return iam_it_next(&it->oi_it);
6175 }
6176
6177 /**
6178  * Return pointer to the key under iterator.
6179  */
6180
6181 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6182                                      const struct dt_it *di)
6183 {
6184         struct osd_thread_info *oti = osd_oti_get(env);
6185         struct osd_it_iam *it = (struct osd_it_iam *)di;
6186         struct osd_object *obj = it->oi_obj;
6187         struct dt_key *key;
6188
6189         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6190
6191         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6192                 /* swab quota uid/gid */
6193                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6194                 key = (struct dt_key *)&oti->oti_quota_id;
6195         }
6196
6197         return key;
6198 }
6199
6200 /**
6201  * Return size of key under iterator (in bytes)
6202  */
6203
6204 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6205 {
6206         struct osd_it_iam *it = (struct osd_it_iam *)di;
6207
6208         return iam_it_key_size(&it->oi_it);
6209 }
6210
6211 static inline void
6212 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6213 {
6214         /* check if file type is required */
6215         if (ent->lde_attrs & LUDA_TYPE) {
6216                 struct luda_type *lt;
6217                 int align = sizeof(*lt) - 1;
6218
6219                 len = (len + align) & ~align;
6220                 lt = (struct luda_type *)(ent->lde_name + len);
6221                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6222         }
6223
6224         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6225 }
6226
6227 /**
6228  * build lu direct from backend fs dirent.
6229  */
6230
6231 static inline void
6232 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6233                    char *name, __u16 namelen, __u16 type, __u32 attr)
6234 {
6235         ent->lde_attrs = attr | LUDA_FID;
6236         fid_cpu_to_le(&ent->lde_fid, fid);
6237
6238         ent->lde_hash = cpu_to_le64(offset);
6239         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6240
6241         strncpy(ent->lde_name, name, namelen);
6242         ent->lde_name[namelen] = '\0';
6243         ent->lde_namelen = cpu_to_le16(namelen);
6244
6245         /* append lustre attributes */
6246         osd_it_append_attrs(ent, namelen, type);
6247 }
6248
6249 /**
6250  * Return pointer to the record under iterator.
6251  */
6252 static int osd_it_iam_rec(const struct lu_env *env,
6253                           const struct dt_it *di,
6254                           struct dt_rec *dtrec, __u32 attr)
6255 {
6256         struct osd_it_iam *it = (struct osd_it_iam *)di;
6257         struct osd_thread_info *info = osd_oti_get(env);
6258
6259         ENTRY;
6260
6261         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6262                 const struct osd_fid_pack *rec;
6263                 struct lu_fid *fid = &info->oti_fid;
6264                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6265                 char *name;
6266                 int namelen;
6267                 __u64 hash;
6268                 int rc;
6269
6270                 name = (char *)iam_it_key_get(&it->oi_it);
6271                 if (IS_ERR(name))
6272                         RETURN(PTR_ERR(name));
6273
6274                 namelen = iam_it_key_size(&it->oi_it);
6275
6276                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6277                 if (IS_ERR(rec))
6278                         RETURN(PTR_ERR(rec));
6279
6280                 rc = osd_fid_unpack(fid, rec);
6281                 if (rc)
6282                         RETURN(rc);
6283
6284                 hash = iam_it_store(&it->oi_it);
6285
6286                 /* IAM does not store object type in IAM index (dir) */
6287                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6288                                    0, LUDA_FID);
6289         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6290                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6291                            (struct iam_rec *)dtrec);
6292                 osd_quota_unpack(it->oi_obj, dtrec);
6293         } else {
6294                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6295                            (struct iam_rec *)dtrec);
6296         }
6297
6298         RETURN(0);
6299 }
6300
6301 /**
6302  * Returns cookie for current Iterator position.
6303  */
6304 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6305 {
6306         struct osd_it_iam *it = (struct osd_it_iam *)di;
6307
6308         return iam_it_store(&it->oi_it);
6309 }
6310
6311 /**
6312  * Restore iterator from cookie.
6313  *
6314  * \param  di      osd iterator
6315  * \param  hash    Iterator location cookie
6316  *
6317  * \retval +ve  di points to record with least key not larger than key.
6318  * \retval  0   di points to exact matched key
6319  * \retval -ve  failure
6320  */
6321
6322 static int osd_it_iam_load(const struct lu_env *env,
6323                            const struct dt_it *di, __u64 hash)
6324 {
6325         struct osd_it_iam *it = (struct osd_it_iam *)di;
6326
6327         return iam_it_load(&it->oi_it, hash);
6328 }
6329
6330 static const struct dt_index_operations osd_index_iam_ops = {
6331         .dio_lookup         = osd_index_iam_lookup,
6332         .dio_declare_insert = osd_index_declare_iam_insert,
6333         .dio_insert         = osd_index_iam_insert,
6334         .dio_declare_delete = osd_index_declare_iam_delete,
6335         .dio_delete         = osd_index_iam_delete,
6336         .dio_it     = {
6337                 .init     = osd_it_iam_init,
6338                 .fini     = osd_it_iam_fini,
6339                 .get      = osd_it_iam_get,
6340                 .put      = osd_it_iam_put,
6341                 .next     = osd_it_iam_next,
6342                 .key      = osd_it_iam_key,
6343                 .key_size = osd_it_iam_key_size,
6344                 .rec      = osd_it_iam_rec,
6345                 .store    = osd_it_iam_store,
6346                 .load     = osd_it_iam_load
6347         }
6348 };
6349
6350
6351 /**
6352  * Creates or initializes iterator context.
6353  *
6354  * \retval struct osd_it_ea, iterator structure on success
6355  *
6356  */
6357 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6358                                     struct dt_object *dt,
6359                                     __u32 attr)
6360 {
6361         struct osd_object *obj = osd_dt_obj(dt);
6362         struct osd_thread_info *info = osd_oti_get(env);
6363         struct osd_it_ea *oie;
6364         struct file *file;
6365         struct lu_object *lo = &dt->do_lu;
6366         struct dentry *obj_dentry;
6367
6368         ENTRY;
6369
6370         if (!dt_object_exists(dt) || obj->oo_destroyed)
6371                 RETURN(ERR_PTR(-ENOENT));
6372
6373         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6374         if (oie == NULL)
6375                 RETURN(ERR_PTR(-ENOMEM));
6376         obj_dentry = &oie->oie_dentry;
6377
6378         obj_dentry->d_inode = obj->oo_inode;
6379         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6380         obj_dentry->d_name.hash = 0;
6381
6382         oie->oie_rd_dirent       = 0;
6383         oie->oie_it_dirent       = 0;
6384         oie->oie_dirent          = NULL;
6385         if (unlikely(!info->oti_it_ea_buf_used)) {
6386                 oie->oie_buf = info->oti_it_ea_buf;
6387                 info->oti_it_ea_buf_used = 1;
6388         } else {
6389                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6390                 if (oie->oie_buf == NULL)
6391                         RETURN(ERR_PTR(-ENOMEM));
6392         }
6393         oie->oie_obj = obj;
6394
6395         file = &oie->oie_file;
6396
6397         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6398         if (attr & LUDA_64BITHASH)
6399                 file->f_mode    = FMODE_64BITHASH;
6400         else
6401                 file->f_mode    = FMODE_32BITHASH;
6402         file->f_path.dentry     = obj_dentry;
6403         file->f_mapping         = obj->oo_inode->i_mapping;
6404         file->f_op              = obj->oo_inode->i_fop;
6405         set_file_inode(file, obj->oo_inode);
6406
6407         lu_object_get(lo);
6408         RETURN((struct dt_it *)oie);
6409 }
6410
6411 /**
6412  * Destroy or finishes iterator context.
