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
2373         ENTRY;
2374
2375         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2376         RETURN(s->s_op->sync_fs(s, 0));
2377 }
2378
2379 /* Our own copy of the set readonly functions if present, or NU if not. */
2380 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2381 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2382 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2383
2384 /*
2385  * Concurrency: shouldn't matter.
2386  */
2387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2388 {
2389         struct super_block *sb = osd_sb(osd_dt_dev(d));
2390         struct block_device *dev = sb->s_bdev;
2391         int rc = -EOPNOTSUPP;
2392
2393         ENTRY;
2394
2395         if (priv_dev_set_rdonly) {
2396                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2397
2398                 rc = 0;
2399                 CERROR("*** setting %s read-only ***\n",
2400                        osd_dt_dev(d)->od_svname);
2401
2402                 if (sb->s_op->freeze_fs) {
2403                         rc = sb->s_op->freeze_fs(sb);
2404                         if (rc)
2405                                 goto out;
2406                 }
2407
2408                 if (jdev && (jdev != dev)) {
2409                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2410                                (long)jdev);
2411                         priv_dev_set_rdonly(jdev);
2412                 }
2413                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2414                 priv_dev_set_rdonly(dev);
2415
2416                 if (sb->s_op->unfreeze_fs)
2417                         sb->s_op->unfreeze_fs(sb);
2418         }
2419
2420 out:
2421         if (rc)
2422                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2423                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2424
2425         RETURN(rc);
2426 }
2427
2428 /**
2429  * Note: we do not count into QUOTA here.
2430  * If we mount with --data_journal we may need more.
2431  */
2432 const int osd_dto_credits_noquota[DTO_NR] = {
2433         /**
2434          * Insert.
2435          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2436          * SINGLEDATA_TRANS_BLOCKS(8)
2437          * XXX Note: maybe iam need more, since iam have more level than
2438          *           EXT3 htree.
2439          */
2440         [DTO_INDEX_INSERT]  = 16,
2441         /**
2442          * Delete
2443          * just modify a single entry, probably merge few within a block
2444          */
2445         [DTO_INDEX_DELETE]  = 1,
2446         /**
2447          * Used for OI scrub
2448          */
2449         [DTO_INDEX_UPDATE]  = 16,
2450         /**
2451          * 4(inode, inode bits, groups, GDT)
2452          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2453          */
2454         [DTO_OBJECT_CREATE] = 4,
2455         /**
2456          * 4(inode, inode bits, groups, GDT)
2457          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2458          */
2459         [DTO_OBJECT_DELETE] = 4,
2460         /**
2461          * Attr set credits (inode)
2462          */
2463         [DTO_ATTR_SET_BASE] = 1,
2464         /**
2465          * Xattr set. The same as xattr of EXT3.
2466          * DATA_TRANS_BLOCKS(14)
2467          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2468          * are also counted in. Do not know why?
2469          */
2470         [DTO_XATTR_SET]     = 14,
2471         /**
2472          * credits for inode change during write.
2473          */
2474         [DTO_WRITE_BASE]    = 3,
2475         /**
2476          * credits for single block write.
2477          */
2478         [DTO_WRITE_BLOCK]   = 14,
2479         /**
2480          * Attr set credits for chown.
2481          * This is extra credits for setattr, and it is null without quota
2482          */
2483         [DTO_ATTR_SET_CHOWN] = 0
2484 };
2485
2486 static const struct dt_device_operations osd_dt_ops = {
2487         .dt_root_get       = osd_root_get,
2488         .dt_statfs         = osd_statfs,
2489         .dt_trans_create   = osd_trans_create,
2490         .dt_trans_start    = osd_trans_start,
2491         .dt_trans_stop     = osd_trans_stop,
2492         .dt_trans_cb_add   = osd_trans_cb_add,
2493         .dt_conf_get       = osd_conf_get,
2494         .dt_sync           = osd_sync,
2495         .dt_ro             = osd_ro,
2496         .dt_commit_async   = osd_commit_async,
2497 };
2498
2499 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2500                           unsigned int role)
2501 {
2502         struct osd_object *obj = osd_dt_obj(dt);
2503         struct osd_thread_info *oti = osd_oti_get(env);
2504
2505         LINVRNT(osd_invariant(obj));
2506
2507         LASSERT(obj->oo_owner != env);
2508         down_read_nested(&obj->oo_sem, role);
2509
2510         LASSERT(obj->oo_owner == NULL);
2511         oti->oti_r_locks++;
2512 }
2513
2514 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2515                            unsigned int role)
2516 {
2517         struct osd_object *obj = osd_dt_obj(dt);
2518         struct osd_thread_info *oti = osd_oti_get(env);
2519
2520         LINVRNT(osd_invariant(obj));
2521
2522         LASSERT(obj->oo_owner != env);
2523         down_write_nested(&obj->oo_sem, role);
2524
2525         LASSERT(obj->oo_owner == NULL);
2526         obj->oo_owner = env;
2527         oti->oti_w_locks++;
2528 }
2529
2530 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2531 {
2532         struct osd_object *obj = osd_dt_obj(dt);
2533         struct osd_thread_info *oti = osd_oti_get(env);
2534
2535         LINVRNT(osd_invariant(obj));
2536
2537         LASSERT(oti->oti_r_locks > 0);
2538         oti->oti_r_locks--;
2539         up_read(&obj->oo_sem);
2540 }
2541
2542 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2543 {
2544         struct osd_object *obj = osd_dt_obj(dt);
2545         struct osd_thread_info *oti = osd_oti_get(env);
2546
2547         LINVRNT(osd_invariant(obj));
2548
2549         LASSERT(obj->oo_owner == env);
2550         LASSERT(oti->oti_w_locks > 0);
2551         oti->oti_w_locks--;
2552         obj->oo_owner = NULL;
2553         up_write(&obj->oo_sem);
2554 }
2555
2556 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2557 {
2558         struct osd_object *obj = osd_dt_obj(dt);
2559
2560         LINVRNT(osd_invariant(obj));
2561
2562         return obj->oo_owner == env;
2563 }
2564
2565 static struct timespec *osd_inode_time(const struct lu_env *env,
2566                                        struct inode *inode, __u64 seconds)
2567 {
2568         struct osd_thread_info *oti = osd_oti_get(env);
2569         struct timespec *t = &oti->oti_time;
2570
2571         t->tv_sec = seconds;
2572         t->tv_nsec = 0;
2573         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2574         return t;
2575 }
2576
2577 static void osd_inode_getattr(const struct lu_env *env,
2578                               struct inode *inode, struct lu_attr *attr)
2579 {
2580         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2581                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2582                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2583                            LA_BLKSIZE | LA_TYPE;
2584
2585         attr->la_atime   = LTIME_S(inode->i_atime);
2586         attr->la_mtime   = LTIME_S(inode->i_mtime);
2587         attr->la_ctime   = LTIME_S(inode->i_ctime);
2588         attr->la_mode    = inode->i_mode;
2589         attr->la_size    = i_size_read(inode);
2590         attr->la_blocks  = inode->i_blocks;
2591         attr->la_uid     = i_uid_read(inode);
2592         attr->la_gid     = i_gid_read(inode);
2593         attr->la_projid  = i_projid_read(inode);
2594         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2595         attr->la_nlink   = inode->i_nlink;
2596         attr->la_rdev    = inode->i_rdev;
2597         attr->la_blksize = 1 << inode->i_blkbits;
2598         attr->la_blkbits = inode->i_blkbits;
2599         /*
2600          * Ext4 did not transfer inherit flags from raw inode
2601          * to inode flags, and ext4 internally test raw inode
2602          * @i_flags directly. Instead of patching ext4, we do it here.
2603          */
2604         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2605                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2606 }
2607
2608 static int osd_attr_get(const struct lu_env *env,
2609                         struct dt_object *dt,
2610                         struct lu_attr *attr)
2611 {
2612         struct osd_object *obj = osd_dt_obj(dt);
2613
2614         if (unlikely(!dt_object_exists(dt)))
2615                 return -ENOENT;
2616         if (unlikely(obj->oo_destroyed))
2617                 return -ENOENT;
2618
2619         LASSERT(!dt_object_remote(dt));
2620         LINVRNT(osd_invariant(obj));
2621
2622         spin_lock(&obj->oo_guard);
2623         osd_inode_getattr(env, obj->oo_inode, attr);
2624         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2625                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2626         spin_unlock(&obj->oo_guard);
2627
2628         return 0;
2629 }
2630
2631 static int osd_declare_attr_qid(const struct lu_env *env,
2632                                 struct osd_object *obj,
2633                                 struct osd_thandle *oh, long long bspace,
2634                                 qid_t old_id, qid_t new_id, bool enforce,
2635                                 unsigned int type, bool ignore_edquot)
2636 {
2637         int rc;
2638         struct osd_thread_info *info = osd_oti_get(env);
2639         struct lquota_id_info  *qi = &info->oti_qi;
2640
2641         qi->lqi_type = type;
2642         /* inode accounting */
2643         qi->lqi_is_blk = false;
2644
2645         /* one more inode for the new id ... */
2646         qi->lqi_id.qid_uid = new_id;
2647         qi->lqi_space      = 1;
2648         /* Reserve credits for the new id */
2649         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2650         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2651                 rc = 0;
2652         if (rc)
2653                 RETURN(rc);
2654
2655         /* and one less inode for the current id */
2656         qi->lqi_id.qid_uid = old_id;
2657         qi->lqi_space = -1;
2658         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2659         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2660                 rc = 0;
2661         if (rc)
2662                 RETURN(rc);
2663
2664         /* block accounting */
2665         qi->lqi_is_blk = true;
2666
2667         /* more blocks for the new id ... */
2668         qi->lqi_id.qid_uid = new_id;
2669         qi->lqi_space      = bspace;
2670         /*
2671          * Credits for the new uid has been reserved, re-use "obj"
2672          * to save credit reservation.
2673          */
2674         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2675         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2676                 rc = 0;
2677         if (rc)
2678                 RETURN(rc);
2679
2680         /* and finally less blocks for the current uid */
2681         qi->lqi_id.qid_uid = old_id;
2682         qi->lqi_space      = -bspace;
2683         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2684         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2685                 rc = 0;
2686
2687         RETURN(rc);
2688 }
2689
2690 static int osd_declare_attr_set(const struct lu_env *env,
2691                                 struct dt_object *dt,
2692                                 const struct lu_attr *attr,
2693                                 struct thandle *handle)
2694 {
2695         struct osd_thandle *oh;
2696         struct osd_object *obj;
2697         qid_t uid;
2698         qid_t gid;
2699         long long bspace;
2700         int rc = 0;
2701         bool enforce;
2702
2703         ENTRY;
2704
2705         LASSERT(dt != NULL);
2706         LASSERT(handle != NULL);
2707
2708         obj = osd_dt_obj(dt);
2709         LASSERT(osd_invariant(obj));
2710
2711         oh = container_of0(handle, struct osd_thandle, ot_super);
2712         LASSERT(oh->ot_handle == NULL);
2713
2714         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2715                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2716
2717         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2718                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2719
2720         if (attr == NULL || obj->oo_inode == NULL)
2721                 RETURN(rc);
2722
2723         bspace   = obj->oo_inode->i_blocks << 9;
2724         bspace   = toqb(bspace);
2725
2726         /*
2727          * Changing ownership is always preformed by super user, it should not
2728          * fail with EDQUOT unless required explicitly.
2729          *
2730          * We still need to call the osd_declare_qid() to calculate the journal
2731          * credits for updating quota accounting files and to trigger quota
2732          * space adjustment once the operation is completed.
2733          */
2734         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2735                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2736
2737                 if (!ignore_edquot)
2738                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2739                                "(the quota space is %lld)\n",
2740                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2741                                attr->la_gid, bspace);
2742
2743                 /* USERQUOTA */
2744                 uid = i_uid_read(obj->oo_inode);
2745                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2746                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2747                                           attr->la_uid, enforce, USRQUOTA,
2748                                           true);
2749                 if (rc)
2750                         RETURN(rc);
2751
2752                 gid = i_gid_read(obj->oo_inode);
2753                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2754                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2755                                           i_gid_read(obj->oo_inode),
2756                                           attr->la_gid, enforce, GRPQUOTA,
2757                                           ignore_edquot);
2758                 if (rc)
2759                         RETURN(rc);
2760
2761         }
2762 #ifdef HAVE_PROJECT_QUOTA
2763         if (attr->la_valid & LA_PROJID) {
2764                 __u32 projid = i_projid_read(obj->oo_inode);
2765
2766                 enforce = (attr->la_valid & LA_PROJID) &&
2767                                         (attr->la_projid != projid);
2768                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2769                                           (qid_t)projid, (qid_t)attr->la_projid,
2770                                           enforce, PRJQUOTA, true);
2771                 if (rc)
2772                         RETURN(rc);
2773         }
2774 #endif
2775         RETURN(rc);
2776 }
2777
2778 static int osd_inode_setattr(const struct lu_env *env,
2779                              struct inode *inode, const struct lu_attr *attr)
2780 {
2781         __u64 bits = attr->la_valid;
2782
2783         /* Only allow set size for regular file */
2784         if (!S_ISREG(inode->i_mode))
2785                 bits &= ~(LA_SIZE | LA_BLOCKS);
2786
2787         if (bits == 0)
2788                 return 0;
2789
2790         if (bits & LA_ATIME)
2791                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2792         if (bits & LA_CTIME)
2793                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2794         if (bits & LA_MTIME)
2795                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2796         if (bits & LA_SIZE) {
2797                 spin_lock(&inode->i_lock);
2798                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2799                 i_size_write(inode, attr->la_size);
2800                 spin_unlock(&inode->i_lock);
2801         }
2802
2803         /*
2804          * OSD should not change "i_blocks" which is used by quota.
2805          * "i_blocks" should be changed by ldiskfs only.
2806          */
2807         if (bits & LA_MODE)
2808                 inode->i_mode = (inode->i_mode & S_IFMT) |
2809                                 (attr->la_mode & ~S_IFMT);
2810         if (bits & LA_UID)
2811                 i_uid_write(inode, attr->la_uid);
2812         if (bits & LA_GID)
2813                 i_gid_write(inode, attr->la_gid);
2814         if (bits & LA_PROJID)
2815                 i_projid_write(inode, attr->la_projid);
2816         if (bits & LA_NLINK)
2817                 set_nlink(inode, attr->la_nlink);
2818         if (bits & LA_RDEV)
2819                 inode->i_rdev = attr->la_rdev;
2820
2821         if (bits & LA_FLAGS) {
2822                 /* always keep S_NOCMTIME */
2823                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2824                                  S_NOCMTIME;
2825                 /*
2826                  * Ext4 did not transfer inherit flags from
2827                  * @inode->i_flags to raw inode i_flags when writing
2828                  * flags, we do it explictly here.
