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         /*
1898          * XXX temporary stuff. Some abstraction layer should
1899          * be used.
1900          */
1901         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1902         osd_th_started(oh);
1903         if (!IS_ERR(jh)) {
1904                 oh->ot_handle = jh;
1905                 LASSERT(oti->oti_txns == 0);
1906
1907                 lu_device_get(&d->dd_lu_dev);
1908                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1909                               "osd-tx", th);
1910                 oti->oti_txns++;
1911                 rc = 0;
1912         } else {
1913                 rc = PTR_ERR(jh);
1914         }
1915 out:
1916         RETURN(rc);
1917 }
1918
1919 static int osd_seq_exists(const struct lu_env *env,
1920                           struct osd_device *osd, u64 seq)
1921 {
1922         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1923         struct seq_server_site *ss = osd_seq_site(osd);
1924         int rc;
1925
1926         ENTRY;
1927
1928         LASSERT(ss != NULL);
1929         LASSERT(ss->ss_server_fld != NULL);
1930
1931         rc = osd_fld_lookup(env, osd, seq, range);
1932         if (rc != 0) {
1933                 if (rc != -ENOENT)
1934                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1935                                osd_name(osd), seq, rc);
1936                 RETURN(0);
1937         }
1938
1939         RETURN(ss->ss_node_id == range->lsr_index);
1940 }
1941
1942 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1943 {
1944         struct dt_txn_commit_cb *dcb;
1945         struct dt_txn_commit_cb *tmp;
1946
1947         /* call per-transaction stop callbacks if any */
1948         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1949                                  dcb_linkage) {
1950                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1951                          "commit callback entry: magic=%x name='%s'\n",
1952                          dcb->dcb_magic, dcb->dcb_name);
1953                 list_del_init(&dcb->dcb_linkage);
1954                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1955         }
1956 }
1957
1958 /*
1959  * Concurrency: shouldn't matter.
1960  */
1961 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1962                           struct thandle *th)
1963 {
1964         struct osd_thread_info *oti = osd_oti_get(env);
1965         struct osd_thandle *oh;
1966         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1967         struct osd_device *osd = osd_dt_dev(th->th_dev);
1968         struct qsd_instance *qsd = osd->od_quota_slave;
1969         struct lquota_trans *qtrans;
1970         struct list_head truncates = LIST_HEAD_INIT(truncates);
1971         int rc = 0, remove_agents = 0;
1972
1973         ENTRY;
1974
1975         oh = container_of0(th, struct osd_thandle, ot_super);
1976
1977         remove_agents = oh->ot_remove_agents;
1978
1979         qtrans = oh->ot_quota_trans;
1980         oh->ot_quota_trans = NULL;
1981
1982         /* move locks to local list, stop tx, execute truncates */
1983         list_splice(&oh->ot_trunc_locks, &truncates);
1984
1985         if (oh->ot_handle != NULL) {
1986                 int rc2;
1987
1988                 handle_t *hdl = oh->ot_handle;
1989
1990                 /*
1991                  * add commit callback
1992                  * notice we don't do this in osd_trans_start()
1993                  * as underlying transaction can change during truncate
1994                  */
1995                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1996                                              &oh->ot_jcb);
1997
1998                 LASSERT(oti->oti_txns == 1);
1999                 oti->oti_txns--;
2000
2001                 rc = dt_txn_hook_stop(env, th);
2002                 if (rc != 0)
2003                         CERROR("%s: failed in transaction hook: rc = %d\n",
2004                                osd_name(osd), rc);
2005
2006                 osd_trans_stop_cb(oh, rc);
2007                 /* hook functions might modify th_sync */
2008                 hdl->h_sync = th->th_sync;
2009
2010                 oh->ot_handle = NULL;
2011                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2012                 if (rc2 != 0)
2013                         CERROR("%s: failed to stop transaction: rc = %d\n",
2014                                osd_name(osd), rc2);
2015                 if (!rc)
2016                         rc = rc2;
2017
2018                 osd_process_truncates(&truncates);
2019         } else {
2020                 osd_trans_stop_cb(oh, th->th_result);
2021                 OBD_FREE_PTR(oh);
2022         }
2023
2024         osd_trunc_unlock_all(&truncates);
2025
2026         /* inform the quota slave device that the transaction is stopping */
2027         qsd_op_end(env, qsd, qtrans);
2028
2029         /*
2030          * as we want IO to journal and data IO be concurrent, we don't block
2031          * awaiting data IO completion in osd_do_bio(), instead we wait here
2032          * once transaction is submitted to the journal. all reqular requests
2033          * don't do direct IO (except read/write), thus this wait_event becomes
2034          * no-op for them.
2035          *
2036          * IMPORTANT: we have to wait till any IO submited by the thread is
2037          * completed otherwise iobuf may be corrupted by different request
2038          */
2039         wait_event(iobuf->dr_wait,
2040                        atomic_read(&iobuf->dr_numreqs) == 0);
2041         osd_fini_iobuf(osd, iobuf);
2042         if (!rc)
2043                 rc = iobuf->dr_error;
2044
2045         if (unlikely(remove_agents != 0))
2046                 osd_process_scheduled_agent_removals(env, osd);
2047
2048         oti->oti_ins_cache_depth--;
2049         /* reset OI cache for safety */
2050         if (oti->oti_ins_cache_depth == 0)
2051                 oti->oti_ins_cache_used = 0;
2052
2053         sb_end_write(osd_sb(osd));
2054
2055         RETURN(rc);
2056 }
2057
2058 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2059 {
2060         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2061                                                ot_super);
2062
2063         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2064         LASSERT(&dcb->dcb_func != NULL);
2065         if (dcb->dcb_flags & DCB_TRANS_STOP)
2066                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2067         else
2068                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2069
2070         return 0;
2071 }
2072
2073 /*
2074  * Called just before object is freed. Releases all resources except for
2075  * object itself (that is released by osd_object_free()).
2076  *
2077  * Concurrency: no concurrent access is possible that late in object
2078  * life-cycle.
2079  */
2080 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2081 {
2082         struct osd_object *obj = osd_obj(l);
2083         struct inode *inode = obj->oo_inode;
2084
2085         LINVRNT(osd_invariant(obj));
2086
2087         /*
2088          * If object is unlinked remove fid->ino mapping from object index.
2089          */
2090
2091         osd_index_fini(obj);
2092         if (inode != NULL) {
2093                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
2094                 qid_t                    uid = i_uid_read(inode);
2095                 qid_t                    gid = i_gid_read(inode);
2096
2097                 obj->oo_inode = NULL;
2098                 iput(inode);
2099                 if (!obj->oo_header && qsd) {
2100                         struct osd_thread_info *info = osd_oti_get(env);
2101                         struct lquota_id_info *qi = &info->oti_qi;
2102
2103                         /* Release granted quota to master if necessary */
2104                         qi->lqi_id.qid_uid = uid;
2105                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2106
2107                         qi->lqi_id.qid_uid = gid;
2108                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2109
2110                         qi->lqi_id.qid_uid = i_projid_read(inode);
2111                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2112                 }
2113         }
2114 }
2115
2116 /*
2117  * Concurrency: ->loo_object_release() is called under site spin-lock.
2118  */
2119 static void osd_object_release(const struct lu_env *env,
2120                                struct lu_object *l)
2121 {
2122         struct osd_object *o = osd_obj(l);
2123
2124         /*
2125          * nobody should be releasing a non-destroyed object with nlink=0
2126          * the API allows this, but ldiskfs doesn't like and then report
2127          * this inode as deleted
2128          */
2129         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2130                   o->oo_inode->i_nlink == 0));
2131 }
2132
2133 /*
2134  * Concurrency: shouldn't matter.
2135  */
2136 static int osd_object_print(const struct lu_env *env, void *cookie,
2137                             lu_printer_t p, const struct lu_object *l)
2138 {
2139         struct osd_object *o = osd_obj(l);
2140         struct iam_descr *d;
2141
2142         if (o->oo_dir != NULL)
2143                 d = o->oo_dir->od_container.ic_descr;
2144         else
2145                 d = NULL;
2146         return (*p)(env, cookie,
2147                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2148                     o, o->oo_inode,
2149                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2150                     o->oo_inode ? o->oo_inode->i_generation : 0,
2151                     d ? d->id_ops->id_name : "plain");
2152 }
2153
2154 /*
2155  * Concurrency: shouldn't matter.
2156  */
2157 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2158                 struct obd_statfs *sfs)
2159 {
2160         struct osd_device *osd = osd_dt_dev(d);
2161         struct super_block *sb = osd_sb(osd);
2162         struct kstatfs *ksfs;
2163         __u64 reserved;
2164         int result = 0;
2165
2166         if (unlikely(osd->od_mnt == NULL))
2167                 return -EINPROGRESS;
2168
2169         /* osd_lproc.c call this without env, allocate ksfs for that case */
2170         if (unlikely(env == NULL)) {
2171                 OBD_ALLOC_PTR(ksfs);
2172                 if (ksfs == NULL)
2173                         return -ENOMEM;
2174         } else {
2175                 ksfs = &osd_oti_get(env)->oti_ksfs;
2176         }
2177
2178         result = sb->s_op->statfs(sb->s_root, ksfs);
2179         if (result)
2180                 goto out;
2181
2182         statfs_pack(sfs, ksfs);
2183         if (unlikely(sb->s_flags & MS_RDONLY))
2184                 sfs->os_state |= OS_STATE_READONLY;
2185         if (ldiskfs_has_feature_extents(sb))
2186                 sfs->os_maxbytes = sb->s_maxbytes;
2187         else
2188                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2189
2190         /*
2191          * Reserve some space so to avoid fragmenting the filesystem too much.
2192          * Fragmentation not only impacts performance, but can also increase
2193          * metadata overhead significantly, causing grant calculation to be
2194          * wrong.
2195          *
2196          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2197          */
2198         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2199         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2200         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2201                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2202
2203         sfs->os_blocks -= reserved;
2204         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2205         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2206
2207 out:
2208         if (unlikely(env == NULL))
2209                 OBD_FREE_PTR(ksfs);
2210         return result;
2211 }
2212
2213 /**
2214  * Estimate space needed for file creations. We assume the largest filename
2215  * which is 2^64 - 1, hence a filename of 20 chars.
2216  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2217  */
2218 #ifdef __LDISKFS_DIR_REC_LEN
2219 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2220 #else
2221 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2222 #endif
2223
2224 /*
2225  * Concurrency: doesn't access mutable data.
2226  */
2227 static void osd_conf_get(const struct lu_env *env,
2228                          const struct dt_device *dev,
2229                          struct dt_device_param *param)
2230 {
2231         struct osd_device *d = osd_dt_dev(dev);
2232         struct super_block *sb = osd_sb(d);
2233         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2234         const char *name;
2235         int ea_overhead;
2236
2237         /*
2238          * XXX should be taken from not-yet-existing fs abstraction layer.
2239          */
2240         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2241         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2242         param->ddp_symlink_max  = sb->s_blocksize;
2243         param->ddp_mount_type   = LDD_MT_LDISKFS;
2244         if (ldiskfs_has_feature_extents(sb))
2245                 param->ddp_maxbytes = sb->s_maxbytes;
2246         else
2247                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2248         /*
2249          * inode are statically allocated, so per-inode space consumption
2250          * is the space consumed by the directory entry
2251          */
2252         param->ddp_inodespace     = PER_OBJ_USAGE;
2253         /*
2254          * EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2255          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2256          * maximum length to not under count the actual number of extents
2257          * needed for writing a file.
2258          */
2259         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2260         /* worst-case extent insertion metadata overhead */
2261         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2262         param->ddp_mntopts = 0;
2263         if (test_opt(sb, XATTR_USER))
2264                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2265         if (test_opt(sb, POSIX_ACL))
2266                 param->ddp_mntopts |= MNTOPT_ACL;
2267
2268         /*
2269          * LOD might calculate the max stripe count based on max_ea_size,
2270          * so we need take account in the overhead as well,
2271          * xattr_header + magic + xattr_entry_head
2272          */
2273         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2274                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2275
2276 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2277         if (ldiskfs_has_feature_ea_inode(sb))
2278                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2279                                                                 ea_overhead;
2280         else
2281 #endif
2282                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2283
2284         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2285                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2286
2287         /*
2288          * Preferred RPC size for efficient disk IO.  4MB shows good
2289          * all-around performance for ldiskfs, but use bigalloc chunk size
2290          * by default if larger.
2291          */
2292 #if defined(LDISKFS_CLUSTER_SIZE)
2293         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2294                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2295         else
2296 #endif
2297                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2298
2299         param->ddp_t10_cksum_type = 0;
2300         if (bi) {
2301                 unsigned short interval = blk_integrity_interval(bi);
2302                 name = blk_integrity_name(bi);
2303                 /*
2304                  * Expected values:
2305                  * T10-DIF-TYPE1-CRC
2306                  * T10-DIF-TYPE3-CRC
2307                  * T10-DIF-TYPE1-IP
2308                  * T10-DIF-TYPE3-IP
2309                  */
2310                 if (strncmp(name, "T10-DIF-TYPE",
2311                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2312                         /* also skip "1/3-" at end */
2313                         const int type_off = sizeof("T10-DIF-TYPE.");
2314
2315                         if (interval != 512 && interval != 4096)
2316                                 CERROR("%s: unsupported T10PI sector size %u\n",
2317                                        d->od_svname, interval);
2318                         else if (strcmp(name + type_off, "CRC") == 0)
2319                                 param->ddp_t10_cksum_type = interval == 512 ?
2320                                         OBD_CKSUM_T10CRC512 :
2321                                         OBD_CKSUM_T10CRC4K;
2322                         else if (strcmp(name + type_off, "IP") == 0)
2323                                 param->ddp_t10_cksum_type = interval == 512 ?
2324                                         OBD_CKSUM_T10IP512 :
2325                                         OBD_CKSUM_T10IP4K;
2326                         else
2327                                 CERROR("%s: unsupported checksum type of "
2328                                        "T10PI type '%s'",
2329                                        d->od_svname, name);
2330                 } else {
2331                         CERROR("%s: unsupported T10PI type '%s'",
2332                                d->od_svname, name);
2333                 }
2334         }
2335 }
2336
2337 /*
2338  * Concurrency: shouldn't matter.
2339  */
2340 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2341 {
2342         int rc;
2343         struct super_block *s = osd_sb(osd_dt_dev(d));
2344         ENTRY;
2345
2346         down_read(&s->s_umount);
2347         rc = s->s_op->sync_fs(s, 1);
2348         up_read(&s->s_umount);
2349
2350         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2351                rc);
2352
2353         return rc;
2354 }
2355
2356 /**
2357  * Start commit for OSD device.
2358  *
2359  * An implementation of dt_commit_async method for OSD device.
2360  * Asychronously starts underlayng fs sync and thereby a transaction
2361  * commit.
2362  *
2363  * \param env environment
2364  * \param d dt device
2365  *
2366  * \see dt_device_operations
2367  */
2368 static int osd_commit_async(const struct lu_env *env,
2369                             struct dt_device *d)
2370 {
2371         struct super_block *s = osd_sb(osd_dt_dev(d));
2372         int rc;
2373
2374         ENTRY;
2375
2376         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2377         down_read(&s->s_umount);
2378         rc = s->s_op->sync_fs(s, 0);
2379         up_read(&s->s_umount);
2380
2381         RETURN(rc);
2382 }
2383
2384 /* Our own copy of the set readonly functions if present, or NU if not. */
2385 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2386 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2387 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2388
2389 /*
2390  * Concurrency: shouldn't matter.
2391  */
2392 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2393 {
2394         struct super_block *sb = osd_sb(osd_dt_dev(d));
2395         struct block_device *dev = sb->s_bdev;
2396         int rc = -EOPNOTSUPP;
2397
2398         ENTRY;
2399
2400         if (priv_dev_set_rdonly) {
2401                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2402
2403                 rc = 0;
2404                 CERROR("*** setting %s read-only ***\n",
2405                        osd_dt_dev(d)->od_svname);
2406
2407                 if (sb->s_op->freeze_fs) {
2408                         rc = sb->s_op->freeze_fs(sb);
2409                         if (rc)
2410                                 goto out;
2411                 }
2412
2413                 if (jdev && (jdev != dev)) {
2414                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2415                                (long)jdev);
2416                         priv_dev_set_rdonly(jdev);
2417                 }
2418                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2419                 priv_dev_set_rdonly(dev);
2420
2421                 if (sb->s_op->unfreeze_fs)
2422                         sb->s_op->unfreeze_fs(sb);
2423         }
2424
2425 out:
2426         if (rc)
2427                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2428                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2429
2430         RETURN(rc);
2431 }
2432
2433 /**
2434  * Note: we do not count into QUOTA here.
2435  * If we mount with --data_journal we may need more.
2436  */
2437 const int osd_dto_credits_noquota[DTO_NR] = {
2438         /**
2439          * Insert.
2440          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2441          * SINGLEDATA_TRANS_BLOCKS(8)
2442          * XXX Note: maybe iam need more, since iam have more level than
2443          *           EXT3 htree.
2444          */
2445         [DTO_INDEX_INSERT]  = 16,
2446         /**
2447          * Delete
2448          * just modify a single entry, probably merge few within a block
2449          */
2450         [DTO_INDEX_DELETE]  = 1,
2451         /**
2452          * Used for OI scrub
2453          */
2454         [DTO_INDEX_UPDATE]  = 16,
2455         /**
2456          * 4(inode, inode bits, groups, GDT)
2457          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2458          */
2459         [DTO_OBJECT_CREATE] = 4,
2460         /**
2461          * 4(inode, inode bits, groups, GDT)
2462          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2463          */
2464         [DTO_OBJECT_DELETE] = 4,
2465         /**
2466          * Attr set credits (inode)
2467          */
2468         [DTO_ATTR_SET_BASE] = 1,
2469         /**
2470          * Xattr set. The same as xattr of EXT3.
2471          * DATA_TRANS_BLOCKS(14)
2472          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2473          * are also counted in. Do not know why?
2474          */
2475         [DTO_XATTR_SET]     = 14,
2476         /**
2477          * credits for inode change during write.
2478          */
2479         [DTO_WRITE_BASE]    = 3,
2480         /**
2481          * credits for single block write.
2482          */
2483         [DTO_WRITE_BLOCK]   = 14,
2484         /**
2485          * Attr set credits for chown.
2486          * This is extra credits for setattr, and it is null without quota
2487          */
2488         [DTO_ATTR_SET_CHOWN] = 0
2489 };
2490
2491 static const struct dt_device_operations osd_dt_ops = {
2492         .dt_root_get       = osd_root_get,
2493         .dt_statfs         = osd_statfs,
2494         .dt_trans_create   = osd_trans_create,
2495         .dt_trans_start    = osd_trans_start,
2496         .dt_trans_stop     = osd_trans_stop,
2497         .dt_trans_cb_add   = osd_trans_cb_add,
2498         .dt_conf_get       = osd_conf_get,
2499         .dt_sync           = osd_sync,
2500         .dt_ro             = osd_ro,
2501         .dt_commit_async   = osd_commit_async,
2502 };
2503
2504 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2505                           unsigned int role)
2506 {
2507         struct osd_object *obj = osd_dt_obj(dt);
2508         struct osd_thread_info *oti = osd_oti_get(env);
2509
2510         LINVRNT(osd_invariant(obj));
2511
2512         LASSERT(obj->oo_owner != env);
2513         down_read_nested(&obj->oo_sem, role);
2514
2515         LASSERT(obj->oo_owner == NULL);
2516         oti->oti_r_locks++;
2517 }
2518
2519 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2520                            unsigned int role)
2521 {
2522         struct osd_object *obj = osd_dt_obj(dt);
2523         struct osd_thread_info *oti = osd_oti_get(env);
2524
2525         LINVRNT(osd_invariant(obj));
2526
2527         LASSERT(obj->oo_owner != env);
2528         down_write_nested(&obj->oo_sem, role);
2529
2530         LASSERT(obj->oo_owner == NULL);
2531         obj->oo_owner = env;
2532         oti->oti_w_locks++;
2533 }
2534
2535 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2536 {
2537         struct osd_object *obj = osd_dt_obj(dt);
2538         struct osd_thread_info *oti = osd_oti_get(env);
2539
2540         LINVRNT(osd_invariant(obj));
2541
2542         LASSERT(oti->oti_r_locks > 0);
2543         oti->oti_r_locks--;
2544         up_read(&obj->oo_sem);
2545 }
2546
2547 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2548 {
2549         struct osd_object *obj = osd_dt_obj(dt);
2550         struct osd_thread_info *oti = osd_oti_get(env);
2551
2552         LINVRNT(osd_invariant(obj));
2553
2554         LASSERT(obj->oo_owner == env);
2555         LASSERT(oti->oti_w_locks > 0);
2556         oti->oti_w_locks--;
2557         obj->oo_owner = NULL;
2558         up_write(&obj->oo_sem);
2559 }
2560
2561 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2562 {
2563         struct osd_object *obj = osd_dt_obj(dt);
2564
2565         LINVRNT(osd_invariant(obj));
2566
2567         return obj->oo_owner == env;
2568 }
2569
2570 static struct timespec *osd_inode_time(const struct lu_env *env,
2571                                        struct inode *inode, __u64 seconds)
2572 {
2573         struct osd_thread_info *oti = osd_oti_get(env);
2574         struct timespec *t = &oti->oti_time;
2575
2576         t->tv_sec = seconds;
2577         t->tv_nsec = 0;
2578         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2579         return t;
2580 }
2581
2582 static void osd_inode_getattr(const struct lu_env *env,
2583                               struct inode *inode, struct lu_attr *attr)
2584 {
2585         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2586                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2587                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2588                            LA_BLKSIZE | LA_TYPE;
2589
2590         attr->la_atime   = LTIME_S(inode->i_atime);
2591         attr->la_mtime   = LTIME_S(inode->i_mtime);
2592         attr->la_ctime   = LTIME_S(inode->i_ctime);
2593         attr->la_mode    = inode->i_mode;
2594         attr->la_size    = i_size_read(inode);
2595         attr->la_blocks  = inode->i_blocks;
2596         attr->la_uid     = i_uid_read(inode);
2597         attr->la_gid     = i_gid_read(inode);
2598         attr->la_projid  = i_projid_read(inode);
2599         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2600         attr->la_nlink   = inode->i_nlink;
2601         attr->la_rdev    = inode->i_rdev;
2602         attr->la_blksize = 1 << inode->i_blkbits;
2603         attr->la_blkbits = inode->i_blkbits;
2604         /*
2605          * Ext4 did not transfer inherit flags from raw inode
2606          * to inode flags, and ext4 internally test raw inode
2607          * @i_flags directly. Instead of patching ext4, we do it here.
2608          */
2609         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2610                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2611 }
2612
2613 static int osd_attr_get(const struct lu_env *env,
2614                         struct dt_object *dt,
2615                         struct lu_attr *attr)
2616 {
2617         struct osd_object *obj = osd_dt_obj(dt);
2618
2619         if (unlikely(!dt_object_exists(dt)))
2620                 return -ENOENT;
2621         if (unlikely(obj->oo_destroyed))
2622                 return -ENOENT;
2623
2624         LASSERT(!dt_object_remote(dt));
2625         LINVRNT(osd_invariant(obj));
2626
2627         spin_lock(&obj->oo_guard);
2628         osd_inode_getattr(env, obj->oo_inode, attr);
2629         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2630                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2631         spin_unlock(&obj->oo_guard);
2632
2633         return 0;
2634 }
2635
2636 static int osd_declare_attr_qid(const struct lu_env *env,
2637                                 struct osd_object *obj,
2638                                 struct osd_thandle *oh, long long bspace,
2639                                 qid_t old_id, qid_t new_id, bool enforce,
2640                                 unsigned int type, bool ignore_edquot)
2641 {
2642         int rc;
2643         struct osd_thread_info *info = osd_oti_get(env);
2644         struct lquota_id_info  *qi = &info->oti_qi;
2645
2646         qi->lqi_type = type;
2647         /* inode accounting */
2648         qi->lqi_is_blk = false;
2649
2650         /* one more inode for the new id ... */
2651         qi->lqi_id.qid_uid = new_id;
2652         qi->lqi_space      = 1;
2653         /* Reserve credits for the new id */
2654         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2655         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2656                 rc = 0;
2657         if (rc)
2658                 RETURN(rc);
2659
2660         /* and one less inode for the current id */
2661         qi->lqi_id.qid_uid = old_id;
2662         qi->lqi_space = -1;
2663         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2664         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2665                 rc = 0;
2666         if (rc)
2667                 RETURN(rc);
2668
2669         /* block accounting */
2670         qi->lqi_is_blk = true;
2671
2672         /* more blocks for the new id ... */
2673         qi->lqi_id.qid_uid = new_id;
2674         qi->lqi_space      = bspace;
2675         /*
2676          * Credits for the new uid has been reserved, re-use "obj"
2677          * to save credit reservation.