6413  *
6414  * \param di iterator structure to be destroyed
6415  */
6416 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6417 {
6418         struct osd_thread_info *info = osd_oti_get(env);
6419         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6420         struct osd_object *obj = oie->oie_obj;
6421         struct inode *inode = obj->oo_inode;
6422
6423         ENTRY;
6424         oie->oie_file.f_op->release(inode, &oie->oie_file);
6425         osd_object_put(env, obj);
6426         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6427                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6428         else
6429                 info->oti_it_ea_buf_used = 0;
6430         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6431         EXIT;
6432 }
6433
6434 /**
6435  * It position the iterator at given key, so that next lookup continues from
6436  * that key Or it is similar to dio_it->load() but based on a key,
6437  * rather than file position.
6438  *
6439  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6440  * to the beginning.
6441  *
6442  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6443  */
6444 static int osd_it_ea_get(const struct lu_env *env,
6445                          struct dt_it *di, const struct dt_key *key)
6446 {
6447         struct osd_it_ea *it = (struct osd_it_ea *)di;
6448
6449         ENTRY;
6450         LASSERT(((const char *)key)[0] == '\0');
6451         it->oie_file.f_pos = 0;
6452         it->oie_rd_dirent = 0;
6453         it->oie_it_dirent = 0;
6454         it->oie_dirent = NULL;
6455
6456         RETURN(+1);
6457 }
6458
6459 /**
6460  * Does nothing
6461  */
6462 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6463 {
6464 }
6465
6466 struct osd_filldir_cbs {
6467 #ifdef HAVE_DIR_CONTEXT
6468         struct dir_context ctx;
6469 #endif
6470         struct osd_it_ea  *it;
6471 };
6472 /**
6473  * It is called internally by ->readdir(). It fills the
6474  * iterator's in-memory data structure with required
6475  * information i.e. name, namelen, rec_size etc.
6476  *
6477  * \param buf in which information to be filled in.
6478  * \param name name of the file in given dir
6479  *
6480  * \retval 0 on success
6481  * \retval 1 on buffer full
6482  */
6483 #ifdef HAVE_FILLDIR_USE_CTX
6484 static int osd_ldiskfs_filldir(struct dir_context *buf,
6485 #else
6486 static int osd_ldiskfs_filldir(void *buf,
6487 #endif
6488                                const char *name, int namelen,
6489                                loff_t offset, __u64 ino, unsigned int d_type)
6490 {
6491         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6492         struct osd_object *obj = it->oie_obj;
6493         struct osd_it_ea_dirent *ent = it->oie_dirent;
6494         struct lu_fid *fid = &ent->oied_fid;
6495         struct osd_fid_pack *rec;
6496
6497         ENTRY;
6498
6499 /* this should never happen */
6500         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6501                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6502                 RETURN(-EIO);
6503         }
6504
6505         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6506             OSD_IT_EA_BUFSIZE)
6507                 RETURN(1);
6508
6509         /* "." is just the object itself. */
6510         if (namelen == 1 && name[0] == '.') {
6511                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6512         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6513                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6514                 if (osd_fid_unpack(fid, rec) != 0)
6515                         fid_zero(fid);
6516         } else {
6517                 fid_zero(fid);
6518         }
6519         d_type &= ~LDISKFS_DIRENT_LUFID;
6520
6521         /* NOT export local root. */
6522         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6523                 ino = obj->oo_inode->i_ino;
6524                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6525         }
6526
6527         ent->oied_ino     = ino;
6528         ent->oied_off     = offset;
6529         ent->oied_namelen = namelen;
6530         ent->oied_type    = d_type;
6531
6532         memcpy(ent->oied_name, name, namelen);
6533
6534         it->oie_rd_dirent++;
6535         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6536         RETURN(0);
6537 }
6538
6539 /**
6540  * Calls ->readdir() to load a directory entry at a time
6541  * and stored it in iterator's in-memory data structure.
6542  *
6543  * \param di iterator's in memory structure
6544  *
6545  * \retval   0 on success
6546  * \retval -ve on error
6547  * \retval +1 reach the end of entry
6548  */
6549 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6550                                const struct dt_it *di)
6551 {
6552         struct osd_it_ea *it = (struct osd_it_ea *)di;
6553         struct osd_object *obj = it->oie_obj;
6554         struct inode *inode = obj->oo_inode;
6555         struct htree_lock *hlock = NULL;
6556         struct file *filp = &it->oie_file;
6557         int rc = 0;
6558         struct osd_filldir_cbs buf = {
6559 #ifdef HAVE_DIR_CONTEXT
6560                 .ctx.actor = osd_ldiskfs_filldir,
6561 #endif
6562                 .it = it
6563         };
6564
6565         ENTRY;
6566         it->oie_dirent = it->oie_buf;
6567         it->oie_rd_dirent = 0;
6568
6569         if (obj->oo_hl_head != NULL) {
6570                 hlock = osd_oti_get(env)->oti_hlock;
6571                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6572                                    inode, LDISKFS_HLOCK_READDIR);
6573         } else {
6574                 down_read(&obj->oo_ext_idx_sem);
6575         }
6576
6577 #ifdef HAVE_DIR_CONTEXT
6578         buf.ctx.pos = filp->f_pos;
6579 #ifdef HAVE_ITERATE_SHARED
6580         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6581 #else
6582         rc = inode->i_fop->iterate(filp, &buf.ctx);
6583 #endif
6584         filp->f_pos = buf.ctx.pos;
6585 #else
6586         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6587 #endif
6588
6589         if (hlock != NULL)
6590                 ldiskfs_htree_unlock(hlock);
6591         else
6592                 up_read(&obj->oo_ext_idx_sem);
6593
6594         if (it->oie_rd_dirent == 0) {
6595                 /*
6596                  * If it does not get any dirent, it means it has been reached
6597                  * to the end of the dir
6598                  */
6599                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6600                 if (rc == 0)
6601                         rc = 1;
6602         } else {
6603                 it->oie_dirent = it->oie_buf;
6604                 it->oie_it_dirent = 1;
6605         }
6606
6607         RETURN(rc);
6608 }
6609
6610 /**
6611  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6612  * to load a directory entry at a time and stored it in
6613  * iterator's in-memory data structure.
6614  *
6615  * \param di iterator's in memory structure
6616  *
6617  * \retval +ve iterator reached to end
6618  * \retval   0 iterator not reached to end
6619  * \retval -ve on error
6620  */
6621 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6622 {
6623         struct osd_it_ea *it = (struct osd_it_ea *)di;
6624         int rc;
6625
6626         ENTRY;
6627
6628         if (it->oie_it_dirent < it->oie_rd_dirent) {
6629                 it->oie_dirent =
6630                         (void *)it->oie_dirent +
6631                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6632                                        it->oie_dirent->oied_namelen);
6633                 it->oie_it_dirent++;
6634                 rc = 0;
6635         } else {
6636                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6637                         rc = 1;
6638                 else
6639                         rc = osd_ldiskfs_it_fill(env, di);
6640         }
6641
6642         RETURN(rc);
6643 }
6644
6645 /**
6646  * Returns the key at current position from iterator's in memory structure.
6647  *
6648  * \param di iterator's in memory structure
6649  *
6650  * \retval key i.e. struct dt_key on success
6651  */
6652 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6653                                     const struct dt_it *di)
6654 {
6655         struct osd_it_ea *it = (struct osd_it_ea *)di;
6656
6657         return (struct dt_key *)it->oie_dirent->oied_name;
6658 }
6659
6660 /**
6661  * Returns key's size at current position from iterator's in memory structure.