2829                  */
2830                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2831                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2832                 else
2833                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2834         }
2835         return 0;
2836 }
2837
2838 #ifdef HAVE_PROJECT_QUOTA
2839 static int osd_transfer_project(struct inode *inode, __u32 projid)
2840 {
2841         struct super_block *sb = inode->i_sb;
2842         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2843         int err;
2844         kprojid_t kprojid;
2845         struct ldiskfs_iloc iloc;
2846         struct ldiskfs_inode *raw_inode;
2847         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2848
2849         if (!ldiskfs_has_feature_project(sb)) {
2850                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2851                         == LDISKFS_DEF_PROJID);
2852                 if (projid != LDISKFS_DEF_PROJID)
2853                         return -EOPNOTSUPP;
2854                 else
2855                         return 0;
2856         }
2857
2858         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2859                 return -EOPNOTSUPP;
2860
2861         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2862         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2863                 return 0;
2864
2865         err = ldiskfs_get_inode_loc(inode, &iloc);
2866         if (err)
2867                 return err;
2868
2869         raw_inode = ldiskfs_raw_inode(&iloc);
2870         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2871                 err = -EOVERFLOW;
2872                 brelse(iloc.bh);
2873                 return err;
2874         }
2875         brelse(iloc.bh);
2876
2877         dquot_initialize(inode);
2878         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2879         if (transfer_to[PRJQUOTA]) {
2880                 err = __dquot_transfer(inode, transfer_to);
2881                 dqput(transfer_to[PRJQUOTA]);
2882                 if (err)
2883                         return err;
2884         }
2885
2886         return err;
2887 }
2888 #endif
2889
2890 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2891 {
2892         int rc;
2893
2894         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2895             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2896                 struct iattr iattr;
2897
2898                 ll_vfs_dq_init(inode);
2899                 iattr.ia_valid = 0;
2900                 if (attr->la_valid & LA_UID)
2901                         iattr.ia_valid |= ATTR_UID;
2902                 if (attr->la_valid & LA_GID)
2903                         iattr.ia_valid |= ATTR_GID;
2904                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2905                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2906
2907                 rc = ll_vfs_dq_transfer(inode, &iattr);
2908                 if (rc) {
2909                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2910                                "enforcement enabled on the ldiskfs "
2911                                "filesystem?\n", inode->i_sb->s_id, rc);
2912                         return rc;
2913                 }
2914         }
2915
2916         /* Handle project id transfer here properly */
2917         if (attr->la_valid & LA_PROJID &&
2918             attr->la_projid != i_projid_read(inode)) {
2919 #ifdef HAVE_PROJECT_QUOTA
2920                 rc = osd_transfer_project(inode, attr->la_projid);
2921 #else
2922                 rc = -ENOTSUPP;
2923 #endif
2924                 if (rc) {
2925                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2926                                "enforcement enabled on the ldiskfs "
2927                                "filesystem?\n", inode->i_sb->s_id, rc);
2928                         return rc;
2929                 }
2930         }
2931         return 0;
2932 }
2933
2934 static int osd_attr_set(const struct lu_env *env,
2935                         struct dt_object *dt,
2936                         const struct lu_attr *attr,
2937                         struct thandle *handle)
2938 {
2939         struct osd_object *obj = osd_dt_obj(dt);
2940         struct inode *inode;
2941         int rc;
2942
2943         if (!dt_object_exists(dt))
2944                 return -ENOENT;
2945
2946         LASSERT(handle != NULL);
2947         LASSERT(!dt_object_remote(dt));
2948         LASSERT(osd_invariant(obj));
2949
2950         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2951
2952         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2953             !osd_obj2dev(obj)->od_is_ost) {
2954                 struct osd_thread_info *oti = osd_oti_get(env);
2955                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2956                 struct lu_fid *fid1 = &oti->oti_fid;
2957                 struct osd_inode_id *id = &oti->oti_id;
2958                 struct iam_path_descr *ipd;
2959                 struct iam_container *bag;
2960                 struct osd_thandle *oh;
2961                 int rc;
2962
2963                 fid_cpu_to_be(fid1, fid0);
2964                 memset(id, 1, sizeof(*id));
2965                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2966                                   fid0)->oi_dir.od_container;
2967                 ipd = osd_idx_ipd_get(env, bag);
2968                 if (unlikely(ipd == NULL))
2969                         RETURN(-ENOMEM);
2970
2971                 oh = container_of0(handle, struct osd_thandle, ot_super);
2972                 rc = iam_update(oh->ot_handle, bag,
2973                                 (const struct iam_key *)fid1,
2974                                 (const struct iam_rec *)id, ipd);
2975                 osd_ipd_put(env, bag, ipd);
2976                 return(rc > 0 ? 0 : rc);
2977         }
2978
2979         inode = obj->oo_inode;
2980
2981         rc = osd_quota_transfer(inode, attr);
2982         if (rc)
2983                 return rc;
2984
2985         spin_lock(&obj->oo_guard);
2986         rc = osd_inode_setattr(env, inode, attr);
2987         spin_unlock(&obj->oo_guard);
2988         if (rc != 0)
2989                 GOTO(out, rc);
2990
2991         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2992
2993         if (!(attr->la_valid & LA_FLAGS))
2994                 GOTO(out, rc);
2995
2996         /* Let's check if there are extra flags need to be set into LMA */
2997         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2998                 struct osd_thread_info *info = osd_oti_get(env);
2999                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3000
3001                 LASSERT(!obj->oo_pfid_in_lma);
3002
3003                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3004                                  &info->oti_ost_attrs);
3005                 if (rc)
3006                         GOTO(out, rc);
3007
3008                 lma->lma_incompat |=
3009                         lustre_to_lma_flags(attr->la_flags);
3010                 lustre_lma_swab(lma);
3011                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3012                                      lma, sizeof(*lma), XATTR_REPLACE);
3013                 if (rc != 0) {
3014                         struct osd_device *osd = osd_obj2dev(obj);
3015
3016                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3017                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3018                               lma->lma_incompat, rc);
3019                 } else {
3020                         obj->oo_lma_flags =
3021                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3022                 }
3023                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3024         }
3025 out:
3026         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3027
3028         return rc;
3029 }
3030
3031 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3032                                            struct osd_object *obj,
3033                                            const char *name, const int namelen)
3034 {
3035         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3036 }
3037
3038 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3039                       umode_t mode, struct dt_allocation_hint *hint,
3040                       struct thandle *th)
3041 {
3042         int result;
3043         struct osd_device *osd = osd_obj2dev(obj);
3044         struct osd_thandle *oth;
3045         struct dt_object *parent = NULL;
3046         struct inode *inode;
3047
3048         LINVRNT(osd_invariant(obj));
3049         LASSERT(obj->oo_inode == NULL);
3050         LASSERT(obj->oo_hl_head == NULL);
3051
3052         if (S_ISDIR(mode) && ldiskfs_pdo) {
3053                 obj->oo_hl_head =
3054                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3055                 if (obj->oo_hl_head == NULL)
3056                         return -ENOMEM;
3057         }
3058
3059         oth = container_of(th, struct osd_thandle, ot_super);
3060         LASSERT(oth->ot_handle->h_transaction != NULL);
3061
3062         if (hint != NULL && hint->dah_parent != NULL &&
3063             !dt_object_remote(hint->dah_parent))
3064                 parent = hint->dah_parent;
3065
3066         inode = ldiskfs_create_inode(oth->ot_handle,
3067                                      parent ? osd_dt_obj(parent)->oo_inode :
3068                                               osd_sb(osd)->s_root->d_inode,
3069                                      mode);
3070         if (!IS_ERR(inode)) {
3071                 /* Do not update file c/mtime in ldiskfs. */
3072                 inode->i_flags |= S_NOCMTIME;
3073
3074                 /*
3075                  * For new created object, it must be consistent,
3076                  * and it is unnecessary to scrub against it.
3077                  */
3078                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3079
3080                 obj->oo_inode = inode;
3081                 result = 0;
3082         } else {
3083                 if (obj->oo_hl_head != NULL) {
3084                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3085                         obj->oo_hl_head = NULL;
3086                 }
3087                 result = PTR_ERR(inode);
3088         }
3089         LINVRNT(osd_invariant(obj));
3090         return result;
3091 }
3092
3093 enum {
3094         OSD_NAME_LEN = 255
3095 };
3096
3097 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3098                      struct lu_attr *attr,
3099                      struct dt_allocation_hint *hint,
3100                      struct dt_object_format *dof,
3101                      struct thandle *th)
3102 {
3103         int result;
3104         struct osd_thandle *oth;
3105         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3106
3107         LASSERT(S_ISDIR(attr->la_mode));
3108
3109         oth = container_of(th, struct osd_thandle, ot_super);
3110         LASSERT(oth->ot_handle->h_transaction != NULL);
3111         result = osd_mkfile(info, obj, mode, hint, th);
3112
3113         return result;
3114 }
3115
3116 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3117                         struct lu_attr *attr,
3118                         struct dt_allocation_hint *hint,
3119                         struct dt_object_format *dof,
3120                         struct thandle *th)
3121 {
3122         int result;
3123         struct osd_thandle *oth;
3124         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3125
3126         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3127
3128         LASSERT(S_ISREG(attr->la_mode));
3129
3130         oth = container_of(th, struct osd_thandle, ot_super);
3131         LASSERT(oth->ot_handle->h_transaction != NULL);
3132
3133         result = osd_mkfile(info, obj, mode, hint, th);
3134         if (result == 0) {
3135                 LASSERT(obj->oo_inode != NULL);
3136                 if (feat->dif_flags & DT_IND_VARKEY)
3137                         result = iam_lvar_create(obj->oo_inode,
3138                                                  feat->dif_keysize_max,
3139                                                  feat->dif_ptrsize,
3140                                                  feat->dif_recsize_max,
3141                                                  oth->ot_handle);
3142                 else
3143                         result = iam_lfix_create(obj->oo_inode,
3144                                                  feat->dif_keysize_max,
3145                                                  feat->dif_ptrsize,
3146                                                  feat->dif_recsize_max,
3147                                                  oth->ot_handle);
3148         }
3149         return result;
3150 }
3151
3152 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3153                      struct lu_attr *attr,
3154                      struct dt_allocation_hint *hint,
3155                      struct dt_object_format *dof,
3156                      struct thandle *th)
3157 {
3158         LASSERT(S_ISREG(attr->la_mode));
3159         return osd_mkfile(info, obj, (attr->la_mode &
3160                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3161 }
3162
3163 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3164                      struct lu_attr *attr,
3165                      struct dt_allocation_hint *hint,
3166                      struct dt_object_format *dof,
3167                      struct thandle *th)
3168 {
3169         LASSERT(S_ISLNK(attr->la_mode));
3170         return osd_mkfile(info, obj, (attr->la_mode &
3171                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3172 }
3173
3174 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3175                      struct lu_attr *attr,
3176                      struct dt_allocation_hint *hint,
3177                      struct dt_object_format *dof,
3178                      struct thandle *th)
3179 {
3180         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3181         int result;
3182
3183         LINVRNT(osd_invariant(obj));
3184         LASSERT(obj->oo_inode == NULL);
3185         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3186                 S_ISFIFO(mode) || S_ISSOCK(mode));
3187
3188         result = osd_mkfile(info, obj, mode, hint, th);
3189         if (result == 0) {
3190                 LASSERT(obj->oo_inode != NULL);
3191                 /*
3192                  * This inode should be marked dirty for i_rdev.  Currently
3193                  * that is done in the osd_attr_init().
3194                  */
3195                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3196                                    attr->la_rdev);
3197         }
3198         LINVRNT(osd_invariant(obj));
3199         return result;
3200 }
3201
3202 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3203                               struct lu_attr *,
3204                               struct dt_allocation_hint *hint,
3205                               struct dt_object_format *dof,
3206                               struct thandle *);
3207
3208 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3209 {
3210         osd_obj_type_f result;
3211
3212         switch (type) {
3213         case DFT_DIR:
3214                 result = osd_mkdir;
3215                 break;
3216         case DFT_REGULAR:
3217                 result = osd_mkreg;
3218                 break;
3219         case DFT_SYM:
3220                 result = osd_mksym;
3221                 break;
3222         case DFT_NODE:
3223                 result = osd_mknod;
3224                 break;
3225         case DFT_INDEX:
3226                 result = osd_mk_index;
3227                 break;
3228
3229         default:
3230                 LBUG();
3231                 break;
3232         }
3233         return result;
3234 }
3235
3236
3237 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3238                         struct dt_object *parent, struct dt_object *child,
3239                         umode_t child_mode)
3240 {
3241         LASSERT(ah);
3242
3243         ah->dah_parent = parent;
3244         ah->dah_mode = child_mode;
3245
3246         if (parent != NULL && !dt_object_remote(parent)) {
3247                 /* will help to find FID->ino at dt_insert("..") */
3248                 struct osd_object *pobj = osd_dt_obj(parent);
3249
3250                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3251         }
3252 }
3253
3254 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3255                           struct lu_attr *attr, struct dt_object_format *dof)
3256 {
3257         struct inode *inode = obj->oo_inode;
3258         __u64 valid = attr->la_valid;
3259         int result;
3260
3261         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3262
3263         if (dof->dof_type != DFT_NODE)
3264                 attr->la_valid &= ~LA_RDEV;
3265         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3266                 attr->la_valid &= ~LA_ATIME;
3267         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3268                 attr->la_valid &= ~LA_CTIME;
3269         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3270                 attr->la_valid &= ~LA_MTIME;
3271
3272         result = osd_quota_transfer(inode, attr);
3273         if (result)
3274                 return;
3275
3276         if (attr->la_valid != 0) {
3277                 result = osd_inode_setattr(info->oti_env, inode, attr);
3278                 /*
3279                  * The osd_inode_setattr() should always succeed here.  The
3280                  * only error that could be returned is EDQUOT when we are
3281                  * trying to change the UID or GID of the inode. However, this
3282                  * should not happen since quota enforcement is no longer
3283                  * enabled on ldiskfs (lquota takes care of it).