2678          */
2679         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2680         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2681                 rc = 0;
2682         if (rc)
2683                 RETURN(rc);
2684
2685         /* and finally less blocks for the current uid */
2686         qi->lqi_id.qid_uid = old_id;
2687         qi->lqi_space      = -bspace;
2688         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2689         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2690                 rc = 0;
2691
2692         RETURN(rc);
2693 }
2694
2695 static int osd_declare_attr_set(const struct lu_env *env,
2696                                 struct dt_object *dt,
2697                                 const struct lu_attr *attr,
2698                                 struct thandle *handle)
2699 {
2700         struct osd_thandle *oh;
2701         struct osd_object *obj;
2702         qid_t uid;
2703         qid_t gid;
2704         long long bspace;
2705         int rc = 0;
2706         bool enforce;
2707
2708         ENTRY;
2709
2710         LASSERT(dt != NULL);
2711         LASSERT(handle != NULL);
2712
2713         obj = osd_dt_obj(dt);
2714         LASSERT(osd_invariant(obj));
2715
2716         oh = container_of0(handle, struct osd_thandle, ot_super);
2717         LASSERT(oh->ot_handle == NULL);
2718
2719         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2720                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2721
2722         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2723                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2724
2725         if (attr == NULL || obj->oo_inode == NULL)
2726                 RETURN(rc);
2727
2728         bspace   = obj->oo_inode->i_blocks << 9;
2729         bspace   = toqb(bspace);
2730
2731         /*
2732          * Changing ownership is always preformed by super user, it should not
2733          * fail with EDQUOT unless required explicitly.
2734          *
2735          * We still need to call the osd_declare_qid() to calculate the journal
2736          * credits for updating quota accounting files and to trigger quota
2737          * space adjustment once the operation is completed.
2738          */
2739         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2740                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2741
2742                 if (!ignore_edquot)
2743                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2744                                "(the quota space is %lld)\n",
2745                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2746                                attr->la_gid, bspace);
2747
2748                 /* USERQUOTA */
2749                 uid = i_uid_read(obj->oo_inode);
2750                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2751                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2752                                           attr->la_uid, enforce, USRQUOTA,
2753                                           true);
2754                 if (rc)
2755                         RETURN(rc);
2756
2757                 gid = i_gid_read(obj->oo_inode);
2758                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2759                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2760                                           i_gid_read(obj->oo_inode),
2761                                           attr->la_gid, enforce, GRPQUOTA,
2762                                           ignore_edquot);
2763                 if (rc)
2764                         RETURN(rc);
2765
2766         }
2767 #ifdef HAVE_PROJECT_QUOTA
2768         if (attr->la_valid & LA_PROJID) {
2769                 __u32 projid = i_projid_read(obj->oo_inode);
2770
2771                 enforce = (attr->la_valid & LA_PROJID) &&
2772                                         (attr->la_projid != projid);
2773                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2774                                           (qid_t)projid, (qid_t)attr->la_projid,
2775                                           enforce, PRJQUOTA, true);
2776                 if (rc)
2777                         RETURN(rc);
2778         }
2779 #endif
2780         RETURN(rc);
2781 }
2782
2783 static int osd_inode_setattr(const struct lu_env *env,
2784                              struct inode *inode, const struct lu_attr *attr)
2785 {
2786         __u64 bits = attr->la_valid;
2787
2788         /* Only allow set size for regular file */
2789         if (!S_ISREG(inode->i_mode))
2790                 bits &= ~(LA_SIZE | LA_BLOCKS);
2791
2792         if (bits == 0)
2793                 return 0;
2794
2795         if (bits & LA_ATIME)
2796                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2797         if (bits & LA_CTIME)
2798                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2799         if (bits & LA_MTIME)
2800                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2801         if (bits & LA_SIZE) {
2802                 spin_lock(&inode->i_lock);
2803                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2804                 i_size_write(inode, attr->la_size);
2805                 spin_unlock(&inode->i_lock);
2806         }
2807
2808         /*
2809          * OSD should not change "i_blocks" which is used by quota.
2810          * "i_blocks" should be changed by ldiskfs only.
2811          */
2812         if (bits & LA_MODE)
2813                 inode->i_mode = (inode->i_mode & S_IFMT) |
2814                                 (attr->la_mode & ~S_IFMT);
2815         if (bits & LA_UID)
2816                 i_uid_write(inode, attr->la_uid);
2817         if (bits & LA_GID)
2818                 i_gid_write(inode, attr->la_gid);
2819         if (bits & LA_PROJID)
2820                 i_projid_write(inode, attr->la_projid);
2821         if (bits & LA_NLINK)
2822                 set_nlink(inode, attr->la_nlink);
2823         if (bits & LA_RDEV)
2824                 inode->i_rdev = attr->la_rdev;
2825
2826         if (bits & LA_FLAGS) {
2827                 /* always keep S_NOCMTIME */
2828                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2829                                  S_NOCMTIME;
2830                 /*
2831                  * Ext4 did not transfer inherit flags from
2832                  * @inode->i_flags to raw inode i_flags when writing
2833                  * flags, we do it explictly here.
2834                  */
2835                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2836                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2837                 else
2838                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2839         }
2840         return 0;
2841 }
2842
2843 #ifdef HAVE_PROJECT_QUOTA
2844 static int osd_transfer_project(struct inode *inode, __u32 projid)
2845 {
2846         struct super_block *sb = inode->i_sb;
2847         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2848         int err;
2849         kprojid_t kprojid;
2850         struct ldiskfs_iloc iloc;
2851         struct ldiskfs_inode *raw_inode;
2852         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2853
2854         if (!ldiskfs_has_feature_project(sb)) {
2855                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2856                         == LDISKFS_DEF_PROJID);
2857                 if (projid != LDISKFS_DEF_PROJID)
2858                         return -EOPNOTSUPP;
2859                 else
2860                         return 0;
2861         }
2862
2863         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2864                 return -EOPNOTSUPP;
2865
2866         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2867         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2868                 return 0;
2869
2870         err = ldiskfs_get_inode_loc(inode, &iloc);
2871         if (err)
2872                 return err;
2873
2874         raw_inode = ldiskfs_raw_inode(&iloc);
2875         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2876                 err = -EOVERFLOW;
2877                 brelse(iloc.bh);
2878                 return err;
2879         }
2880         brelse(iloc.bh);
2881
2882         dquot_initialize(inode);
2883         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2884         if (transfer_to[PRJQUOTA]) {
2885                 err = __dquot_transfer(inode, transfer_to);
2886                 dqput(transfer_to[PRJQUOTA]);
2887                 if (err)
2888                         return err;
2889         }
2890
2891         return err;
2892 }
2893 #endif
2894
2895 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2896 {
2897         int rc;
2898
2899         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2900             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2901                 struct iattr iattr;
2902
2903                 ll_vfs_dq_init(inode);
2904                 iattr.ia_valid = 0;
2905                 if (attr->la_valid & LA_UID)
2906                         iattr.ia_valid |= ATTR_UID;
2907                 if (attr->la_valid & LA_GID)
2908                         iattr.ia_valid |= ATTR_GID;
2909                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2910                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2911
2912                 rc = ll_vfs_dq_transfer(inode, &iattr);
2913                 if (rc) {
2914                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2915                                "enforcement enabled on the ldiskfs "
2916                                "filesystem?\n", inode->i_sb->s_id, rc);
2917                         return rc;
2918                 }
2919         }
2920
2921         /* Handle project id transfer here properly */
2922         if (attr->la_valid & LA_PROJID &&
2923             attr->la_projid != i_projid_read(inode)) {
2924 #ifdef HAVE_PROJECT_QUOTA
2925                 rc = osd_transfer_project(inode, attr->la_projid);
2926 #else
2927                 rc = -ENOTSUPP;
2928 #endif
2929                 if (rc) {
2930                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2931                                "enforcement enabled on the ldiskfs "
2932                                "filesystem?\n", inode->i_sb->s_id, rc);
2933                         return rc;
2934                 }
2935         }
2936         return 0;
2937 }
2938
2939 static int osd_attr_set(const struct lu_env *env,
2940                         struct dt_object *dt,
2941                         const struct lu_attr *attr,
2942                         struct thandle *handle)
2943 {
2944         struct osd_object *obj = osd_dt_obj(dt);
2945         struct inode *inode;
2946         int rc;
2947
2948         if (!dt_object_exists(dt))
2949                 return -ENOENT;
2950
2951         LASSERT(handle != NULL);
2952         LASSERT(!dt_object_remote(dt));
2953         LASSERT(osd_invariant(obj));
2954
2955         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2956
2957         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2958             !osd_obj2dev(obj)->od_is_ost) {
2959                 struct osd_thread_info *oti = osd_oti_get(env);
2960                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2961                 struct lu_fid *fid1 = &oti->oti_fid;
2962                 struct osd_inode_id *id = &oti->oti_id;
2963                 struct iam_path_descr *ipd;
2964                 struct iam_container *bag;
2965                 struct osd_thandle *oh;
2966                 int rc;
2967
2968                 fid_cpu_to_be(fid1, fid0);
2969                 memset(id, 1, sizeof(*id));
2970                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2971                                   fid0)->oi_dir.od_container;
2972                 ipd = osd_idx_ipd_get(env, bag);
2973                 if (unlikely(ipd == NULL))
2974                         RETURN(-ENOMEM);
2975
2976                 oh = container_of0(handle, struct osd_thandle, ot_super);
2977                 rc = iam_update(oh->ot_handle, bag,
2978                                 (const struct iam_key *)fid1,
2979                                 (const struct iam_rec *)id, ipd);
2980                 osd_ipd_put(env, bag, ipd);
2981                 return(rc > 0 ? 0 : rc);
2982         }
2983
2984         inode = obj->oo_inode;
2985
2986         rc = osd_quota_transfer(inode, attr);
2987         if (rc)
2988                 return rc;
2989
2990         spin_lock(&obj->oo_guard);
2991         rc = osd_inode_setattr(env, inode, attr);
2992         spin_unlock(&obj->oo_guard);
2993         if (rc != 0)
2994                 GOTO(out, rc);
2995
2996         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2997
2998         if (!(attr->la_valid & LA_FLAGS))
2999                 GOTO(out, rc);
3000
3001         /* Let's check if there are extra flags need to be set into LMA */
3002         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3003                 struct osd_thread_info *info = osd_oti_get(env);
3004                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3005
3006                 LASSERT(!obj->oo_pfid_in_lma);
3007
3008                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3009                                  &info->oti_ost_attrs);
3010                 if (rc)
3011                         GOTO(out, rc);
3012
3013                 lma->lma_incompat |=
3014                         lustre_to_lma_flags(attr->la_flags);
3015                 lustre_lma_swab(lma);
3016                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3017                                      lma, sizeof(*lma), XATTR_REPLACE);
3018                 if (rc != 0) {
3019                         struct osd_device *osd = osd_obj2dev(obj);
3020
3021                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3022                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3023                               lma->lma_incompat, rc);
3024                 } else {
3025                         obj->oo_lma_flags =
3026                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3027                 }
3028                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3029         }
3030 out:
3031         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3032
3033         return rc;
3034 }
3035
3036 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3037                                            struct osd_object *obj,
3038                                            const char *name, const int namelen)
3039 {
3040         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3041 }
3042
3043 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3044                       umode_t mode, struct dt_allocation_hint *hint,
3045                       struct thandle *th)
3046 {
3047         int result;
3048         struct osd_device *osd = osd_obj2dev(obj);
3049         struct osd_thandle *oth;
3050         struct dt_object *parent = NULL;
3051         struct inode *inode;
3052
3053         LINVRNT(osd_invariant(obj));
3054         LASSERT(obj->oo_inode == NULL);
3055         LASSERT(obj->oo_hl_head == NULL);
3056
3057         if (S_ISDIR(mode) && ldiskfs_pdo) {
3058                 obj->oo_hl_head =
3059                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3060                 if (obj->oo_hl_head == NULL)
3061                         return -ENOMEM;
3062         }
3063
3064         oth = container_of(th, struct osd_thandle, ot_super);
3065         LASSERT(oth->ot_handle->h_transaction != NULL);
3066
3067         if (hint != NULL && hint->dah_parent != NULL &&
3068             !dt_object_remote(hint->dah_parent))
3069                 parent = hint->dah_parent;
3070
3071         inode = ldiskfs_create_inode(oth->ot_handle,
3072                                      parent ? osd_dt_obj(parent)->oo_inode :
3073                                               osd_sb(osd)->s_root->d_inode,
3074                                      mode);
3075         if (!IS_ERR(inode)) {
3076                 /* Do not update file c/mtime in ldiskfs. */
3077                 inode->i_flags |= S_NOCMTIME;
3078
3079                 /*
3080                  * For new created object, it must be consistent,
3081                  * and it is unnecessary to scrub against it.
3082                  */
3083                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3084
3085                 obj->oo_inode = inode;
3086                 result = 0;
3087         } else {
3088                 if (obj->oo_hl_head != NULL) {
3089                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3090                         obj->oo_hl_head = NULL;
3091                 }
3092                 result = PTR_ERR(inode);
3093         }
3094         LINVRNT(osd_invariant(obj));
3095         return result;
3096 }
3097
3098 enum {
3099         OSD_NAME_LEN = 255
3100 };
3101
3102 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3103                      struct lu_attr *attr,
3104                      struct dt_allocation_hint *hint,
3105                      struct dt_object_format *dof,
3106                      struct thandle *th)
3107 {
3108         int result;
3109         struct osd_thandle *oth;
3110         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3111
3112         LASSERT(S_ISDIR(attr->la_mode));
3113
3114         oth = container_of(th, struct osd_thandle, ot_super);
3115         LASSERT(oth->ot_handle->h_transaction != NULL);
3116         result = osd_mkfile(info, obj, mode, hint, th);
3117
3118         return result;
3119 }
3120
3121 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3122                         struct lu_attr *attr,
3123                         struct dt_allocation_hint *hint,
3124                         struct dt_object_format *dof,
3125                         struct thandle *th)
3126 {
3127         int result;
3128         struct osd_thandle *oth;
3129         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3130
3131         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3132
3133         LASSERT(S_ISREG(attr->la_mode));
3134
3135         oth = container_of(th, struct osd_thandle, ot_super);
3136         LASSERT(oth->ot_handle->h_transaction != NULL);
3137
3138         result = osd_mkfile(info, obj, mode, hint, th);
3139         if (result == 0) {
3140                 LASSERT(obj->oo_inode != NULL);
3141                 if (feat->dif_flags & DT_IND_VARKEY)
3142                         result = iam_lvar_create(obj->oo_inode,
3143                                                  feat->dif_keysize_max,
3144                                                  feat->dif_ptrsize,
3145                                                  feat->dif_recsize_max,
3146                                                  oth->ot_handle);
3147                 else
3148                         result = iam_lfix_create(obj->oo_inode,
3149                                                  feat->dif_keysize_max,
3150                                                  feat->dif_ptrsize,
3151                                                  feat->dif_recsize_max,
3152                                                  oth->ot_handle);
3153         }
3154         return result;
3155 }
3156
3157 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3158                      struct lu_attr *attr,
3159                      struct dt_allocation_hint *hint,
3160                      struct dt_object_format *dof,
3161                      struct thandle *th)
3162 {
3163         LASSERT(S_ISREG(attr->la_mode));
3164         return osd_mkfile(info, obj, (attr->la_mode &
3165                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3166 }
3167
3168 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3169                      struct lu_attr *attr,
3170                      struct dt_allocation_hint *hint,
3171                      struct dt_object_format *dof,
3172                      struct thandle *th)
3173 {
3174         LASSERT(S_ISLNK(attr->la_mode));
3175         return osd_mkfile(info, obj, (attr->la_mode &
3176                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3177 }
3178
3179 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3180                      struct lu_attr *attr,
3181                      struct dt_allocation_hint *hint,
3182                      struct dt_object_format *dof,
3183                      struct thandle *th)
3184 {
3185         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3186         int result;
3187
3188         LINVRNT(osd_invariant(obj));
3189         LASSERT(obj->oo_inode == NULL);
3190         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3191                 S_ISFIFO(mode) || S_ISSOCK(mode));
3192
3193         result = osd_mkfile(info, obj, mode, hint, th);
3194         if (result == 0) {
3195                 LASSERT(obj->oo_inode != NULL);
3196                 /*
3197                  * This inode should be marked dirty for i_rdev.  Currently
3198                  * that is done in the osd_attr_init().
3199                  */
3200                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3201                                    attr->la_rdev);
3202         }
3203         LINVRNT(osd_invariant(obj));
3204         return result;
3205 }
3206
3207 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3208                               struct lu_attr *,
3209                               struct dt_allocation_hint *hint,
3210                               struct dt_object_format *dof,
3211                               struct thandle *);
3212
3213 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3214 {
3215         osd_obj_type_f result;
3216
3217         switch (type) {
3218         case DFT_DIR:
3219                 result = osd_mkdir;
3220                 break;
3221         case DFT_REGULAR:
3222                 result = osd_mkreg;
3223                 break;
3224         case DFT_SYM:
3225                 result = osd_mksym;
3226                 break;
3227         case DFT_NODE:
3228                 result = osd_mknod;
3229                 break;
3230         case DFT_INDEX:
3231                 result = osd_mk_index;
3232                 break;
3233
3234         default:
3235                 LBUG();
3236                 break;
3237         }
3238         return result;
3239 }
3240
3241
3242 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3243                         struct dt_object *parent, struct dt_object *child,
3244                         umode_t child_mode)
3245 {
3246         LASSERT(ah);
3247
3248         ah->dah_parent = parent;
3249         ah->dah_mode = child_mode;
3250
3251         if (parent != NULL && !dt_object_remote(parent)) {
3252                 /* will help to find FID->ino at dt_insert("..") */
3253                 struct osd_object *pobj = osd_dt_obj(parent);
3254
3255                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3256         }
3257 }
3258
3259 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3260                           struct lu_attr *attr, struct dt_object_format *dof)
3261 {
3262         struct inode *inode = obj->oo_inode;
3263         __u64 valid = attr->la_valid;
3264         int result;
3265
3266         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3267
3268         if (dof->dof_type != DFT_NODE)
3269                 attr->la_valid &= ~LA_RDEV;
3270         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3271                 attr->la_valid &= ~LA_ATIME;
3272         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3273                 attr->la_valid &= ~LA_CTIME;
3274         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3275                 attr->la_valid &= ~LA_MTIME;
3276
3277         result = osd_quota_transfer(inode, attr);
3278         if (result)
3279                 return;
3280
3281         if (attr->la_valid != 0) {
3282                 result = osd_inode_setattr(info->oti_env, inode, attr);
3283                 /*
3284                  * The osd_inode_setattr() should always succeed here.  The
3285                  * only error that could be returned is EDQUOT when we are
3286                  * trying to change the UID or GID of the inode. However, this
3287                  * should not happen since quota enforcement is no longer
3288                  * enabled on ldiskfs (lquota takes care of it).
3289                  */
3290                 LASSERTF(result == 0, "%d\n", result);
3291                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3292         }
3293
3294         attr->la_valid = valid;
3295 }
3296
3297 /**
3298  * Helper function for osd_create()
3299  *
3300  * \retval 0, on success
3301  */
3302 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3303                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3304                         struct dt_object_format *dof, struct thandle *th)
3305 {
3306         int result;
3307         __u32 umask;
3308
3309         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3310
3311         /* we drop umask so that permissions we pass are not affected */
3312         umask = current->fs->umask;
3313         current->fs->umask = 0;
3314
3315         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3316                                                   th);
3317         if (likely(obj->oo_inode != NULL)) {
3318                 LASSERT(obj->oo_inode->i_state & I_NEW);
3319
3320                 /*
3321                  * Unlock the inode before attr initialization to avoid
3322                  * unnecessary dqget operations. LU-6378
3323                  */
3324                 unlock_new_inode(obj->oo_inode);
3325         }
3326
3327         if (likely(result == 0)) {
3328                 osd_attr_init(info, obj, attr, dof);
3329                 osd_object_init0(obj);
3330         }
3331
3332         /* restore previous umask value */
3333         current->fs->umask = umask;
3334
3335         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3336
3337         return result;
3338 }
3339
3340 /**
3341  * Helper function for osd_create()
3342  *
3343  * \retval 0, on success
3344  */
3345 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3346                            const struct lu_fid *fid, struct thandle *th)
3347 {
3348         struct osd_thread_info *info = osd_oti_get(env);
3349         struct osd_inode_id    *id   = &info->oti_id;
3350         struct osd_device      *osd  = osd_obj2dev(obj);
3351         struct osd_thandle     *oh;
3352         int rc;
3353
3354         LASSERT(obj->oo_inode != NULL);
3355
3356         oh = container_of0(th, struct osd_thandle, ot_super);
3357         LASSERT(oh->ot_handle);
3358         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3359
3360         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3361         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3362                            OI_CHECK_FLD, NULL);
3363         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3364
3365         return rc;
3366 }
3367
3368 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3369                    u64 seq, struct lu_seq_range *range)
3370 {
3371         struct seq_server_site *ss = osd_seq_site(osd);
3372
3373         if (fid_seq_is_idif(seq)) {
3374                 fld_range_set_ost(range);
3375                 range->lsr_index = idif_ost_idx(seq);
3376                 return 0;
3377         }
3378
3379         if (!fid_seq_in_fldb(seq)) {
3380                 fld_range_set_mdt(range);
3381                 if (ss != NULL)
3382                         /*
3383                          * FIXME: If ss is NULL, it suppose not get lsr_index
3384                          * at all
3385                          */
3386                         range->lsr_index = ss->ss_node_id;
3387                 return 0;
3388         }
3389
3390         LASSERT(ss != NULL);
3391         fld_range_set_any(range);
3392         /* OSD will only do local fld lookup */
3393         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3394 }
3395
3396 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3397                               struct lu_attr *attr,
3398                               struct dt_allocation_hint *hint,
3399                               struct dt_object_format *dof,
3400                               struct thandle *handle)
3401 {
3402         struct osd_thandle *oh;
3403         int rc;
3404
3405         ENTRY;
3406
3407         LASSERT(handle != NULL);
3408
3409         oh = container_of0(handle, struct osd_thandle, ot_super);
3410         LASSERT(oh->ot_handle == NULL);
3411
3412         /*
3413          * EA object consumes more credits than regular object: osd_mk_index
3414          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3415          * leaf_node, could involves the block, block bitmap, groups, GDT
3416          * change for each block, so add 4 * 2 credits in that case.
3417          */
3418         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3419                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3420                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3421         /*
3422          * Reuse idle OI block may cause additional one OI block
3423          * to be changed.