6662  *
6663  * \param di iterator's in memory structure
6664  *
6665  * \retval key_size i.e. struct dt_key on success
6666  */
6667 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6668 {
6669         struct osd_it_ea *it = (struct osd_it_ea *)di;
6670
6671         return it->oie_dirent->oied_namelen;
6672 }
6673
6674 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6675 {
6676         if (LDISKFS_DIR_REC_LEN(de) >=
6677             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6678                 return true;
6679
6680         return false;
6681 }
6682
6683 static inline bool
6684 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6685                      unsigned int blocksize, bool dotdot)
6686 {
6687         if (dotdot)
6688                 return osd_dotdot_has_space(de);
6689
6690         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6691             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6692                 return true;
6693
6694         return false;
6695 }
6696
6697 static int
6698 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6699                     handle_t *jh, struct dentry *dentry,
6700                     const struct lu_fid *fid, struct buffer_head *bh,
6701                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6702                     bool dotdot)
6703 {
6704         struct inode *dir = dentry->d_parent->d_inode;
6705         struct inode *inode = dentry->d_inode;
6706         struct osd_fid_pack *rec;
6707         struct ldiskfs_dentry_param *ldp;
6708         int namelen = dentry->d_name.len;
6709         int rc;
6710         struct osd_thread_info *info = osd_oti_get(env);
6711
6712         ENTRY;
6713
6714         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6715                 RETURN(0);
6716
6717         /* There is enough space to hold the FID-in-dirent. */
6718         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6719                 rc = ldiskfs_journal_get_write_access(jh, bh);
6720                 if (rc != 0)
6721                         RETURN(rc);
6722
6723                 de->name[namelen] = 0;
6724                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6725                 rec->fp_len = sizeof(struct lu_fid) + 1;
6726                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6727                 de->file_type |= LDISKFS_DIRENT_LUFID;
6728                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6729
6730                 RETURN(rc);
6731         }
6732
6733         LASSERT(!dotdot);
6734
6735         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6736         if (rc != 0)
6737                 RETURN(rc);
6738
6739         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6740         osd_get_ldiskfs_dirent_param(ldp, fid);
6741         dentry->d_fsdata = (void *)ldp;
6742         ll_vfs_dq_init(dir);
6743         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6744         /*
6745          * It is too bad, we cannot reinsert the name entry back.
6746          * That means we lose it!
6747          */
6748         if (rc != 0)
6749                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6750                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6751                        osd_ino2name(inode),
6752                        dir->i_ino, dir->i_generation, namelen,
6753                        dentry->d_name.name, PFID(fid), rc);
6754
6755         RETURN(rc);
6756 }
6757
6758 static int
6759 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6760                         struct osd_it_ea *it, struct lu_fid *fid,
6761                         struct osd_inode_id *id, __u32 *attr)
6762 {
6763         struct osd_thread_info *info = osd_oti_get(env);
6764         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6765         struct osd_device *dev = osd_obj2dev(obj);
6766         struct super_block *sb = osd_sb(dev);
6767         const char *devname = osd_name(dev);
6768         struct osd_it_ea_dirent *ent = it->oie_dirent;
6769         struct inode *dir = obj->oo_inode;
6770         struct htree_lock *hlock = NULL;
6771         struct buffer_head *bh = NULL;
6772         handle_t *jh = NULL;
6773         struct ldiskfs_dir_entry_2 *de;
6774         struct dentry *dentry;
6775         struct inode *inode;
6776         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6777         int credits;
6778         int rc;
6779         bool dotdot = false;
6780         bool dirty = false;
6781
6782         ENTRY;
6783
6784         if (ent->oied_name[0] == '.') {
6785                 if (ent->oied_namelen == 1)
6786                         RETURN(0);
6787
6788                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6789                         dotdot = true;
6790         }
6791
6792         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6793         inode = osd_iget(info, dev, id);
6794         if (IS_ERR(inode)) {
6795                 rc = PTR_ERR(inode);
6796                 if (rc == -ENOENT || rc == -ESTALE) {
6797                         /*
6798                          * Maybe dangling name entry, or
6799                          * corrupted directory entry.
6800                          */
6801                         *attr |= LUDA_UNKNOWN;
6802                         rc = 0;
6803                 } else {
6804                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6805                                "check_repair, dir = %lu/%u, name = %.*s, "
6806                                "ino = %llu, rc = %d\n",
6807                                devname, dir->i_ino, dir->i_generation,
6808                                ent->oied_namelen, ent->oied_name,
6809                                ent->oied_ino, rc);
6810                 }
6811
6812                 RETURN(rc);
6813         }
6814
6815         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6816                                            ent->oied_namelen);
6817         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6818         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6819                 lma = NULL;
6820         else if (rc != 0)
6821                 GOTO(out, rc);
6822
6823         /*
6824          * We need to ensure that the name entry is still valid.
6825          * Because it may be removed or renamed by other already.
6826          *
6827          * The unlink or rename operation will start journal before PDO lock,
6828          * so to avoid deadlock, here we need to start journal handle before
6829          * related PDO lock also. But because we do not know whether there
6830          * will be something to be repaired before PDO lock, we just start
6831          * journal without conditions.
6832          *
6833          * We may need to remove the name entry firstly, then insert back.
6834          * One credit is for user quota file update.
6835          * One credit is for group quota file update.
6836          * Two credits are for dirty inode.
6837          */
6838         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6839                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6840
6841         if (dev->od_dirent_journal != 0) {
6842
6843 again:
6844                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6845                 if (IS_ERR(jh)) {
6846                         rc = PTR_ERR(jh);
6847                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6848                                "check_repair, dir = %lu/%u, credits = %d, "
6849                                "name = %.*s, ino = %llu: rc = %d\n",
6850                                devname, dir->i_ino, dir->i_generation, credits,
6851                                ent->oied_namelen, ent->oied_name,
6852                                ent->oied_ino, rc);
6853
6854                         GOTO(out_inode, rc);
6855                 }
6856
6857                 if (obj->oo_hl_head != NULL) {
6858                         hlock = osd_oti_get(env)->oti_hlock;
6859                         /*
6860                          * "0" means exclusive lock for the whole directory.
6861                          * We need to prevent others access such name entry
6862                          * during the delete + insert. Neither HLOCK_ADD nor
6863                          * HLOCK_DEL cannot guarantee the atomicity.
6864                          */
6865                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6866                 } else {
6867                         down_write(&obj->oo_ext_idx_sem);
6868                 }
6869         } else {
6870                 if (obj->oo_hl_head != NULL) {
6871                         hlock = osd_oti_get(env)->oti_hlock;
6872                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6873                                            LDISKFS_HLOCK_LOOKUP);
6874                 } else {
6875                         down_read(&obj->oo_ext_idx_sem);
6876                 }
6877         }
6878
6879         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6880         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6881                 *attr |= LUDA_IGNORE;
6882
6883                 GOTO(out, rc = 0);
6884         }
6885
6886         /*
6887          * For dotdot entry, if there is not enough space to hold the
6888          * FID-in-dirent, just keep them there. It only happens when the
6889          * device upgraded from 1.8 or restored from MDT file-level backup.
6890          * For the whole directory, only dotdot entry have no FID-in-dirent
6891          * and needs to get FID from LMA when readdir, it will not affect the
6892          * performance much.