3284                  */
3285                 LASSERTF(result == 0, "%d\n", result);
3286                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3287         }
3288
3289         attr->la_valid = valid;
3290 }
3291
3292 /**
3293  * Helper function for osd_create()
3294  *
3295  * \retval 0, on success
3296  */
3297 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3298                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3299                         struct dt_object_format *dof, struct thandle *th)
3300 {
3301         int result;
3302         __u32 umask;
3303
3304         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3305
3306         /* we drop umask so that permissions we pass are not affected */
3307         umask = current->fs->umask;
3308         current->fs->umask = 0;
3309
3310         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3311                                                   th);
3312         if (likely(obj->oo_inode != NULL)) {
3313                 LASSERT(obj->oo_inode->i_state & I_NEW);
3314
3315                 /*
3316                  * Unlock the inode before attr initialization to avoid
3317                  * unnecessary dqget operations. LU-6378
3318                  */
3319                 unlock_new_inode(obj->oo_inode);
3320         }
3321
3322         if (likely(result == 0)) {
3323                 osd_attr_init(info, obj, attr, dof);
3324                 osd_object_init0(obj);
3325         }
3326
3327         /* restore previous umask value */
3328         current->fs->umask = umask;
3329
3330         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3331
3332         return result;
3333 }
3334
3335 /**
3336  * Helper function for osd_create()
3337  *
3338  * \retval 0, on success
3339  */
3340 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3341                            const struct lu_fid *fid, struct thandle *th)
3342 {
3343         struct osd_thread_info *info = osd_oti_get(env);
3344         struct osd_inode_id    *id   = &info->oti_id;
3345         struct osd_device      *osd  = osd_obj2dev(obj);
3346         struct osd_thandle     *oh;
3347         int rc;
3348
3349         LASSERT(obj->oo_inode != NULL);
3350
3351         oh = container_of0(th, struct osd_thandle, ot_super);
3352         LASSERT(oh->ot_handle);
3353         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3354
3355         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3356         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3357                            OI_CHECK_FLD, NULL);
3358         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3359
3360         return rc;
3361 }
3362
3363 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3364                    u64 seq, struct lu_seq_range *range)
3365 {
3366         struct seq_server_site *ss = osd_seq_site(osd);
3367
3368         if (fid_seq_is_idif(seq)) {
3369                 fld_range_set_ost(range);
3370                 range->lsr_index = idif_ost_idx(seq);
3371                 return 0;
3372         }
3373
3374         if (!fid_seq_in_fldb(seq)) {
3375                 fld_range_set_mdt(range);
3376                 if (ss != NULL)
3377                         /*
3378                          * FIXME: If ss is NULL, it suppose not get lsr_index
3379                          * at all
3380                          */
3381                         range->lsr_index = ss->ss_node_id;
3382                 return 0;
3383         }
3384
3385         LASSERT(ss != NULL);
3386         fld_range_set_any(range);
3387         /* OSD will only do local fld lookup */
3388         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3389 }
3390
3391 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3392                               struct lu_attr *attr,
3393                               struct dt_allocation_hint *hint,
3394                               struct dt_object_format *dof,
3395                               struct thandle *handle)
3396 {
3397         struct osd_thandle *oh;
3398         int rc;
3399
3400         ENTRY;
3401
3402         LASSERT(handle != NULL);
3403
3404         oh = container_of0(handle, struct osd_thandle, ot_super);
3405         LASSERT(oh->ot_handle == NULL);
3406
3407         /*
3408          * EA object consumes more credits than regular object: osd_mk_index
3409          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3410          * leaf_node, could involves the block, block bitmap, groups, GDT
3411          * change for each block, so add 4 * 2 credits in that case.
3412          */
3413         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3414                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3415                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3416         /*
3417          * Reuse idle OI block may cause additional one OI block
3418          * to be changed.
3419          */
3420         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3421                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3422
3423         /* will help to find FID->ino mapping at dt_insert() */
3424         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3425                                    osd_dt_obj(dt));
3426         if (rc != 0)
3427                 RETURN(rc);
3428
3429         if (!attr)
3430                 RETURN(0);
3431
3432         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3433                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3434                                    NULL, OSD_QID_INODE);
3435         if (rc != 0)
3436                 RETURN(rc);
3437
3438         RETURN(rc);
3439 }
3440
3441 /**
3442  * Called to destroy on-disk representation of the object
3443  *
3444  * Concurrency: must be locked
3445  */
3446 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3447                                struct thandle *th)
3448 {
3449         struct osd_object *obj = osd_dt_obj(dt);
3450         struct inode *inode = obj->oo_inode;
3451         struct osd_thandle *oh;
3452         int rc;
3453
3454         ENTRY;
3455
3456         if (inode == NULL)
3457                 RETURN(-ENOENT);
3458
3459         oh = container_of0(th, struct osd_thandle, ot_super);
3460         LASSERT(oh->ot_handle == NULL);
3461
3462         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3463                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3464
3465         /* For removing agent entry */
3466         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3467                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3468
3469         /*
3470          * Recycle idle OI leaf may cause additional three OI blocks
3471          * to be changed.
3472          */
3473         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3474                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3475                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3476         /* one less inode */
3477         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3478                                    i_projid_read(inode), -1, oh, obj, NULL,
3479                                    OSD_QID_INODE);
3480         if (rc)
3481                 RETURN(rc);
3482         /* data to be truncated */
3483         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3484                                    i_projid_read(inode), 0, oh, obj, NULL,
3485                                    OSD_QID_BLK);
3486         if (rc)
3487                 RETURN(rc);
3488
3489         /*
3490          * will help to find FID->ino when this object is being
3491          * added to PENDING
3492          */
3493         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3494
3495         RETURN(rc);
3496 }
3497
3498 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3499                        struct thandle *th)
3500 {
3501         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3502         struct osd_object *obj = osd_dt_obj(dt);
3503         struct inode *inode = obj->oo_inode;
3504         struct osd_device *osd = osd_obj2dev(obj);
3505         struct osd_thandle *oh;
3506         int result;
3507
3508         ENTRY;
3509
3510         oh = container_of0(th, struct osd_thandle, ot_super);
3511         LASSERT(oh->ot_handle);
3512         LASSERT(inode);
3513         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3514
3515         if (unlikely(fid_is_acct(fid)))
3516                 RETURN(-EPERM);
3517
3518         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3519                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3520                 if (result != 0)
3521                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3522                                osd_name(osd), PFID(fid), result);
3523         }
3524
3525         if (S_ISDIR(inode->i_mode)) {
3526                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3527                         inode->i_nlink == 2);
3528
3529                 spin_lock(&obj->oo_guard);
3530                 clear_nlink(inode);
3531                 spin_unlock(&obj->oo_guard);
3532                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3533         }
3534
3535         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3536
3537         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3538
3539         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3540                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3541                                        oh->ot_handle, OI_CHECK_FLD);
3542
3543         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3544         /* XXX: add to ext3 orphan list */
3545         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3546
3547         /* not needed in the cache anymore */
3548         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3549         obj->oo_destroyed = 1;
3550
3551         RETURN(0);
3552 }
3553
3554 /**
3555  * Put the fid into lustre_mdt_attrs, and then place the structure
3556  * inode's ea. This fid should not be altered during the life time
3557  * of the inode.
3558  *
3559  * \retval +ve, on success
3560  * \retval -ve, on error
3561  *
3562  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3563  */
3564 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3565                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3566 {
3567         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3568         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3569         int rc;
3570
3571         ENTRY;
3572
3573         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3574                 RETURN(0);
3575
3576         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3577                 rc = -ENOMEM;
3578
3579         lustre_loa_init(loa, fid, compat, incompat);
3580         lustre_loa_swab(loa, false);
3581
3582         /*
3583          * For the OST device with 256 bytes inode size by default,
3584          * the PFID EA will be stored together with LMA EA to avoid
3585          * performance trouble. Otherwise the PFID EA can be stored
3586          * independently. LU-8998
3587          */
3588         if ((compat & LMAC_FID_ON_OST) &&
3589             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3590                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3591                                      sizeof(*loa), XATTR_CREATE);
3592         else
3593                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3594                                      sizeof(*lma), XATTR_CREATE);
3595         /*
3596          * LMA may already exist, but we need to check that all the
3597          * desired compat/incompat flags have been added.
3598          */
3599         if (unlikely(rc == -EEXIST)) {
3600                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3601                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3602                 if (rc < 0)
3603                         RETURN(rc);
3604
3605                 if (rc < sizeof(*lma))
3606                         RETURN(-EINVAL);
3607
3608                 lustre_loa_swab(loa, true);
3609                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3610                     ((compat == 0 && incompat == 0) ||
3611                      (!(~lma->lma_compat & compat) &&
3612                       !(~lma->lma_incompat & incompat))))
3613                         RETURN(0);
3614
3615                 lma->lma_self_fid = *fid;
3616                 lma->lma_compat |= compat;
3617                 lma->lma_incompat |= incompat;
3618                 if (rc == sizeof(*lma)) {
3619                         lustre_lma_swab(lma);
3620                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3621                                              sizeof(*lma), XATTR_REPLACE);
3622                 } else {
3623                         lustre_loa_swab(loa, false);
3624                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3625                                              sizeof(*loa), XATTR_REPLACE);
3626                 }
3627         }
3628
3629         RETURN(rc);
3630 }
3631
3632 /**
3633  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3634  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3635  * To have compatilibility with 1.8 ldiskfs driver we need to have
3636  * magic number at start of fid data.
3637  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3638  * its inmemory API.
3639  */
3640 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3641                                          const struct lu_fid *fid)
3642 {
3643         if (!fid_is_namespace_visible(fid) ||
3644             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3645                 param->edp_magic = 0;
3646                 return;
3647         }
3648
3649         param->edp_magic = LDISKFS_LUFID_MAGIC;
3650         param->edp_len =  sizeof(struct lu_fid) + 1;
3651         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3652 }
3653
3654 /**
3655  * Try to read the fid from inode ea into dt_rec.
3656  *
3657  * \param fid object fid.
3658  *
3659  * \retval 0 on success
3660  */
3661 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3662                           __u32 ino, struct lu_fid *fid,
3663                           struct osd_inode_id *id)
3664 {
3665         struct osd_thread_info *info  = osd_oti_get(env);
3666         struct inode *inode;
3667
3668         ENTRY;
3669
3670         osd_id_gen(id, ino, OSD_OII_NOGEN);
3671         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3672         if (IS_ERR(inode))
3673                 RETURN(PTR_ERR(inode));
3674
3675         iput(inode);
3676         RETURN(0);
3677 }
3678
3679 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3680                                         struct inode *dir,
3681                                         struct inode *parent_dir,
3682                                         const struct lu_fid *dot_fid,
3683                                         const struct lu_fid *dot_dot_fid,
3684                                         struct osd_thandle *oth)
3685 {
3686         struct ldiskfs_dentry_param *dot_ldp;
3687         struct ldiskfs_dentry_param *dot_dot_ldp;
3688         __u32 saved_nlink = dir->i_nlink;
3689         int rc;
3690
3691         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3692         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3693
3694         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3695         dot_ldp->edp_magic = 0;
3696
3697         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3698                                     dir, dot_ldp, dot_dot_ldp);
3699         /*
3700          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3701          * the subseqent ref_add() will increase the dir->i_nlink
3702          * as 3. That is incorrect for new created directory.
3703          *
3704          * It looks like hack, because we want to make the OSD API
3705          * to be order-independent for new created directory object
3706          * between dt_insert(..) and ref_add() operations.
3707          *
3708          * Here, we only restore the in-RAM dir-inode's nlink attr,
3709          * becuase if the nlink attr is not 2, then there will be
3710          * ref_add() called following the dt_insert(..), such call
3711          * will make both the in-RAM and on-disk dir-inode's nlink
3712          * attr to be set as 2. LU-7447
3713          */
3714         set_nlink(dir, saved_nlink);
3715         return rc;
3716 }
3717
3718 /**
3719  * Create an local agent inode for remote entry
3720  */
3721 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3722                                                   struct osd_device *osd,
3723                                                   struct osd_object *pobj,
3724                                                   const struct lu_fid *fid,
3725                                                   __u32 type,
3726                                                   struct thandle *th)
3727 {
3728         struct osd_thread_info *info = osd_oti_get(env);
3729         struct inode *local;
3730         struct osd_thandle *oh;
3731         int rc;
3732
3733         ENTRY;
3734
3735         LASSERT(th);
3736         oh = container_of(th, struct osd_thandle, ot_super);
3737         LASSERT(oh->ot_handle->h_transaction != NULL);
3738
3739         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3740         if (IS_ERR(local)) {
3741                 CERROR("%s: create local error %d\n", osd_name(osd),
3742                        (int)PTR_ERR(local));
3743                 RETURN(local);
3744         }
3745
3746         /*
3747          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3748          * correct gid on remote file, not agent here
3749          */
3750         local->i_gid = current_fsgid();
3751         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3752         unlock_new_inode(local);
3753
3754         /* Agent inode should not have project ID */
3755 #ifdef  HAVE_PROJECT_QUOTA
3756         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3757             i_projid_read(pobj->oo_inode) != 0) {
3758                 rc = osd_transfer_project(local, 0);
3759                 if (rc) {
3760                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3761                                "quota enforcement enabled on the ldiskfs "
3762                                "filesystem?\n", local->i_sb->s_id, rc);
3763                         RETURN(ERR_PTR(rc));
3764                 }
3765         }
3766 #endif
3767         /* Set special LMA flag for local agent inode */
3768         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3769         if (rc != 0) {
3770                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3771                        osd_name(osd), PFID(fid), rc);
3772                 RETURN(ERR_PTR(rc));
3773         }
3774
3775         if (!S_ISDIR(type))
3776                 RETURN(local);
3777
3778         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3779                                          lu_object_fid(&pobj->oo_dt.do_lu),
3780                                          fid, oh);
3781         if (rc != 0) {
3782                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3783                         osd_name(osd), PFID(fid), rc);
3784                 RETURN(ERR_PTR(rc));
3785         }
3786
3787         RETURN(local);
3788 }
3789
3790 /**
3791  * when direntry is deleted, we have to take care of possible agent inode
3792  * referenced by that. unfortunately we can't do this at that point:
3793  * iget() within a running transaction leads to deadlock and we better do
3794  * not call that every delete declaration to save performance. so we put
3795  * a potention agent inode on a list and process that once the transaction
3796  * is over. Notice it's not any worse in terms of real orphans as regular
3797  * object destroy doesn't put inodes on the on-disk orphan list. this should
3798  * be addressed separately
3799  */
3800 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3801                                             struct osd_thandle *oh,
3802                                             __u32 ino)
3803 {
3804         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3805         struct osd_obj_orphan *oor;
3806
3807         OBD_ALLOC_PTR(oor);
3808         if (oor == NULL)
3809                 return -ENOMEM;
3810
3811         oor->oor_ino = ino;
3812         oor->oor_env = (struct lu_env *)env;
3813         spin_lock(&osd->od_osfs_lock);
3814         list_add(&oor->oor_list, &osd->od_orphan_list);
3815         spin_unlock(&osd->od_osfs_lock);
3816
3817         oh->ot_remove_agents = 1;
3818
3819         return 0;
3820
3821 }
3822
3823 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3824                                                 struct osd_device *osd)
3825 {
3826         struct osd_thread_info *info = osd_oti_get(env);
3827         struct osd_obj_orphan *oor, *tmp;
3828         struct osd_inode_id id;
3829         struct list_head list;
3830         struct inode *inode;
3831         struct lu_fid fid;
3832         handle_t *jh;
3833         __u32 ino;
3834
3835         INIT_LIST_HEAD(&list);
3836
3837         spin_lock(&osd->od_osfs_lock);
3838         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3839                 if (oor->oor_env == env) {
3840                         list_del(&oor->oor_list);
3841                         list_add(&oor->oor_list, &list);
3842                 }
3843         }
3844         spin_unlock(&osd->od_osfs_lock);
3845
3846         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3847
3848                 ino = oor->oor_ino;
3849
3850                 list_del(&oor->oor_list);
3851                 OBD_FREE_PTR(oor);
3852
3853                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3854                 inode = osd_iget_fid(info, osd, &id, &fid);
3855                 if (IS_ERR(inode))
3856                         continue;
3857
3858                 if (!osd_remote_fid(env, osd, &fid)) {
3859                         iput(inode);
3860                         continue;
3861                 }
3862
3863                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3864                 clear_nlink(inode);
3865                 mark_inode_dirty(inode);
3866                 ldiskfs_journal_stop(jh);
3867                 iput(inode);
3868         }
3869
3870         return 0;
3871 }
3872
3873 /**
3874  * OSD layer object create function for OST objects (b=11826).