3424          */
3425         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3426                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3427
3428         /* will help to find FID->ino mapping at dt_insert() */
3429         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3430                                    osd_dt_obj(dt));
3431         if (rc != 0)
3432                 RETURN(rc);
3433
3434         if (!attr)
3435                 RETURN(0);
3436
3437         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3438                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3439                                    NULL, OSD_QID_INODE);
3440         if (rc != 0)
3441                 RETURN(rc);
3442
3443         RETURN(rc);
3444 }
3445
3446 /**
3447  * Called to destroy on-disk representation of the object
3448  *
3449  * Concurrency: must be locked
3450  */
3451 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3452                                struct thandle *th)
3453 {
3454         struct osd_object *obj = osd_dt_obj(dt);
3455         struct inode *inode = obj->oo_inode;
3456         struct osd_thandle *oh;
3457         int rc;
3458
3459         ENTRY;
3460
3461         if (inode == NULL)
3462                 RETURN(-ENOENT);
3463
3464         oh = container_of0(th, struct osd_thandle, ot_super);
3465         LASSERT(oh->ot_handle == NULL);
3466
3467         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3468                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3469
3470         /* For removing agent entry */
3471         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3472                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3473
3474         /*
3475          * Recycle idle OI leaf may cause additional three OI blocks
3476          * to be changed.
3477          */
3478         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3479                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3480                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3481         /* one less inode */
3482         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3483                                    i_projid_read(inode), -1, oh, obj, NULL,
3484                                    OSD_QID_INODE);
3485         if (rc)
3486                 RETURN(rc);
3487         /* data to be truncated */
3488         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3489                                    i_projid_read(inode), 0, oh, obj, NULL,
3490                                    OSD_QID_BLK);
3491         if (rc)
3492                 RETURN(rc);
3493
3494         /*
3495          * will help to find FID->ino when this object is being
3496          * added to PENDING
3497          */
3498         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3499
3500         RETURN(rc);
3501 }
3502
3503 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3504                        struct thandle *th)
3505 {
3506         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3507         struct osd_object *obj = osd_dt_obj(dt);
3508         struct inode *inode = obj->oo_inode;
3509         struct osd_device *osd = osd_obj2dev(obj);
3510         struct osd_thandle *oh;
3511         int result;
3512
3513         ENTRY;
3514
3515         oh = container_of0(th, struct osd_thandle, ot_super);
3516         LASSERT(oh->ot_handle);
3517         LASSERT(inode);
3518         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3519
3520         if (unlikely(fid_is_acct(fid)))
3521                 RETURN(-EPERM);
3522
3523         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3524                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3525                 if (result != 0)
3526                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3527                                osd_name(osd), PFID(fid), result);
3528         }
3529
3530         if (S_ISDIR(inode->i_mode)) {
3531                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3532                         inode->i_nlink == 2);
3533
3534                 spin_lock(&obj->oo_guard);
3535                 clear_nlink(inode);
3536                 spin_unlock(&obj->oo_guard);
3537                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3538         }
3539
3540         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3541
3542         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3543
3544         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3545                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3546                                        oh->ot_handle, OI_CHECK_FLD);
3547
3548         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3549         /* XXX: add to ext3 orphan list */
3550         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3551
3552         /* not needed in the cache anymore */
3553         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3554         obj->oo_destroyed = 1;
3555
3556         RETURN(0);
3557 }
3558
3559 /**
3560  * Put the fid into lustre_mdt_attrs, and then place the structure
3561  * inode's ea. This fid should not be altered during the life time
3562  * of the inode.
3563  *
3564  * \retval +ve, on success
3565  * \retval -ve, on error
3566  *
3567  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3568  */
3569 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3570                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3571 {
3572         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3573         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3574         int rc;
3575
3576         ENTRY;
3577
3578         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3579                 RETURN(0);
3580
3581         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3582                 rc = -ENOMEM;
3583
3584         lustre_loa_init(loa, fid, compat, incompat);
3585         lustre_loa_swab(loa, false);
3586
3587         /*
3588          * For the OST device with 256 bytes inode size by default,
3589          * the PFID EA will be stored together with LMA EA to avoid
3590          * performance trouble. Otherwise the PFID EA can be stored
3591          * independently. LU-8998
3592          */
3593         if ((compat & LMAC_FID_ON_OST) &&
3594             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3595                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3596                                      sizeof(*loa), XATTR_CREATE);
3597         else
3598                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3599                                      sizeof(*lma), XATTR_CREATE);
3600         /*
3601          * LMA may already exist, but we need to check that all the
3602          * desired compat/incompat flags have been added.
3603          */
3604         if (unlikely(rc == -EEXIST)) {
3605                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3606                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3607                 if (rc < 0)
3608                         RETURN(rc);
3609
3610                 if (rc < sizeof(*lma))
3611                         RETURN(-EINVAL);
3612
3613                 lustre_loa_swab(loa, true);
3614                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3615                     ((compat == 0 && incompat == 0) ||
3616                      (!(~lma->lma_compat & compat) &&
3617                       !(~lma->lma_incompat & incompat))))
3618                         RETURN(0);
3619
3620                 lma->lma_self_fid = *fid;
3621                 lma->lma_compat |= compat;
3622                 lma->lma_incompat |= incompat;
3623                 if (rc == sizeof(*lma)) {
3624                         lustre_lma_swab(lma);
3625                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3626                                              sizeof(*lma), XATTR_REPLACE);
3627                 } else {
3628                         lustre_loa_swab(loa, false);
3629                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3630                                              sizeof(*loa), XATTR_REPLACE);
3631                 }
3632         }
3633
3634         RETURN(rc);
3635 }
3636
3637 /**
3638  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3639  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3640  * To have compatilibility with 1.8 ldiskfs driver we need to have
3641  * magic number at start of fid data.
3642  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3643  * its inmemory API.
3644  */
3645 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3646                                          const struct lu_fid *fid)
3647 {
3648         if (!fid_is_namespace_visible(fid) ||
3649             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3650                 param->edp_magic = 0;
3651                 return;
3652         }
3653
3654         param->edp_magic = LDISKFS_LUFID_MAGIC;
3655         param->edp_len =  sizeof(struct lu_fid) + 1;
3656         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3657 }
3658
3659 /**
3660  * Try to read the fid from inode ea into dt_rec.
3661  *
3662  * \param fid object fid.
3663  *
3664  * \retval 0 on success
3665  */
3666 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3667                           __u32 ino, struct lu_fid *fid,
3668                           struct osd_inode_id *id)
3669 {
3670         struct osd_thread_info *info  = osd_oti_get(env);
3671         struct inode *inode;
3672
3673         ENTRY;
3674
3675         osd_id_gen(id, ino, OSD_OII_NOGEN);
3676         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3677         if (IS_ERR(inode))
3678                 RETURN(PTR_ERR(inode));
3679
3680         iput(inode);
3681         RETURN(0);
3682 }
3683
3684 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3685                                         struct inode *dir,
3686                                         struct inode *parent_dir,
3687                                         const struct lu_fid *dot_fid,
3688                                         const struct lu_fid *dot_dot_fid,
3689                                         struct osd_thandle *oth)
3690 {
3691         struct ldiskfs_dentry_param *dot_ldp;
3692         struct ldiskfs_dentry_param *dot_dot_ldp;
3693         __u32 saved_nlink = dir->i_nlink;
3694         int rc;
3695
3696         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3697         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3698
3699         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3700         dot_ldp->edp_magic = 0;
3701
3702         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3703                                     dir, dot_ldp, dot_dot_ldp);
3704         /*
3705          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3706          * the subseqent ref_add() will increase the dir->i_nlink
3707          * as 3. That is incorrect for new created directory.
3708          *
3709          * It looks like hack, because we want to make the OSD API
3710          * to be order-independent for new created directory object
3711          * between dt_insert(..) and ref_add() operations.
3712          *
3713          * Here, we only restore the in-RAM dir-inode's nlink attr,
3714          * becuase if the nlink attr is not 2, then there will be
3715          * ref_add() called following the dt_insert(..), such call
3716          * will make both the in-RAM and on-disk dir-inode's nlink
3717          * attr to be set as 2. LU-7447
3718          */
3719         set_nlink(dir, saved_nlink);
3720         return rc;
3721 }
3722
3723 /**
3724  * Create an local agent inode for remote entry
3725  */
3726 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3727                                                   struct osd_device *osd,
3728                                                   struct osd_object *pobj,
3729                                                   const struct lu_fid *fid,
3730                                                   __u32 type,
3731                                                   struct thandle *th)
3732 {
3733         struct osd_thread_info *info = osd_oti_get(env);
3734         struct inode *local;
3735         struct osd_thandle *oh;
3736         int rc;
3737
3738         ENTRY;
3739
3740         LASSERT(th);
3741         oh = container_of(th, struct osd_thandle, ot_super);
3742         LASSERT(oh->ot_handle->h_transaction != NULL);
3743
3744         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3745         if (IS_ERR(local)) {
3746                 CERROR("%s: create local error %d\n", osd_name(osd),
3747                        (int)PTR_ERR(local));
3748                 RETURN(local);
3749         }
3750
3751         /*
3752          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3753          * correct gid on remote file, not agent here
3754          */
3755         local->i_gid = current_fsgid();
3756         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3757         unlock_new_inode(local);
3758
3759         /* Agent inode should not have project ID */
3760 #ifdef  HAVE_PROJECT_QUOTA
3761         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3762             i_projid_read(pobj->oo_inode) != 0) {
3763                 rc = osd_transfer_project(local, 0);
3764                 if (rc) {
3765                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3766                                "quota enforcement enabled on the ldiskfs "
3767                                "filesystem?\n", local->i_sb->s_id, rc);
3768                         RETURN(ERR_PTR(rc));
3769                 }
3770         }
3771 #endif
3772         /* Set special LMA flag for local agent inode */
3773         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3774         if (rc != 0) {
3775                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3776                        osd_name(osd), PFID(fid), rc);
3777                 RETURN(ERR_PTR(rc));
3778         }
3779
3780         if (!S_ISDIR(type))
3781                 RETURN(local);
3782
3783         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3784                                          lu_object_fid(&pobj->oo_dt.do_lu),
3785                                          fid, oh);
3786         if (rc != 0) {
3787                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3788                         osd_name(osd), PFID(fid), rc);
3789                 RETURN(ERR_PTR(rc));
3790         }
3791
3792         RETURN(local);
3793 }
3794
3795 /**
3796  * when direntry is deleted, we have to take care of possible agent inode
3797  * referenced by that. unfortunately we can't do this at that point:
3798  * iget() within a running transaction leads to deadlock and we better do
3799  * not call that every delete declaration to save performance. so we put
3800  * a potention agent inode on a list and process that once the transaction
3801  * is over. Notice it's not any worse in terms of real orphans as regular
3802  * object destroy doesn't put inodes on the on-disk orphan list. this should
3803  * be addressed separately
3804  */
3805 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3806                                             struct osd_thandle *oh,
3807                                             __u32 ino)
3808 {
3809         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3810         struct osd_obj_orphan *oor;
3811
3812         OBD_ALLOC_PTR(oor);
3813         if (oor == NULL)
3814                 return -ENOMEM;
3815
3816         oor->oor_ino = ino;
3817         oor->oor_env = (struct lu_env *)env;
3818         spin_lock(&osd->od_osfs_lock);
3819         list_add(&oor->oor_list, &osd->od_orphan_list);
3820         spin_unlock(&osd->od_osfs_lock);
3821
3822         oh->ot_remove_agents = 1;
3823
3824         return 0;
3825
3826 }
3827
3828 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3829                                                 struct osd_device *osd)
3830 {
3831         struct osd_thread_info *info = osd_oti_get(env);
3832         struct osd_obj_orphan *oor, *tmp;
3833         struct osd_inode_id id;
3834         struct list_head list;
3835         struct inode *inode;
3836         struct lu_fid fid;
3837         handle_t *jh;
3838         __u32 ino;
3839
3840         INIT_LIST_HEAD(&list);
3841
3842         spin_lock(&osd->od_osfs_lock);
3843         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3844                 if (oor->oor_env == env) {
3845                         list_del(&oor->oor_list);
3846                         list_add(&oor->oor_list, &list);
3847                 }
3848         }
3849         spin_unlock(&osd->od_osfs_lock);
3850
3851         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3852
3853                 ino = oor->oor_ino;
3854
3855                 list_del(&oor->oor_list);
3856                 OBD_FREE_PTR(oor);
3857
3858                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3859                 inode = osd_iget_fid(info, osd, &id, &fid);
3860                 if (IS_ERR(inode))
3861                         continue;
3862
3863                 if (!osd_remote_fid(env, osd, &fid)) {
3864                         iput(inode);
3865                         continue;
3866                 }
3867
3868                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3869                 clear_nlink(inode);
3870                 mark_inode_dirty(inode);
3871                 ldiskfs_journal_stop(jh);
3872                 iput(inode);
3873         }
3874
3875         return 0;
3876 }
3877
3878 /**
3879  * OSD layer object create function for OST objects (b=11826).
3880  *
3881  * The FID is inserted into inode xattr here.
3882  *
3883  * \retval   0, on success
3884  * \retval -ve, on error
3885  */
3886 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3887                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3888                       struct dt_object_format *dof, struct thandle *th)
3889 {
3890         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3891         struct osd_object *obj = osd_dt_obj(dt);
3892         struct osd_thread_info *info = osd_oti_get(env);
3893         int result, on_ost = 0;
3894
3895         ENTRY;
3896
3897         if (dt_object_exists(dt))
3898                 RETURN(-EEXIST);
3899
3900         LINVRNT(osd_invariant(obj));
3901         LASSERT(!dt_object_remote(dt));
3902         LASSERT(osd_is_write_locked(env, obj));
3903         LASSERT(th != NULL);
3904
3905         if (unlikely(fid_is_acct(fid)))
3906                 /*
3907                  * Quota files can't be created from the kernel any more,
3908                  * 'tune2fs -O quota' will take care of creating them
3909                  */
3910                 RETURN(-EPERM);
3911
3912         result = __osd_create(info, obj, attr, hint, dof, th);
3913         if (result == 0) {
3914                 if (fid_is_idif(fid) &&
3915                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3916                         struct lu_fid *tfid = &info->oti_fid;
3917                         struct ost_id *oi   = &info->oti_ostid;
3918
3919                         fid_to_ostid(fid, oi);
3920                         ostid_to_fid(tfid, oi, 0);
3921                         on_ost = 1;
3922                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3923                                                 LMAC_FID_ON_OST, 0);
3924                 } else {
3925                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3926                                                fid, OI_CHECK_FLD);
3927                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3928                                                 on_ost ? LMAC_FID_ON_OST : 0,
3929                                                 0);
3930                 }
3931                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3932                         obj->oo_dt.do_body_ops = &osd_body_ops;
3933         }
3934
3935         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3936                 result = __osd_oi_insert(env, obj, fid, th);
3937
3938         /*
3939          * a small optimization - dt_insert() isn't usually applied
3940          * to OST objects, so we don't need to cache OI mapping for
3941          * OST objects
3942          */
3943         if (result == 0 && on_ost == 0) {
3944                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3945
3946                 result = osd_idc_find_and_init(env, osd, obj);
3947                 LASSERT(result == 0);
3948         }
3949
3950         LASSERT(ergo(result == 0,
3951                      dt_object_exists(dt) && !dt_object_remote(dt)));
3952         LINVRNT(osd_invariant(obj));
3953         RETURN(result);
3954 }
3955
3956 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3957                                struct thandle *handle)
3958 {
3959         struct osd_thandle *oh;
3960
3961         /* it's possible that object doesn't exist yet */
3962         LASSERT(handle != NULL);
3963
3964         oh = container_of0(handle, struct osd_thandle, ot_super);
3965         LASSERT(oh->ot_handle == NULL);
3966
3967         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3968                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3969
3970         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3971
3972         return 0;
3973 }
3974
3975 /*
3976  * Concurrency: @dt is write locked.
3977  */
3978 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3979                        struct thandle *th)
3980 {
3981         struct osd_object *obj = osd_dt_obj(dt);
3982         struct inode *inode = obj->oo_inode;
3983         struct osd_thandle *oh;
3984         int rc = 0;
3985
3986         if (!dt_object_exists(dt) || obj->oo_destroyed)
3987                 return -ENOENT;
3988
3989         LINVRNT(osd_invariant(obj));
3990         LASSERT(!dt_object_remote(dt));
3991         LASSERT(osd_is_write_locked(env, obj));
3992         LASSERT(th != NULL);
3993
3994         oh = container_of0(th, struct osd_thandle, ot_super);
3995         LASSERT(oh->ot_handle != NULL);
3996
3997         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3998
3999         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4000                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4001         /*
4002          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4003          * (65000) subdirectories by storing "1" in i_nlink if the link count
4004          * would otherwise overflow. Directory tranversal tools understand
4005          * that (st_nlink == 1) indicates that the filesystem dose not track
4006          * hard links count on the directory, and will not abort subdirectory
4007          * scanning early once (st_nlink - 2) subdirs have been found.
4008          *
4009          * This also has to properly handle the case of inodes with nlink == 0
4010          * in case they are being linked into the PENDING directory
4011          */
4012         spin_lock(&obj->oo_guard);
4013         if (unlikely(inode->i_nlink == 0))
4014                 /* inc_nlink from 0 may cause WARN_ON */
4015                 set_nlink(inode, 1);
4016         else {
4017                 ldiskfs_inc_count(oh->ot_handle, inode);
4018                 if (!S_ISDIR(inode->i_mode))
4019                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4020         }
4021         spin_unlock(&obj->oo_guard);
4022
4023         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4024         LINVRNT(osd_invariant(obj));
4025
4026         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4027
4028         return rc;
4029 }
4030
4031 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4032                                struct thandle *handle)
4033 {
4034         struct osd_thandle *oh;
4035
4036         if (!dt_object_exists(dt))
4037                 return -ENOENT;
4038
4039         LASSERT(!dt_object_remote(dt));
4040         LASSERT(handle != NULL);
4041
4042         oh = container_of0(handle, struct osd_thandle, ot_super);
4043         LASSERT(oh->ot_handle == NULL);
4044
4045         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4046                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4047
4048         return 0;
4049 }
4050
4051 /*
4052  * Concurrency: @dt is write locked.
4053  */
4054 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4055                        struct thandle *th)
4056 {
4057         struct osd_object *obj = osd_dt_obj(dt);
4058         struct inode *inode = obj->oo_inode;
4059         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4060         struct osd_thandle *oh;
4061
4062         if (!dt_object_exists(dt))
4063                 return -ENOENT;
4064
4065         LINVRNT(osd_invariant(obj));
4066         LASSERT(!dt_object_remote(dt));
4067         LASSERT(osd_is_write_locked(env, obj));
4068         LASSERT(th != NULL);
4069
4070         oh = container_of0(th, struct osd_thandle, ot_super);
4071         LASSERT(oh->ot_handle != NULL);
4072
4073         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4074
4075         spin_lock(&obj->oo_guard);
4076         /*
4077          * That can be result of upgrade from old Lustre version and
4078          * applied only to local files.  Just skip this ref_del call.
4079          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4080          */
4081         if (inode->i_nlink == 0) {
4082                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4083                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4084                              ", maybe an upgraded file? (LU-3915)\n",
4085                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4086                 spin_unlock(&obj->oo_guard);
4087                 return 0;
4088         }
4089
4090         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4091                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4092
4093         ldiskfs_dec_count(oh->ot_handle, inode);
4094         spin_unlock(&obj->oo_guard);
4095
4096         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4097         LINVRNT(osd_invariant(obj));
4098
4099         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4100
4101         return 0;
4102 }
4103
4104 /*
4105  * Concurrency: @dt is read locked.
4106  */
4107 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4108                          struct lu_buf *buf, const char *name)
4109 {
4110         struct osd_object      *obj    = osd_dt_obj(dt);
4111         struct inode           *inode  = obj->oo_inode;
4112         struct osd_thread_info *info   = osd_oti_get(env);
4113         struct dentry          *dentry = &info->oti_obj_dentry;
4114         bool cache_xattr = false;
4115         int rc;
4116
4117         LASSERT(buf);
4118
4119         /* version get is not real XATTR but uses xattr API */
4120         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4121                 dt_obj_version_t *ver = buf->lb_buf;
4122
4123                 /*
4124                  * for version we are just using xattr API but change inode
4125                  * field instead
4126                  */
4127                 if (buf->lb_len == 0)
4128                         return sizeof(dt_obj_version_t);
4129
4130                 if (buf->lb_len < sizeof(dt_obj_version_t))
4131                         return -ERANGE;
4132
4133                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4134                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4135
4136                 *ver = LDISKFS_I(inode)->i_fs_version;
4137
4138                 return sizeof(dt_obj_version_t);
4139         }
4140
4141         if (!dt_object_exists(dt))
4142                 return -ENOENT;
4143
4144         LASSERT(!dt_object_remote(dt));
4145         LASSERT(inode->i_op != NULL);
4146 #ifdef HAVE_IOP_XATTR
4147         LASSERT(inode->i_op->getxattr != NULL);
4148 #endif
4149
4150         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4151             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4152                 cache_xattr = true;
4153
4154         if (cache_xattr) {
4155                 rc = osd_oxc_get(obj, name, buf);
4156                 if (rc != -ENOENT)
4157                         return rc;
4158         }
4159
4160         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4161                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4162                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4163                 struct filter_fid *ff;
4164                 struct ost_layout *ol;
4165
4166                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4167                 if (rc)
4168                         return rc;
4169
4170                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4171
4172                 rc = sizeof(*ff);
4173                 if (buf->lb_len == 0 || !buf->lb_buf)
4174                         return rc;
4175
4176                 if (buf->lb_len < rc)
4177                         return -ERANGE;
4178
4179                 ff = buf->lb_buf;
4180                 ol = &ff->ff_layout;
4181                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4182                                                   PFID_STRIPE_IDX_BITS);
4183                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4184                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4185                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4186                 if (lma->lma_compat & LMAC_COMP_INFO) {
4187                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4188                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4189                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4190                 } else {
4191                         ol->ol_comp_start = 0;
4192                         ol->ol_comp_end = 0;
4193                         ol->ol_comp_id = 0;
4194                 }
4195         } else {
4196                 rc = __osd_xattr_get(inode, dentry, name,
4197                                      buf->lb_buf, buf->lb_len);
4198         }
4199
4200         if (cache_xattr) {
4201                 if (rc == -ENOENT || rc == -ENODATA)
4202                         osd_oxc_add(obj, name, NULL, 0);
4203                 else if (rc > 0 && buf->lb_buf != NULL)
4204                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4205         }
4206
4207         return rc;
4208 }
4209
4210 static int osd_declare_xattr_set(const struct lu_env *env,
4211                                  struct dt_object *dt,
4212                                  const struct lu_buf *buf, const char *name,
4213                                  int fl, struct thandle *handle)
4214 {
4215         struct osd_thandle *oh;
4216         int credits = 0;
4217         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4218
4219         LASSERT(handle != NULL);
4220
4221         oh = container_of0(handle, struct osd_thandle, ot_super);
4222         LASSERT(oh->ot_handle == NULL);
4223
4224         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4225                 /*
4226                  * For non-upgrading case, the LMA is set first and
4227                  * usually fit inode. But for upgrade case, the LMA
4228                  * may be in another separated EA block.
4229                  */
4230                 if (dt_object_exists(dt)) {
4231                         if (fl == LU_XATTR_REPLACE)
4232                                 credits = 1;
4233                         else
4234                                 goto upgrade;
4235                 }
4236         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4237                 credits = 1;
4238         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4239                 /* We may need to delete the old PFID EA. */
4240                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4241                 if (fl == LU_XATTR_REPLACE)
4242                         credits += 1;
4243                 else
4244                         goto upgrade;
4245         } else {
4246                 /*
4247                  * If some name entry resides on remote MDT, then will create
4248                  * agent entry under remote parent. On the other hand, if the
4249                  * remote entry will be removed, then related agent entry may
4250                  * need to be removed from the remote parent. So there may be
4251                  * kinds of cases, let's declare enough credits. The credits
4252                  * for create agent entry is enough for remove case.