6893          */
6894         if (dotdot && !osd_dotdot_has_space(de)) {
6895                 *attr |= LUDA_UNKNOWN;
6896
6897                 GOTO(out, rc = 0);
6898         }
6899
6900         if (lma != NULL) {
6901                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6902                         GOTO(out, rc = 0);
6903
6904                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6905                         struct lu_fid *tfid = &lma->lma_self_fid;
6906
6907                         if (likely(dotdot &&
6908                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6909                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6910                                 /*
6911                                  * It must be REMOTE_PARENT_DIR and as the
6912                                  * 'dotdot' entry of remote directory
6913                                  */
6914                                 *attr |= LUDA_IGNORE;
6915                         } else {
6916                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6917                                        "parent directory, but got %.*s under "
6918                                        "dir = %lu/%u with the FID "DFID"\n",
6919                                        devname, ent->oied_namelen,
6920                                        ent->oied_name, dir->i_ino,
6921                                        dir->i_generation, PFID(tfid));
6922
6923                                 *attr |= LUDA_UNKNOWN;
6924                         }
6925
6926                         GOTO(out, rc = 0);
6927                 }
6928         }
6929
6930         if (!fid_is_zero(fid)) {
6931                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6932                                               ent->oied_namelen);
6933                 if (rc == -ENOENT ||
6934                     (rc == -ENODATA &&
6935                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6936                         /*
6937                          * linkEA does not recognize the dirent entry,
6938                          * it may because the dirent entry corruption
6939                          * and points to other's inode.
6940                          */
6941                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6942                                "recognize the dirent, dir = %lu/%u, "
6943                                " name = %.*s, ino = %llu, "
6944                                DFID": rc = %d\n", devname, dir->i_ino,
6945                                dir->i_generation, ent->oied_namelen,
6946                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6947                         *attr |= LUDA_UNKNOWN;
6948
6949                         GOTO(out, rc = 0);
6950                 }
6951
6952                 if (rc && rc != -ENODATA) {
6953                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6954                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6955                                DFID": rc = %d\n", devname, dir->i_ino,
6956                                dir->i_generation, ent->oied_namelen,
6957                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6958                         *attr |= LUDA_UNKNOWN;
6959
6960                         GOTO(out, rc = 0);
6961                 }
6962         }
6963
6964         if (lma != NULL) {
6965                 /*
6966                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6967                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6968                  */
6969                 if (*attr & LUDA_VERIFY_DRYRUN) {
6970                         *fid = lma->lma_self_fid;
6971                         *attr |= LUDA_REPAIR;
6972
6973                         GOTO(out, rc = 0);
6974                 }
6975
6976                 if (jh == NULL) {
6977                         brelse(bh);
6978                         dev->od_dirent_journal = 1;
6979                         if (hlock != NULL) {
6980                                 ldiskfs_htree_unlock(hlock);
6981                                 hlock = NULL;
6982                         } else {
6983                                 up_read(&obj->oo_ext_idx_sem);
6984                         }
6985
6986                         goto again;
6987                 }
6988
6989                 *fid = lma->lma_self_fid;
6990                 dirty = true;
6991                 /* Update or append the FID-in-dirent. */
6992                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6993                                          bh, de, hlock, dotdot);
6994                 if (rc == 0)
6995                         *attr |= LUDA_REPAIR;
6996                 else
6997                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6998                                "the dirent, dir = %lu/%u, name = %.*s, "
6999                                "ino = %llu, "DFID": rc = %d\n",
7000                                devname, dir->i_ino, dir->i_generation,
7001                                ent->oied_namelen, ent->oied_name,
7002                                ent->oied_ino, PFID(fid), rc);
7003         } else {
7004                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7005                 if (*attr & LUDA_VERIFY_DRYRUN) {
7006                         if (fid_is_sane(fid)) {
7007                                 *attr |= LUDA_REPAIR;
7008                         } else if (dev->od_index == 0) {
7009                                 lu_igif_build(fid, inode->i_ino,
7010                                               inode->i_generation);
7011                                 *attr |= LUDA_UPGRADE;
7012                         }
7013
7014                         GOTO(out, rc = 0);
7015                 }
7016
7017                 if (jh == NULL) {
7018                         brelse(bh);
7019                         dev->od_dirent_journal = 1;
7020                         if (hlock != NULL) {
7021                                 ldiskfs_htree_unlock(hlock);
7022                                 hlock = NULL;
7023                         } else {
7024                                 up_read(&obj->oo_ext_idx_sem);
7025                         }
7026
7027                         goto again;
7028                 }
7029
7030                 dirty = true;
7031                 if (unlikely(fid_is_sane(fid))) {
7032                         /*
7033                          * FID-in-dirent exists, but FID-in-LMA is lost.
7034                          * Trust the FID-in-dirent, and add FID-in-LMA.
7035                          */
7036                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7037                         if (rc == 0)
7038                                 *attr |= LUDA_REPAIR;
7039                         else
7040                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7041                                        "update dirent, dir = %lu/%u, "
7042                                        "name = %.*s, ino = %llu, "
7043                                        DFID": rc = %d\n",
7044                                        devname, dir->i_ino, dir->i_generation,
7045                                        ent->oied_namelen, ent->oied_name,
7046                                        ent->oied_ino, PFID(fid), rc);
7047                 } else if (dev->od_index == 0) {
7048                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7049                         /*
7050                          * It is probably IGIF object. Only aappend the
7051                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7052                          */
7053                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7054                                                  bh, de, hlock, dotdot);
7055                         if (rc == 0)
7056                                 *attr |= LUDA_UPGRADE;
7057                         else
7058                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7059                                        "after the dirent, dir = %lu/%u, "
7060                                        "name = %.*s, ino = %llu, "
7061                                        DFID": rc = %d\n",
7062                                        devname, dir->i_ino, dir->i_generation,
7063                                        ent->oied_namelen, ent->oied_name,
7064                                        ent->oied_ino, PFID(fid), rc);
7065                 }
7066         }
7067
7068         GOTO(out, rc);
7069
7070 out:
7071         if (!IS_ERR(bh))
7072                 brelse(bh);
7073         if (hlock != NULL) {
7074                 ldiskfs_htree_unlock(hlock);
7075         } else {
7076                 if (dev->od_dirent_journal != 0)
7077                         up_write(&obj->oo_ext_idx_sem);
7078                 else
7079                         up_read(&obj->oo_ext_idx_sem);
7080         }
7081
7082         if (jh != NULL)
7083                 ldiskfs_journal_stop(jh);
7084
7085 out_inode:
7086         iput(inode);
7087         if (rc >= 0 && !dirty)
7088                 dev->od_dirent_journal = 0;
7089
7090         return rc;
7091 }
7092
7093 /**
7094  * Returns the value at current position from iterator's in memory structure.
7095  *
7096  * \param di struct osd_it_ea, iterator's in memory structure
7097  * \param attr attr requested for dirent.
7098  * \param lde lustre dirent
7099  *
7100  * \retval   0 no error and \param lde has correct lustre dirent.
7101  * \retval -ve on error
7102  */
7103 static inline int osd_it_ea_rec(const struct lu_env *env,
7104                                 const struct dt_it *di,
7105                                 struct dt_rec *dtrec, __u32 attr)
7106 {
7107         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7108         struct osd_object      *obj   = it->oie_obj;
7109         struct osd_device      *dev   = osd_obj2dev(obj);
7110         struct osd_thread_info *oti   = osd_oti_get(env);
7111         struct osd_inode_id    *id    = &oti->oti_id;
7112         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7113         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7114         __u32 ino = it->oie_dirent->oied_ino;
7115         int rc = 0;
7116
7117         ENTRY;
7118
7119         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7120
7121         if (attr & LUDA_VERIFY) {
7122                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7123                         attr |= LUDA_IGNORE;
7124                         /*
7125                          * If the parent is on remote MDT, and there
7126                          * is no FID-in-dirent, then we have to get
7127                          * the parent FID from the linkEA.
7128                          */
7129                         if (!fid_is_sane(fid) &&
7130                             it->oie_dirent->oied_namelen == 2 &&
7131                             it->oie_dirent->oied_name[0] == '.' &&
7132                             it->oie_dirent->oied_name[1] == '.')