3875  *
3876  * The FID is inserted into inode xattr here.
3877  *
3878  * \retval   0, on success
3879  * \retval -ve, on error
3880  */
3881 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3882                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3883                       struct dt_object_format *dof, struct thandle *th)
3884 {
3885         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3886         struct osd_object *obj = osd_dt_obj(dt);
3887         struct osd_thread_info *info = osd_oti_get(env);
3888         int result, on_ost = 0;
3889
3890         ENTRY;
3891
3892         if (dt_object_exists(dt))
3893                 RETURN(-EEXIST);
3894
3895         LINVRNT(osd_invariant(obj));
3896         LASSERT(!dt_object_remote(dt));
3897         LASSERT(osd_is_write_locked(env, obj));
3898         LASSERT(th != NULL);
3899
3900         if (unlikely(fid_is_acct(fid)))
3901                 /*
3902                  * Quota files can't be created from the kernel any more,
3903                  * 'tune2fs -O quota' will take care of creating them
3904                  */
3905                 RETURN(-EPERM);
3906
3907         result = __osd_create(info, obj, attr, hint, dof, th);
3908         if (result == 0) {
3909                 if (fid_is_idif(fid) &&
3910                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3911                         struct lu_fid *tfid = &info->oti_fid;
3912                         struct ost_id *oi   = &info->oti_ostid;
3913
3914                         fid_to_ostid(fid, oi);
3915                         ostid_to_fid(tfid, oi, 0);
3916                         on_ost = 1;
3917                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3918                                                 LMAC_FID_ON_OST, 0);
3919                 } else {
3920                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3921                                                fid, OI_CHECK_FLD);
3922                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3923                                                 on_ost ? LMAC_FID_ON_OST : 0,
3924                                                 0);
3925                 }
3926                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3927                         obj->oo_dt.do_body_ops = &osd_body_ops;
3928         }
3929
3930         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3931                 result = __osd_oi_insert(env, obj, fid, th);
3932
3933         /*
3934          * a small optimization - dt_insert() isn't usually applied
3935          * to OST objects, so we don't need to cache OI mapping for
3936          * OST objects
3937          */
3938         if (result == 0 && on_ost == 0) {
3939                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3940
3941                 result = osd_idc_find_and_init(env, osd, obj);
3942                 LASSERT(result == 0);
3943         }
3944
3945         LASSERT(ergo(result == 0,
3946                      dt_object_exists(dt) && !dt_object_remote(dt)));
3947         LINVRNT(osd_invariant(obj));
3948         RETURN(result);
3949 }
3950
3951 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3952                                struct thandle *handle)
3953 {
3954         struct osd_thandle *oh;
3955
3956         /* it's possible that object doesn't exist yet */
3957         LASSERT(handle != NULL);
3958
3959         oh = container_of0(handle, struct osd_thandle, ot_super);
3960         LASSERT(oh->ot_handle == NULL);
3961
3962         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3963                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3964
3965         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3966
3967         return 0;
3968 }
3969
3970 /*
3971  * Concurrency: @dt is write locked.
3972  */
3973 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3974                        struct thandle *th)
3975 {
3976         struct osd_object *obj = osd_dt_obj(dt);
3977         struct inode *inode = obj->oo_inode;
3978         struct osd_thandle *oh;
3979         int rc = 0;
3980
3981         if (!dt_object_exists(dt) || obj->oo_destroyed)
3982                 return -ENOENT;
3983
3984         LINVRNT(osd_invariant(obj));
3985         LASSERT(!dt_object_remote(dt));
3986         LASSERT(osd_is_write_locked(env, obj));
3987         LASSERT(th != NULL);
3988
3989         oh = container_of0(th, struct osd_thandle, ot_super);
3990         LASSERT(oh->ot_handle != NULL);
3991
3992         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3993
3994         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3995                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3996         /*
3997          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3998          * (65000) subdirectories by storing "1" in i_nlink if the link count
3999          * would otherwise overflow. Directory tranversal tools understand
4000          * that (st_nlink == 1) indicates that the filesystem dose not track
4001          * hard links count on the directory, and will not abort subdirectory
4002          * scanning early once (st_nlink - 2) subdirs have been found.
4003          *
4004          * This also has to properly handle the case of inodes with nlink == 0
4005          * in case they are being linked into the PENDING directory
4006          */
4007         spin_lock(&obj->oo_guard);
4008         if (unlikely(inode->i_nlink == 0))
4009                 /* inc_nlink from 0 may cause WARN_ON */
4010                 set_nlink(inode, 1);
4011         else {
4012                 ldiskfs_inc_count(oh->ot_handle, inode);
4013                 if (!S_ISDIR(inode->i_mode))
4014                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4015         }
4016         spin_unlock(&obj->oo_guard);
4017
4018         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4019         LINVRNT(osd_invariant(obj));
4020
4021         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4022
4023         return rc;
4024 }
4025
4026 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4027                                struct thandle *handle)
4028 {
4029         struct osd_thandle *oh;
4030
4031         if (!dt_object_exists(dt))
4032                 return -ENOENT;
4033
4034         LASSERT(!dt_object_remote(dt));
4035         LASSERT(handle != NULL);
4036
4037         oh = container_of0(handle, struct osd_thandle, ot_super);
4038         LASSERT(oh->ot_handle == NULL);
4039
4040         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4041                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4042
4043         return 0;
4044 }
4045
4046 /*
4047  * Concurrency: @dt is write locked.
4048  */
4049 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4050                        struct thandle *th)
4051 {
4052         struct osd_object *obj = osd_dt_obj(dt);
4053         struct inode *inode = obj->oo_inode;
4054         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4055         struct osd_thandle *oh;
4056
4057         if (!dt_object_exists(dt))
4058                 return -ENOENT;
4059
4060         LINVRNT(osd_invariant(obj));
4061         LASSERT(!dt_object_remote(dt));
4062         LASSERT(osd_is_write_locked(env, obj));
4063         LASSERT(th != NULL);
4064
4065         oh = container_of0(th, struct osd_thandle, ot_super);
4066         LASSERT(oh->ot_handle != NULL);
4067
4068         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4069
4070         spin_lock(&obj->oo_guard);
4071         /*
4072          * That can be result of upgrade from old Lustre version and
4073          * applied only to local files.  Just skip this ref_del call.
4074          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4075          */
4076         if (inode->i_nlink == 0) {
4077                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4078                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4079                              ", maybe an upgraded file? (LU-3915)\n",
4080                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4081                 spin_unlock(&obj->oo_guard);
4082                 return 0;
4083         }
4084
4085         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4086                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4087
4088         ldiskfs_dec_count(oh->ot_handle, inode);
4089         spin_unlock(&obj->oo_guard);
4090
4091         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4092         LINVRNT(osd_invariant(obj));
4093
4094         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4095
4096         return 0;
4097 }
4098
4099 /*
4100  * Concurrency: @dt is read locked.
4101  */
4102 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4103                          struct lu_buf *buf, const char *name)
4104 {
4105         struct osd_object      *obj    = osd_dt_obj(dt);
4106         struct inode           *inode  = obj->oo_inode;
4107         struct osd_thread_info *info   = osd_oti_get(env);
4108         struct dentry          *dentry = &info->oti_obj_dentry;
4109         bool cache_xattr = false;
4110         int rc;
4111
4112         LASSERT(buf);
4113
4114         /* version get is not real XATTR but uses xattr API */
4115         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4116                 dt_obj_version_t *ver = buf->lb_buf;
4117
4118                 /*
4119                  * for version we are just using xattr API but change inode
4120                  * field instead
4121                  */
4122                 if (buf->lb_len == 0)
4123                         return sizeof(dt_obj_version_t);
4124
4125                 if (buf->lb_len < sizeof(dt_obj_version_t))
4126                         return -ERANGE;
4127
4128                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4129                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4130
4131                 *ver = LDISKFS_I(inode)->i_fs_version;
4132
4133                 return sizeof(dt_obj_version_t);
4134         }
4135
4136         if (!dt_object_exists(dt))
4137                 return -ENOENT;
4138
4139         LASSERT(!dt_object_remote(dt));
4140         LASSERT(inode->i_op != NULL);
4141 #ifdef HAVE_IOP_XATTR
4142         LASSERT(inode->i_op->getxattr != NULL);
4143 #endif
4144
4145         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4146             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4147                 cache_xattr = true;
4148
4149         if (cache_xattr) {
4150                 rc = osd_oxc_get(obj, name, buf);
4151                 if (rc != -ENOENT)
4152                         return rc;
4153         }
4154
4155         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4156                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4157                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4158                 struct filter_fid *ff;
4159                 struct ost_layout *ol;
4160
4161                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4162                 if (rc)
4163                         return rc;
4164
4165                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4166
4167                 rc = sizeof(*ff);
4168                 if (buf->lb_len == 0 || !buf->lb_buf)
4169                         return rc;
4170
4171                 if (buf->lb_len < rc)
4172                         return -ERANGE;
4173
4174                 ff = buf->lb_buf;
4175                 ol = &ff->ff_layout;
4176                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4177                                                   PFID_STRIPE_IDX_BITS);
4178                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4179                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4180                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4181                 if (lma->lma_compat & LMAC_COMP_INFO) {
4182                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4183                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4184                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4185                 } else {
4186                         ol->ol_comp_start = 0;
4187                         ol->ol_comp_end = 0;
4188                         ol->ol_comp_id = 0;
4189                 }
4190         } else {
4191                 rc = __osd_xattr_get(inode, dentry, name,
4192                                      buf->lb_buf, buf->lb_len);
4193         }
4194
4195         if (cache_xattr) {
4196                 if (rc == -ENOENT || rc == -ENODATA)
4197                         osd_oxc_add(obj, name, NULL, 0);
4198                 else if (rc > 0 && buf->lb_buf != NULL)
4199                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4200         }
4201
4202         return rc;
4203 }
4204
4205 static int osd_declare_xattr_set(const struct lu_env *env,
4206                                  struct dt_object *dt,
4207                                  const struct lu_buf *buf, const char *name,
4208                                  int fl, struct thandle *handle)
4209 {
4210         struct osd_thandle *oh;
4211         int credits = 0;
4212         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4213
4214         LASSERT(handle != NULL);
4215
4216         oh = container_of0(handle, struct osd_thandle, ot_super);
4217         LASSERT(oh->ot_handle == NULL);
4218
4219         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4220                 /*
4221                  * For non-upgrading case, the LMA is set first and
4222                  * usually fit inode. But for upgrade case, the LMA
4223                  * may be in another separated EA block.
4224                  */
4225                 if (dt_object_exists(dt)) {
4226                         if (fl == LU_XATTR_REPLACE)
4227                                 credits = 1;
4228                         else
4229                                 goto upgrade;
4230                 }
4231         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4232                 credits = 1;
4233         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4234                 /* We may need to delete the old PFID EA. */
4235                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4236                 if (fl == LU_XATTR_REPLACE)
4237                         credits += 1;
4238                 else
4239                         goto upgrade;
4240         } else {
4241                 /*
4242                  * If some name entry resides on remote MDT, then will create
4243                  * agent entry under remote parent. On the other hand, if the
4244                  * remote entry will be removed, then related agent entry may
4245                  * need to be removed from the remote parent. So there may be
4246                  * kinds of cases, let's declare enough credits. The credits
4247                  * for create agent entry is enough for remove case.
4248                  */
4249                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4250                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4251                         if (dt_object_exists(dt))
4252                                 credits += 1; /* For updating LMA */
4253                 }
4254
4255 upgrade:
4256                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4257
4258                 if (buf != NULL) {
4259                         ssize_t buflen;
4260
4261                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4262                                 /*
4263                                  * learn xattr size from osd_xattr_get if
4264                                  * attribute has not been read yet
4265                                  */
4266                                 buflen = __osd_xattr_get(
4267                                     osd_dt_obj(dt)->oo_inode,
4268                                     &osd_oti_get(env)->oti_obj_dentry,
4269                                     name, NULL, 0);
4270                                 if (buflen < 0)
4271                                         buflen = 0;
4272                         } else {
4273                                 buflen = buf->lb_len;
4274                         }
4275
4276                         if (buflen > sb->s_blocksize) {
4277                                 credits += osd_calc_bkmap_credits(
4278                                     sb, NULL, 0, -1,
4279                                     (buflen + sb->s_blocksize - 1) >>
4280                                     sb->s_blocksize_bits);
4281                         }
4282                 }
4283                 /*
4284                  * xattr set may involve inode quota change, reserve credits for
4285                  * dquot_initialize()
4286                  */
4287                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4288         }
4289
4290         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4291
4292         return 0;
4293 }
4294
4295 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4296                               const struct lu_buf *buf, int fl,
4297                               struct thandle *handle)
4298 {
4299         struct osd_thread_info *info = osd_oti_get(env);
4300         struct dentry *dentry = &info->oti_obj_dentry;
4301         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4302         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4303         struct inode *inode = obj->oo_inode;
4304         struct filter_fid *ff = buf->lb_buf;
4305         struct ost_layout *ol = &ff->ff_layout;
4306         int flags = XATTR_REPLACE;
4307         int rc;
4308
4309         ENTRY;
4310
4311         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4312                 RETURN(-EINVAL);
4313
4314         rc = osd_get_lma(info, inode, dentry, loa);
4315         if (rc == -ENODATA) {
4316                 /* Usually for upgarding from old device */
4317                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4318                                 LMAC_FID_ON_OST, 0);
4319                 flags = XATTR_CREATE;
4320         } else if (rc) {
4321                 RETURN(rc);
4322         }
4323
4324         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4325                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4326                         RETURN(-EEXIST);
4327
4328                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4329                         /* Separate PFID EA from LMA */
4330                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4331                         lustre_lma_swab(lma);
4332                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4333                                              sizeof(*lma), XATTR_REPLACE);
4334                         if (!rc) {
4335                                 obj->oo_pfid_in_lma = 0;
4336                                 rc = LU_XATTR_CREATE;
4337                         }
4338
4339                         RETURN(rc);
4340                 }
4341         } else {
4342                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4343                         RETURN(fl);
4344
4345                 /*
4346                  * Old client does not send stripe information,
4347                  * then store the PFID EA on disk separatedly.