4253                  */
4254                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4255                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4256                         if (dt_object_exists(dt))
4257                                 credits += 1; /* For updating LMA */
4258                 }
4259
4260 upgrade:
4261                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4262
4263                 if (buf != NULL) {
4264                         ssize_t buflen;
4265
4266                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4267                                 /*
4268                                  * learn xattr size from osd_xattr_get if
4269                                  * attribute has not been read yet
4270                                  */
4271                                 buflen = __osd_xattr_get(
4272                                     osd_dt_obj(dt)->oo_inode,
4273                                     &osd_oti_get(env)->oti_obj_dentry,
4274                                     name, NULL, 0);
4275                                 if (buflen < 0)
4276                                         buflen = 0;
4277                         } else {
4278                                 buflen = buf->lb_len;
4279                         }
4280
4281                         if (buflen > sb->s_blocksize) {
4282                                 credits += osd_calc_bkmap_credits(
4283                                     sb, NULL, 0, -1,
4284                                     (buflen + sb->s_blocksize - 1) >>
4285                                     sb->s_blocksize_bits);
4286                         }
4287                 }
4288                 /*
4289                  * xattr set may involve inode quota change, reserve credits for
4290                  * dquot_initialize()
4291                  */
4292                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4293         }
4294
4295         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4296
4297         return 0;
4298 }
4299
4300 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4301                               const struct lu_buf *buf, int fl,
4302                               struct thandle *handle)
4303 {
4304         struct osd_thread_info *info = osd_oti_get(env);
4305         struct dentry *dentry = &info->oti_obj_dentry;
4306         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4307         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4308         struct inode *inode = obj->oo_inode;
4309         struct filter_fid *ff = buf->lb_buf;
4310         struct ost_layout *ol = &ff->ff_layout;
4311         int flags = XATTR_REPLACE;
4312         int rc;
4313
4314         ENTRY;
4315
4316         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4317                 RETURN(-EINVAL);
4318
4319         rc = osd_get_lma(info, inode, dentry, loa);
4320         if (rc == -ENODATA) {
4321                 /* Usually for upgarding from old device */
4322                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4323                                 LMAC_FID_ON_OST, 0);
4324                 flags = XATTR_CREATE;
4325         } else if (rc) {
4326                 RETURN(rc);
4327         }
4328
4329         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4330                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4331                         RETURN(-EEXIST);
4332
4333                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4334                         /* Separate PFID EA from LMA */
4335                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4336                         lustre_lma_swab(lma);
4337                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4338                                              sizeof(*lma), XATTR_REPLACE);
4339                         if (!rc) {
4340                                 obj->oo_pfid_in_lma = 0;
4341                                 rc = LU_XATTR_CREATE;
4342                         }
4343
4344                         RETURN(rc);
4345                 }
4346         } else {
4347                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4348                         RETURN(fl);
4349
4350                 /*
4351                  * Old client does not send stripe information,
4352                  * then store the PFID EA on disk separatedly.
4353                  */
4354                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4355                              ol->ol_stripe_size == 0))
4356                         RETURN(fl);
4357
4358                 /* Remove old PFID EA entry firstly. */
4359                 ll_vfs_dq_init(inode);
4360                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4361                 if (rc == -ENODATA) {
4362                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4363                                 RETURN(rc);
4364                 } else if (rc) {
4365                         RETURN(rc);
4366                 }
4367         }
4368
4369         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4370         if (likely(ol->ol_stripe_size != 0)) {
4371                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4372                                              PFID_STRIPE_IDX_BITS;
4373                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4374                 lma->lma_compat |= LMAC_STRIPE_INFO;
4375                 if (ol->ol_comp_id != 0) {
4376                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4377                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4378                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4379                         lma->lma_compat |= LMAC_COMP_INFO;
4380                 }
4381         }
4382
4383         lustre_loa_swab(loa, false);
4384
4385         /* Store the PFID EA inside LMA. */
4386         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4387                              flags);
4388         if (!rc)
4389                 obj->oo_pfid_in_lma = 1;
4390
4391         RETURN(rc);
4392 }
4393
4394 /*
4395  * In DNE environment, the object (in spite of regular file or directory)
4396  * and its name entry may reside on different MDTs. Under such case, we will
4397  * create an agent entry on the MDT where the object resides. The agent entry
4398  * references the object locally, that makes the object to be visible to the
4399  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4400  * such as 'tar' can handle the object properly.
4401  *
4402  * We handle the agent entry during set linkEA that is the common interface
4403  * for both regular file and directroy, can handle kinds of cases, such as
4404  * create/link/unlink/rename, and so on.
4405  *
4406  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4407  *
4408  * XXX: There are two known issues:
4409  * 1. For one object, we will create at most one agent entry even if there
4410  *    may be more than one cross-MDTs hard links on the object. So the local
4411  *    e2fsck may claim that the object's nlink is larger than the name entries
4412  *    that reference such inode. And in further, the e2fsck will fix the nlink
4413  *    attribute to match the local references. Then it will cause the object's
4414  *    nlink attribute to be inconsistent with the global references. it is bad
4415  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4416  *    other hand, the global namespace LFSCK can repair the object's attribute
4417  *    according to the linkEA.
4418  * 2. There may be too many hard links on the object as to its linkEA overflow,
4419  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4420  *    case happened, then at this point, we do not know whether there are some
4421  *    cross-MDTs reference. But there are local references, it guarantees that
4422  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4423  */
4424 static int osd_xattr_handle_linkea(const struct lu_env *env,
4425                                    struct osd_device *osd,
4426                                    struct osd_object *obj,
4427                                    const struct lu_buf *buf,
4428                                    struct thandle *handle)
4429 {
4430         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4431         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4432         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4433         struct lu_name tmpname;
4434         struct osd_thandle *oh;
4435         int rc;
4436         bool remote = false;
4437
4438         ENTRY;
4439
4440         oh = container_of0(handle, struct osd_thandle, ot_super);
4441         LASSERT(oh->ot_handle != NULL);
4442
4443         rc = linkea_init_with_rec(&ldata);
4444         if (!rc) {
4445                 linkea_first_entry(&ldata);
4446                 while (ldata.ld_lee != NULL && !remote) {
4447                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4448                                             &tmpname, tfid);
4449                         if (osd_remote_fid(env, osd, tfid) > 0)
4450                                 remote = true;
4451                         else
4452                                 linkea_next_entry(&ldata);
4453                 }
4454         } else if (rc == -ENODATA) {
4455                 rc = 0;
4456         } else {
4457                 RETURN(rc);
4458         }
4459
4460         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4461                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4462                 if (rc)
4463                         CERROR("%s: failed to remove agent entry for "DFID
4464                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4465         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4466                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4467                 if (rc)
4468                         CERROR("%s: failed to create agent entry for "DFID
4469                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4470         }
4471
4472         RETURN(rc);
4473 }
4474
4475 /*
4476  * Concurrency: @dt is write locked.
4477  */
4478 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4479                          const struct lu_buf *buf, const char *name, int fl,
4480                          struct thandle *handle)
4481 {
4482         struct osd_object *obj = osd_dt_obj(dt);
4483         struct osd_device *osd = osd_obj2dev(obj);
4484         struct inode *inode = obj->oo_inode;
4485         struct osd_thread_info *info = osd_oti_get(env);
4486         int fs_flags = 0;
4487         int len;
4488         int rc;
4489
4490         ENTRY;
4491
4492         LASSERT(handle);
4493         LASSERT(buf);
4494
4495         /* version set is not real XATTR */
4496         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4497                 dt_obj_version_t *version = buf->lb_buf;
4498
4499                 /*
4500                  * for version we are just using xattr API but change inode
4501                  * field instead
4502                  */
4503                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4504
4505                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4506                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4507
4508                 LDISKFS_I(inode)->i_fs_version = *version;
4509                 /*
4510                  * Version is set after all inode operations are finished,
4511                  * so we should mark it dirty here
4512                  */
4513                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4514
4515                 RETURN(0);
4516         }
4517
4518         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4519                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4520
4521         len = buf->lb_len;
4522         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4523
4524         /*
4525          * For the OST device with 256 bytes inode size by default,
4526          * the PFID EA will be stored together with LMA EA to avoid
4527          * performance trouble. Otherwise the PFID EA can be stored
4528          * independently. LU-8998
4529          */
4530         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4531             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4532                 LASSERT(buf->lb_buf);
4533
4534                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4535                 if (fl <= 0)
4536                         RETURN(fl);
4537         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4538                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4539                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4540
4541                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4542                 if (rc)
4543                         RETURN(rc);
4544
4545                 lma->lma_incompat |= LMAI_STRIPED;
4546                 lustre_lma_swab(lma);
4547                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4548                                      sizeof(*lma), XATTR_REPLACE);
4549                 if (rc != 0)
4550                         RETURN(rc);
4551         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4552                 LASSERT(!osd->od_is_ost);
4553
4554                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4555                 if (rc)
4556                         RETURN(rc);
4557         }
4558
4559         if (fl & LU_XATTR_REPLACE)
4560                 fs_flags |= XATTR_REPLACE;
4561
4562         if (fl & LU_XATTR_CREATE)
4563                 fs_flags |= XATTR_CREATE;
4564
4565         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4566         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4567
4568         if (rc == 0 &&
4569             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4570              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4571                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4572
4573         return rc;
4574 }
4575
4576 /*
4577  * Concurrency: @dt is read locked.
4578  */
4579 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4580                           const struct lu_buf *buf)
4581 {
4582         struct osd_object *obj = osd_dt_obj(dt);
4583         struct inode *inode = obj->oo_inode;
4584         struct osd_thread_info *info = osd_oti_get(env);
4585         struct dentry *dentry = &info->oti_obj_dentry;
4586
4587         if (!dt_object_exists(dt))
4588                 return -ENOENT;
4589
4590         LASSERT(!dt_object_remote(dt));
4591         LASSERT(inode->i_op != NULL);
4592         LASSERT(inode->i_op->listxattr != NULL);
4593
4594         dentry->d_inode = inode;
4595         dentry->d_sb = inode->i_sb;
4596         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4597 }
4598
4599 static int osd_declare_xattr_del(const struct lu_env *env,
4600                                  struct dt_object *dt, const char *name,
4601                                  struct thandle *handle)
4602 {
4603         struct osd_thandle *oh;
4604         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4605
4606         LASSERT(!dt_object_remote(dt));
4607         LASSERT(handle != NULL);
4608
4609         oh = container_of0(handle, struct osd_thandle, ot_super);
4610         LASSERT(oh->ot_handle == NULL);
4611
4612         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4613                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4614         /*
4615          * xattr del may involve inode quota change, reserve credits for
4616          * dquot_initialize()
4617          */
4618         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4619
4620         return 0;
4621 }
4622
4623 /*
4624  * Concurrency: @dt is write locked.
4625  */
4626 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4627                          const char *name, struct thandle *handle)
4628 {
4629         struct osd_object *obj = osd_dt_obj(dt);
4630         struct inode *inode = obj->oo_inode;
4631         struct osd_thread_info *info = osd_oti_get(env);
4632         struct dentry *dentry = &info->oti_obj_dentry;
4633         int rc;
4634
4635         if (!dt_object_exists(dt))
4636                 return -ENOENT;
4637
4638         LASSERT(!dt_object_remote(dt));
4639         LASSERT(inode->i_op != NULL);
4640         LASSERT(handle != NULL);
4641 #ifdef HAVE_IOP_XATTR
4642         LASSERT(inode->i_op->removexattr != NULL);
4643 #endif
4644
4645         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4646
4647         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4648                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4649
4650                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4651                                  &info->oti_ost_attrs);
4652                 if (!rc) {
4653                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4654
4655                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4656                         lustre_lma_swab(lma);
4657                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4658                                              sizeof(*lma), XATTR_REPLACE);
4659                         if (!rc)
4660                                 obj->oo_pfid_in_lma = 0;
4661                 }
4662         } else {
4663                 ll_vfs_dq_init(inode);
4664                 dentry->d_inode = inode;
4665                 dentry->d_sb = inode->i_sb;
4666                 rc = osd_removexattr(dentry, inode, name);
4667         }
4668
4669         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4670
4671         if (rc == 0 &&
4672             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4673              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4674                 osd_oxc_del(obj, name);
4675
4676         return rc;
4677 }
4678
4679 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4680                            __u64 start, __u64 end)
4681 {
4682         struct osd_object *obj = osd_dt_obj(dt);
4683         struct inode *inode = obj->oo_inode;
4684         struct osd_thread_info *info = osd_oti_get(env);
4685         struct dentry *dentry = &info->oti_obj_dentry;
4686         struct file *file = &info->oti_file;
4687         int rc;
4688
4689         ENTRY;
4690
4691         dentry->d_inode = inode;
4692         dentry->d_sb = inode->i_sb;
4693         file->f_path.dentry = dentry;
4694         file->f_mapping = inode->i_mapping;
4695         file->f_op = inode->i_fop;
4696         set_file_inode(file, inode);
4697
4698         rc = ll_vfs_fsync_range(file, start, end, 0);
4699
4700         RETURN(rc);
4701 }
4702
4703 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4704 {
4705         return 0;
4706 }
4707
4708 /*
4709  * Index operations.
4710  */
4711
4712 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4713                                const struct dt_index_features *feat)
4714 {
4715         struct iam_descr *descr;
4716
4717         if (osd_object_is_root(o))
4718                 return feat == &dt_directory_features;
4719
4720         LASSERT(o->oo_dir != NULL);
4721
4722         descr = o->oo_dir->od_container.ic_descr;
4723         if (feat == &dt_directory_features) {
4724                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4725                         return 1;
4726                 else
4727                         return 0;
4728         } else {
4729                 return feat->dif_keysize_min <= descr->id_key_size &&
4730                        descr->id_key_size <= feat->dif_keysize_max &&
4731                        feat->dif_recsize_min <= descr->id_rec_size &&
4732                        descr->id_rec_size <= feat->dif_recsize_max &&
4733                        !(feat->dif_flags & (DT_IND_VARKEY |
4734                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4735                        ergo(feat->dif_flags & DT_IND_UPDATE,
4736                             1 /* XXX check that object (and fs) is writable */);
4737         }
4738 }
4739
4740 static int osd_iam_container_init(const struct lu_env *env,
4741                                   struct osd_object *obj,
4742                                   struct osd_directory *dir)
4743 {
4744         struct iam_container *bag = &dir->od_container;
4745         int result;
4746
4747         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4748         if (result != 0)
4749                 return result;
4750
4751         result = iam_container_setup(bag);
4752         if (result == 0)
4753                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4754         else
4755                 iam_container_fini(bag);
4756
4757         return result;
4758 }
4759
4760
4761 /*
4762  * Concurrency: no external locking is necessary.
4763  */
4764 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4765                          const struct dt_index_features *feat)
4766 {
4767         int result;
4768         int skip_iam = 0;
4769         struct osd_object *obj = osd_dt_obj(dt);
4770
4771         LINVRNT(osd_invariant(obj));
4772
4773         if (osd_object_is_root(obj)) {
4774                 dt->do_index_ops = &osd_index_ea_ops;
4775                 result = 0;
4776         } else if (feat == &dt_directory_features) {
4777                 dt->do_index_ops = &osd_index_ea_ops;
4778                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4779                         result = 0;
4780                 else
4781                         result = -ENOTDIR;
4782                 skip_iam = 1;
4783         } else if (unlikely(feat == &dt_otable_features)) {
4784                 dt->do_index_ops = &osd_otable_ops;
4785                 return 0;
4786         } else if (unlikely(feat == &dt_acct_features)) {
4787                 dt->do_index_ops = &osd_acct_index_ops;
4788                 result = 0;
4789                 skip_iam = 1;
4790         } else if (!osd_has_index(obj)) {
4791                 struct osd_directory *dir;
4792                 struct osd_device *osd = osd_obj2dev(obj);
4793                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4794
4795                 OBD_ALLOC_PTR(dir);
4796                 if (dir) {
4797
4798                         spin_lock(&obj->oo_guard);
4799                         if (obj->oo_dir == NULL)
4800                                 obj->oo_dir = dir;
4801                         else
4802                                 /*
4803                                  * Concurrent thread allocated container data.
4804                                  */
4805                                 OBD_FREE_PTR(dir);
4806                         spin_unlock(&obj->oo_guard);
4807                         /*
4808                          * Now, that we have container data, serialize its
4809                          * initialization.
4810                          */
4811                         down_write(&obj->oo_ext_idx_sem);
4812                         /*
4813                          * recheck under lock.
4814                          */
4815
4816                         if (osd_has_index(obj)) {
4817                                 result = 0;
4818                                 goto unlock;
4819                         }
4820
4821                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4822                         if (result || feat == &dt_lfsck_namespace_features ||
4823                             feat == &dt_lfsck_layout_orphan_features ||
4824                             feat == &dt_lfsck_layout_dangling_features)
4825                                 goto unlock;
4826
4827                         result = osd_index_register(osd, fid,
4828                                                     feat->dif_keysize_max,
4829                                                     feat->dif_recsize_max);
4830                         if (result < 0)
4831                                 CWARN("%s: failed to register index "
4832                                       DFID": rc = %d\n",
4833                                       osd_name(osd), PFID(fid), result);
4834                         else if (result > 0)
4835                                 result = 0;
4836                         else
4837                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4838                                        " (%d/%d) registered\n",
4839                                        osd_name(osd), PFID(fid),
4840                                        (int)feat->dif_keysize_max,
4841                                        (int)feat->dif_recsize_max);
4842
4843 unlock:
4844                         up_write(&obj->oo_ext_idx_sem);
4845                 } else {
4846                         result = -ENOMEM;
4847                 }
4848         } else {
4849                 result = 0;
4850         }
4851
4852         if (result == 0 && skip_iam == 0) {
4853                 if (!osd_iam_index_probe(env, obj, feat))
4854                         result = -ENOTDIR;
4855         }
4856         LINVRNT(osd_invariant(obj));
4857
4858         return result;
4859 }
4860
4861 static int osd_otable_it_attr_get(const struct lu_env *env,
4862                                  struct dt_object *dt,
4863                                  struct lu_attr *attr)
4864 {
4865         attr->la_valid = 0;
4866         return 0;
4867 }
4868
4869 static const struct dt_object_operations osd_obj_ops = {
4870         .do_read_lock           = osd_read_lock,
4871         .do_write_lock          = osd_write_lock,
4872         .do_read_unlock         = osd_read_unlock,
4873         .do_write_unlock        = osd_write_unlock,
4874         .do_write_locked        = osd_write_locked,
4875         .do_attr_get            = osd_attr_get,
4876         .do_declare_attr_set    = osd_declare_attr_set,
4877         .do_attr_set            = osd_attr_set,
4878         .do_ah_init             = osd_ah_init,
4879         .do_declare_create      = osd_declare_create,
4880         .do_create              = osd_create,
4881         .do_declare_destroy     = osd_declare_destroy,
4882         .do_destroy             = osd_destroy,
4883         .do_index_try           = osd_index_try,
4884         .do_declare_ref_add     = osd_declare_ref_add,
4885         .do_ref_add             = osd_ref_add,
4886         .do_declare_ref_del     = osd_declare_ref_del,
4887         .do_ref_del             = osd_ref_del,
4888         .do_xattr_get           = osd_xattr_get,
4889         .do_declare_xattr_set   = osd_declare_xattr_set,
4890         .do_xattr_set           = osd_xattr_set,
4891         .do_declare_xattr_del   = osd_declare_xattr_del,
4892         .do_xattr_del           = osd_xattr_del,
4893         .do_xattr_list          = osd_xattr_list,
4894         .do_object_sync         = osd_object_sync,
4895         .do_invalidate          = osd_invalidate,
4896 };
4897
4898 static const struct dt_object_operations osd_obj_otable_it_ops = {
4899         .do_attr_get    = osd_otable_it_attr_get,
4900         .do_index_try   = osd_index_try,
4901 };
4902
4903 static int osd_index_declare_iam_delete(const struct lu_env *env,
4904                                         struct dt_object *dt,
4905                                         const struct dt_key *key,
4906                                         struct thandle *handle)
4907 {
4908         struct osd_thandle *oh;
4909
4910         oh = container_of0(handle, struct osd_thandle, ot_super);
4911         LASSERT(oh->ot_handle == NULL);
4912
4913         /* Recycle  may cause additional three blocks to be changed. */
4914         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4915                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4916
4917         return 0;
4918 }
4919
4920 /**
4921  *      delete a (key, value) pair from index \a dt specified by \a key
4922  *
4923  *      \param  dt      osd index object
4924  *      \param  key     key for index
4925  *      \param  rec     record reference
4926  *      \param  handle  transaction handler
4927  *
4928  *      \retval  0  success
4929  *      \retval -ve   failure
4930  */
4931 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4932                                 const struct dt_key *key,
4933                                 struct thandle *handle)
4934 {
4935         struct osd_thread_info *oti = osd_oti_get(env);
4936         struct osd_object *obj = osd_dt_obj(dt);
4937         struct osd_thandle *oh;
4938         struct iam_path_descr *ipd;
4939         struct iam_container *bag = &obj->oo_dir->od_container;
4940         int rc;
4941
4942         ENTRY;
4943
4944         if (!dt_object_exists(dt))
4945                 RETURN(-ENOENT);
4946
4947         LINVRNT(osd_invariant(obj));
4948         LASSERT(!dt_object_remote(dt));
4949         LASSERT(bag->ic_object == obj->oo_inode);
4950         LASSERT(handle != NULL);
4951
4952         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4953
4954         ipd = osd_idx_ipd_get(env, bag);
4955         if (unlikely(ipd == NULL))
4956                 RETURN(-ENOMEM);
4957
4958         oh = container_of0(handle, struct osd_thandle, ot_super);
4959         LASSERT(oh->ot_handle != NULL);
4960         LASSERT(oh->ot_handle->h_transaction != NULL);
4961
4962         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4963                 /* swab quota uid/gid provided by caller */
4964                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4965                 key = (const struct dt_key *)&oti->oti_quota_id;
4966         }
4967
4968         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4969         osd_ipd_put(env, bag, ipd);
4970         LINVRNT(osd_invariant(obj));
4971         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4972         RETURN(rc);
4973 }
4974
4975 static int osd_index_declare_ea_delete(const struct lu_env *env,
4976                                        struct dt_object *dt,
4977                                        const struct dt_key *key,
4978                                        struct thandle *handle)
4979 {
4980         struct osd_thandle *oh;
4981         struct inode *inode;
4982         int rc, credits;
4983
4984         ENTRY;
4985
4986         LASSERT(!dt_object_remote(dt));
4987         LASSERT(handle != NULL);
4988
4989         oh = container_of0(handle, struct osd_thandle, ot_super);
4990         LASSERT(oh->ot_handle == NULL);
4991
4992         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4993         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4994
4995         inode = osd_dt_obj(dt)->oo_inode;
4996         if (inode == NULL)
4997                 RETURN(-ENOENT);
4998
4999         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5000                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5001                                    NULL, OSD_QID_BLK);
5002         RETURN(rc);
5003 }
5004
5005 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5006                                           struct dt_rec *fid)
5007 {
5008         struct osd_fid_pack *rec;
5009         int rc = -ENODATA;
5010
5011         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5012                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5013                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5014                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5015                         rc = -EINVAL;
5016         }
5017         return rc;
5018 }
5019
5020 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5021                           const struct lu_fid *fid)
5022 {
5023         struct seq_server_site *ss = osd_seq_site(osd);
5024
5025         ENTRY;
5026
5027         /* FID seqs not in FLDB, must be local seq */
5028         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5029                 RETURN(0);
5030
5031         /*
5032          * If FLD is not being initialized yet, it only happens during the
5033          * initialization, likely during mgs initialization, and we assume
5034          * this is local FID.