7133                                 osd_get_pfid_from_linkea(env, obj, fid);
7134                 } else {
7135                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7136                                                      &attr);
7137                 }
7138
7139                 if (!fid_is_sane(fid))
7140                         attr |= LUDA_UNKNOWN;
7141         } else {
7142                 attr &= ~LU_DIRENT_ATTRS_MASK;
7143                 if (!fid_is_sane(fid)) {
7144                         bool is_dotdot = false;
7145
7146                         if (it->oie_dirent->oied_namelen == 2 &&
7147                             it->oie_dirent->oied_name[0] == '.' &&
7148                             it->oie_dirent->oied_name[1] == '.')
7149                                 is_dotdot = true;
7150                         /*
7151                          * If the parent is on remote MDT, and there
7152                          * is no FID-in-dirent, then we have to get
7153                          * the parent FID from the linkEA.
7154                          */
7155                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7156                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7157                         } else {
7158                                 if (is_dotdot == false &&
7159                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7160                                         RETURN(-ENOENT);
7161
7162                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7163                         }
7164                 } else {
7165                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7166                 }
7167         }
7168
7169         /* Pack the entry anyway, at least the offset is right. */
7170         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7171                            it->oie_dirent->oied_name,
7172                            it->oie_dirent->oied_namelen,
7173                            it->oie_dirent->oied_type, attr);
7174
7175         if (rc < 0)
7176                 RETURN(rc);
7177
7178         if (osd_remote_fid(env, dev, fid))
7179                 RETURN(0);
7180
7181         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7182                 osd_add_oi_cache(oti, dev, id, fid);
7183
7184         RETURN(rc > 0 ? 0 : rc);
7185 }
7186
7187 /**
7188  * Returns the record size size at current position.
7189  *
7190  * This function will return record(lu_dirent) size in bytes.
7191  *
7192  * \param[in] env       execution environment
7193  * \param[in] di        iterator's in memory structure
7194  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7195  *                      calculate the lu_dirent size.
7196  *
7197  * \retval      record size(in bytes & in memory) of the current lu_dirent
7198  *              entry.
7199  */
7200 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7201                               __u32 attr)
7202 {
7203         struct osd_it_ea *it = (struct osd_it_ea *)di;
7204
7205         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7206 }
7207
7208 /**
7209  * Returns a cookie for current position of the iterator head, so that
7210  * user can use this cookie to load/start the iterator next time.
7211  *
7212  * \param di iterator's in memory structure
7213  *
7214  * \retval cookie for current position, on success
7215  */
7216 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7217 {
7218         struct osd_it_ea *it = (struct osd_it_ea *)di;
7219
7220         return it->oie_dirent->oied_off;
7221 }
7222
7223 /**
7224  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7225  * to load a directory entry at a time and stored it i inn,
7226  * in iterator's in-memory data structure.
7227  *
7228  * \param di struct osd_it_ea, iterator's in memory structure
7229  *
7230  * \retval +ve on success
7231  * \retval -ve on error
7232  */
7233 static int osd_it_ea_load(const struct lu_env *env,
7234                           const struct dt_it *di, __u64 hash)
7235 {
7236         struct osd_it_ea *it = (struct osd_it_ea *)di;
7237         int rc;
7238
7239         ENTRY;
7240         it->oie_file.f_pos = hash;
7241
7242         rc =  osd_ldiskfs_it_fill(env, di);
7243         if (rc > 0)
7244                 rc = -ENODATA;
7245
7246         if (rc == 0)
7247                 rc = 1;
7248
7249         RETURN(rc);
7250 }
7251
7252 /**
7253  * Index lookup function for interoperability mode (b11826).
7254  *
7255  * \param key,  key i.e. file name to be searched
7256  *
7257  * \retval +ve, on success
7258  * \retval -ve, on error
7259  */
7260 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7261                                struct dt_rec *rec, const struct dt_key *key)
7262 {
7263         struct osd_object *obj = osd_dt_obj(dt);
7264         int rc = 0;
7265
7266         ENTRY;
7267
7268         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7269         LINVRNT(osd_invariant(obj));
7270
7271         rc = osd_ea_lookup_rec(env, obj, rec, key);
7272         if (rc == 0)
7273                 rc = 1;
7274         RETURN(rc);
7275 }
7276
7277 /**
7278  * Index and Iterator operations for interoperability
7279  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7280  */
7281 static const struct dt_index_operations osd_index_ea_ops = {
7282         .dio_lookup         = osd_index_ea_lookup,
7283         .dio_declare_insert = osd_index_declare_ea_insert,
7284         .dio_insert         = osd_index_ea_insert,
7285         .dio_declare_delete = osd_index_declare_ea_delete,
7286         .dio_delete         = osd_index_ea_delete,
7287         .dio_it     = {
7288                 .init     = osd_it_ea_init,
7289                 .fini     = osd_it_ea_fini,
7290                 .get      = osd_it_ea_get,
7291                 .put      = osd_it_ea_put,
7292                 .next     = osd_it_ea_next,
7293                 .key      = osd_it_ea_key,
7294                 .key_size = osd_it_ea_key_size,
7295                 .rec      = osd_it_ea_rec,
7296                 .rec_size = osd_it_ea_rec_size,
7297                 .store    = osd_it_ea_store,
7298                 .load     = osd_it_ea_load
7299         }
7300 };
7301
7302 static void *osd_key_init(const struct lu_context *ctx,
7303                           struct lu_context_key *key)
7304 {
7305         struct osd_thread_info *info;
7306
7307         OBD_ALLOC_PTR(info);
7308         if (info == NULL)
7309                 return ERR_PTR(-ENOMEM);
7310
7311         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7312         if (info->oti_it_ea_buf == NULL)
7313                 goto out_free_info;
7314
7315         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7316
7317         info->oti_hlock = ldiskfs_htree_lock_alloc();
7318         if (info->oti_hlock == NULL)
7319                 goto out_free_ea;
7320
7321         return info;
7322
7323 out_free_ea:
7324         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7325 out_free_info:
7326         OBD_FREE_PTR(info);
7327         return ERR_PTR(-ENOMEM);
7328 }
7329
7330 static void osd_key_fini(const struct lu_context *ctx,
7331                          struct lu_context_key *key, void *data)
7332 {
7333         struct osd_thread_info *info = data;
7334         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7335         struct osd_idmap_cache *idc = info->oti_ins_cache;
7336
7337         if (info->oti_dio_pages) {
7338                 int i;
7339                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7340                         if (info->oti_dio_pages[i])
7341                                 __free_page(info->oti_dio_pages[i]);
7342                 }
7343                 OBD_FREE(info->oti_dio_pages,
7344                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7345         }
7346
7347         if (info->oti_inode != NULL)
7348                 OBD_FREE_PTR(lli);
7349         if (info->oti_hlock != NULL)
7350                 ldiskfs_htree_lock_free(info->oti_hlock);
7351         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7352         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7353         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7354         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7355         lu_buf_free(&info->oti_big_buf);
7356         if (idc != NULL) {
7357                 LASSERT(info->oti_ins_cache_size > 0);
7358                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7359                 info->oti_ins_cache = NULL;
7360                 info->oti_ins_cache_size = 0;
7361         }
7362         OBD_FREE_PTR(info);
7363 }
7364
7365 static void osd_key_exit(const struct lu_context *ctx,
7366                          struct lu_context_key *key, void *data)
7367 {
7368         struct osd_thread_info *info = data;
7369
7370         LASSERT(info->oti_r_locks == 0);
7371         LASSERT(info->oti_w_locks == 0);
7372         LASSERT(info->oti_txns    == 0);
7373 }
7374
7375 /* type constructor/destructor: osd_type_init, osd_type_fini */
7376 LU_TYPE_INIT_FINI(osd, &osd_key);
7377
7378 struct lu_context_key osd_key = {
7379         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7380         .lct_init = osd_key_init,
7381         .lct_fini = osd_key_fini,
7382         .lct_exit = osd_key_exit
7383 };
7384
7385
7386 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7387                            const char *name, struct lu_device *next)
7388 {
7389         struct osd_device *osd = osd_dev(d);
7390
7391         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7392             sizeof(osd->od_svname))
7393                 return -E2BIG;
7394         return osd_procfs_init(osd, name);
7395 }
7396
7397 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7398 {
7399         struct seq_server_site *ss = osd_seq_site(osd);
7400         int rc;
7401
7402         ENTRY;
7403
7404         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7405                 RETURN(0);
7406
7407         if (unlikely(ss == NULL))
7408                 RETURN(-ENODEV);
7409
7410         OBD_ALLOC_PTR(osd->od_cl_seq);
7411         if (osd->od_cl_seq == NULL)
7412                 RETURN(-ENOMEM);
7413
7414         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7415                              osd->od_svname, ss->ss_server_seq);
7416         if (rc != 0) {
7417                 OBD_FREE_PTR(osd->od_cl_seq);
7418                 osd->od_cl_seq = NULL;
7419                 RETURN(rc);
7420         }
7421
7422         if (ss->ss_node_id == 0) {
7423                 /*
7424                  * If the OSD on the sequence controller(MDT0), then allocate
7425                  * sequence here, otherwise allocate sequence after connected
7426                  * to MDT0 (see mdt_register_lwp_callback()).