4348                  */
4349                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4350                              ol->ol_stripe_size == 0))
4351                         RETURN(fl);
4352
4353                 /* Remove old PFID EA entry firstly. */
4354                 ll_vfs_dq_init(inode);
4355                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4356                 if (rc == -ENODATA) {
4357                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4358                                 RETURN(rc);
4359                 } else if (rc) {
4360                         RETURN(rc);
4361                 }
4362         }
4363
4364         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4365         if (likely(ol->ol_stripe_size != 0)) {
4366                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4367                                              PFID_STRIPE_IDX_BITS;
4368                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4369                 lma->lma_compat |= LMAC_STRIPE_INFO;
4370                 if (ol->ol_comp_id != 0) {
4371                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4372                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4373                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4374                         lma->lma_compat |= LMAC_COMP_INFO;
4375                 }
4376         }
4377
4378         lustre_loa_swab(loa, false);
4379
4380         /* Store the PFID EA inside LMA. */
4381         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4382                              flags);
4383         if (!rc)
4384                 obj->oo_pfid_in_lma = 1;
4385
4386         RETURN(rc);
4387 }
4388
4389 /*
4390  * In DNE environment, the object (in spite of regular file or directory)
4391  * and its name entry may reside on different MDTs. Under such case, we will
4392  * create an agent entry on the MDT where the object resides. The agent entry
4393  * references the object locally, that makes the object to be visible to the
4394  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4395  * such as 'tar' can handle the object properly.
4396  *
4397  * We handle the agent entry during set linkEA that is the common interface
4398  * for both regular file and directroy, can handle kinds of cases, such as
4399  * create/link/unlink/rename, and so on.
4400  *
4401  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4402  *
4403  * XXX: There are two known issues:
4404  * 1. For one object, we will create at most one agent entry even if there
4405  *    may be more than one cross-MDTs hard links on the object. So the local
4406  *    e2fsck may claim that the object's nlink is larger than the name entries
4407  *    that reference such inode. And in further, the e2fsck will fix the nlink
4408  *    attribute to match the local references. Then it will cause the object's
4409  *    nlink attribute to be inconsistent with the global references. it is bad
4410  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4411  *    other hand, the global namespace LFSCK can repair the object's attribute
4412  *    according to the linkEA.
4413  * 2. There may be too many hard links on the object as to its linkEA overflow,
4414  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4415  *    case happened, then at this point, we do not know whether there are some
4416  *    cross-MDTs reference. But there are local references, it guarantees that
4417  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4418  */
4419 static int osd_xattr_handle_linkea(const struct lu_env *env,
4420                                    struct osd_device *osd,
4421                                    struct osd_object *obj,
4422                                    const struct lu_buf *buf,
4423                                    struct thandle *handle)
4424 {
4425         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4426         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4427         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4428         struct lu_name tmpname;
4429         struct osd_thandle *oh;
4430         int rc;
4431         bool remote = false;
4432
4433         ENTRY;
4434
4435         oh = container_of0(handle, struct osd_thandle, ot_super);
4436         LASSERT(oh->ot_handle != NULL);
4437
4438         rc = linkea_init_with_rec(&ldata);
4439         if (!rc) {
4440                 linkea_first_entry(&ldata);
4441                 while (ldata.ld_lee != NULL && !remote) {
4442                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4443                                             &tmpname, tfid);
4444                         if (osd_remote_fid(env, osd, tfid) > 0)
4445                                 remote = true;
4446                         else
4447                                 linkea_next_entry(&ldata);
4448                 }
4449         } else if (rc == -ENODATA) {
4450                 rc = 0;
4451         } else {
4452                 RETURN(rc);
4453         }
4454
4455         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4456                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4457                 if (rc)
4458                         CERROR("%s: failed to remove agent entry for "DFID
4459                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4460         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4461                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4462                 if (rc)
4463                         CERROR("%s: failed to create agent entry for "DFID
4464                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4465         }
4466
4467         RETURN(rc);
4468 }
4469
4470 /*
4471  * Concurrency: @dt is write locked.
4472  */
4473 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4474                          const struct lu_buf *buf, const char *name, int fl,
4475                          struct thandle *handle)
4476 {
4477         struct osd_object *obj = osd_dt_obj(dt);
4478         struct osd_device *osd = osd_obj2dev(obj);
4479         struct inode *inode = obj->oo_inode;
4480         struct osd_thread_info *info = osd_oti_get(env);
4481         int fs_flags = 0;
4482         int len;
4483         int rc;
4484
4485         ENTRY;
4486
4487         LASSERT(handle);
4488         LASSERT(buf);
4489
4490         /* version set is not real XATTR */
4491         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4492                 dt_obj_version_t *version = buf->lb_buf;
4493
4494                 /*
4495                  * for version we are just using xattr API but change inode
4496                  * field instead
4497                  */
4498                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4499
4500                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4501                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4502
4503                 LDISKFS_I(inode)->i_fs_version = *version;
4504                 /*
4505                  * Version is set after all inode operations are finished,
4506                  * so we should mark it dirty here
4507                  */
4508                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4509
4510                 RETURN(0);
4511         }
4512
4513         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4514                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4515
4516         len = buf->lb_len;
4517         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4518
4519         /*
4520          * For the OST device with 256 bytes inode size by default,
4521          * the PFID EA will be stored together with LMA EA to avoid
4522          * performance trouble. Otherwise the PFID EA can be stored
4523          * independently. LU-8998
4524          */
4525         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4526             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4527                 LASSERT(buf->lb_buf);
4528
4529                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4530                 if (fl <= 0)
4531                         RETURN(fl);
4532         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4533                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4534                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4535
4536                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4537                 if (rc)
4538                         RETURN(rc);
4539
4540                 lma->lma_incompat |= LMAI_STRIPED;
4541                 lustre_lma_swab(lma);
4542                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4543                                      sizeof(*lma), XATTR_REPLACE);
4544                 if (rc != 0)
4545                         RETURN(rc);
4546         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4547                 LASSERT(!osd->od_is_ost);
4548
4549                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4550                 if (rc)
4551                         RETURN(rc);
4552         }
4553
4554         if (fl & LU_XATTR_REPLACE)
4555                 fs_flags |= XATTR_REPLACE;
4556
4557         if (fl & LU_XATTR_CREATE)
4558                 fs_flags |= XATTR_CREATE;
4559
4560         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4561         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4562
4563         if (rc == 0 &&
4564             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4565              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4566                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4567
4568         return rc;
4569 }
4570
4571 /*
4572  * Concurrency: @dt is read locked.
4573  */
4574 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4575                           const struct lu_buf *buf)
4576 {
4577         struct osd_object *obj = osd_dt_obj(dt);
4578         struct inode *inode = obj->oo_inode;
4579         struct osd_thread_info *info = osd_oti_get(env);
4580         struct dentry *dentry = &info->oti_obj_dentry;
4581
4582         if (!dt_object_exists(dt))
4583                 return -ENOENT;
4584
4585         LASSERT(!dt_object_remote(dt));
4586         LASSERT(inode->i_op != NULL);
4587         LASSERT(inode->i_op->listxattr != NULL);
4588
4589         dentry->d_inode = inode;
4590         dentry->d_sb = inode->i_sb;
4591         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4592 }
4593
4594 static int osd_declare_xattr_del(const struct lu_env *env,
4595                                  struct dt_object *dt, const char *name,
4596                                  struct thandle *handle)
4597 {
4598         struct osd_thandle *oh;
4599         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4600
4601         LASSERT(!dt_object_remote(dt));
4602         LASSERT(handle != NULL);
4603
4604         oh = container_of0(handle, struct osd_thandle, ot_super);
4605         LASSERT(oh->ot_handle == NULL);
4606
4607         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4608                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4609         /*
4610          * xattr del may involve inode quota change, reserve credits for
4611          * dquot_initialize()
4612          */
4613         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4614
4615         return 0;
4616 }
4617
4618 /*
4619  * Concurrency: @dt is write locked.
4620  */
4621 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4622                          const char *name, struct thandle *handle)
4623 {
4624         struct osd_object *obj = osd_dt_obj(dt);
4625         struct inode *inode = obj->oo_inode;
4626         struct osd_thread_info *info = osd_oti_get(env);
4627         struct dentry *dentry = &info->oti_obj_dentry;
4628         int rc;
4629
4630         if (!dt_object_exists(dt))
4631                 return -ENOENT;
4632
4633         LASSERT(!dt_object_remote(dt));
4634         LASSERT(inode->i_op != NULL);
4635         LASSERT(handle != NULL);
4636 #ifdef HAVE_IOP_XATTR
4637         LASSERT(inode->i_op->removexattr != NULL);
4638 #endif
4639
4640         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4641
4642         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4643                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4644
4645                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4646                                  &info->oti_ost_attrs);
4647                 if (!rc) {
4648                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4649
4650                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4651                         lustre_lma_swab(lma);
4652                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4653                                              sizeof(*lma), XATTR_REPLACE);
4654                         if (!rc)
4655                                 obj->oo_pfid_in_lma = 0;
4656                 }
4657         } else {
4658                 ll_vfs_dq_init(inode);
4659                 dentry->d_inode = inode;
4660                 dentry->d_sb = inode->i_sb;
4661                 rc = osd_removexattr(dentry, inode, name);
4662         }
4663
4664         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4665
4666         if (rc == 0 &&
4667             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4668              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4669                 osd_oxc_del(obj, name);
4670
4671         return rc;
4672 }
4673
4674 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4675                            __u64 start, __u64 end)
4676 {
4677         struct osd_object *obj = osd_dt_obj(dt);
4678         struct inode *inode = obj->oo_inode;
4679         struct osd_thread_info *info = osd_oti_get(env);
4680         struct dentry *dentry = &info->oti_obj_dentry;
4681         struct file *file = &info->oti_file;
4682         int rc;
4683
4684         ENTRY;
4685
4686         dentry->d_inode = inode;
4687         dentry->d_sb = inode->i_sb;
4688         file->f_path.dentry = dentry;
4689         file->f_mapping = inode->i_mapping;
4690         file->f_op = inode->i_fop;
4691         set_file_inode(file, inode);
4692
4693         rc = ll_vfs_fsync_range(file, start, end, 0);
4694
4695         RETURN(rc);
4696 }
4697
4698 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4699 {
4700         return 0;
4701 }
4702
4703 /*
4704  * Index operations.
4705  */
4706
4707 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4708                                const struct dt_index_features *feat)
4709 {
4710         struct iam_descr *descr;
4711
4712         if (osd_object_is_root(o))
4713                 return feat == &dt_directory_features;
4714
4715         LASSERT(o->oo_dir != NULL);
4716
4717         descr = o->oo_dir->od_container.ic_descr;
4718         if (feat == &dt_directory_features) {
4719                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4720                         return 1;
4721                 else
4722                         return 0;
4723         } else {
4724                 return feat->dif_keysize_min <= descr->id_key_size &&
4725                        descr->id_key_size <= feat->dif_keysize_max &&
4726                        feat->dif_recsize_min <= descr->id_rec_size &&
4727                        descr->id_rec_size <= feat->dif_recsize_max &&
4728                        !(feat->dif_flags & (DT_IND_VARKEY |
4729                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4730                        ergo(feat->dif_flags & DT_IND_UPDATE,
4731                             1 /* XXX check that object (and fs) is writable */);
4732         }
4733 }
4734
4735 static int osd_iam_container_init(const struct lu_env *env,
4736                                   struct osd_object *obj,
4737                                   struct osd_directory *dir)
4738 {
4739         struct iam_container *bag = &dir->od_container;
4740         int result;
4741
4742         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4743         if (result != 0)
4744                 return result;
4745
4746         result = iam_container_setup(bag);
4747         if (result == 0)
4748                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4749         else
4750                 iam_container_fini(bag);
4751
4752         return result;
4753 }
4754
4755
4756 /*
4757  * Concurrency: no external locking is necessary.
4758  */
4759 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4760                          const struct dt_index_features *feat)
4761 {
4762         int result;
4763         int skip_iam = 0;
4764         struct osd_object *obj = osd_dt_obj(dt);
4765
4766         LINVRNT(osd_invariant(obj));
4767
4768         if (osd_object_is_root(obj)) {
4769                 dt->do_index_ops = &osd_index_ea_ops;
4770                 result = 0;
4771         } else if (feat == &dt_directory_features) {
4772                 dt->do_index_ops = &osd_index_ea_ops;
4773                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4774                         result = 0;
4775                 else
4776                         result = -ENOTDIR;
4777                 skip_iam = 1;
4778         } else if (unlikely(feat == &dt_otable_features)) {
4779                 dt->do_index_ops = &osd_otable_ops;
4780                 return 0;
4781         } else if (unlikely(feat == &dt_acct_features)) {
4782                 dt->do_index_ops = &osd_acct_index_ops;
4783                 result = 0;
4784                 skip_iam = 1;
4785         } else if (!osd_has_index(obj)) {
4786                 struct osd_directory *dir;
4787                 struct osd_device *osd = osd_obj2dev(obj);
4788                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4789
4790                 OBD_ALLOC_PTR(dir);
4791                 if (dir) {
4792
4793                         spin_lock(&obj->oo_guard);
4794                         if (obj->oo_dir == NULL)
4795                                 obj->oo_dir = dir;
4796                         else
4797                                 /*
4798                                  * Concurrent thread allocated container data.
4799                                  */
4800                                 OBD_FREE_PTR(dir);
4801                         spin_unlock(&obj->oo_guard);
4802                         /*
4803                          * Now, that we have container data, serialize its
4804                          * initialization.
4805                          */
4806                         down_write(&obj->oo_ext_idx_sem);
4807                         /*
4808                          * recheck under lock.