5035          */
5036         if (ss == NULL || ss->ss_server_fld == NULL)
5037                 RETURN(0);
5038
5039         /* Only check the local FLDB here */
5040         if (osd_seq_exists(env, osd, fid_seq(fid)))
5041                 RETURN(0);
5042
5043         RETURN(1);
5044 }
5045
5046 static void osd_take_care_of_agent(const struct lu_env *env,
5047                                    struct osd_device *osd,
5048                                    struct osd_thandle *oh,
5049                                    struct ldiskfs_dir_entry_2 *de)
5050 {
5051         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5052         struct osd_idmap_cache *idc;
5053         int rc, schedule = 0;
5054
5055         LASSERT(de != NULL);
5056
5057         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5058         if (likely(rc == 0)) {
5059                 idc = osd_idc_find_or_init(env, osd, fid);
5060                 if (IS_ERR(idc) || idc->oic_remote)
5061                         schedule = 1;
5062         } else if (rc == -ENODATA) {
5063                 /*
5064                  * can't get FID, postpone to the end of the
5065                  * transaction when iget() is safe
5066                  */
5067                 schedule = 1;
5068         } else {
5069                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5070         }
5071         if (schedule)
5072                 osd_schedule_agent_inode_removal(env, oh,
5073                                                  le32_to_cpu(de->inode));
5074 }
5075
5076 /**
5077  * Index delete function for interoperability mode (b11826).
5078  * It will remove the directory entry added by osd_index_ea_insert().
5079  * This entry is needed to maintain name->fid mapping.
5080  *
5081  * \param key,  key i.e. file entry to be deleted
5082  *
5083  * \retval   0, on success
5084  * \retval -ve, on error
5085  */
5086 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5087                                const struct dt_key *key, struct thandle *handle)
5088 {
5089         struct osd_object *obj = osd_dt_obj(dt);
5090         struct inode *dir = obj->oo_inode;
5091         struct dentry *dentry;
5092         struct osd_thandle *oh;
5093         struct ldiskfs_dir_entry_2 *de = NULL;
5094         struct buffer_head *bh;
5095         struct htree_lock *hlock = NULL;
5096         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5097         int rc;
5098
5099         ENTRY;
5100
5101         if (!dt_object_exists(dt))
5102                 RETURN(-ENOENT);
5103
5104         LINVRNT(osd_invariant(obj));
5105         LASSERT(!dt_object_remote(dt));
5106         LASSERT(handle != NULL);
5107
5108         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5109
5110         oh = container_of(handle, struct osd_thandle, ot_super);
5111         LASSERT(oh->ot_handle != NULL);
5112         LASSERT(oh->ot_handle->h_transaction != NULL);
5113
5114         ll_vfs_dq_init(dir);
5115         dentry = osd_child_dentry_get(env, obj,
5116                                       (char *)key, strlen((char *)key));
5117
5118         if (obj->oo_hl_head != NULL) {
5119                 hlock = osd_oti_get(env)->oti_hlock;
5120                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5121                                    dir, LDISKFS_HLOCK_DEL);
5122         } else {
5123                 down_write(&obj->oo_ext_idx_sem);
5124         }
5125
5126         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5127         if (!IS_ERR(bh)) {
5128                 /*
5129                  * If this is not the ".." entry, it might be a remote DNE
5130                  * entry and  we need to check if the FID is for a remote
5131                  * MDT.  If the FID is  not in the directory entry (e.g.
5132                  * upgraded 1.8 filesystem without dirdata enabled) then
5133                  * we need to get the FID from the LMA. For a remote directory
5134                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5135                  * case.
5136                  *
5137                  * Delete the entry before the agent inode in order to
5138                  * simplify error handling.  At worst an error after deleting
5139                  * the entry first might leak the agent inode afterward. The
5140                  * reverse would need filesystem abort in case of error deleting
5141                  * the entry after the agent had been removed, or leave a
5142                  * dangling entry pointing at a random inode.
5143                  */
5144                 if (strcmp((char *)key, dotdot) != 0)
5145                         osd_take_care_of_agent(env, osd, oh, de);
5146                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5147                 brelse(bh);
5148         } else {
5149                 rc = PTR_ERR(bh);
5150         }
5151         if (hlock != NULL)
5152                 ldiskfs_htree_unlock(hlock);
5153         else
5154                 up_write(&obj->oo_ext_idx_sem);
5155
5156         GOTO(out, rc);
5157 out:
5158         LASSERT(osd_invariant(obj));
5159         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5160         RETURN(rc);
5161 }
5162
5163 /**
5164  *      Lookup index for \a key and copy record to \a rec.
5165  *
5166  *      \param  dt      osd index object
5167  *      \param  key     key for index
5168  *      \param  rec     record reference
5169  *
5170  *      \retval  +ve  success : exact mach
5171  *      \retval  0    return record with key not greater than \a key
5172  *      \retval -ve   failure
5173  */
5174 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5175                                 struct dt_rec *rec, const struct dt_key *key)
5176 {
5177         struct osd_object *obj = osd_dt_obj(dt);
5178         struct iam_path_descr *ipd;
5179         struct iam_container *bag = &obj->oo_dir->od_container;
5180         struct osd_thread_info *oti = osd_oti_get(env);
5181         struct iam_iterator *it = &oti->oti_idx_it;
5182         struct iam_rec *iam_rec;
5183         int rc;
5184
5185         ENTRY;
5186
5187         if (!dt_object_exists(dt))
5188                 RETURN(-ENOENT);
5189
5190         LASSERT(osd_invariant(obj));
5191         LASSERT(!dt_object_remote(dt));
5192         LASSERT(bag->ic_object == obj->oo_inode);
5193
5194         ipd = osd_idx_ipd_get(env, bag);
5195         if (IS_ERR(ipd))
5196                 RETURN(-ENOMEM);
5197
5198         /* got ipd now we can start iterator. */
5199         iam_it_init(it, bag, 0, ipd);
5200
5201         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5202                 /* swab quota uid/gid provided by caller */
5203                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5204                 key = (const struct dt_key *)&oti->oti_quota_id;
5205         }
5206
5207         rc = iam_it_get(it, (struct iam_key *)key);
5208         if (rc >= 0) {
5209                 if (S_ISDIR(obj->oo_inode->i_mode))
5210                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5211                 else
5212                         iam_rec = (struct iam_rec *)rec;
5213
5214                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5215
5216                 if (S_ISDIR(obj->oo_inode->i_mode))
5217                         osd_fid_unpack((struct lu_fid *)rec,
5218                                        (struct osd_fid_pack *)iam_rec);
5219                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5220                         osd_quota_unpack(obj, rec);
5221         }
5222
5223         iam_it_put(it);
5224         iam_it_fini(it);
5225         osd_ipd_put(env, bag, ipd);
5226
5227         LINVRNT(osd_invariant(obj));
5228
5229         RETURN(rc);
5230 }
5231
5232 static int osd_index_declare_iam_insert(const struct lu_env *env,
5233                                         struct dt_object *dt,
5234                                         const struct dt_rec *rec,
5235                                         const struct dt_key *key,
5236                                         struct thandle *handle)
5237 {
5238         struct osd_thandle *oh;
5239
5240         LASSERT(handle != NULL);
5241
5242         oh = container_of0(handle, struct osd_thandle, ot_super);
5243         LASSERT(oh->ot_handle == NULL);
5244
5245         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5246                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5247
5248         return 0;
5249 }
5250
5251 /**
5252  *      Inserts (key, value) pair in \a dt index object.
5253  *
5254  *      \param  dt      osd index object
5255  *      \param  key     key for index
5256  *      \param  rec     record reference
5257  *      \param  th      transaction handler
5258  *
5259  *      \retval  0  success
5260  *      \retval -ve failure
5261  */
5262 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5263                                 const struct dt_rec *rec,
5264                                 const struct dt_key *key, struct thandle *th)
5265 {
5266         struct osd_object *obj = osd_dt_obj(dt);
5267         struct iam_path_descr *ipd;
5268         struct osd_thandle *oh;
5269         struct iam_container *bag;
5270         struct osd_thread_info *oti = osd_oti_get(env);
5271         struct iam_rec *iam_rec;
5272         int rc;
5273
5274         ENTRY;
5275
5276         if (!dt_object_exists(dt))
5277                 RETURN(-ENOENT);
5278
5279         LINVRNT(osd_invariant(obj));
5280         LASSERT(!dt_object_remote(dt));
5281
5282         bag = &obj->oo_dir->od_container;
5283         LASSERT(bag->ic_object == obj->oo_inode);
5284         LASSERT(th != NULL);
5285
5286         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5287
5288         ipd = osd_idx_ipd_get(env, bag);
5289         if (unlikely(ipd == NULL))
5290                 RETURN(-ENOMEM);
5291
5292         oh = container_of0(th, struct osd_thandle, ot_super);
5293         LASSERT(oh->ot_handle != NULL);
5294         LASSERT(oh->ot_handle->h_transaction != NULL);
5295         if (S_ISDIR(obj->oo_inode->i_mode)) {
5296                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5297                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5298                              &oti->oti_fid);
5299         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5300                 /* pack quota uid/gid */
5301                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5302                 key = (const struct dt_key *)&oti->oti_quota_id;
5303                 /* pack quota record */
5304                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5305                 iam_rec = (struct iam_rec *)rec;
5306         } else {
5307                 iam_rec = (struct iam_rec *)rec;
5308         }
5309
5310         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5311                         iam_rec, ipd);
5312         osd_ipd_put(env, bag, ipd);
5313         LINVRNT(osd_invariant(obj));
5314         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5315         RETURN(rc);
5316 }
5317
5318 /**
5319  * Calls ldiskfs_add_entry() to add directory entry
5320  * into the directory. This is required for
5321  * interoperability mode (b11826)
5322  *
5323  * \retval   0, on success
5324  * \retval -ve, on error
5325  */
5326 static int __osd_ea_add_rec(struct osd_thread_info *info,
5327                             struct osd_object *pobj, struct inode  *cinode,
5328                             const char *name, const struct lu_fid *fid,
5329                             struct htree_lock *hlock, struct thandle *th)
5330 {
5331         struct ldiskfs_dentry_param *ldp;
5332         struct dentry *child;
5333         struct osd_thandle *oth;
5334         int rc;
5335
5336         oth = container_of(th, struct osd_thandle, ot_super);
5337         LASSERT(oth->ot_handle != NULL);
5338         LASSERT(oth->ot_handle->h_transaction != NULL);
5339         LASSERT(pobj->oo_inode);
5340
5341         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5342         if (unlikely(pobj->oo_inode ==
5343                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5344                 ldp->edp_magic = 0;
5345         else
5346                 osd_get_ldiskfs_dirent_param(ldp, fid);
5347         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5348         child->d_fsdata = (void *)ldp;
5349         ll_vfs_dq_init(pobj->oo_inode);
5350         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5351                                    child, cinode, hlock);
5352         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5353                 struct ldiskfs_dir_entry_2      *de;
5354                 struct buffer_head              *bh;
5355                 int                              rc1;
5356
5357                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5358                                             NULL, hlock);
5359                 if (!IS_ERR(bh)) {
5360                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5361                                                                bh);
5362                         if (rc1 == 0) {
5363                                 if (S_ISDIR(cinode->i_mode))
5364                                         de->file_type = LDISKFS_DIRENT_LUFID |
5365                                                         LDISKFS_FT_REG_FILE;
5366                                 else
5367                                         de->file_type = LDISKFS_DIRENT_LUFID |
5368                                                         LDISKFS_FT_DIR;
5369                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5370                                                               NULL, bh);
5371                         }
5372                         brelse(bh);
5373                 }
5374         }
5375
5376         RETURN(rc);
5377 }
5378
5379 /**
5380  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5381  * into the directory.Also sets flags into osd object to
5382  * indicate dot and dotdot are created. This is required for
5383  * interoperability mode (b11826)
5384  *
5385  * \param dir   directory for dot and dotdot fixup.
5386  * \param obj   child object for linking
5387  *
5388  * \retval   0, on success
5389  * \retval -ve, on error
5390  */
5391 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5392                               struct osd_object *dir,
5393                               struct inode *parent_dir, const char *name,
5394                               const struct lu_fid *dot_fid,
5395                               const struct lu_fid *dot_dot_fid,
5396                               struct thandle *th)
5397 {
5398         struct inode *inode = dir->oo_inode;
5399         struct osd_thandle *oth;
5400         int result = 0;
5401
5402         oth = container_of(th, struct osd_thandle, ot_super);
5403         LASSERT(oth->ot_handle->h_transaction != NULL);
5404         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5405
5406         if (strcmp(name, dot) == 0) {
5407                 if (dir->oo_compat_dot_created) {
5408                         result = -EEXIST;
5409                 } else {
5410                         LASSERT(inode->i_ino == parent_dir->i_ino);
5411                         dir->oo_compat_dot_created = 1;
5412                         result = 0;
5413                 }
5414         } else if (strcmp(name, dotdot) == 0) {
5415                 if (!dir->oo_compat_dot_created)
5416                         return -EINVAL;
5417                 /* in case of rename, dotdot is already created */
5418                 if (dir->oo_compat_dotdot_created) {
5419                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5420                                                 dot_dot_fid, NULL, th);
5421                 }
5422
5423                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5424                         struct lu_fid tfid = *dot_dot_fid;
5425
5426                         tfid.f_oid--;
5427                         result = osd_add_dot_dotdot_internal(info,
5428                                         dir->oo_inode, parent_dir, dot_fid,
5429                                         &tfid, oth);
5430                 } else {
5431                         result = osd_add_dot_dotdot_internal(info,
5432                                         dir->oo_inode, parent_dir, dot_fid,
5433                                         dot_dot_fid, oth);
5434                 }
5435
5436                 if (result == 0)
5437                         dir->oo_compat_dotdot_created = 1;
5438         }
5439
5440         return result;
5441 }
5442
5443
5444 /**
5445  * It will call the appropriate osd_add* function and return the
5446  * value, return by respective functions.
5447  */
5448 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5449                           struct inode *cinode, const char *name,
5450                           const struct lu_fid *fid, struct thandle *th)
5451 {
5452         struct osd_thread_info *info = osd_oti_get(env);
5453         struct htree_lock *hlock;
5454         int rc;
5455
5456         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5457
5458         if (name[0] == '.' && (name[1] == '\0' ||
5459                                (name[1] == '.' && name[2] == '\0'))) {
5460                 if (hlock != NULL) {
5461                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5462                                            pobj->oo_inode, 0);
5463                 } else {
5464                         down_write(&pobj->oo_ext_idx_sem);
5465                 }
5466
5467                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5468                                         lu_object_fid(&pobj->oo_dt.do_lu),
5469                                         fid, th);
5470         } else {
5471                 if (hlock != NULL) {
5472                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5473                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5474                 } else {
5475                         down_write(&pobj->oo_ext_idx_sem);
5476                 }
5477
5478                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5479                         struct lu_fid *tfid = &info->oti_fid;
5480
5481                         *tfid = *fid;
5482                         tfid->f_ver = ~0;
5483                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5484                                               tfid, hlock, th);
5485                 } else {
5486                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5487                                               hlock, th);
5488                 }
5489         }
5490         if (hlock != NULL)
5491                 ldiskfs_htree_unlock(hlock);
5492         else
5493                 up_write(&pobj->oo_ext_idx_sem);
5494
5495         return rc;
5496 }
5497
5498 static int
5499 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5500                       struct osd_idmap_cache *oic)
5501 {
5502         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5503         struct lu_fid *fid = &oic->oic_fid;
5504         struct osd_inode_id *id = &oic->oic_lid;
5505         struct inode *inode = NULL;
5506         int once = 0;
5507         bool insert;
5508         int rc;
5509
5510         ENTRY;
5511
5512         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5513                 RETURN(0);
5514
5515         if (thread_is_running(&scrub->os_thread) &&
5516             scrub->os_pos_current > id->oii_ino)
5517                 RETURN(0);
5518
5519         if (dev->od_auto_scrub_interval == AS_NEVER ||
5520             ktime_get_real_seconds() <
5521             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5522                 RETURN(0);
5523
5524 again:
5525         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5526         if (rc == -ENOENT) {
5527                 __u32 gen = id->oii_gen;
5528
5529                 insert = true;
5530                 if (inode != NULL)
5531                         goto trigger;
5532
5533                 inode = osd_iget(oti, dev, id);
5534                 /* The inode has been removed (by race maybe). */
5535                 if (IS_ERR(inode)) {
5536                         rc = PTR_ERR(inode);
5537
5538                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5539                 }
5540
5541                 /* The OI mapping is lost. */
5542                 if (gen != OSD_OII_NOGEN)
5543                         goto trigger;
5544
5545                 /*
5546                  * The inode may has been reused by others, we do not know,
5547                  * leave it to be handled by subsequent osd_fid_lookup().
5548                  */
5549                 GOTO(out, rc = 0);
5550         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5551                 GOTO(out, rc);
5552         }
5553
5554         insert = false;
5555
5556 trigger:
5557         if (thread_is_running(&scrub->os_thread)) {
5558                 if (inode == NULL) {
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
5568                 rc = osd_oii_insert(dev, oic, insert);
5569                 /*
5570                  * There is race condition between osd_oi_lookup and OI scrub.
5571                  * The OI scrub finished just after osd_oi_lookup() failure.
5572                  * Under such case, it is unnecessary to trigger OI scrub again,
5573                  * but try to call osd_oi_lookup() again.
5574                  */
5575                 if (unlikely(rc == -EAGAIN))
5576                         goto again;
5577
5578                 if (!S_ISDIR(inode->i_mode))
5579                         rc = 0;
5580                 else
5581                         rc = osd_check_lmv(oti, dev, inode, oic);
5582
5583                 GOTO(out, rc);
5584         }
5585
5586         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5587                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5588                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5589                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5590                        "%s: trigger partial OI scrub for RPC inconsistency "
5591                        "checking FID "DFID": rc = %d\n",
5592                        osd_dev2name(dev), PFID(fid), rc);
5593                 if (rc == 0 || rc == -EALREADY)
5594                         goto again;
5595         }
5596
5597         GOTO(out, rc);
5598
5599 out:
5600         if (inode)
5601                 iput(inode);
5602
5603         RETURN(rc);
5604 }
5605
5606 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5607                                struct osd_device *dev,
5608                                struct osd_idmap_cache *oic,
5609                                struct lu_fid *fid, __u32 ino)
5610 {
5611         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5612         struct inode *inode;
5613         int rc;
5614
5615         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5616         inode = osd_iget(oti, dev, &oic->oic_lid);
5617         if (IS_ERR(inode)) {
5618                 fid_zero(&oic->oic_fid);
5619                 return PTR_ERR(inode);
5620         }
5621
5622         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5623         iput(inode);
5624         if (rc != 0)
5625                 fid_zero(&oic->oic_fid);
5626         else
5627                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5628         return rc;
5629 }
5630
5631 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5632                       struct osd_inode_id *id, const struct lu_fid *fid)
5633 {
5634         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5635                id->oii_ino, id->oii_gen, info);
5636         info->oti_cache.oic_lid = *id;
5637         info->oti_cache.oic_fid = *fid;
5638         info->oti_cache.oic_dev = osd;
5639 }
5640
5641 /**
5642  * Get parent FID from the linkEA.
5643  *
5644  * For a directory which parent resides on remote MDT, to satisfy the
5645  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5646  * the other hand, to make the lookup(..) on the directory can return
5647  * the real parent FID, we append the real parent FID after its ".."
5648  * name entry in the /REMOTE_PARENT_DIR.
5649  *
5650  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5651  * backup. So after the restore, we cannot get the right parent FID from
5652  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5653  * we have stored the real parent FID in the directory object's linkEA,
5654  * we can parse the linkEA for the real parent FID.
5655  *
5656  * \param[in] env       pointer to the thread context
5657  * \param[in] obj       pointer to the object to be handled
5658  * \param[out]fid       pointer to the buffer to hold the parent FID
5659  *
5660  * \retval              0 for getting the real parent FID successfully
5661  * \retval              negative error number on failure
5662  */
5663 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5664                                     struct osd_object *obj,
5665                                     struct lu_fid *fid)
5666 {
5667         struct osd_thread_info *oti = osd_oti_get(env);
5668         struct lu_buf *buf = &oti->oti_big_buf;
5669         struct dentry *dentry = &oti->oti_obj_dentry;
5670         struct inode *inode = obj->oo_inode;
5671         struct linkea_data ldata = { NULL };
5672         int rc;
5673
5674         ENTRY;
5675
5676         fid_zero(fid);
5677         if (!S_ISDIR(inode->i_mode))
5678                 RETURN(-EIO);
5679
5680 again:
5681         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5682                              buf->lb_buf, buf->lb_len);
5683         if (rc == -ERANGE) {
5684                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5685                                      NULL, 0);
5686                 if (rc > 0) {
5687                         lu_buf_realloc(buf, rc);
5688                         if (buf->lb_buf == NULL)
5689                                 RETURN(-ENOMEM);
5690
5691                         goto again;
5692                 }
5693         }
5694
5695         if (unlikely(rc == 0))
5696                 RETURN(-ENODATA);
5697
5698         if (rc < 0)
5699                 RETURN(rc);
5700
5701         if (unlikely(buf->lb_buf == NULL)) {
5702                 lu_buf_realloc(buf, rc);
5703                 if (buf->lb_buf == NULL)
5704                         RETURN(-ENOMEM);
5705
5706                 goto again;
5707         }
5708
5709         ldata.ld_buf = buf;
5710         rc = linkea_init_with_rec(&ldata);
5711         if (!rc) {
5712                 linkea_first_entry(&ldata);
5713                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5714         }
5715
5716         RETURN(rc);
5717 }
5718
5719 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5720                                     struct inode *inode,
5721                                     const struct lu_fid *pfid,
5722                                     const char *name, const int namelen)
5723 {
5724         struct osd_thread_info *oti = osd_oti_get(env);
5725         struct lu_buf *buf = &oti->oti_big_buf;
5726         struct dentry *dentry = &oti->oti_obj_dentry;
5727         struct linkea_data ldata = { NULL };
5728         struct lu_name cname = { .ln_name = name,
5729                                  .ln_namelen = namelen };
5730         int rc;
5731
5732         ENTRY;
5733
5734 again:
5735         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5736                              buf->lb_buf, buf->lb_len);
5737         if (rc == -ERANGE)
5738                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5739
5740         if (rc < 0)
5741                 RETURN(rc);
5742
5743         if (unlikely(rc == 0))
5744                 RETURN(-ENODATA);
5745
5746         if (buf->lb_len < rc) {
5747                 lu_buf_realloc(buf, rc);
5748                 if (buf->lb_buf == NULL)
5749                         RETURN(-ENOMEM);
5750
5751                 goto again;
5752         }
5753
5754         ldata.ld_buf = buf;
5755         rc = linkea_init_with_rec(&ldata);
5756         if (!rc)
5757                 rc = linkea_links_find(&ldata, &cname, pfid);
5758
5759         RETURN(rc);
5760 }
5761
5762 /**
5763  * Calls ->lookup() to find dentry. From dentry get inode and
5764  * read inode's ea to get fid. This is required for  interoperability
5765  * mode (b11826)
5766  *
5767  * \retval   0, on success
5768  * \retval -ve, on error
5769  */
5770 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5771                              struct dt_rec *rec, const struct dt_key *key)
5772 {
5773         struct inode *dir = obj->oo_inode;
5774         struct dentry *dentry;
5775         struct ldiskfs_dir_entry_2 *de;
5776         struct buffer_head *bh;
5777         struct lu_fid *fid = (struct lu_fid *)rec;
5778         struct htree_lock *hlock = NULL;
5779         int ino;
5780         int rc;
5781
5782         ENTRY;
5783
5784         LASSERT(dir->i_op != NULL);
5785         LASSERT(dir->i_op->lookup != NULL);
5786
5787         dentry = osd_child_dentry_get(env, obj,
5788                                       (char *)key, strlen((char *)key));
5789
5790         if (obj->oo_hl_head != NULL) {
5791                 hlock = osd_oti_get(env)->oti_hlock;
5792                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5793                                    dir, LDISKFS_HLOCK_LOOKUP);
5794         } else {
5795                 down_read(&obj->oo_ext_idx_sem);
5796         }
5797
5798         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5799         if (!IS_ERR(bh)) {
5800                 struct osd_thread_info *oti = osd_oti_get(env);
5801                 struct osd_inode_id *id = &oti->oti_id;
5802                 struct osd_idmap_cache *oic = &oti->oti_cache;
5803                 struct osd_device *dev = osd_obj2dev(obj);
5804
5805                 ino = le32_to_cpu(de->inode);
5806                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5807                         brelse(bh);
5808                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5809                         GOTO(out, rc);
5810                 }
5811
5812                 rc = osd_get_fid_from_dentry(de, rec);
5813
5814                 /* done with de, release bh */
5815                 brelse(bh);
5816                 if (rc != 0) {
5817                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5818                                 const char *name = (const char *)key;
5819
5820                                 /*
5821                                  * If the parent is on remote MDT, and there
5822                                  * is no FID-in-dirent, then we have to get
5823                                  * the parent FID from the linkEA.