7427                  */
7428                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7429                                    &osd->od_cl_seq->lcs_space, env);
7430         }
7431
7432         RETURN(rc);
7433 }
7434
7435 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7436 {
7437         if (osd->od_cl_seq == NULL)
7438                 return;
7439
7440         seq_client_fini(osd->od_cl_seq);
7441         OBD_FREE_PTR(osd->od_cl_seq);
7442         osd->od_cl_seq = NULL;
7443 }
7444
7445 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7446 {
7447         ENTRY;
7448
7449         /* shutdown quota slave instance associated with the device */
7450         if (o->od_quota_slave != NULL) {
7451                 struct qsd_instance *qsd = o->od_quota_slave;
7452
7453                 o->od_quota_slave = NULL;
7454                 qsd_fini(env, qsd);
7455         }
7456
7457         osd_fid_fini(env, o);
7458         osd_scrub_cleanup(env, o);
7459
7460         RETURN(0);
7461 }
7462
7463 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7464 {
7465         ENTRY;
7466
7467         if (o->od_mnt != NULL) {
7468                 shrink_dcache_sb(osd_sb(o));
7469                 osd_sync(env, &o->od_dt_dev);
7470
7471                 mntput(o->od_mnt);
7472                 o->od_mnt = NULL;
7473         }
7474
7475         EXIT;
7476 }
7477
7478 static int osd_mount(const struct lu_env *env,
7479                      struct osd_device *o, struct lustre_cfg *cfg)
7480 {
7481         const char *name = lustre_cfg_string(cfg, 0);
7482         const char *dev = lustre_cfg_string(cfg, 1);
7483         const char *opts;
7484         unsigned long page, s_flags, lmd_flags = 0;
7485         struct page *__page;
7486         struct file_system_type *type;
7487         char *options = NULL;
7488         char *str;
7489         struct osd_thread_info *info = osd_oti_get(env);
7490         struct lu_fid *fid = &info->oti_fid;
7491         struct inode *inode;
7492         int rc = 0, force_over_512tb = 0;
7493
7494         ENTRY;
7495
7496         if (o->od_mnt != NULL)
7497                 RETURN(0);
7498
7499         if (strlen(dev) >= sizeof(o->od_mntdev))
7500                 RETURN(-E2BIG);
7501         strcpy(o->od_mntdev, dev);
7502
7503         str = lustre_cfg_string(cfg, 2);
7504         s_flags = simple_strtoul(str, NULL, 0);
7505         str = strstr(str, ":");
7506         if (str)
7507                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7508         opts = lustre_cfg_string(cfg, 3);
7509 #ifdef __BIG_ENDIAN
7510         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7511                 CERROR("%s: device %s extents feature is not guaranteed to "
7512                        "work on big-endian systems. Use \"bigendian_extents\" "
7513                        "mount option to override.\n", name, dev);
7514                 RETURN(-EINVAL);
7515         }
7516 #endif
7517 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7518         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7519                 CWARN("force_over_128tb option is deprecated. "
7520                       "Filesystems less than 512TB can be created without any "
7521                       "force options. Use force_over_512tb option for "
7522                       "filesystems greater than 512TB.\n");
7523         }
7524 #endif
7525 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7526         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7527                 CWARN("force_over_256tb option is deprecated. "
7528                       "Filesystems less than 512TB can be created without any "
7529                       "force options. Use force_over_512tb option for "
7530                       "filesystems greater than 512TB.\n");
7531         }
7532 #endif
7533
7534         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7535                 force_over_512tb = 1;
7536
7537         __page = alloc_page(GFP_KERNEL);
7538         if (__page == NULL)
7539                 GOTO(out, rc = -ENOMEM);
7540         page = (unsigned long)page_address(__page);
7541         options = (char *)page;
7542         *options = '\0';
7543         if (opts != NULL) {
7544                 /* strip out the options for back compatiblity */
7545                 static char *sout[] = {
7546                         "mballoc",
7547                         "iopen",
7548                         "noiopen",
7549                         "iopen_nopriv",
7550                         "extents",
7551                         "noextents",
7552                         /* strip out option we processed in osd */
7553                         "bigendian_extents",
7554                         "force_over_128tb",
7555                         "force_over_256tb",
7556                         "force_over_512tb",
7557                         NULL
7558                 };
7559                 strcat(options, opts);
7560                 for (rc = 0, str = options; sout[rc]; ) {
7561                         char *op = strstr(str, sout[rc]);
7562
7563                         if (op == NULL) {
7564                                 rc++;
7565                                 str = options;
7566                                 continue;
7567                         }
7568                         if (op == options || *(op - 1) == ',') {
7569                                 str = op + strlen(sout[rc]);
7570                                 if (*str == ',' || *str == '\0') {
7571                                         *str == ',' ? str++ : str;
7572                                         memmove(op, str, strlen(str) + 1);
7573                                 }
7574                         }
7575                         for (str = op; *str != ',' && *str != '\0'; str++)
7576                                 ;
7577                 }
7578         } else {
7579                 strncat(options, "user_xattr,acl", 14);
7580         }
7581
7582         /* Glom up mount options */
7583         if (*options != '\0')
7584                 strcat(options, ",");
7585         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7586
7587         type = get_fs_type("ldiskfs");
7588         if (!type) {
7589                 CERROR("%s: cannot find ldiskfs module\n", name);
7590                 GOTO(out, rc = -ENODEV);
7591         }
7592
7593         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7594         module_put(type->owner);
7595
7596         if (IS_ERR(o->od_mnt)) {
7597                 rc = PTR_ERR(o->od_mnt);
7598                 o->od_mnt = NULL;
7599                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7600                 GOTO(out, rc);
7601         }
7602
7603         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7604                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7605                                  force_over_512tb == 0) {
7606                 CERROR("%s: device %s LDISKFS does not support filesystems "
7607                        "greater than 512TB and can cause data corruption. "
7608                        "Use \"force_over_512tb\" mount option to override.\n",
7609                        name, dev);
7610                 GOTO(out_mnt, rc = -EINVAL);
7611         }
7612
7613         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7614                 if (priv_dev_set_rdonly) {
7615                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7616                         o->od_dt_dev.dd_rdonly = 1;
7617                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7618                                       name);
7619                 } else {
7620                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7621                                       name);
7622
7623                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7624                 }
7625         } else if (priv_dev_check_rdonly &&
7626                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7627                 CERROR("%s: underlying device %s is marked as "
7628                        "read-only. Setup failed\n", name, dev);
7629
7630                 GOTO(out_mnt, rc = -EROFS);
7631         }
7632