4809                          */
4810
4811                         if (osd_has_index(obj)) {
4812                                 result = 0;
4813                                 goto unlock;
4814                         }
4815
4816                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4817                         if (result || feat == &dt_lfsck_namespace_features ||
4818                             feat == &dt_lfsck_layout_orphan_features ||
4819                             feat == &dt_lfsck_layout_dangling_features)
4820                                 goto unlock;
4821
4822                         result = osd_index_register(osd, fid,
4823                                                     feat->dif_keysize_max,
4824                                                     feat->dif_recsize_max);
4825                         if (result < 0)
4826                                 CWARN("%s: failed to register index "
4827                                       DFID": rc = %d\n",
4828                                       osd_name(osd), PFID(fid), result);
4829                         else if (result > 0)
4830                                 result = 0;
4831                         else
4832                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4833                                        " (%d/%d) registered\n",
4834                                        osd_name(osd), PFID(fid),
4835                                        (int)feat->dif_keysize_max,
4836                                        (int)feat->dif_recsize_max);
4837
4838 unlock:
4839                         up_write(&obj->oo_ext_idx_sem);
4840                 } else {
4841                         result = -ENOMEM;
4842                 }
4843         } else {
4844                 result = 0;
4845         }
4846
4847         if (result == 0 && skip_iam == 0) {
4848                 if (!osd_iam_index_probe(env, obj, feat))
4849                         result = -ENOTDIR;
4850         }
4851         LINVRNT(osd_invariant(obj));
4852
4853         return result;
4854 }
4855
4856 static int osd_otable_it_attr_get(const struct lu_env *env,
4857                                  struct dt_object *dt,
4858                                  struct lu_attr *attr)
4859 {
4860         attr->la_valid = 0;
4861         return 0;
4862 }
4863
4864 static const struct dt_object_operations osd_obj_ops = {
4865         .do_read_lock           = osd_read_lock,
4866         .do_write_lock          = osd_write_lock,
4867         .do_read_unlock         = osd_read_unlock,
4868         .do_write_unlock        = osd_write_unlock,
4869         .do_write_locked        = osd_write_locked,
4870         .do_attr_get            = osd_attr_get,
4871         .do_declare_attr_set    = osd_declare_attr_set,
4872         .do_attr_set            = osd_attr_set,
4873         .do_ah_init             = osd_ah_init,
4874         .do_declare_create      = osd_declare_create,
4875         .do_create              = osd_create,
4876         .do_declare_destroy     = osd_declare_destroy,
4877         .do_destroy             = osd_destroy,
4878         .do_index_try           = osd_index_try,
4879         .do_declare_ref_add     = osd_declare_ref_add,
4880         .do_ref_add             = osd_ref_add,
4881         .do_declare_ref_del     = osd_declare_ref_del,
4882         .do_ref_del             = osd_ref_del,
4883         .do_xattr_get           = osd_xattr_get,
4884         .do_declare_xattr_set   = osd_declare_xattr_set,
4885         .do_xattr_set           = osd_xattr_set,
4886         .do_declare_xattr_del   = osd_declare_xattr_del,
4887         .do_xattr_del           = osd_xattr_del,
4888         .do_xattr_list          = osd_xattr_list,
4889         .do_object_sync         = osd_object_sync,
4890         .do_invalidate          = osd_invalidate,
4891 };
4892
4893 static const struct dt_object_operations osd_obj_otable_it_ops = {
4894         .do_attr_get    = osd_otable_it_attr_get,
4895         .do_index_try   = osd_index_try,
4896 };
4897
4898 static int osd_index_declare_iam_delete(const struct lu_env *env,
4899                                         struct dt_object *dt,
4900                                         const struct dt_key *key,
4901                                         struct thandle *handle)
4902 {
4903         struct osd_thandle *oh;
4904
4905         oh = container_of0(handle, struct osd_thandle, ot_super);
4906         LASSERT(oh->ot_handle == NULL);
4907
4908         /* Recycle  may cause additional three blocks to be changed. */
4909         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4910                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4911
4912         return 0;
4913 }
4914
4915 /**
4916  *      delete a (key, value) pair from index \a dt specified by \a key
4917  *
4918  *      \param  dt      osd index object
4919  *      \param  key     key for index
4920  *      \param  rec     record reference
4921  *      \param  handle  transaction handler
4922  *
4923  *      \retval  0  success
4924  *      \retval -ve   failure
4925  */
4926 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4927                                 const struct dt_key *key,
4928                                 struct thandle *handle)
4929 {
4930         struct osd_thread_info *oti = osd_oti_get(env);
4931         struct osd_object *obj = osd_dt_obj(dt);
4932         struct osd_thandle *oh;
4933         struct iam_path_descr *ipd;
4934         struct iam_container *bag = &obj->oo_dir->od_container;
4935         int rc;
4936
4937         ENTRY;
4938
4939         if (!dt_object_exists(dt))
4940                 RETURN(-ENOENT);
4941
4942         LINVRNT(osd_invariant(obj));
4943         LASSERT(!dt_object_remote(dt));
4944         LASSERT(bag->ic_object == obj->oo_inode);
4945         LASSERT(handle != NULL);
4946
4947         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4948
4949         ipd = osd_idx_ipd_get(env, bag);
4950         if (unlikely(ipd == NULL))
4951                 RETURN(-ENOMEM);
4952
4953         oh = container_of0(handle, struct osd_thandle, ot_super);
4954         LASSERT(oh->ot_handle != NULL);
4955         LASSERT(oh->ot_handle->h_transaction != NULL);
4956
4957         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4958                 /* swab quota uid/gid provided by caller */
4959                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4960                 key = (const struct dt_key *)&oti->oti_quota_id;
4961         }
4962
4963         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4964         osd_ipd_put(env, bag, ipd);
4965         LINVRNT(osd_invariant(obj));
4966         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4967         RETURN(rc);
4968 }
4969
4970 static int osd_index_declare_ea_delete(const struct lu_env *env,
4971                                        struct dt_object *dt,
4972                                        const struct dt_key *key,
4973                                        struct thandle *handle)
4974 {
4975         struct osd_thandle *oh;
4976         struct inode *inode;
4977         int rc, credits;
4978
4979         ENTRY;
4980
4981         LASSERT(!dt_object_remote(dt));
4982         LASSERT(handle != NULL);
4983
4984         oh = container_of0(handle, struct osd_thandle, ot_super);
4985         LASSERT(oh->ot_handle == NULL);
4986
4987         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4988         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4989
4990         inode = osd_dt_obj(dt)->oo_inode;
4991         if (inode == NULL)
4992                 RETURN(-ENOENT);
4993
4994         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4995                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4996                                    NULL, OSD_QID_BLK);
4997         RETURN(rc);
4998 }
4999
5000 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5001                                           struct dt_rec *fid)
5002 {
5003         struct osd_fid_pack *rec;
5004         int rc = -ENODATA;
5005
5006         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5007                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5008                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5009                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5010                         rc = -EINVAL;
5011         }
5012         return rc;
5013 }
5014
5015 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5016                           const struct lu_fid *fid)
5017 {
5018         struct seq_server_site *ss = osd_seq_site(osd);
5019
5020         ENTRY;
5021
5022         /* FID seqs not in FLDB, must be local seq */
5023         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5024                 RETURN(0);
5025
5026         /*
5027          * If FLD is not being initialized yet, it only happens during the
5028          * initialization, likely during mgs initialization, and we assume
5029          * this is local FID.
5030          */
5031         if (ss == NULL || ss->ss_server_fld == NULL)
5032                 RETURN(0);
5033
5034         /* Only check the local FLDB here */
5035         if (osd_seq_exists(env, osd, fid_seq(fid)))
5036                 RETURN(0);
5037
5038         RETURN(1);
5039 }
5040
5041 static void osd_take_care_of_agent(const struct lu_env *env,
5042                                    struct osd_device *osd,
5043                                    struct osd_thandle *oh,
5044                                    struct ldiskfs_dir_entry_2 *de)
5045 {
5046         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5047         struct osd_idmap_cache *idc;
5048         int rc, schedule = 0;
5049
5050         LASSERT(de != NULL);
5051
5052         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5053         if (likely(rc == 0)) {
5054                 idc = osd_idc_find_or_init(env, osd, fid);
5055                 if (IS_ERR(idc) || idc->oic_remote)
5056                         schedule = 1;
5057         } else if (rc == -ENODATA) {
5058                 /*
5059                  * can't get FID, postpone to the end of the
5060                  * transaction when iget() is safe
5061                  */
5062                 schedule = 1;
5063         } else {
5064                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5065         }
5066         if (schedule)
5067                 osd_schedule_agent_inode_removal(env, oh,
5068                                                  le32_to_cpu(de->inode));
5069 }
5070
5071 /**
5072  * Index delete function for interoperability mode (b11826).
5073  * It will remove the directory entry added by osd_index_ea_insert().
5074  * This entry is needed to maintain name->fid mapping.
5075  *
5076  * \param key,  key i.e. file entry to be deleted
5077  *
5078  * \retval   0, on success
5079  * \retval -ve, on error
5080  */
5081 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5082                                const struct dt_key *key, struct thandle *handle)
5083 {
5084         struct osd_object *obj = osd_dt_obj(dt);
5085         struct inode *dir = obj->oo_inode;
5086         struct dentry *dentry;
5087         struct osd_thandle *oh;
5088         struct ldiskfs_dir_entry_2 *de = NULL;
5089         struct buffer_head *bh;
5090         struct htree_lock *hlock = NULL;
5091         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5092         int rc;
5093
5094         ENTRY;
5095
5096         if (!dt_object_exists(dt))
5097                 RETURN(-ENOENT);
5098
5099         LINVRNT(osd_invariant(obj));
5100         LASSERT(!dt_object_remote(dt));
5101         LASSERT(handle != NULL);
5102
5103         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5104
5105         oh = container_of(handle, struct osd_thandle, ot_super);
5106         LASSERT(oh->ot_handle != NULL);
5107         LASSERT(oh->ot_handle->h_transaction != NULL);
5108
5109         ll_vfs_dq_init(dir);
5110         dentry = osd_child_dentry_get(env, obj,
5111                                       (char *)key, strlen((char *)key));
5112
5113         if (obj->oo_hl_head != NULL) {
5114                 hlock = osd_oti_get(env)->oti_hlock;
5115                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5116                                    dir, LDISKFS_HLOCK_DEL);
5117         } else {
5118                 down_write(&obj->oo_ext_idx_sem);
5119         }
5120
5121         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5122         if (!IS_ERR(bh)) {
5123                 /*
5124                  * If this is not the ".." entry, it might be a remote DNE
5125                  * entry and  we need to check if the FID is for a remote
5126                  * MDT.  If the FID is  not in the directory entry (e.g.
5127                  * upgraded 1.8 filesystem without dirdata enabled) then
5128                  * we need to get the FID from the LMA. For a remote directory
5129                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5130                  * case.
5131                  *
5132                  * Delete the entry before the agent inode in order to
5133                  * simplify error handling.  At worst an error after deleting
5134                  * the entry first might leak the agent inode afterward. The
5135                  * reverse would need filesystem abort in case of error deleting
5136                  * the entry after the agent had been removed, or leave a
5137                  * dangling entry pointing at a random inode.
5138                  */
5139                 if (strcmp((char *)key, dotdot) != 0)
5140                         osd_take_care_of_agent(env, osd, oh, de);
5141                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5142                 brelse(bh);
5143         } else {
5144                 rc = PTR_ERR(bh);
5145         }
5146         if (hlock != NULL)
5147                 ldiskfs_htree_unlock(hlock);
5148         else
5149                 up_write(&obj->oo_ext_idx_sem);
5150
5151         GOTO(out, rc);
5152 out:
5153         LASSERT(osd_invariant(obj));
5154         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5155         RETURN(rc);
5156 }
5157
5158 /**
5159  *      Lookup index for \a key and copy record to \a rec.
5160  *
5161  *      \param  dt      osd index object
5162  *      \param  key     key for index
5163  *      \param  rec     record reference
5164  *
5165  *      \retval  +ve  success : exact mach
5166  *      \retval  0    return record with key not greater than \a key
5167  *      \retval -ve   failure
5168  */
5169 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5170                                 struct dt_rec *rec, const struct dt_key *key)
5171 {
5172         struct osd_object *obj = osd_dt_obj(dt);
5173         struct iam_path_descr *ipd;
5174         struct iam_container *bag = &obj->oo_dir->od_container;
5175         struct osd_thread_info *oti = osd_oti_get(env);
5176         struct iam_iterator *it = &oti->oti_idx_it;
5177         struct iam_rec *iam_rec;
5178         int rc;
5179
5180         ENTRY;
5181
5182         if (!dt_object_exists(dt))
5183                 RETURN(-ENOENT);
5184
5185         LASSERT(osd_invariant(obj));
5186         LASSERT(!dt_object_remote(dt));
5187         LASSERT(bag->ic_object == obj->oo_inode);
5188
5189         ipd = osd_idx_ipd_get(env, bag);
5190         if (IS_ERR(ipd))
5191                 RETURN(-ENOMEM);
5192
5193         /* got ipd now we can start iterator. */
5194         iam_it_init(it, bag, 0, ipd);
5195
5196         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5197                 /* swab quota uid/gid provided by caller */
5198                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5199                 key = (const struct dt_key *)&oti->oti_quota_id;
5200         }
5201
5202         rc = iam_it_get(it, (struct iam_key *)key);
5203         if (rc >= 0) {
5204                 if (S_ISDIR(obj->oo_inode->i_mode))
5205                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5206                 else
5207                         iam_rec = (struct iam_rec *)rec;
5208
5209                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5210
5211                 if (S_ISDIR(obj->oo_inode->i_mode))
5212                         osd_fid_unpack((struct lu_fid *)rec,
5213                                        (struct osd_fid_pack *)iam_rec);
5214                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5215                         osd_quota_unpack(obj, rec);
5216         }
5217
5218         iam_it_put(it);
5219         iam_it_fini(it);
5220         osd_ipd_put(env, bag, ipd);
5221
5222         LINVRNT(osd_invariant(obj));
5223
5224         RETURN(rc);
5225 }
5226
5227 static int osd_index_declare_iam_insert(const struct lu_env *env,
5228                                         struct dt_object *dt,
5229                                         const struct dt_rec *rec,
5230                                         const struct dt_key *key,
5231                                         struct thandle *handle)
5232 {
5233         struct osd_thandle *oh;
5234
5235         LASSERT(handle != NULL);
5236
5237         oh = container_of0(handle, struct osd_thandle, ot_super);
5238         LASSERT(oh->ot_handle == NULL);
5239
5240         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5241                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5242
5243         return 0;
5244 }
5245
5246 /**
5247  *      Inserts (key, value) pair in \a dt index object.