5824                                  */
5825                                 if (likely(strlen(name) == 2 &&
5826                                            name[0] == '.' && name[1] == '.'))
5827                                         rc = osd_get_pfid_from_linkea(env, obj,
5828                                                                       fid);
5829                         } else {
5830                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5831                         }
5832                 } else {
5833                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5834                 }
5835
5836                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5837                         fid_zero(&oic->oic_fid);
5838
5839                         GOTO(out, rc);
5840                 }
5841
5842                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5843                 rc = osd_consistency_check(oti, dev, oic);
5844                 if (rc == -ENOENT)
5845                         fid_zero(&oic->oic_fid);
5846                 else
5847                         /* Other error should not affect lookup result. */
5848                         rc = 0;
5849         } else {
5850                 rc = PTR_ERR(bh);
5851         }
5852
5853         GOTO(out, rc);
5854
5855 out:
5856         if (hlock != NULL)
5857                 ldiskfs_htree_unlock(hlock);
5858         else
5859                 up_read(&obj->oo_ext_idx_sem);
5860         return rc;
5861 }
5862
5863 static int osd_index_declare_ea_insert(const struct lu_env *env,
5864                                        struct dt_object *dt,
5865                                        const struct dt_rec *rec,
5866                                        const struct dt_key *key,
5867                                        struct thandle *handle)
5868 {
5869         struct osd_thandle *oh;
5870         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5871         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5872         const struct lu_fid *fid = rec1->rec_fid;
5873         int credits, rc = 0;
5874         struct osd_idmap_cache *idc;
5875
5876         ENTRY;
5877
5878         LASSERT(!dt_object_remote(dt));
5879         LASSERT(handle != NULL);
5880         LASSERT(fid != NULL);
5881         LASSERT(rec1->rec_type != 0);
5882
5883         oh = container_of0(handle, struct osd_thandle, ot_super);
5884         LASSERT(oh->ot_handle == NULL);
5885
5886         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5887
5888         /*
5889          * we can't call iget() while a transactions is running
5890          * (this can lead to a deadlock), but we need to know
5891          * inum and object type. so we find this information at
5892          * declaration and cache in per-thread info
5893          */
5894         idc = osd_idc_find_or_init(env, osd, fid);
5895         if (IS_ERR(idc))
5896                 RETURN(PTR_ERR(idc));
5897         if (idc->oic_remote) {
5898                 /*
5899                  * a reference to remote inode is represented by an
5900                  * agent inode which we have to create
5901                  */
5902                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5903                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5904         }
5905
5906         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5907
5908         if (osd_dt_obj(dt)->oo_inode != NULL) {
5909                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5910
5911                 /*
5912                  * We ignore block quota on meta pool (MDTs), so needn't
5913                  * calculate how many blocks will be consumed by this index
5914                  * insert
5915                  */
5916                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5917                                            i_gid_read(inode),
5918                                            i_projid_read(inode), 0,
5919                                            oh, osd_dt_obj(dt), NULL,
5920                                            OSD_QID_BLK);
5921                 if (rc)
5922                         RETURN(rc);
5923
5924 #ifdef HAVE_PROJECT_QUOTA
5925                 /*
5926                  * Reserve credits for local agent inode to transfer
5927                  * to 0, quota enforcement is ignored in this case.
5928                  */
5929                 if (idc->oic_remote &&
5930                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5931                     i_projid_read(inode) != 0)
5932                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5933                                                   0, i_projid_read(inode),
5934                                                   0, false, PRJQUOTA, true);
5935 #endif
5936         }
5937
5938         RETURN(rc);
5939 }
5940
5941 /**
5942  * Index add function for interoperability mode (b11826).
5943  * It will add the directory entry.This entry is needed to
5944  * maintain name->fid mapping.
5945  *
5946  * \param key it is key i.e. file entry to be inserted
5947  * \param rec it is value of given key i.e. fid
5948  *
5949  * \retval   0, on success
5950  * \retval -ve, on error
5951  */
5952 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5953                                const struct dt_rec *rec,
5954                                const struct dt_key *key, struct thandle *th)
5955 {
5956         struct osd_object *obj = osd_dt_obj(dt);
5957         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5958         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5959         const struct lu_fid *fid = rec1->rec_fid;
5960         const char *name = (const char *)key;
5961         struct osd_thread_info *oti = osd_oti_get(env);
5962         struct inode *child_inode = NULL;
5963         struct osd_idmap_cache *idc;
5964         int rc;
5965
5966         ENTRY;
5967
5968         if (!dt_object_exists(dt))
5969                 RETURN(-ENOENT);
5970
5971         LASSERT(osd_invariant(obj));
5972         LASSERT(!dt_object_remote(dt));
5973         LASSERT(th != NULL);
5974
5975         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5976
5977         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5978
5979         idc = osd_idc_find(env, osd, fid);
5980         if (unlikely(idc == NULL)) {
5981                 idc = osd_idc_find_or_init(env, osd, fid);
5982                 if (IS_ERR(idc)) {
5983                         /*
5984                          * this dt_insert() wasn't declared properly, so
5985                          * FID is missing in OI cache. we better do not
5986                          * lookup FID in FLDB/OI and don't risk to deadlock,
5987                          * but in some special cases (lfsck testing, etc)
5988                          * it's much simpler than fixing a caller.
5989                          *
5990                          * normally this error should be placed after the first
5991                          * find, but migrate may attach source stripes to
5992                          * target, which doesn't create stripes.
5993                          */
5994                         CERROR("%s: "DFID" wasn't declared for insert\n",
5995                                osd_name(osd), PFID(fid));
5996                         dump_stack();
5997                         RETURN(PTR_ERR(idc));
5998                 }
5999         }
6000
6001         if (idc->oic_remote) {
6002                 /* Insert remote entry */
6003                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6004                         child_inode =
6005                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6006                 } else {
6007                         child_inode = osd_create_local_agent_inode(env, osd,
6008                                         obj, fid, rec1->rec_type & S_IFMT, th);
6009                         if (IS_ERR(child_inode))
6010                                 RETURN(PTR_ERR(child_inode));
6011                 }
6012         } else {
6013                 /* Insert local entry */
6014                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6015                         /* for a reason OI cache wasn't filled properly */
6016                         CERROR("%s: OIC for "DFID" isn't filled\n",
6017                                osd_name(osd), PFID(fid));
6018                         RETURN(-EINVAL);
6019                 }
6020                 child_inode = oti->oti_inode;
6021                 if (unlikely(child_inode == NULL)) {
6022                         struct ldiskfs_inode_info *lii;
6023
6024                         OBD_ALLOC_PTR(lii);
6025                         if (lii == NULL)
6026                                 RETURN(-ENOMEM);
6027                         child_inode = oti->oti_inode = &lii->vfs_inode;
6028                 }
6029                 child_inode->i_sb = osd_sb(osd);
6030                 child_inode->i_ino = idc->oic_lid.oii_ino;
6031                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6032         }
6033
6034         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6035
6036         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6037                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6038
6039         if (child_inode && child_inode != oti->oti_inode)
6040                 iput(child_inode);
6041         LASSERT(osd_invariant(obj));
6042         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6043         RETURN(rc);
6044 }
6045
6046 /**
6047  *  Initialize osd Iterator for given osd index object.
6048  *
6049  *  \param  dt      osd index object
6050  */
6051
6052 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6053                                      struct dt_object *dt,
6054                                      __u32 unused)
6055 {
6056         struct osd_it_iam *it;
6057         struct osd_object *obj = osd_dt_obj(dt);
6058         struct lu_object *lo = &dt->do_lu;
6059         struct iam_path_descr *ipd;
6060         struct iam_container *bag = &obj->oo_dir->od_container;
6061
6062         if (!dt_object_exists(dt))
6063                 return ERR_PTR(-ENOENT);
6064
6065         OBD_ALLOC_PTR(it);
6066         if (it == NULL)
6067                 return ERR_PTR(-ENOMEM);
6068
6069         ipd = osd_it_ipd_get(env, bag);
6070         if (likely(ipd != NULL)) {
6071                 it->oi_obj = obj;
6072                 it->oi_ipd = ipd;
6073                 lu_object_get(lo);
6074                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6075                 return (struct dt_it *)it;
6076         } else {
6077                 OBD_FREE_PTR(it);
6078                 return ERR_PTR(-ENOMEM);
6079         }
6080 }
6081
6082 /**
6083  * free given Iterator.
6084  */
6085 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6086 {
6087         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6088         struct osd_object *obj = it->oi_obj;
6089
6090         iam_it_fini(&it->oi_it);
6091         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6092         osd_object_put(env, obj);
6093         OBD_FREE_PTR(it);
6094 }
6095
6096 /**
6097  *  Move Iterator to record specified by \a key
6098  *
6099  *  \param  di      osd iterator
6100  *  \param  key     key for index
6101  *
6102  *  \retval +ve  di points to record with least key not larger than key
6103  *  \retval  0   di points to exact matched key
6104  *  \retval -ve  failure
6105  */
6106
6107 static int osd_it_iam_get(const struct lu_env *env,
6108                           struct dt_it *di, const struct dt_key *key)
6109 {
6110         struct osd_thread_info *oti = osd_oti_get(env);
6111         struct osd_it_iam *it = (struct osd_it_iam *)di;
6112
6113         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6114                 /* swab quota uid/gid */
6115                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6116                 key = (struct dt_key *)&oti->oti_quota_id;
6117         }
6118
6119         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6120 }
6121
6122 /**
6123  *  Release Iterator
6124  *
6125  *  \param  di      osd iterator
6126  */
6127 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6128 {
6129         struct osd_it_iam *it = (struct osd_it_iam *)di;
6130
6131         iam_it_put(&it->oi_it);
6132 }
6133
6134 /**
6135  *  Move iterator by one record
6136  *
6137  *  \param  di      osd iterator
6138  *
6139  *  \retval +1   end of container reached
6140  *  \retval  0   success
6141  *  \retval -ve  failure
6142  */
6143
6144 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6145 {
6146         struct osd_it_iam *it = (struct osd_it_iam *)di;
6147
6148         return iam_it_next(&it->oi_it);
6149 }
6150
6151 /**
6152  * Return pointer to the key under iterator.
6153  */
6154
6155 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6156                                      const struct dt_it *di)
6157 {
6158         struct osd_thread_info *oti = osd_oti_get(env);
6159         struct osd_it_iam *it = (struct osd_it_iam *)di;
6160         struct osd_object *obj = it->oi_obj;
6161         struct dt_key *key;
6162
6163         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6164
6165         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6166                 /* swab quota uid/gid */
6167                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6168                 key = (struct dt_key *)&oti->oti_quota_id;
6169         }
6170
6171         return key;
6172 }
6173
6174 /**
6175  * Return size of key under iterator (in bytes)
6176  */
6177
6178 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6179 {
6180         struct osd_it_iam *it = (struct osd_it_iam *)di;
6181
6182         return iam_it_key_size(&it->oi_it);
6183 }
6184
6185 static inline void
6186 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6187 {
6188         /* check if file type is required */
6189         if (ent->lde_attrs & LUDA_TYPE) {
6190                 struct luda_type *lt;
6191                 int align = sizeof(*lt) - 1;
6192
6193                 len = (len + align) & ~align;
6194                 lt = (struct luda_type *)(ent->lde_name + len);
6195                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6196         }
6197
6198         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6199 }
6200
6201 /**
6202  * build lu direct from backend fs dirent.
6203  */
6204
6205 static inline void
6206 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6207                    char *name, __u16 namelen, __u16 type, __u32 attr)
6208 {
6209         ent->lde_attrs = attr | LUDA_FID;
6210         fid_cpu_to_le(&ent->lde_fid, fid);
6211
6212         ent->lde_hash = cpu_to_le64(offset);
6213         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6214
6215         strncpy(ent->lde_name, name, namelen);
6216         ent->lde_name[namelen] = '\0';
6217         ent->lde_namelen = cpu_to_le16(namelen);
6218
6219         /* append lustre attributes */
6220         osd_it_append_attrs(ent, namelen, type);
6221 }
6222
6223 /**
6224  * Return pointer to the record under iterator.
6225  */
6226 static int osd_it_iam_rec(const struct lu_env *env,
6227                           const struct dt_it *di,
6228                           struct dt_rec *dtrec, __u32 attr)
6229 {
6230         struct osd_it_iam *it = (struct osd_it_iam *)di;
6231         struct osd_thread_info *info = osd_oti_get(env);
6232
6233         ENTRY;
6234
6235         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6236                 const struct osd_fid_pack *rec;
6237                 struct lu_fid *fid = &info->oti_fid;
6238                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6239                 char *name;
6240                 int namelen;
6241                 __u64 hash;
6242                 int rc;
6243
6244                 name = (char *)iam_it_key_get(&it->oi_it);
6245                 if (IS_ERR(name))
6246                         RETURN(PTR_ERR(name));
6247
6248                 namelen = iam_it_key_size(&it->oi_it);
6249
6250                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6251                 if (IS_ERR(rec))
6252                         RETURN(PTR_ERR(rec));
6253
6254                 rc = osd_fid_unpack(fid, rec);
6255                 if (rc)
6256                         RETURN(rc);
6257
6258                 hash = iam_it_store(&it->oi_it);
6259
6260                 /* IAM does not store object type in IAM index (dir) */
6261                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6262                                    0, LUDA_FID);
6263         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6264                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6265                            (struct iam_rec *)dtrec);
6266                 osd_quota_unpack(it->oi_obj, dtrec);
6267         } else {
6268                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6269                            (struct iam_rec *)dtrec);
6270         }
6271
6272         RETURN(0);
6273 }
6274
6275 /**
6276  * Returns cookie for current Iterator position.
6277  */
6278 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6279 {
6280         struct osd_it_iam *it = (struct osd_it_iam *)di;
6281
6282         return iam_it_store(&it->oi_it);
6283 }
6284
6285 /**
6286  * Restore iterator from cookie.
6287  *
6288  * \param  di      osd iterator
6289  * \param  hash    Iterator location cookie
6290  *
6291  * \retval +ve  di points to record with least key not larger than key.
6292  * \retval  0   di points to exact matched key
6293  * \retval -ve  failure
6294  */
6295
6296 static int osd_it_iam_load(const struct lu_env *env,
6297                            const struct dt_it *di, __u64 hash)
6298 {
6299         struct osd_it_iam *it = (struct osd_it_iam *)di;
6300
6301         return iam_it_load(&it->oi_it, hash);
6302 }
6303
6304 static const struct dt_index_operations osd_index_iam_ops = {
6305         .dio_lookup         = osd_index_iam_lookup,
6306         .dio_declare_insert = osd_index_declare_iam_insert,
6307         .dio_insert         = osd_index_iam_insert,
6308         .dio_declare_delete = osd_index_declare_iam_delete,
6309         .dio_delete         = osd_index_iam_delete,
6310         .dio_it     = {
6311                 .init     = osd_it_iam_init,
6312                 .fini     = osd_it_iam_fini,
6313                 .get      = osd_it_iam_get,
6314                 .put      = osd_it_iam_put,
6315                 .next     = osd_it_iam_next,
6316                 .key      = osd_it_iam_key,
6317                 .key_size = osd_it_iam_key_size,
6318                 .rec      = osd_it_iam_rec,
6319                 .store    = osd_it_iam_store,
6320                 .load     = osd_it_iam_load
6321         }
6322 };
6323
6324
6325 /**
6326  * Creates or initializes iterator context.
6327  *
6328  * \retval struct osd_it_ea, iterator structure on success
6329  *
6330  */
6331 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6332                                     struct dt_object *dt,
6333                                     __u32 attr)
6334 {
6335         struct osd_object *obj = osd_dt_obj(dt);
6336         struct osd_thread_info *info = osd_oti_get(env);
6337         struct osd_it_ea *oie;
6338         struct file *file;
6339         struct lu_object *lo = &dt->do_lu;
6340         struct dentry *obj_dentry;
6341
6342         ENTRY;
6343
6344         if (!dt_object_exists(dt) || obj->oo_destroyed)
6345                 RETURN(ERR_PTR(-ENOENT));
6346
6347         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6348         if (oie == NULL)
6349                 RETURN(ERR_PTR(-ENOMEM));
6350         obj_dentry = &oie->oie_dentry;
6351
6352         obj_dentry->d_inode = obj->oo_inode;
6353         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6354         obj_dentry->d_name.hash = 0;
6355
6356         oie->oie_rd_dirent       = 0;
6357         oie->oie_it_dirent       = 0;
6358         oie->oie_dirent          = NULL;
6359         if (unlikely(!info->oti_it_ea_buf_used)) {
6360                 oie->oie_buf = info->oti_it_ea_buf;
6361                 info->oti_it_ea_buf_used = 1;
6362         } else {
6363                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6364                 if (oie->oie_buf == NULL)
6365                         RETURN(ERR_PTR(-ENOMEM));
6366         }
6367         oie->oie_obj = obj;
6368
6369         file = &oie->oie_file;
6370
6371         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6372         if (attr & LUDA_64BITHASH)
6373                 file->f_mode    = FMODE_64BITHASH;
6374         else
6375                 file->f_mode    = FMODE_32BITHASH;
6376         file->f_path.dentry     = obj_dentry;
6377         file->f_mapping         = obj->oo_inode->i_mapping;
6378         file->f_op              = obj->oo_inode->i_fop;
6379         set_file_inode(file, obj->oo_inode);
6380
6381         lu_object_get(lo);
6382         RETURN((struct dt_it *)oie);
6383 }
6384
6385 /**
6386  * Destroy or finishes iterator context.
6387  *
6388  * \param di iterator structure to be destroyed
6389  */
6390 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6391 {
6392         struct osd_thread_info *info = osd_oti_get(env);
6393         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6394         struct osd_object *obj = oie->oie_obj;
6395         struct inode *inode = obj->oo_inode;
6396
6397         ENTRY;
6398         oie->oie_file.f_op->release(inode, &oie->oie_file);
6399         osd_object_put(env, obj);
6400         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6401                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6402         else
6403                 info->oti_it_ea_buf_used = 0;
6404         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6405         EXIT;
6406 }
6407
6408 /**
6409  * It position the iterator at given key, so that next lookup continues from
6410  * that key Or it is similar to dio_it->load() but based on a key,
6411  * rather than file position.
6412  *
6413  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6414  * to the beginning.
6415  *
6416  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6417  */
6418 static int osd_it_ea_get(const struct lu_env *env,
6419                          struct dt_it *di, const struct dt_key *key)
6420 {
6421         struct osd_it_ea *it = (struct osd_it_ea *)di;
6422
6423         ENTRY;
6424         LASSERT(((const char *)key)[0] == '\0');
6425         it->oie_file.f_pos = 0;
6426         it->oie_rd_dirent = 0;
6427         it->oie_it_dirent = 0;
6428         it->oie_dirent = NULL;
6429
6430         RETURN(+1);
6431 }
6432
6433 /**
6434  * Does nothing
6435  */
6436 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6437 {
6438 }
6439
6440 struct osd_filldir_cbs {
6441 #ifdef HAVE_DIR_CONTEXT
6442         struct dir_context ctx;
6443 #endif
6444         struct osd_it_ea  *it;
6445 };
6446 /**
6447  * It is called internally by ->readdir(). It fills the
6448  * iterator's in-memory data structure with required
6449  * information i.e. name, namelen, rec_size etc.
6450  *
6451  * \param buf in which information to be filled in.
6452  * \param name name of the file in given dir
6453  *
6454  * \retval 0 on success
6455  * \retval 1 on buffer full
6456  */
6457 #ifdef HAVE_FILLDIR_USE_CTX
6458 static int osd_ldiskfs_filldir(struct dir_context *buf,
6459 #else
6460 static int osd_ldiskfs_filldir(void *buf,
6461 #endif
6462                                const char *name, int namelen,
6463                                loff_t offset, __u64 ino, unsigned int d_type)
6464 {
6465         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6466         struct osd_object *obj = it->oie_obj;
6467         struct osd_it_ea_dirent *ent = it->oie_dirent;
6468         struct lu_fid *fid = &ent->oied_fid;
6469         struct osd_fid_pack *rec;
6470
6471         ENTRY;
6472
6473 /* this should never happen */
6474         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6475                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6476                 RETURN(-EIO);
6477         }
6478
6479         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6480             OSD_IT_EA_BUFSIZE)
6481                 RETURN(1);
6482
6483         /* "." is just the object itself. */
6484         if (namelen == 1 && name[0] == '.') {
6485                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6486         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6487                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6488                 if (osd_fid_unpack(fid, rec) != 0)
6489                         fid_zero(fid);
6490         } else {
6491                 fid_zero(fid);
6492         }
6493         d_type &= ~LDISKFS_DIRENT_LUFID;
6494
6495         /* NOT export local root. */
6496         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6497                 ino = obj->oo_inode->i_ino;
6498                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6499         }
6500
6501         ent->oied_ino     = ino;
6502         ent->oied_off     = offset;
6503         ent->oied_namelen = namelen;
6504         ent->oied_type    = d_type;
6505
6506         memcpy(ent->oied_name, name, namelen);
6507
6508         it->oie_rd_dirent++;
6509         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6510         RETURN(0);
6511 }
6512
6513 /**
6514  * Calls ->readdir() to load a directory entry at a time
6515  * and stored it in iterator's in-memory data structure.
6516  *
6517  * \param di iterator's in memory structure
6518  *
6519  * \retval   0 on success
6520  * \retval -ve on error
6521  * \retval +1 reach the end of entry
6522  */
6523 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6524                                const struct dt_it *di)
6525 {
6526         struct osd_it_ea *it = (struct osd_it_ea *)di;
6527         struct osd_object *obj = it->oie_obj;
6528         struct inode *inode = obj->oo_inode;
6529         struct htree_lock *hlock = NULL;
6530         struct file *filp = &it->oie_file;
6531         int rc = 0;
6532         struct osd_filldir_cbs buf = {
6533 #ifdef HAVE_DIR_CONTEXT
6534                 .ctx.actor = osd_ldiskfs_filldir,
6535 #endif
6536                 .it = it
6537         };
6538
6539         ENTRY;
6540         it->oie_dirent = it->oie_buf;
6541         it->oie_rd_dirent = 0;
6542
6543         if (obj->oo_hl_head != NULL) {
6544                 hlock = osd_oti_get(env)->oti_hlock;
6545                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6546                                    inode, LDISKFS_HLOCK_READDIR);
6547         } else {
6548                 down_read(&obj->oo_ext_idx_sem);
6549         }
6550
6551 #ifdef HAVE_DIR_CONTEXT
6552         buf.ctx.pos = filp->f_pos;
6553 #ifdef HAVE_ITERATE_SHARED
6554         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6555 #else
6556         rc = inode->i_fop->iterate(filp, &buf.ctx);
6557 #endif
6558         filp->f_pos = buf.ctx.pos;
6559 #else
6560         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6561 #endif
6562
6563         if (hlock != NULL)
6564                 ldiskfs_htree_unlock(hlock);
6565         else
6566                 up_read(&obj->oo_ext_idx_sem);
6567
6568         if (it->oie_rd_dirent == 0) {
6569                 /*
6570                  * If it does not get any dirent, it means it has been reached
6571                  * to the end of the dir
6572                  */
6573                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6574                 if (rc == 0)
6575                         rc = 1;
6576         } else {
6577                 it->oie_dirent = it->oie_buf;
6578                 it->oie_it_dirent = 1;
6579         }
6580
6581         RETURN(rc);
6582 }
6583
6584 /**
6585  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6586  * to load a directory entry at a time and stored it in
6587  * iterator's in-memory data structure.