7633         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7634                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7635                 GOTO(out_mnt, rc = -EINVAL);
7636         }
7637
7638 #ifdef LDISKFS_MOUNT_DIRDATA
7639         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7640                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7641         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7642                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7643                       "enabling the dirdata feature. If you do not want to "
7644                       "downgrade to Lustre-1.x again, you can enable it via "
7645                       "'tune2fs -O dirdata device'\n", name, dev);
7646 #endif
7647         inode = osd_sb(o)->s_root->d_inode;
7648         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7649         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7650         if (rc != 0) {
7651                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7652                 GOTO(out_mnt, rc);
7653         }
7654
7655         if (lmd_flags & LMD_FLG_NOSCRUB)
7656                 o->od_auto_scrub_interval = AS_NEVER;
7657
7658         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7659                 /* do not use pagecache with flash-backed storage */
7660                 o->od_writethrough_cache = 0;
7661                 o->od_read_cache = 0;
7662         }
7663
7664         GOTO(out, rc = 0);
7665
7666 out_mnt:
7667         mntput(o->od_mnt);
7668         o->od_mnt = NULL;
7669
7670 out:
7671         if (__page)
7672                 __free_page(__page);
7673
7674         return rc;
7675 }
7676
7677 static struct lu_device *osd_device_fini(const struct lu_env *env,
7678                                          struct lu_device *d)
7679 {
7680         struct osd_device *o = osd_dev(d);
7681
7682         ENTRY;
7683
7684         osd_index_backup(env, o, false);
7685         osd_shutdown(env, o);
7686         osd_procfs_fini(o);
7687         osd_obj_map_fini(o);
7688         osd_umount(env, o);
7689
7690         RETURN(NULL);
7691 }
7692
7693 static int osd_device_init0(const struct lu_env *env,
7694                             struct osd_device *o,
7695                             struct lustre_cfg *cfg)
7696 {
7697         struct lu_device *l = osd2lu_dev(o);
7698         struct osd_thread_info *info;
7699         int rc;
7700         int cplen = 0;
7701
7702         /* if the module was re-loaded, env can loose its keys */
7703         rc = lu_env_refill((struct lu_env *)env);
7704         if (rc)
7705                 GOTO(out, rc);
7706         info = osd_oti_get(env);
7707         LASSERT(info);
7708
7709         l->ld_ops = &osd_lu_ops;
7710         o->od_dt_dev.dd_ops = &osd_dt_ops;
7711
7712         spin_lock_init(&o->od_osfs_lock);
7713         mutex_init(&o->od_otable_mutex);
7714         INIT_LIST_HEAD(&o->od_orphan_list);
7715         INIT_LIST_HEAD(&o->od_index_backup_list);
7716         INIT_LIST_HEAD(&o->od_index_restore_list);
7717         spin_lock_init(&o->od_lock);
7718         o->od_index_backup_policy = LIBP_NONE;
7719         o->od_t10_type = 0;
7720
7721         o->od_read_cache = 1;
7722         o->od_writethrough_cache = 1;
7723         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7724         o->od_auto_scrub_interval = AS_DEFAULT;
7725
7726         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7727                         sizeof(o->od_svname));
7728         if (cplen >= sizeof(o->od_svname)) {
7729                 rc = -E2BIG;
7730                 GOTO(out, rc);
7731         }
7732
7733         o->od_index_backup_stop = 0;
7734         o->od_index = -1; /* -1 means index is invalid */
7735         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7736         if (rc == LDD_F_SV_TYPE_OST)
7737                 o->od_is_ost = 1;
7738
7739         o->od_full_scrub_ratio = OFSR_DEFAULT;
7740         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7741         rc = osd_mount(env, o, cfg);
7742         if (rc != 0)
7743                 GOTO(out, rc);
7744
7745         rc = osd_obj_map_init(env, o);
7746         if (rc != 0)
7747                 GOTO(out_mnt, rc);
7748
7749         rc = lu_site_init(&o->od_site, l);
7750         if (rc != 0)
7751                 GOTO(out_compat, rc);
7752         o->od_site.ls_bottom_dev = l;
7753
7754         rc = lu_site_init_finish(&o->od_site);
7755         if (rc != 0)
7756                 GOTO(out_site, rc);
7757
7758         INIT_LIST_HEAD(&o->od_ios_list);
7759         /* setup scrub, including OI files initialization */
7760         o->od_in_init = 1;
7761         rc = osd_scrub_setup(env, o);
7762         o->od_in_init = 0;
7763         if (rc < 0)
7764                 GOTO(out_site, rc);
7765
7766         rc = osd_procfs_init(o, o->od_svname);
7767         if (rc != 0) {
7768                 CERROR("%s: can't initialize procfs: rc = %d\n",
7769                        o->od_svname, rc);
7770                 GOTO(out_scrub, rc);
7771         }
7772
7773         LASSERT(l->ld_site->ls_linkage.next != NULL);
7774         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7775
7776         /* initialize quota slave instance */
7777         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7778                                      o->od_proc_entry);
7779         if (IS_ERR(o->od_quota_slave)) {
7780                 rc = PTR_ERR(o->od_quota_slave);
7781                 o->od_quota_slave = NULL;
7782                 GOTO(out_procfs, rc);
7783         }
7784
7785         RETURN(0);
7786
7787 out_procfs:
7788         osd_procfs_fini(o);
7789 out_scrub:
7790         osd_scrub_cleanup(env, o);
7791 out_site:
7792         lu_site_fini(&o->od_site);
7793 out_compat:
7794         osd_obj_map_fini(o);
7795 out_mnt:
7796         osd_umount(env, o);
7797 out:
7798         return rc;
7799 }
7800
7801 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7802                                           struct lu_device_type *t,
7803                                           struct lustre_cfg *cfg)
7804 {
7805         struct osd_device *o;
7806         int rc;
7807
7808         OBD_ALLOC_PTR(o);
7809         if (o == NULL)
7810                 return ERR_PTR(-ENOMEM);
7811
7812         rc = dt_device_init(&o->od_dt_dev, t);
7813         if (rc == 0) {
7814                 /*
7815                  * Because the ctx might be revived in dt_device_init,
7816                  * refill the env here
7817                  */
7818                 lu_env_refill((struct lu_env *)env);
7819                 rc = osd_device_init0(env, o, cfg);
7820                 if (rc)
7821                         dt_device_fini(&o->od_dt_dev);
7822         }
7823
7824         if (unlikely(rc != 0))
7825                 OBD_FREE_PTR(o);
7826
7827         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7828 }
7829
7830 static struct lu_device *osd_device_free(const struct lu_env *env,
7831                                          struct lu_device *d)
7832 {
7833         struct osd_device *o = osd_dev(d);
7834
7835         ENTRY;
7836
7837         /* XXX: make osd top device in order to release reference */
7838         d->ld_site->ls_top_dev = d;
7839         lu_site_purge(env, d->ld_site, -1);
7840         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7841                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7842                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7843         }
7844         lu_site_fini(&o->od_site);
7845         dt_device_fini(&o->od_dt_dev);
7846         OBD_FREE_PTR(o);
7847         RETURN(NULL);
7848 }
7849
7850 static int osd_process_config(const struct lu_env *env,
7851                               struct lu_device *d, struct lustre_cfg *cfg)
7852 {
7853         struct osd_device *o = osd_dev(d);
7854         int rc;
7855
7856         ENTRY;
7857
7858         switch (cfg->lcfg_command) {
7859         case LCFG_SETUP:
7860                 rc = osd_mount(env, o, cfg);
7861                 break;
7862         case LCFG_CLEANUP:
7863                 /*
7864                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7865                  * that may happend if hit failure during mount process.