5248  *
5249  *      \param  dt      osd index object
5250  *      \param  key     key for index
5251  *      \param  rec     record reference
5252  *      \param  th      transaction handler
5253  *
5254  *      \retval  0  success
5255  *      \retval -ve failure
5256  */
5257 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5258                                 const struct dt_rec *rec,
5259                                 const struct dt_key *key, struct thandle *th,
5260                                 int ignore_quota)
5261 {
5262         struct osd_object *obj = osd_dt_obj(dt);
5263         struct iam_path_descr *ipd;
5264         struct osd_thandle *oh;
5265         struct iam_container *bag;
5266         struct osd_thread_info *oti = osd_oti_get(env);
5267         struct iam_rec *iam_rec;
5268         int rc;
5269
5270         ENTRY;
5271
5272         if (!dt_object_exists(dt))
5273                 RETURN(-ENOENT);
5274
5275         LINVRNT(osd_invariant(obj));
5276         LASSERT(!dt_object_remote(dt));
5277
5278         bag = &obj->oo_dir->od_container;
5279         LASSERT(bag->ic_object == obj->oo_inode);
5280         LASSERT(th != NULL);
5281
5282         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5283
5284         ipd = osd_idx_ipd_get(env, bag);
5285         if (unlikely(ipd == NULL))
5286                 RETURN(-ENOMEM);
5287
5288         oh = container_of0(th, struct osd_thandle, ot_super);
5289         LASSERT(oh->ot_handle != NULL);
5290         LASSERT(oh->ot_handle->h_transaction != NULL);
5291         if (S_ISDIR(obj->oo_inode->i_mode)) {
5292                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5293                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5294                              &oti->oti_fid);
5295         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5296                 /* pack quota uid/gid */
5297                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5298                 key = (const struct dt_key *)&oti->oti_quota_id;
5299                 /* pack quota record */
5300                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5301                 iam_rec = (struct iam_rec *)rec;
5302         } else {
5303                 iam_rec = (struct iam_rec *)rec;
5304         }
5305
5306         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5307                         iam_rec, ipd);
5308         osd_ipd_put(env, bag, ipd);
5309         LINVRNT(osd_invariant(obj));
5310         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5311         RETURN(rc);
5312 }
5313
5314 /**
5315  * Calls ldiskfs_add_entry() to add directory entry
5316  * into the directory. This is required for
5317  * interoperability mode (b11826)
5318  *
5319  * \retval   0, on success
5320  * \retval -ve, on error
5321  */
5322 static int __osd_ea_add_rec(struct osd_thread_info *info,
5323                             struct osd_object *pobj, struct inode  *cinode,
5324                             const char *name, const struct lu_fid *fid,
5325                             struct htree_lock *hlock, struct thandle *th)
5326 {
5327         struct ldiskfs_dentry_param *ldp;
5328         struct dentry *child;
5329         struct osd_thandle *oth;
5330         int rc;
5331
5332         oth = container_of(th, struct osd_thandle, ot_super);
5333         LASSERT(oth->ot_handle != NULL);
5334         LASSERT(oth->ot_handle->h_transaction != NULL);
5335         LASSERT(pobj->oo_inode);
5336
5337         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5338         if (unlikely(pobj->oo_inode ==
5339                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5340                 ldp->edp_magic = 0;
5341         else
5342                 osd_get_ldiskfs_dirent_param(ldp, fid);
5343         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5344         child->d_fsdata = (void *)ldp;
5345         ll_vfs_dq_init(pobj->oo_inode);
5346         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5347                                    child, cinode, hlock);
5348         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5349                 struct ldiskfs_dir_entry_2      *de;
5350                 struct buffer_head              *bh;
5351                 int                              rc1;
5352
5353                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5354                                             NULL, hlock);
5355                 if (!IS_ERR(bh)) {
5356                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5357                                                                bh);
5358                         if (rc1 == 0) {
5359                                 if (S_ISDIR(cinode->i_mode))
5360                                         de->file_type = LDISKFS_DIRENT_LUFID |
5361                                                         LDISKFS_FT_REG_FILE;
5362                                 else
5363                                         de->file_type = LDISKFS_DIRENT_LUFID |
5364                                                         LDISKFS_FT_DIR;
5365                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5366                                                               NULL, bh);
5367                         }
5368                         brelse(bh);
5369                 }
5370         }
5371
5372         RETURN(rc);
5373 }
5374
5375 /**
5376  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5377  * into the directory.Also sets flags into osd object to
5378  * indicate dot and dotdot are created. This is required for
5379  * interoperability mode (b11826)
5380  *
5381  * \param dir   directory for dot and dotdot fixup.
5382  * \param obj   child object for linking
5383  *
5384  * \retval   0, on success
5385  * \retval -ve, on error
5386  */
5387 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5388                               struct osd_object *dir,
5389                               struct inode *parent_dir, const char *name,
5390                               const struct lu_fid *dot_fid,
5391                               const struct lu_fid *dot_dot_fid,
5392                               struct thandle *th)
5393 {
5394         struct inode *inode = dir->oo_inode;
5395         struct osd_thandle *oth;
5396         int result = 0;
5397
5398         oth = container_of(th, struct osd_thandle, ot_super);
5399         LASSERT(oth->ot_handle->h_transaction != NULL);
5400         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5401
5402         if (strcmp(name, dot) == 0) {
5403                 if (dir->oo_compat_dot_created) {
5404                         result = -EEXIST;
5405                 } else {
5406                         LASSERT(inode->i_ino == parent_dir->i_ino);
5407                         dir->oo_compat_dot_created = 1;
5408                         result = 0;
5409                 }
5410         } else if (strcmp(name, dotdot) == 0) {
5411                 if (!dir->oo_compat_dot_created)
5412                         return -EINVAL;
5413                 /* in case of rename, dotdot is already created */
5414                 if (dir->oo_compat_dotdot_created) {
5415                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5416                                                 dot_dot_fid, NULL, th);
5417                 }
5418
5419                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5420                         struct lu_fid tfid = *dot_dot_fid;
5421
5422                         tfid.f_oid--;
5423                         result = osd_add_dot_dotdot_internal(info,
5424                                         dir->oo_inode, parent_dir, dot_fid,
5425                                         &tfid, oth);
5426                 } else {
5427                         result = osd_add_dot_dotdot_internal(info,
5428                                         dir->oo_inode, parent_dir, dot_fid,
5429                                         dot_dot_fid, oth);
5430                 }
5431
5432                 if (result == 0)
5433                         dir->oo_compat_dotdot_created = 1;
5434         }
5435
5436         return result;
5437 }
5438
5439
5440 /**
5441  * It will call the appropriate osd_add* function and return the
5442  * value, return by respective functions.
5443  */
5444 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5445                           struct inode *cinode, const char *name,
5446                           const struct lu_fid *fid, struct thandle *th)
5447 {
5448         struct osd_thread_info *info = osd_oti_get(env);
5449         struct htree_lock *hlock;
5450         int rc;
5451
5452         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5453
5454         if (name[0] == '.' && (name[1] == '\0' ||
5455                                (name[1] == '.' && name[2] == '\0'))) {
5456                 if (hlock != NULL) {
5457                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5458                                            pobj->oo_inode, 0);
5459                 } else {
5460                         down_write(&pobj->oo_ext_idx_sem);
5461                 }
5462
5463                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5464                                         lu_object_fid(&pobj->oo_dt.do_lu),
5465                                         fid, th);
5466         } else {
5467                 if (hlock != NULL) {
5468                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5469                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5470                 } else {
5471                         down_write(&pobj->oo_ext_idx_sem);
5472                 }
5473
5474                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5475                         struct lu_fid *tfid = &info->oti_fid;
5476
5477                         *tfid = *fid;
5478                         tfid->f_ver = ~0;
5479                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5480                                               tfid, hlock, th);
5481                 } else {
5482                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5483                                               hlock, th);
5484                 }
5485         }
5486         if (hlock != NULL)
5487                 ldiskfs_htree_unlock(hlock);
5488         else
5489                 up_write(&pobj->oo_ext_idx_sem);
5490
5491         return rc;
5492 }
5493
5494 static int
5495 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5496                       struct osd_idmap_cache *oic)
5497 {
5498         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5499         struct lu_fid *fid = &oic->oic_fid;
5500         struct osd_inode_id *id = &oic->oic_lid;
5501         struct inode *inode = NULL;
5502         int once = 0;
5503         bool insert;
5504         int rc;
5505
5506         ENTRY;
5507
5508         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5509                 RETURN(0);
5510
5511         if (thread_is_running(&scrub->os_thread) &&
5512             scrub->os_pos_current > id->oii_ino)
5513                 RETURN(0);
5514
5515         if (dev->od_auto_scrub_interval == AS_NEVER ||
5516             ktime_get_real_seconds() <
5517             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5518                 RETURN(0);
5519
5520 again:
5521         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5522         if (rc == -ENOENT) {
5523                 __u32 gen = id->oii_gen;
5524
5525                 insert = true;
5526                 if (inode != NULL)
5527                         goto trigger;
5528
5529                 inode = osd_iget(oti, dev, id);
5530                 /* The inode has been removed (by race maybe). */
5531                 if (IS_ERR(inode)) {
5532                         rc = PTR_ERR(inode);
5533
5534                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5535                 }
5536
5537                 /* The OI mapping is lost. */
5538                 if (gen != OSD_OII_NOGEN)
5539                         goto trigger;
5540
5541                 /*
5542                  * The inode may has been reused by others, we do not know,
5543                  * leave it to be handled by subsequent osd_fid_lookup().
5544                  */
5545                 GOTO(out, rc = 0);
5546         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5547                 GOTO(out, rc);
5548         }
5549
5550         insert = false;
5551
5552 trigger:
5553         if (thread_is_running(&scrub->os_thread)) {
5554                 if (inode == NULL) {
5555                         inode = osd_iget(oti, dev, id);
5556                         /* The inode has been removed (by race maybe). */
5557                         if (IS_ERR(inode)) {
5558                                 rc = PTR_ERR(inode);
5559
5560                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5561                         }
5562                 }
5563
5564                 rc = osd_oii_insert(dev, oic, insert);
5565                 /*
5566                  * There is race condition between osd_oi_lookup and OI scrub.
5567                  * The OI scrub finished just after osd_oi_lookup() failure.
5568                  * Under such case, it is unnecessary to trigger OI scrub again,
5569                  * but try to call osd_oi_lookup() again.
5570                  */
5571                 if (unlikely(rc == -EAGAIN))
5572                         goto again;
5573
5574                 if (!S_ISDIR(inode->i_mode))
5575                         rc = 0;
5576                 else
5577                         rc = osd_check_lmv(oti, dev, inode, oic);
5578
5579                 GOTO(out, rc);
5580         }
5581
5582         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5583                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5584                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5585                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5586                        "%s: trigger partial OI scrub for RPC inconsistency "
5587                        "checking FID "DFID": rc = %d\n",
5588                        osd_dev2name(dev), PFID(fid), rc);
5589                 if (rc == 0 || rc == -EALREADY)
5590                         goto again;
5591         }
5592
5593         GOTO(out, rc);
5594
5595 out:
5596         if (inode)
5597                 iput(inode);
5598
5599         RETURN(rc);
5600 }
5601
5602 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5603                                struct osd_device *dev,
5604                                struct osd_idmap_cache *oic,
5605                                struct lu_fid *fid, __u32 ino)
5606 {
5607         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5608         struct inode *inode;
5609         int rc;
5610
5611         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5612         inode = osd_iget(oti, dev, &oic->oic_lid);
5613         if (IS_ERR(inode)) {
5614                 fid_zero(&oic->oic_fid);
5615                 return PTR_ERR(inode);
5616         }
5617
5618         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5619         iput(inode);
5620         if (rc != 0)
5621                 fid_zero(&oic->oic_fid);
5622         else
5623                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5624         return rc;
5625 }
5626
5627 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5628                       struct osd_inode_id *id, const struct lu_fid *fid)
5629 {
5630         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5631                id->oii_ino, id->oii_gen, info);
5632         info->oti_cache.oic_lid = *id;
5633         info->oti_cache.oic_fid = *fid;
5634         info->oti_cache.oic_dev = osd;
5635 }
5636
5637 /**
5638  * Get parent FID from the linkEA.
5639  *
5640  * For a directory which parent resides on remote MDT, to satisfy the
5641  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5642  * the other hand, to make the lookup(..) on the directory can return
5643  * the real parent FID, we append the real parent FID after its ".."
5644  * name entry in the /REMOTE_PARENT_DIR.
5645  *
5646  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5647  * backup. So after the restore, we cannot get the right parent FID from
5648  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5649  * we have stored the real parent FID in the directory object's linkEA,
5650  * we can parse the linkEA for the real parent FID.
5651  *
5652  * \param[in] env       pointer to the thread context
5653  * \param[in] obj       pointer to the object to be handled
5654  * \param[out]fid       pointer to the buffer to hold the parent FID
5655  *
5656  * \retval              0 for getting the real parent FID successfully
5657  * \retval              negative error number on failure
5658  */
5659 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5660                                     struct osd_object *obj,
5661                                     struct lu_fid *fid)
5662 {
5663         struct osd_thread_info *oti = osd_oti_get(env);
5664         struct lu_buf *buf = &oti->oti_big_buf;
5665         struct dentry *dentry = &oti->oti_obj_dentry;
5666         struct inode *inode = obj->oo_inode;
5667         struct linkea_data ldata = { NULL };
5668         int rc;
5669
5670         ENTRY;
5671
5672         fid_zero(fid);
5673         if (!S_ISDIR(inode->i_mode))
5674                 RETURN(-EIO);
5675
5676 again:
5677         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5678                              buf->lb_buf, buf->lb_len);
5679         if (rc == -ERANGE) {
5680                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5681                                      NULL, 0);
5682                 if (rc > 0) {
5683                         lu_buf_realloc(buf, rc);
5684                         if (buf->lb_buf == NULL)
5685                                 RETURN(-ENOMEM);
5686
5687                         goto again;
5688                 }
5689         }
5690
5691         if (unlikely(rc == 0))
5692                 RETURN(-ENODATA);
5693
5694         if (rc < 0)
5695                 RETURN(rc);
5696
5697         if (unlikely(buf->lb_buf == NULL)) {
5698                 lu_buf_realloc(buf, rc);
5699                 if (buf->lb_buf == NULL)
5700                         RETURN(-ENOMEM);
5701
5702                 goto again;
5703         }
5704
5705         ldata.ld_buf = buf;
5706         rc = linkea_init_with_rec(&ldata);
5707         if (!rc) {
5708                 linkea_first_entry(&ldata);
5709                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5710         }
5711
5712         RETURN(rc);
5713 }
5714
5715 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5716                                     struct inode *inode,
5717                                     const struct lu_fid *pfid,
5718                                     const char *name, const int namelen)
5719 {
5720         struct osd_thread_info *oti = osd_oti_get(env);
5721         struct lu_buf *buf = &oti->oti_big_buf;
5722         struct dentry *dentry = &oti->oti_obj_dentry;
5723         struct linkea_data ldata = { NULL };
5724         struct lu_name cname = { .ln_name = name,
5725                                  .ln_namelen = namelen };
5726         int rc;
5727
5728         ENTRY;
5729
5730 again:
5731         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5732                              buf->lb_buf, buf->lb_len);
5733         if (rc == -ERANGE)
5734                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5735
5736         if (rc < 0)
5737                 RETURN(rc);
5738
5739         if (unlikely(rc == 0))
5740                 RETURN(-ENODATA);
5741
5742         if (buf->lb_len < rc) {
5743                 lu_buf_realloc(buf, rc);
5744                 if (buf->lb_buf == NULL)
5745                         RETURN(-ENOMEM);
5746
5747                 goto again;
5748         }
5749
5750         ldata.ld_buf = buf;
5751         rc = linkea_init_with_rec(&ldata);
5752         if (!rc)
5753                 rc = linkea_links_find(&ldata, &cname, pfid);
5754
5755         RETURN(rc);
5756 }
5757
5758 /**
5759  * Calls ->lookup() to find dentry. From dentry get inode and
5760  * read inode's ea to get fid. This is required for  interoperability
5761  * mode (b11826)
5762  *
5763  * \retval   0, on success
5764  * \retval -ve, on error
5765  */
5766 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5767                              struct dt_rec *rec, const struct dt_key *key)
5768 {
5769         struct inode *dir = obj->oo_inode;
5770         struct dentry *dentry;
5771         struct ldiskfs_dir_entry_2 *de;
5772         struct buffer_head *bh;
5773         struct lu_fid *fid = (struct lu_fid *)rec;
5774         struct htree_lock *hlock = NULL;
5775         int ino;
5776         int rc;
5777
5778         ENTRY;
5779
5780         LASSERT(dir->i_op != NULL);
5781         LASSERT(dir->i_op->lookup != NULL);
5782
5783         dentry = osd_child_dentry_get(env, obj,
5784                                       (char *)key, strlen((char *)key));
5785
5786         if (obj->oo_hl_head != NULL) {
5787                 hlock = osd_oti_get(env)->oti_hlock;
5788                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5789                                    dir, LDISKFS_HLOCK_LOOKUP);
5790         } else {
5791                 down_read(&obj->oo_ext_idx_sem);
5792         }
5793
5794         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5795         if (!IS_ERR(bh)) {
5796                 struct osd_thread_info *oti = osd_oti_get(env);
5797                 struct osd_inode_id *id = &oti->oti_id;
5798                 struct osd_idmap_cache *oic = &oti->oti_cache;
5799                 struct osd_device *dev = osd_obj2dev(obj);
5800
5801                 ino = le32_to_cpu(de->inode);
5802                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5803                         brelse(bh);
5804                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5805                         GOTO(out, rc);
5806                 }
5807
5808                 rc = osd_get_fid_from_dentry(de, rec);
5809
5810                 /* done with de, release bh */
5811                 brelse(bh);
5812                 if (rc != 0) {
5813                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5814                                 const char *name = (const char *)key;
5815
5816                                 /*
5817                                  * If the parent is on remote MDT, and there
5818                                  * is no FID-in-dirent, then we have to get
5819                                  * the parent FID from the linkEA.