6588  *
6589  * \param di iterator's in memory structure
6590  *
6591  * \retval +ve iterator reached to end
6592  * \retval   0 iterator not reached to end
6593  * \retval -ve on error
6594  */
6595 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6596 {
6597         struct osd_it_ea *it = (struct osd_it_ea *)di;
6598         int rc;
6599
6600         ENTRY;
6601
6602         if (it->oie_it_dirent < it->oie_rd_dirent) {
6603                 it->oie_dirent =
6604                         (void *)it->oie_dirent +
6605                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6606                                        it->oie_dirent->oied_namelen);
6607                 it->oie_it_dirent++;
6608                 rc = 0;
6609         } else {
6610                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6611                         rc = 1;
6612                 else
6613                         rc = osd_ldiskfs_it_fill(env, di);
6614         }
6615
6616         RETURN(rc);
6617 }
6618
6619 /**
6620  * Returns the key at current position from iterator's in memory structure.
6621  *
6622  * \param di iterator's in memory structure
6623  *
6624  * \retval key i.e. struct dt_key on success
6625  */
6626 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6627                                     const struct dt_it *di)
6628 {
6629         struct osd_it_ea *it = (struct osd_it_ea *)di;
6630
6631         return (struct dt_key *)it->oie_dirent->oied_name;
6632 }
6633
6634 /**
6635  * Returns key's size at current position from iterator's in memory structure.
6636  *
6637  * \param di iterator's in memory structure
6638  *
6639  * \retval key_size i.e. struct dt_key on success
6640  */
6641 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6642 {
6643         struct osd_it_ea *it = (struct osd_it_ea *)di;
6644
6645         return it->oie_dirent->oied_namelen;
6646 }
6647
6648 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6649 {
6650         if (LDISKFS_DIR_REC_LEN(de) >=
6651             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6652                 return true;
6653
6654         return false;
6655 }
6656
6657 static inline bool
6658 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6659                      unsigned int blocksize, bool dotdot)
6660 {
6661         if (dotdot)
6662                 return osd_dotdot_has_space(de);
6663
6664         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6665             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6666                 return true;
6667
6668         return false;
6669 }
6670
6671 static int
6672 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6673                     handle_t *jh, struct dentry *dentry,
6674                     const struct lu_fid *fid, struct buffer_head *bh,
6675                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6676                     bool dotdot)
6677 {
6678         struct inode *dir = dentry->d_parent->d_inode;
6679         struct inode *inode = dentry->d_inode;
6680         struct osd_fid_pack *rec;
6681         struct ldiskfs_dentry_param *ldp;
6682         int namelen = dentry->d_name.len;
6683         int rc;
6684         struct osd_thread_info *info = osd_oti_get(env);
6685
6686         ENTRY;
6687
6688         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6689                 RETURN(0);
6690
6691         /* There is enough space to hold the FID-in-dirent. */
6692         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6693                 rc = ldiskfs_journal_get_write_access(jh, bh);
6694                 if (rc != 0)
6695                         RETURN(rc);
6696
6697                 de->name[namelen] = 0;
6698                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6699                 rec->fp_len = sizeof(struct lu_fid) + 1;
6700                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6701                 de->file_type |= LDISKFS_DIRENT_LUFID;
6702                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6703
6704                 RETURN(rc);
6705         }
6706
6707         LASSERT(!dotdot);
6708
6709         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6710         if (rc != 0)
6711                 RETURN(rc);
6712
6713         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6714         osd_get_ldiskfs_dirent_param(ldp, fid);
6715         dentry->d_fsdata = (void *)ldp;
6716         ll_vfs_dq_init(dir);
6717         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6718         /*
6719          * It is too bad, we cannot reinsert the name entry back.
6720          * That means we lose it!
6721          */
6722         if (rc != 0)
6723                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6724                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6725                        osd_ino2name(inode),
6726                        dir->i_ino, dir->i_generation, namelen,
6727                        dentry->d_name.name, PFID(fid), rc);
6728
6729         RETURN(rc);
6730 }
6731
6732 static int
6733 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6734                         struct osd_it_ea *it, struct lu_fid *fid,
6735                         struct osd_inode_id *id, __u32 *attr)
6736 {
6737         struct osd_thread_info *info = osd_oti_get(env);
6738         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6739         struct osd_device *dev = osd_obj2dev(obj);
6740         struct super_block *sb = osd_sb(dev);
6741         const char *devname = osd_name(dev);
6742         struct osd_it_ea_dirent *ent = it->oie_dirent;
6743         struct inode *dir = obj->oo_inode;
6744         struct htree_lock *hlock = NULL;
6745         struct buffer_head *bh = NULL;
6746         handle_t *jh = NULL;
6747         struct ldiskfs_dir_entry_2 *de;
6748         struct dentry *dentry;
6749         struct inode *inode;
6750         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6751         int credits;
6752         int rc;
6753         bool dotdot = false;
6754         bool dirty = false;
6755
6756         ENTRY;
6757
6758         if (ent->oied_name[0] == '.') {
6759                 if (ent->oied_namelen == 1)
6760                         RETURN(0);
6761
6762                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6763                         dotdot = true;
6764         }
6765
6766         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6767         inode = osd_iget(info, dev, id);
6768         if (IS_ERR(inode)) {
6769                 rc = PTR_ERR(inode);
6770                 if (rc == -ENOENT || rc == -ESTALE) {
6771                         /*
6772                          * Maybe dangling name entry, or
6773                          * corrupted directory entry.
6774                          */
6775                         *attr |= LUDA_UNKNOWN;
6776                         rc = 0;
6777                 } else {
6778                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6779                                "check_repair, dir = %lu/%u, name = %.*s, "
6780                                "ino = %llu, rc = %d\n",
6781                                devname, dir->i_ino, dir->i_generation,
6782                                ent->oied_namelen, ent->oied_name,
6783                                ent->oied_ino, rc);
6784                 }
6785
6786                 RETURN(rc);
6787         }
6788
6789         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6790                                            ent->oied_namelen);
6791         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6792         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6793                 lma = NULL;
6794         else if (rc != 0)
6795                 GOTO(out, rc);
6796
6797         /*
6798          * We need to ensure that the name entry is still valid.
6799          * Because it may be removed or renamed by other already.
6800          *
6801          * The unlink or rename operation will start journal before PDO lock,
6802          * so to avoid deadlock, here we need to start journal handle before
6803          * related PDO lock also. But because we do not know whether there
6804          * will be something to be repaired before PDO lock, we just start
6805          * journal without conditions.
6806          *
6807          * We may need to remove the name entry firstly, then insert back.
6808          * One credit is for user quota file update.
6809          * One credit is for group quota file update.
6810          * Two credits are for dirty inode.
6811          */
6812         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6813                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6814
6815         if (dev->od_dirent_journal != 0) {
6816
6817 again:
6818                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6819                 if (IS_ERR(jh)) {
6820                         rc = PTR_ERR(jh);
6821                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6822                                "check_repair, dir = %lu/%u, credits = %d, "
6823                                "name = %.*s, ino = %llu: rc = %d\n",
6824                                devname, dir->i_ino, dir->i_generation, credits,
6825                                ent->oied_namelen, ent->oied_name,
6826                                ent->oied_ino, rc);
6827
6828                         GOTO(out_inode, rc);
6829                 }
6830
6831                 if (obj->oo_hl_head != NULL) {
6832                         hlock = osd_oti_get(env)->oti_hlock;
6833                         /*
6834                          * "0" means exclusive lock for the whole directory.
6835                          * We need to prevent others access such name entry
6836                          * during the delete + insert. Neither HLOCK_ADD nor
6837                          * HLOCK_DEL cannot guarantee the atomicity.
6838                          */
6839                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6840                 } else {
6841                         down_write(&obj->oo_ext_idx_sem);
6842                 }
6843         } else {
6844                 if (obj->oo_hl_head != NULL) {
6845                         hlock = osd_oti_get(env)->oti_hlock;
6846                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6847                                            LDISKFS_HLOCK_LOOKUP);
6848                 } else {
6849                         down_read(&obj->oo_ext_idx_sem);
6850                 }
6851         }
6852
6853         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6854         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6855                 *attr |= LUDA_IGNORE;
6856
6857                 GOTO(out, rc = 0);
6858         }
6859
6860         /*
6861          * For dotdot entry, if there is not enough space to hold the
6862          * FID-in-dirent, just keep them there. It only happens when the
6863          * device upgraded from 1.8 or restored from MDT file-level backup.
6864          * For the whole directory, only dotdot entry have no FID-in-dirent
6865          * and needs to get FID from LMA when readdir, it will not affect the
6866          * performance much.
6867          */
6868         if (dotdot && !osd_dotdot_has_space(de)) {
6869                 *attr |= LUDA_UNKNOWN;
6870
6871                 GOTO(out, rc = 0);
6872         }
6873
6874         if (lma != NULL) {
6875                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6876                         GOTO(out, rc = 0);
6877
6878                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6879                         struct lu_fid *tfid = &lma->lma_self_fid;
6880
6881                         if (likely(dotdot &&
6882                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6883                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6884                                 /*
6885                                  * It must be REMOTE_PARENT_DIR and as the
6886                                  * 'dotdot' entry of remote directory
6887                                  */
6888                                 *attr |= LUDA_IGNORE;
6889                         } else {
6890                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6891                                        "parent directory, but got %.*s under "
6892                                        "dir = %lu/%u with the FID "DFID"\n",
6893                                        devname, ent->oied_namelen,
6894                                        ent->oied_name, dir->i_ino,
6895                                        dir->i_generation, PFID(tfid));
6896
6897                                 *attr |= LUDA_UNKNOWN;
6898                         }
6899
6900                         GOTO(out, rc = 0);
6901                 }
6902         }
6903
6904         if (!fid_is_zero(fid)) {
6905                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6906                                               ent->oied_namelen);
6907                 if (rc == -ENOENT ||
6908                     (rc == -ENODATA &&
6909                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6910                         /*
6911                          * linkEA does not recognize the dirent entry,
6912                          * it may because the dirent entry corruption
6913                          * and points to other's inode.
6914                          */
6915                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6916                                "recognize the dirent, dir = %lu/%u, "
6917                                " name = %.*s, ino = %llu, "
6918                                DFID": rc = %d\n", devname, dir->i_ino,
6919                                dir->i_generation, ent->oied_namelen,
6920                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6921                         *attr |= LUDA_UNKNOWN;
6922
6923                         GOTO(out, rc = 0);
6924                 }
6925
6926                 if (rc && rc != -ENODATA) {
6927                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6928                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6929                                DFID": rc = %d\n", devname, dir->i_ino,
6930                                dir->i_generation, ent->oied_namelen,
6931                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6932                         *attr |= LUDA_UNKNOWN;
6933
6934                         GOTO(out, rc = 0);
6935                 }
6936         }
6937
6938         if (lma != NULL) {
6939                 /*
6940                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6941                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6942                  */
6943                 if (*attr & LUDA_VERIFY_DRYRUN) {
6944                         *fid = lma->lma_self_fid;
6945                         *attr |= LUDA_REPAIR;
6946
6947                         GOTO(out, rc = 0);
6948                 }
6949
6950                 if (jh == NULL) {
6951                         brelse(bh);
6952                         dev->od_dirent_journal = 1;
6953                         if (hlock != NULL) {
6954                                 ldiskfs_htree_unlock(hlock);
6955                                 hlock = NULL;
6956                         } else {
6957                                 up_read(&obj->oo_ext_idx_sem);
6958                         }
6959
6960                         goto again;
6961                 }
6962
6963                 *fid = lma->lma_self_fid;
6964                 dirty = true;
6965                 /* Update or append the FID-in-dirent. */
6966                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6967                                          bh, de, hlock, dotdot);
6968                 if (rc == 0)
6969                         *attr |= LUDA_REPAIR;
6970                 else
6971                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6972                                "the dirent, dir = %lu/%u, name = %.*s, "
6973                                "ino = %llu, "DFID": rc = %d\n",
6974                                devname, dir->i_ino, dir->i_generation,
6975                                ent->oied_namelen, ent->oied_name,
6976                                ent->oied_ino, PFID(fid), rc);
6977         } else {
6978                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6979                 if (*attr & LUDA_VERIFY_DRYRUN) {
6980                         if (fid_is_sane(fid)) {
6981                                 *attr |= LUDA_REPAIR;
6982                         } else if (dev->od_index == 0) {
6983                                 lu_igif_build(fid, inode->i_ino,
6984                                               inode->i_generation);
6985                                 *attr |= LUDA_UPGRADE;
6986                         }
6987
6988                         GOTO(out, rc = 0);
6989                 }
6990
6991                 if (jh == NULL) {
6992                         brelse(bh);
6993                         dev->od_dirent_journal = 1;
6994                         if (hlock != NULL) {
6995                                 ldiskfs_htree_unlock(hlock);
6996                                 hlock = NULL;
6997                         } else {
6998                                 up_read(&obj->oo_ext_idx_sem);
6999                         }
7000
7001                         goto again;
7002                 }
7003
7004                 dirty = true;
7005                 if (unlikely(fid_is_sane(fid))) {
7006                         /*
7007                          * FID-in-dirent exists, but FID-in-LMA is lost.
7008                          * Trust the FID-in-dirent, and add FID-in-LMA.
7009                          */
7010                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7011                         if (rc == 0)
7012                                 *attr |= LUDA_REPAIR;
7013                         else
7014                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7015                                        "update dirent, dir = %lu/%u, "
7016                                        "name = %.*s, ino = %llu, "
7017                                        DFID": rc = %d\n",
7018                                        devname, dir->i_ino, dir->i_generation,
7019                                        ent->oied_namelen, ent->oied_name,
7020                                        ent->oied_ino, PFID(fid), rc);
7021                 } else if (dev->od_index == 0) {
7022                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7023                         /*
7024                          * It is probably IGIF object. Only aappend the
7025                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7026                          */
7027                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7028                                                  bh, de, hlock, dotdot);
7029                         if (rc == 0)
7030                                 *attr |= LUDA_UPGRADE;
7031                         else
7032                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7033                                        "after the dirent, dir = %lu/%u, "
7034                                        "name = %.*s, ino = %llu, "
7035                                        DFID": rc = %d\n",
7036                                        devname, dir->i_ino, dir->i_generation,
7037                                        ent->oied_namelen, ent->oied_name,
7038                                        ent->oied_ino, PFID(fid), rc);
7039                 }
7040         }
7041
7042         GOTO(out, rc);
7043
7044 out:
7045         if (!IS_ERR(bh))
7046                 brelse(bh);
7047         if (hlock != NULL) {
7048                 ldiskfs_htree_unlock(hlock);
7049         } else {
7050                 if (dev->od_dirent_journal != 0)
7051                         up_write(&obj->oo_ext_idx_sem);
7052                 else
7053                         up_read(&obj->oo_ext_idx_sem);
7054         }
7055
7056         if (jh != NULL)
7057                 ldiskfs_journal_stop(jh);
7058
7059 out_inode:
7060         iput(inode);
7061         if (rc >= 0 && !dirty)
7062                 dev->od_dirent_journal = 0;
7063
7064         return rc;
7065 }
7066
7067 /**
7068  * Returns the value at current position from iterator's in memory structure.
7069  *
7070  * \param di struct osd_it_ea, iterator's in memory structure
7071  * \param attr attr requested for dirent.
7072  * \param lde lustre dirent
7073  *
7074  * \retval   0 no error and \param lde has correct lustre dirent.
7075  * \retval -ve on error
7076  */
7077 static inline int osd_it_ea_rec(const struct lu_env *env,
7078                                 const struct dt_it *di,
7079                                 struct dt_rec *dtrec, __u32 attr)
7080 {
7081         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7082         struct osd_object      *obj   = it->oie_obj;
7083         struct osd_device      *dev   = osd_obj2dev(obj);
7084         struct osd_thread_info *oti   = osd_oti_get(env);
7085         struct osd_inode_id    *id    = &oti->oti_id;
7086         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7087         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7088         __u32 ino = it->oie_dirent->oied_ino;
7089         int rc = 0;
7090
7091         ENTRY;
7092
7093         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7094
7095         if (attr & LUDA_VERIFY) {
7096                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7097                         attr |= LUDA_IGNORE;
7098                         /*
7099                          * If the parent is on remote MDT, and there
7100                          * is no FID-in-dirent, then we have to get
7101                          * the parent FID from the linkEA.
7102                          */
7103                         if (!fid_is_sane(fid) &&
7104                             it->oie_dirent->oied_namelen == 2 &&
7105                             it->oie_dirent->oied_name[0] == '.' &&
7106                             it->oie_dirent->oied_name[1] == '.')
7107                                 osd_get_pfid_from_linkea(env, obj, fid);
7108                 } else {
7109                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7110                                                      &attr);
7111                 }
7112
7113                 if (!fid_is_sane(fid))
7114                         attr |= LUDA_UNKNOWN;
7115         } else {
7116                 attr &= ~LU_DIRENT_ATTRS_MASK;
7117                 if (!fid_is_sane(fid)) {
7118                         bool is_dotdot = false;
7119
7120                         if (it->oie_dirent->oied_namelen == 2 &&
7121                             it->oie_dirent->oied_name[0] == '.' &&
7122                             it->oie_dirent->oied_name[1] == '.')
7123                                 is_dotdot = true;
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 (is_remote_parent_ino(dev, ino) && is_dotdot) {
7130                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7131                         } else {
7132                                 if (is_dotdot == false &&
7133                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7134                                         RETURN(-ENOENT);
7135
7136                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7137                         }
7138                 } else {
7139                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7140                 }
7141         }
7142
7143         /* Pack the entry anyway, at least the offset is right. */
7144         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7145                            it->oie_dirent->oied_name,
7146                            it->oie_dirent->oied_namelen,
7147                            it->oie_dirent->oied_type, attr);
7148
7149         if (rc < 0)
7150                 RETURN(rc);
7151
7152         if (osd_remote_fid(env, dev, fid))
7153                 RETURN(0);
7154
7155         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7156                 osd_add_oi_cache(oti, dev, id, fid);
7157
7158         RETURN(rc > 0 ? 0 : rc);
7159 }
7160
7161 /**
7162  * Returns the record size size at current position.
7163  *
7164  * This function will return record(lu_dirent) size in bytes.
7165  *
7166  * \param[in] env       execution environment
7167  * \param[in] di        iterator's in memory structure
7168  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7169  *                      calculate the lu_dirent size.
7170  *
7171  * \retval      record size(in bytes & in memory) of the current lu_dirent
7172  *              entry.
7173  */
7174 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7175                               __u32 attr)
7176 {
7177         struct osd_it_ea *it = (struct osd_it_ea *)di;
7178
7179         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7180 }
7181
7182 /**
7183  * Returns a cookie for current position of the iterator head, so that
7184  * user can use this cookie to load/start the iterator next time.
7185  *
7186  * \param di iterator's in memory structure
7187  *
7188  * \retval cookie for current position, on success
7189  */
7190 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7191 {
7192         struct osd_it_ea *it = (struct osd_it_ea *)di;
7193
7194         return it->oie_dirent->oied_off;
7195 }
7196
7197 /**
7198  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7199  * to load a directory entry at a time and stored it i inn,
7200  * in iterator's in-memory data structure.
7201  *
7202  * \param di struct osd_it_ea, iterator's in memory structure
7203  *
7204  * \retval +ve on success
7205  * \retval -ve on error
7206  */
7207 static int osd_it_ea_load(const struct lu_env *env,
7208                           const struct dt_it *di, __u64 hash)
7209 {
7210         struct osd_it_ea *it = (struct osd_it_ea *)di;
7211         int rc;
7212
7213         ENTRY;
7214         it->oie_file.f_pos = hash;
7215
7216         rc =  osd_ldiskfs_it_fill(env, di);
7217         if (rc > 0)
7218                 rc = -ENODATA;
7219
7220         if (rc == 0)
7221                 rc = 1;
7222
7223         RETURN(rc);
7224 }
7225
7226 /**
7227  * Index lookup function for interoperability mode (b11826).
7228  *
7229  * \param key,  key i.e. file name to be searched
7230  *
7231  * \retval +ve, on success
7232  * \retval -ve, on error
7233  */
7234 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7235                                struct dt_rec *rec, const struct dt_key *key)
7236 {
7237         struct osd_object *obj = osd_dt_obj(dt);
7238         int rc = 0;
7239
7240         ENTRY;
7241
7242         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7243         LINVRNT(osd_invariant(obj));
7244
7245         rc = osd_ea_lookup_rec(env, obj, rec, key);
7246         if (rc == 0)
7247                 rc = 1;
7248         RETURN(rc);
7249 }
7250
7251 /**
7252  * Index and Iterator operations for interoperability
7253  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7254  */
7255 static const struct dt_index_operations osd_index_ea_ops = {
7256         .dio_lookup         = osd_index_ea_lookup,
7257         .dio_declare_insert = osd_index_declare_ea_insert,
7258         .dio_insert         = osd_index_ea_insert,
7259         .dio_declare_delete = osd_index_declare_ea_delete,
7260         .dio_delete         = osd_index_ea_delete,
7261         .dio_it     = {
7262                 .init     = osd_it_ea_init,
7263                 .fini     = osd_it_ea_fini,
7264                 .get      = osd_it_ea_get,
7265                 .put      = osd_it_ea_put,
7266                 .next     = osd_it_ea_next,
7267                 .key      = osd_it_ea_key,
7268                 .key_size = osd_it_ea_key_size,
7269                 .rec      = osd_it_ea_rec,
7270                 .rec_size = osd_it_ea_rec_size,
7271                 .store    = osd_it_ea_store,
7272                 .load     = osd_it_ea_load
7273         }
7274 };
7275
7276 static void *osd_key_init(const struct lu_context *ctx,
7277                           struct lu_context_key *key)
7278 {
7279         struct osd_thread_info *info;
7280
7281         OBD_ALLOC_PTR(info);
7282         if (info == NULL)
7283                 return ERR_PTR(-ENOMEM);
7284
7285         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7286         if (info->oti_it_ea_buf == NULL)
7287                 goto out_free_info;
7288
7289         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7290
7291         info->oti_hlock = ldiskfs_htree_lock_alloc();
7292         if (info->oti_hlock == NULL)
7293                 goto out_free_ea;
7294
7295         return info;
7296
7297 out_free_ea:
7298         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7299 out_free_info:
7300         OBD_FREE_PTR(info);
7301         return ERR_PTR(-ENOMEM);
7302 }
7303
7304 static void osd_key_fini(const struct lu_context *ctx,
7305                          struct lu_context_key *key, void *data)
7306 {
7307         struct osd_thread_info *info = data;
7308         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7309         struct osd_idmap_cache *idc = info->oti_ins_cache;
7310
7311         if (info->oti_dio_pages) {
7312                 int i;
7313                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7314                         if (info->oti_dio_pages[i])
7315                                 __free_page(info->oti_dio_pages[i]);
7316                 }
7317                 OBD_FREE(info->oti_dio_pages,
7318                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7319         }
7320
7321         if (info->oti_inode != NULL)
7322                 OBD_FREE_PTR(lli);
7323         if (info->oti_hlock != NULL)
7324                 ldiskfs_htree_lock_free(info->oti_hlock);
7325         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7326         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7327         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7328         lu_buf_free(&info->oti_big_buf);
7329         if (idc != NULL) {
7330                 LASSERT(info->oti_ins_cache_size > 0);
7331                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7332                 info->oti_ins_cache = NULL;
7333                 info->oti_ins_cache_size = 0;
7334         }
7335         OBD_FREE_PTR(info);
7336 }
7337
7338 static void osd_key_exit(const struct lu_context *ctx,
7339                          struct lu_context_key *key, void *data)
7340 {
7341         struct osd_thread_info *info = data;
7342
7343         LASSERT(info->oti_r_locks == 0);
7344         LASSERT(info->oti_w_locks == 0);
7345         LASSERT(info->oti_txns    == 0);
7346 }
7347
7348 /* type constructor/destructor: osd_type_init, osd_type_fini */
7349 LU_TYPE_INIT_FINI(osd, &osd_key);
7350
7351 struct lu_context_key osd_key = {
7352         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7353         .lct_init = osd_key_init,
7354         .lct_fini = osd_key_fini,
7355         .lct_exit = osd_key_exit
7356 };
7357
7358
7359 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7360                            const char *name, struct lu_device *next)
7361 {
7362         struct osd_device *osd = osd_dev(d);
7363
7364         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7365             sizeof(osd->od_svname))
7366                 return -E2BIG;
7367         return osd_procfs_init(osd, name);
7368 }
7369
7370 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7371 {
7372         struct seq_server_site *ss = osd_seq_site(osd);
7373         int rc;
7374
7375         ENTRY;
7376
7377         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7378                 RETURN(0);
7379
7380         if (unlikely(ss == NULL))
7381                 RETURN(-ENODEV);
7382
7383         OBD_ALLOC_PTR(osd->od_cl_seq);
7384         if (osd->od_cl_seq == NULL)
7385                 RETURN(-ENOMEM);
7386
7387         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7388                              osd->od_svname, ss->ss_server_seq);
7389         if (rc != 0) {
7390                 OBD_FREE_PTR(osd->od_cl_seq);
7391                 osd->od_cl_seq = NULL;
7392                 RETURN(rc);
7393         }
7394
7395         if (ss->ss_node_id == 0) {
7396                 /*
7397                  * If the OSD on the sequence controller(MDT0), then allocate
7398                  * sequence here, otherwise allocate sequence after connected
7399                  * to MDT0 (see mdt_register_lwp_callback()).