7866                  */
7867                 osd_index_backup(env, o, false);
7868                 lu_dev_del_linkage(d->ld_site, d);
7869                 rc = osd_shutdown(env, o);
7870                 break;
7871         case LCFG_PARAM:
7872                 LASSERT(&o->od_dt_dev);
7873                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7874                                               cfg, &o->od_dt_dev);
7875                 if (rc > 0 || rc == -ENOSYS) {
7876                         rc = class_process_proc_param(PARAM_OST,
7877                                                       lprocfs_osd_obd_vars,
7878                                                       cfg, &o->od_dt_dev);
7879                         if (rc > 0)
7880                                 rc = 0;
7881                 }
7882                 break;
7883         case LCFG_PRE_CLEANUP:
7884                 osd_scrub_stop(o);
7885                 osd_index_backup(env, o,
7886                                  o->od_index_backup_policy != LIBP_NONE);
7887                 rc = 0;
7888                 break;
7889         default:
7890                 rc = -ENOSYS;
7891         }
7892
7893         RETURN(rc);
7894 }
7895
7896 static int osd_recovery_complete(const struct lu_env *env,
7897                                  struct lu_device *d)
7898 {
7899         struct osd_device *osd = osd_dev(d);
7900         int rc = 0;
7901
7902         ENTRY;
7903
7904         if (osd->od_quota_slave == NULL)
7905                 RETURN(0);
7906
7907         /*
7908          * start qsd instance on recovery completion, this notifies the quota
7909          * slave code that we are about to process new requests now
7910          */
7911         rc = qsd_start(env, osd->od_quota_slave);
7912         RETURN(rc);
7913 }
7914
7915 /*
7916  * we use exports to track all osd users
7917  */
7918 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7919                            struct obd_device *obd, struct obd_uuid *cluuid,
7920                            struct obd_connect_data *data, void *localdata)
7921 {
7922         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7923         struct lustre_handle conn;
7924         int rc;
7925
7926         ENTRY;
7927
7928         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7929
7930         rc = class_connect(&conn, obd, cluuid);
7931         if (rc)
7932                 RETURN(rc);
7933
7934         *exp = class_conn2export(&conn);
7935
7936         spin_lock(&osd->od_osfs_lock);
7937         osd->od_connects++;
7938         spin_unlock(&osd->od_osfs_lock);
7939
7940         RETURN(0);
7941 }
7942
7943 /*
7944  * once last export (we don't count self-export) disappeared
7945  * osd can be released
7946  */
7947 static int osd_obd_disconnect(struct obd_export *exp)
7948 {
7949         struct obd_device *obd = exp->exp_obd;
7950         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7951         int rc, release = 0;
7952
7953         ENTRY;
7954
7955         /* Only disconnect the underlying layers on the final disconnect. */
7956         spin_lock(&osd->od_osfs_lock);
7957         osd->od_connects--;
7958         if (osd->od_connects == 0)
7959                 release = 1;
7960         spin_unlock(&osd->od_osfs_lock);
7961
7962         rc = class_disconnect(exp); /* bz 9811 */
7963
7964         if (rc == 0 && release)
7965                 class_manual_cleanup(obd);
7966         RETURN(rc);
7967 }
7968
7969 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7970                        struct lu_device *dev)
7971 {
7972         struct osd_device *osd = osd_dev(dev);
7973         struct lr_server_data *lsd =
7974                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7975         int result = 0;
7976
7977         ENTRY;
7978
7979         if (osd->od_quota_slave != NULL) {
7980                 /* set up quota slave objects */
7981                 result = qsd_prepare(env, osd->od_quota_slave);
7982                 if (result != 0)
7983                         RETURN(result);
7984         }
7985
7986         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7987 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7988                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7989                         osd->od_index_in_idif = 1;
7990                 } else {
7991                         osd->od_index_in_idif = 0;
7992                         result = osd_register_proc_index_in_idif(osd);
7993                         if (result != 0)
7994                                 RETURN(result);
7995                 }
7996 #else
7997                 osd->od_index_in_idif = 1;
7998 #endif
7999         }
8000
8001         result = osd_fid_init(env, osd);
8002
8003         RETURN(result);
8004 }
8005
8006 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8007                          struct lu_fid *fid, struct md_op_data *op_data)
8008 {
8009         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8010
8011         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8012 }
8013
8014 static const struct lu_object_operations osd_lu_obj_ops = {
8015         .loo_object_init      = osd_object_init,
8016         .loo_object_delete    = osd_object_delete,
8017         .loo_object_release   = osd_object_release,
8018         .loo_object_free      = osd_object_free,
8019         .loo_object_print     = osd_object_print,
8020         .loo_object_invariant = osd_object_invariant
8021 };
8022
8023 const struct lu_device_operations osd_lu_ops = {
8024         .ldo_object_alloc      = osd_object_alloc,
8025         .ldo_process_config    = osd_process_config,
8026         .ldo_recovery_complete = osd_recovery_complete,
8027         .ldo_prepare           = osd_prepare,
8028 };
8029
8030 static const struct lu_device_type_operations osd_device_type_ops = {
8031         .ldto_init = osd_type_init,
8032         .ldto_fini = osd_type_fini,
8033
8034         .ldto_start = osd_type_start,
8035         .ldto_stop  = osd_type_stop,
8036
8037         .ldto_device_alloc = osd_device_alloc,
8038         .ldto_device_free  = osd_device_free,
8039
8040         .ldto_device_init = osd_device_init,
8041         .ldto_device_fini = osd_device_fini
8042 };
8043
8044 static struct lu_device_type osd_device_type = {
8045         .ldt_tags     = LU_DEVICE_DT,
8046         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8047         .ldt_ops      = &osd_device_type_ops,
8048         .ldt_ctx_tags = LCT_LOCAL,
8049 };
8050
8051 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8052 {
8053         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8054         struct super_block *sb = osd_sb(osd);
8055
8056         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8057 }
8058
8059 /*
8060  * lprocfs legacy support.
8061  */
8062 static struct obd_ops osd_obd_device_ops = {
8063         .o_owner = THIS_MODULE,
8064         .o_connect      = osd_obd_connect,
8065         .o_disconnect   = osd_obd_disconnect,
8066         .o_fid_alloc    = osd_fid_alloc,
8067         .o_health_check = osd_health_check,
8068 };
8069
8070 static int __init osd_init(void)
8071 {
8072         int rc;
8073
8074         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8075 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8076         /* please, try to keep osd_thread_info smaller than a page */
8077         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8078 #endif
8079
8080         osd_oi_mod_init();
8081
8082         rc = lu_kmem_init(ldiskfs_caches);
8083         if (rc)
8084                 return rc;
8085
8086 #ifdef CONFIG_KALLSYMS
8087         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8088         priv_dev_check_rdonly =
8089                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8090 #endif
8091
8092         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8093                                  lprocfs_osd_module_vars,
8094                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8095         if (rc)
8096                 lu_kmem_fini(ldiskfs_caches);
8097         return rc;
8098 }
8099
8100 static void __exit osd_exit(void)
8101 {
8102         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8103         lu_kmem_fini(ldiskfs_caches);
8104 }
8105
8106 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8107 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8108 MODULE_VERSION(LUSTRE_VERSION_STRING);
8109 MODULE_LICENSE("GPL");
8110
8111 module_init(osd_init);
8112 module_exit(osd_exit);