7400                  */
7401                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7402                                    &osd->od_cl_seq->lcs_space, env);
7403         }
7404
7405         RETURN(rc);
7406 }
7407
7408 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7409 {
7410         if (osd->od_cl_seq == NULL)
7411                 return;
7412
7413         seq_client_fini(osd->od_cl_seq);
7414         OBD_FREE_PTR(osd->od_cl_seq);
7415         osd->od_cl_seq = NULL;
7416 }
7417
7418 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7419 {
7420         ENTRY;
7421
7422         /* shutdown quota slave instance associated with the device */
7423         if (o->od_quota_slave != NULL) {
7424                 struct qsd_instance *qsd = o->od_quota_slave;
7425
7426                 o->od_quota_slave = NULL;
7427                 qsd_fini(env, qsd);
7428         }
7429
7430         osd_fid_fini(env, o);
7431         osd_scrub_cleanup(env, o);
7432
7433         RETURN(0);
7434 }
7435
7436 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7437 {
7438         ENTRY;
7439
7440         if (o->od_mnt != NULL) {
7441                 shrink_dcache_sb(osd_sb(o));
7442                 osd_sync(env, &o->od_dt_dev);
7443
7444                 mntput(o->od_mnt);
7445                 o->od_mnt = NULL;
7446         }
7447
7448         EXIT;
7449 }
7450
7451 static int osd_mount(const struct lu_env *env,
7452                      struct osd_device *o, struct lustre_cfg *cfg)
7453 {
7454         const char *name = lustre_cfg_string(cfg, 0);
7455         const char *dev = lustre_cfg_string(cfg, 1);
7456         const char *opts;
7457         unsigned long page, s_flags, lmd_flags = 0;
7458         struct page *__page;
7459         struct file_system_type *type;
7460         char *options = NULL;
7461         char *str;
7462         struct osd_thread_info *info = osd_oti_get(env);
7463         struct lu_fid *fid = &info->oti_fid;
7464         struct inode *inode;
7465         int rc = 0, force_over_512tb = 0;
7466
7467         ENTRY;
7468
7469         if (o->od_mnt != NULL)
7470                 RETURN(0);
7471
7472         if (strlen(dev) >= sizeof(o->od_mntdev))
7473                 RETURN(-E2BIG);
7474         strcpy(o->od_mntdev, dev);
7475
7476         str = lustre_cfg_string(cfg, 2);
7477         s_flags = simple_strtoul(str, NULL, 0);
7478         str = strstr(str, ":");
7479         if (str)
7480                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7481         opts = lustre_cfg_string(cfg, 3);
7482 #ifdef __BIG_ENDIAN
7483         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7484                 CERROR("%s: device %s extents feature is not guaranteed to "
7485                        "work on big-endian systems. Use \"bigendian_extents\" "
7486                        "mount option to override.\n", name, dev);
7487                 RETURN(-EINVAL);
7488         }
7489 #endif
7490 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7491         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7492                 CWARN("force_over_128tb option is deprecated. "
7493                       "Filesystems less than 512TB can be created without any "
7494                       "force options. Use force_over_512tb option for "
7495                       "filesystems greater than 512TB.\n");
7496         }
7497 #endif
7498 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7499         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7500                 CWARN("force_over_256tb option is deprecated. "
7501                       "Filesystems less than 512TB can be created without any "
7502                       "force options. Use force_over_512tb option for "
7503                       "filesystems greater than 512TB.\n");
7504         }
7505 #endif
7506
7507         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7508                 force_over_512tb = 1;
7509
7510         __page = alloc_page(GFP_KERNEL);
7511         if (__page == NULL)
7512                 GOTO(out, rc = -ENOMEM);
7513         page = (unsigned long)page_address(__page);
7514         options = (char *)page;
7515         *options = '\0';
7516         if (opts != NULL) {
7517                 /* strip out the options for back compatiblity */
7518                 static char *sout[] = {
7519                         "mballoc",
7520                         "iopen",
7521                         "noiopen",
7522                         "iopen_nopriv",
7523                         "extents",
7524                         "noextents",
7525                         /* strip out option we processed in osd */
7526                         "bigendian_extents",
7527                         "force_over_128tb",
7528                         "force_over_256tb",
7529                         "force_over_512tb",
7530                         NULL
7531                 };
7532                 strcat(options, opts);
7533                 for (rc = 0, str = options; sout[rc]; ) {
7534                         char *op = strstr(str, sout[rc]);
7535
7536                         if (op == NULL) {
7537                                 rc++;
7538                                 str = options;
7539                                 continue;
7540                         }
7541                         if (op == options || *(op - 1) == ',') {
7542                                 str = op + strlen(sout[rc]);
7543                                 if (*str == ',' || *str == '\0') {
7544                                         *str == ',' ? str++ : str;
7545                                         memmove(op, str, strlen(str) + 1);
7546                                 }
7547                         }
7548                         for (str = op; *str != ',' && *str != '\0'; str++)
7549                                 ;
7550                 }
7551         } else {
7552                 strncat(options, "user_xattr,acl", 14);
7553         }
7554
7555         /* Glom up mount options */
7556         if (*options != '\0')
7557                 strcat(options, ",");
7558         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7559
7560         type = get_fs_type("ldiskfs");
7561         if (!type) {
7562                 CERROR("%s: cannot find ldiskfs module\n", name);
7563                 GOTO(out, rc = -ENODEV);
7564         }
7565
7566         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7567         module_put(type->owner);
7568
7569         if (IS_ERR(o->od_mnt)) {
7570                 rc = PTR_ERR(o->od_mnt);
7571                 o->od_mnt = NULL;
7572                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7573                 GOTO(out, rc);
7574         }
7575
7576         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7577                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7578                                  force_over_512tb == 0) {
7579                 CERROR("%s: device %s LDISKFS does not support filesystems "
7580                        "greater than 512TB and can cause data corruption. "
7581                        "Use \"force_over_512tb\" mount option to override.\n",
7582                        name, dev);
7583                 GOTO(out_mnt, rc = -EINVAL);
7584         }
7585
7586         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7587                 if (priv_dev_set_rdonly) {
7588                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7589                         o->od_dt_dev.dd_rdonly = 1;
7590                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7591                                       name);
7592                 } else {
7593                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7594                                       name);
7595
7596                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7597                 }
7598         } else if (priv_dev_check_rdonly &&
7599                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7600                 CERROR("%s: underlying device %s is marked as "
7601                        "read-only. Setup failed\n", name, dev);
7602
7603                 GOTO(out_mnt, rc = -EROFS);
7604         }
7605
7606         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7607                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7608                 GOTO(out_mnt, rc = -EINVAL);
7609         }
7610
7611 #ifdef LDISKFS_MOUNT_DIRDATA
7612         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7613                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7614         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7615                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7616                       "enabling the dirdata feature. If you do not want to "
7617                       "downgrade to Lustre-1.x again, you can enable it via "
7618                       "'tune2fs -O dirdata device'\n", name, dev);
7619 #endif
7620         inode = osd_sb(o)->s_root->d_inode;
7621         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7622         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7623         if (rc != 0) {
7624                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7625                 GOTO(out_mnt, rc);
7626         }
7627
7628         if (lmd_flags & LMD_FLG_NOSCRUB)
7629                 o->od_auto_scrub_interval = AS_NEVER;
7630
7631         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7632                 /* do not use pagecache with flash-backed storage */
7633                 o->od_writethrough_cache = 0;
7634                 o->od_read_cache = 0;
7635         }
7636
7637         GOTO(out, rc = 0);
7638
7639 out_mnt:
7640         mntput(o->od_mnt);
7641         o->od_mnt = NULL;
7642
7643 out:
7644         if (__page)
7645                 __free_page(__page);
7646
7647         return rc;
7648 }
7649
7650 static struct lu_device *osd_device_fini(const struct lu_env *env,
7651                                          struct lu_device *d)
7652 {
7653         struct osd_device *o = osd_dev(d);
7654
7655         ENTRY;
7656
7657         osd_index_backup(env, o, false);
7658         osd_shutdown(env, o);
7659         osd_procfs_fini(o);
7660         osd_obj_map_fini(o);
7661         osd_umount(env, o);
7662
7663         RETURN(NULL);
7664 }
7665
7666 static int osd_device_init0(const struct lu_env *env,
7667                             struct osd_device *o,
7668                             struct lustre_cfg *cfg)
7669 {
7670         struct lu_device *l = osd2lu_dev(o);
7671         struct osd_thread_info *info;
7672         int rc;
7673         int cplen = 0;
7674
7675         /* if the module was re-loaded, env can loose its keys */
7676         rc = lu_env_refill((struct lu_env *)env);
7677         if (rc)
7678                 GOTO(out, rc);
7679         info = osd_oti_get(env);
7680         LASSERT(info);
7681
7682         l->ld_ops = &osd_lu_ops;
7683         o->od_dt_dev.dd_ops = &osd_dt_ops;
7684
7685         spin_lock_init(&o->od_osfs_lock);
7686         mutex_init(&o->od_otable_mutex);
7687         INIT_LIST_HEAD(&o->od_orphan_list);
7688         INIT_LIST_HEAD(&o->od_index_backup_list);
7689         INIT_LIST_HEAD(&o->od_index_restore_list);
7690         spin_lock_init(&o->od_lock);
7691         o->od_index_backup_policy = LIBP_NONE;
7692
7693         o->od_read_cache = 1;
7694         o->od_writethrough_cache = 1;
7695         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7696         o->od_auto_scrub_interval = AS_DEFAULT;
7697
7698         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7699                         sizeof(o->od_svname));
7700         if (cplen >= sizeof(o->od_svname)) {
7701                 rc = -E2BIG;
7702                 GOTO(out, rc);
7703         }
7704
7705         o->od_index_backup_stop = 0;
7706         o->od_index = -1; /* -1 means index is invalid */
7707         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7708         if (rc == LDD_F_SV_TYPE_OST)
7709                 o->od_is_ost = 1;
7710
7711         o->od_full_scrub_ratio = OFSR_DEFAULT;
7712         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7713         rc = osd_mount(env, o, cfg);
7714         if (rc != 0)
7715                 GOTO(out, rc);
7716
7717         rc = osd_obj_map_init(env, o);
7718         if (rc != 0)
7719                 GOTO(out_mnt, rc);
7720
7721         rc = lu_site_init(&o->od_site, l);
7722         if (rc != 0)
7723                 GOTO(out_compat, rc);
7724         o->od_site.ls_bottom_dev = l;
7725
7726         rc = lu_site_init_finish(&o->od_site);
7727         if (rc != 0)
7728                 GOTO(out_site, rc);
7729
7730         INIT_LIST_HEAD(&o->od_ios_list);
7731         /* setup scrub, including OI files initialization */
7732         o->od_in_init = 1;
7733         rc = osd_scrub_setup(env, o);
7734         o->od_in_init = 0;
7735         if (rc < 0)
7736                 GOTO(out_site, rc);
7737
7738         rc = osd_procfs_init(o, o->od_svname);
7739         if (rc != 0) {
7740                 CERROR("%s: can't initialize procfs: rc = %d\n",
7741                        o->od_svname, rc);
7742                 GOTO(out_scrub, rc);
7743         }
7744
7745         LASSERT(l->ld_site->ls_linkage.next != NULL);
7746         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7747
7748         /* initialize quota slave instance */
7749         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7750                                      o->od_proc_entry);
7751         if (IS_ERR(o->od_quota_slave)) {
7752                 rc = PTR_ERR(o->od_quota_slave);
7753                 o->od_quota_slave = NULL;
7754                 GOTO(out_procfs, rc);
7755         }
7756
7757         RETURN(0);
7758
7759 out_procfs:
7760         osd_procfs_fini(o);
7761 out_scrub:
7762         osd_scrub_cleanup(env, o);
7763 out_site:
7764         lu_site_fini(&o->od_site);
7765 out_compat:
7766         osd_obj_map_fini(o);
7767 out_mnt:
7768         osd_umount(env, o);
7769 out:
7770         return rc;
7771 }
7772
7773 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7774                                           struct lu_device_type *t,
7775                                           struct lustre_cfg *cfg)
7776 {
7777         struct osd_device *o;
7778         int rc;
7779
7780         OBD_ALLOC_PTR(o);
7781         if (o == NULL)
7782                 return ERR_PTR(-ENOMEM);
7783
7784         rc = dt_device_init(&o->od_dt_dev, t);
7785         if (rc == 0) {
7786                 /*
7787                  * Because the ctx might be revived in dt_device_init,
7788                  * refill the env here
7789                  */
7790                 lu_env_refill((struct lu_env *)env);
7791                 rc = osd_device_init0(env, o, cfg);
7792                 if (rc)
7793                         dt_device_fini(&o->od_dt_dev);
7794         }
7795
7796         if (unlikely(rc != 0))
7797                 OBD_FREE_PTR(o);
7798
7799         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7800 }
7801
7802 static struct lu_device *osd_device_free(const struct lu_env *env,
7803                                          struct lu_device *d)
7804 {
7805         struct osd_device *o = osd_dev(d);
7806
7807         ENTRY;
7808
7809         /* XXX: make osd top device in order to release reference */
7810         d->ld_site->ls_top_dev = d;
7811         lu_site_purge(env, d->ld_site, -1);
7812         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7813                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7814                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7815         }
7816         lu_site_fini(&o->od_site);
7817         dt_device_fini(&o->od_dt_dev);
7818         OBD_FREE_PTR(o);
7819         RETURN(NULL);
7820 }
7821
7822 static int osd_process_config(const struct lu_env *env,
7823                               struct lu_device *d, struct lustre_cfg *cfg)
7824 {
7825         struct osd_device *o = osd_dev(d);
7826         int rc;
7827
7828         ENTRY;
7829
7830         switch (cfg->lcfg_command) {
7831         case LCFG_SETUP:
7832                 rc = osd_mount(env, o, cfg);
7833                 break;
7834         case LCFG_CLEANUP:
7835                 /*
7836                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7837                  * that may happend if hit failure during mount process.
7838                  */
7839                 osd_index_backup(env, o, false);
7840                 lu_dev_del_linkage(d->ld_site, d);
7841                 rc = osd_shutdown(env, o);
7842                 break;
7843         case LCFG_PARAM:
7844                 LASSERT(&o->od_dt_dev);
7845                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7846                                               cfg, &o->od_dt_dev);
7847                 if (rc > 0 || rc == -ENOSYS) {
7848                         rc = class_process_proc_param(PARAM_OST,
7849                                                       lprocfs_osd_obd_vars,
7850                                                       cfg, &o->od_dt_dev);
7851                         if (rc > 0)
7852                                 rc = 0;
7853                 }
7854                 break;
7855         case LCFG_PRE_CLEANUP:
7856                 osd_scrub_stop(o);
7857                 osd_index_backup(env, o,
7858                                  o->od_index_backup_policy != LIBP_NONE);
7859                 rc = 0;
7860                 break;
7861         default:
7862                 rc = -ENOSYS;
7863         }
7864
7865         RETURN(rc);
7866 }
7867
7868 static int osd_recovery_complete(const struct lu_env *env,
7869                                  struct lu_device *d)
7870 {
7871         struct osd_device *osd = osd_dev(d);
7872         int rc = 0;
7873
7874         ENTRY;
7875
7876         if (osd->od_quota_slave == NULL)
7877                 RETURN(0);
7878
7879         /*
7880          * start qsd instance on recovery completion, this notifies the quota
7881          * slave code that we are about to process new requests now
7882          */
7883         rc = qsd_start(env, osd->od_quota_slave);
7884         RETURN(rc);
7885 }
7886
7887 /*
7888  * we use exports to track all osd users
7889  */
7890 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7891                            struct obd_device *obd, struct obd_uuid *cluuid,
7892                            struct obd_connect_data *data, void *localdata)
7893 {
7894         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7895         struct lustre_handle conn;
7896         int rc;
7897
7898         ENTRY;
7899
7900         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7901
7902         rc = class_connect(&conn, obd, cluuid);
7903         if (rc)
7904                 RETURN(rc);
7905
7906         *exp = class_conn2export(&conn);
7907
7908         spin_lock(&osd->od_osfs_lock);
7909         osd->od_connects++;
7910         spin_unlock(&osd->od_osfs_lock);
7911
7912         RETURN(0);
7913 }
7914
7915 /*
7916  * once last export (we don't count self-export) disappeared
7917  * osd can be released
7918  */
7919 static int osd_obd_disconnect(struct obd_export *exp)
7920 {
7921         struct obd_device *obd = exp->exp_obd;
7922         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7923         int rc, release = 0;
7924
7925         ENTRY;
7926
7927         /* Only disconnect the underlying layers on the final disconnect. */
7928         spin_lock(&osd->od_osfs_lock);
7929         osd->od_connects--;
7930         if (osd->od_connects == 0)
7931                 release = 1;
7932         spin_unlock(&osd->od_osfs_lock);
7933
7934         rc = class_disconnect(exp); /* bz 9811 */
7935
7936         if (rc == 0 && release)
7937                 class_manual_cleanup(obd);
7938         RETURN(rc);
7939 }
7940
7941 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7942                        struct lu_device *dev)
7943 {
7944         struct osd_device *osd = osd_dev(dev);
7945         struct lr_server_data *lsd =
7946                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7947         int result = 0;
7948
7949         ENTRY;
7950
7951         if (osd->od_quota_slave != NULL) {
7952                 /* set up quota slave objects */
7953                 result = qsd_prepare(env, osd->od_quota_slave);
7954                 if (result != 0)
7955                         RETURN(result);
7956         }
7957
7958         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7959 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7960                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7961                         osd->od_index_in_idif = 1;
7962                 } else {
7963                         osd->od_index_in_idif = 0;
7964                         result = osd_register_proc_index_in_idif(osd);
7965                         if (result != 0)
7966                                 RETURN(result);
7967                 }
7968 #else
7969                 osd->od_index_in_idif = 1;
7970 #endif
7971         }
7972
7973         result = osd_fid_init(env, osd);
7974
7975         RETURN(result);
7976 }
7977
7978 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7979                          struct lu_fid *fid, struct md_op_data *op_data)
7980 {
7981         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7982
7983         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7984 }
7985
7986 static const struct lu_object_operations osd_lu_obj_ops = {
7987         .loo_object_init      = osd_object_init,
7988         .loo_object_delete    = osd_object_delete,
7989         .loo_object_release   = osd_object_release,
7990         .loo_object_free      = osd_object_free,
7991         .loo_object_print     = osd_object_print,
7992         .loo_object_invariant = osd_object_invariant
7993 };
7994
7995 const struct lu_device_operations osd_lu_ops = {
7996         .ldo_object_alloc      = osd_object_alloc,
7997         .ldo_process_config    = osd_process_config,
7998         .ldo_recovery_complete = osd_recovery_complete,
7999         .ldo_prepare           = osd_prepare,
8000 };
8001
8002 static const struct lu_device_type_operations osd_device_type_ops = {
8003         .ldto_init = osd_type_init,
8004         .ldto_fini = osd_type_fini,
8005
8006         .ldto_start = osd_type_start,
8007         .ldto_stop  = osd_type_stop,
8008
8009         .ldto_device_alloc = osd_device_alloc,
8010         .ldto_device_free  = osd_device_free,
8011
8012         .ldto_device_init = osd_device_init,
8013         .ldto_device_fini = osd_device_fini
8014 };
8015
8016 static struct lu_device_type osd_device_type = {
8017         .ldt_tags     = LU_DEVICE_DT,
8018         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8019         .ldt_ops      = &osd_device_type_ops,
8020         .ldt_ctx_tags = LCT_LOCAL,
8021 };
8022
8023 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8024 {
8025         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8026         struct super_block *sb = osd_sb(osd);
8027
8028         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8029 }
8030
8031 /*
8032  * lprocfs legacy support.
8033  */
8034 static struct obd_ops osd_obd_device_ops = {
8035         .o_owner = THIS_MODULE,
8036         .o_connect      = osd_obd_connect,
8037         .o_disconnect   = osd_obd_disconnect,
8038         .o_fid_alloc    = osd_fid_alloc,
8039         .o_health_check = osd_health_check,
8040 };
8041
8042 static int __init osd_init(void)
8043 {
8044         int rc;
8045
8046         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8047 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8048         /* please, try to keep osd_thread_info smaller than a page */
8049         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8050 #endif
8051
8052         osd_oi_mod_init();
8053
8054         rc = lu_kmem_init(ldiskfs_caches);
8055         if (rc)
8056                 return rc;
8057
8058 #ifdef CONFIG_KALLSYMS
8059         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8060         priv_dev_check_rdonly =
8061                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8062 #endif
8063
8064         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8065                                  lprocfs_osd_module_vars,
8066                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8067         if (rc)
8068                 lu_kmem_fini(ldiskfs_caches);
8069         return rc;
8070 }
8071
8072 static void __exit osd_exit(void)
8073 {
8074         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8075         lu_kmem_fini(ldiskfs_caches);
8076 }
8077
8078 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8079 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8080 MODULE_VERSION(LUSTRE_VERSION_STRING);
8081 MODULE_LICENSE("GPL");
8082
8083 module_init(osd_init);
8084 module_exit(osd_exit);