5820                                  */
5821                                 if (likely(strlen(name) == 2 &&
5822                                            name[0] == '.' && name[1] == '.'))
5823                                         rc = osd_get_pfid_from_linkea(env, obj,
5824                                                                       fid);
5825                         } else {
5826                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5827                         }
5828                 } else {
5829                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5830                 }
5831
5832                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5833                         fid_zero(&oic->oic_fid);
5834
5835                         GOTO(out, rc);
5836                 }
5837
5838                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5839                 rc = osd_consistency_check(oti, dev, oic);
5840                 if (rc == -ENOENT)
5841                         fid_zero(&oic->oic_fid);
5842                 else
5843                         /* Other error should not affect lookup result. */
5844                         rc = 0;
5845         } else {
5846                 rc = PTR_ERR(bh);
5847         }
5848
5849         GOTO(out, rc);
5850
5851 out:
5852         if (hlock != NULL)
5853                 ldiskfs_htree_unlock(hlock);
5854         else
5855                 up_read(&obj->oo_ext_idx_sem);
5856         return rc;
5857 }
5858
5859 static int osd_index_declare_ea_insert(const struct lu_env *env,
5860                                        struct dt_object *dt,
5861                                        const struct dt_rec *rec,
5862                                        const struct dt_key *key,
5863                                        struct thandle *handle)
5864 {
5865         struct osd_thandle *oh;
5866         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5867         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5868         const struct lu_fid *fid = rec1->rec_fid;
5869         int credits, rc = 0;
5870         struct osd_idmap_cache *idc;
5871
5872         ENTRY;
5873
5874         LASSERT(!dt_object_remote(dt));
5875         LASSERT(handle != NULL);
5876         LASSERT(fid != NULL);
5877         LASSERT(rec1->rec_type != 0);
5878
5879         oh = container_of0(handle, struct osd_thandle, ot_super);
5880         LASSERT(oh->ot_handle == NULL);
5881
5882         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5883
5884         /*
5885          * we can't call iget() while a transactions is running
5886          * (this can lead to a deadlock), but we need to know
5887          * inum and object type. so we find this information at
5888          * declaration and cache in per-thread info
5889          */
5890         idc = osd_idc_find_or_init(env, osd, fid);
5891         if (IS_ERR(idc))
5892                 RETURN(PTR_ERR(idc));
5893         if (idc->oic_remote) {
5894                 /*
5895                  * a reference to remote inode is represented by an
5896                  * agent inode which we have to create
5897                  */
5898                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5899                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5900         }
5901
5902         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5903
5904         if (osd_dt_obj(dt)->oo_inode != NULL) {
5905                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5906
5907                 /*
5908                  * We ignore block quota on meta pool (MDTs), so needn't
5909                  * calculate how many blocks will be consumed by this index
5910                  * insert
5911                  */
5912                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5913                                            i_gid_read(inode),
5914                                            i_projid_read(inode), 0,
5915                                            oh, osd_dt_obj(dt), NULL,
5916                                            OSD_QID_BLK);
5917                 if (rc)
5918                         RETURN(rc);
5919
5920 #ifdef HAVE_PROJECT_QUOTA
5921                 /*
5922                  * Reserve credits for local agent inode to transfer
5923                  * to 0, quota enforcement is ignored in this case.
5924                  */
5925                 if (idc->oic_remote &&
5926                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5927                     i_projid_read(inode) != 0)
5928                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5929                                                   0, i_projid_read(inode),
5930                                                   0, false, PRJQUOTA, true);
5931 #endif
5932         }
5933
5934         RETURN(rc);
5935 }
5936
5937 /**
5938  * Index add function for interoperability mode (b11826).
5939  * It will add the directory entry.This entry is needed to
5940  * maintain name->fid mapping.
5941  *
5942  * \param key it is key i.e. file entry to be inserted
5943  * \param rec it is value of given key i.e. fid
5944  *
5945  * \retval   0, on success
5946  * \retval -ve, on error
5947  */
5948 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5949                                const struct dt_rec *rec,
5950                                const struct dt_key *key, struct thandle *th,
5951                                int ignore_quota)
5952 {
5953         struct osd_object *obj = osd_dt_obj(dt);
5954         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5955         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5956         const struct lu_fid *fid = rec1->rec_fid;
5957         const char *name = (const char *)key;
5958         struct osd_thread_info *oti = osd_oti_get(env);
5959         struct inode *child_inode = NULL;
5960         struct osd_idmap_cache *idc;
5961         int rc;
5962
5963         ENTRY;
5964
5965         if (!dt_object_exists(dt))
5966                 RETURN(-ENOENT);
5967
5968         LASSERT(osd_invariant(obj));
5969         LASSERT(!dt_object_remote(dt));
5970         LASSERT(th != NULL);
5971
5972         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5973
5974         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5975
5976         idc = osd_idc_find(env, osd, fid);
5977         if (unlikely(idc == NULL)) {
5978                 idc = osd_idc_find_or_init(env, osd, fid);
5979                 if (IS_ERR(idc)) {
5980                         /*
5981                          * this dt_insert() wasn't declared properly, so
5982                          * FID is missing in OI cache. we better do not
5983                          * lookup FID in FLDB/OI and don't risk to deadlock,
5984                          * but in some special cases (lfsck testing, etc)
5985                          * it's much simpler than fixing a caller.
5986                          *
5987                          * normally this error should be placed after the first
5988                          * find, but migrate may attach source stripes to
5989                          * target, which doesn't create stripes.
5990                          */
5991                         CERROR("%s: "DFID" wasn't declared for insert\n",
5992                                osd_name(osd), PFID(fid));
5993                         dump_stack();
5994                         RETURN(PTR_ERR(idc));
5995                 }
5996         }
5997
5998         if (idc->oic_remote) {
5999                 /* Insert remote entry */
6000                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6001                         child_inode =
6002                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6003                 } else {
6004                         child_inode = osd_create_local_agent_inode(env, osd,
6005                                         obj, fid, rec1->rec_type & S_IFMT, th);
6006                         if (IS_ERR(child_inode))
6007                                 RETURN(PTR_ERR(child_inode));
6008                 }
6009         } else {
6010                 /* Insert local entry */
6011                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6012                         /* for a reason OI cache wasn't filled properly */
6013                         CERROR("%s: OIC for "DFID" isn't filled\n",
6014                                osd_name(osd), PFID(fid));
6015                         RETURN(-EINVAL);
6016                 }
6017                 child_inode = oti->oti_inode;
6018                 if (unlikely(child_inode == NULL)) {
6019                         struct ldiskfs_inode_info *lii;
6020
6021                         OBD_ALLOC_PTR(lii);
6022                         if (lii == NULL)
6023                                 RETURN(-ENOMEM);
6024                         child_inode = oti->oti_inode = &lii->vfs_inode;
6025                 }
6026                 child_inode->i_sb = osd_sb(osd);
6027                 child_inode->i_ino = idc->oic_lid.oii_ino;
6028                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6029         }
6030
6031         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6032
6033         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6034                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6035
6036         if (child_inode && child_inode != oti->oti_inode)
6037                 iput(child_inode);
6038         LASSERT(osd_invariant(obj));
6039         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6040         RETURN(rc);
6041 }
6042
6043 /**
6044  *  Initialize osd Iterator for given osd index object.
6045  *
6046  *  \param  dt      osd index object
6047  */
6048
6049 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6050                                      struct dt_object *dt,
6051                                      __u32 unused)
6052 {
6053         struct osd_it_iam *it;
6054         struct osd_object *obj = osd_dt_obj(dt);
6055         struct lu_object *lo = &dt->do_lu;
6056         struct iam_path_descr *ipd;
6057         struct iam_container *bag = &obj->oo_dir->od_container;
6058
6059         if (!dt_object_exists(dt))
6060                 return ERR_PTR(-ENOENT);
6061
6062         OBD_ALLOC_PTR(it);
6063         if (it == NULL)
6064                 return ERR_PTR(-ENOMEM);
6065
6066         ipd = osd_it_ipd_get(env, bag);
6067         if (likely(ipd != NULL)) {
6068                 it->oi_obj = obj;
6069                 it->oi_ipd = ipd;
6070                 lu_object_get(lo);
6071                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6072                 return (struct dt_it *)it;
6073         } else {
6074                 OBD_FREE_PTR(it);
6075                 return ERR_PTR(-ENOMEM);
6076         }
6077 }
6078
6079 /**
6080  * free given Iterator.
6081  */
6082 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6083 {
6084         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6085         struct osd_object *obj = it->oi_obj;
6086
6087         iam_it_fini(&it->oi_it);
6088         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6089         osd_object_put(env, obj);
6090         OBD_FREE_PTR(it);
6091 }
6092
6093 /**
6094  *  Move Iterator to record specified by \a key
6095  *
6096  *  \param  di      osd iterator
6097  *  \param  key     key for index
6098  *
6099  *  \retval +ve  di points to record with least key not larger than key
6100  *  \retval  0   di points to exact matched key
6101  *  \retval -ve  failure
6102  */
6103
6104 static int osd_it_iam_get(const struct lu_env *env,
6105                           struct dt_it *di, const struct dt_key *key)
6106 {
6107         struct osd_thread_info *oti = osd_oti_get(env);
6108         struct osd_it_iam *it = (struct osd_it_iam *)di;
6109
6110         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6111                 /* swab quota uid/gid */
6112                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6113                 key = (struct dt_key *)&oti->oti_quota_id;
6114         }
6115
6116         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6117 }
6118
6119 /**
6120  *  Release Iterator
6121  *
6122  *  \param  di      osd iterator
6123  */
6124 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6125 {
6126         struct osd_it_iam *it = (struct osd_it_iam *)di;
6127
6128         iam_it_put(&it->oi_it);
6129 }
6130
6131 /**
6132  *  Move iterator by one record
6133  *
6134  *  \param  di      osd iterator
6135  *
6136  *  \retval +1   end of container reached
6137  *  \retval  0   success
6138  *  \retval -ve  failure
6139  */
6140
6141 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6142 {
6143         struct osd_it_iam *it = (struct osd_it_iam *)di;
6144
6145         return iam_it_next(&it->oi_it);
6146 }
6147
6148 /**
6149  * Return pointer to the key under iterator.
6150  */
6151
6152 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6153                                      const struct dt_it *di)
6154 {
6155         struct osd_thread_info *oti = osd_oti_get(env);
6156         struct osd_it_iam *it = (struct osd_it_iam *)di;
6157         struct osd_object *obj = it->oi_obj;
6158         struct dt_key *key;
6159
6160         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6161
6162         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6163                 /* swab quota uid/gid */
6164                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6165                 key = (struct dt_key *)&oti->oti_quota_id;
6166         }
6167
6168         return key;
6169 }
6170
6171 /**
6172  * Return size of key under iterator (in bytes)
6173  */
6174
6175 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6176 {
6177         struct osd_it_iam *it = (struct osd_it_iam *)di;
6178
6179         return iam_it_key_size(&it->oi_it);
6180 }
6181
6182 static inline void
6183 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6184 {
6185         /* check if file type is required */
6186         if (ent->lde_attrs & LUDA_TYPE) {
6187                 struct luda_type *lt;
6188                 int align = sizeof(*lt) - 1;
6189
6190                 len = (len + align) & ~align;
6191                 lt = (struct luda_type *)(ent->lde_name + len);
6192                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6193         }
6194
6195         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6196 }
6197
6198 /**
6199  * build lu direct from backend fs dirent.
6200  */
6201
6202 static inline void
6203 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6204                    char *name, __u16 namelen, __u16 type, __u32 attr)
6205 {
6206         ent->lde_attrs = attr | LUDA_FID;
6207         fid_cpu_to_le(&ent->lde_fid, fid);
6208
6209         ent->lde_hash = cpu_to_le64(offset);
6210         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6211
6212         strncpy(ent->lde_name, name, namelen);
6213         ent->lde_name[namelen] = '\0';
6214         ent->lde_namelen = cpu_to_le16(namelen);
6215
6216         /* append lustre attributes */
6217         osd_it_append_attrs(ent, namelen, type);
6218 }
6219
6220 /**
6221  * Return pointer to the record under iterator.
6222  */
6223 static int osd_it_iam_rec(const struct lu_env *env,
6224                           const struct dt_it *di,
6225                           struct dt_rec *dtrec, __u32 attr)
6226 {
6227         struct osd_it_iam *it = (struct osd_it_iam *)di;
6228         struct osd_thread_info *info = osd_oti_get(env);
6229
6230         ENTRY;
6231
6232         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6233                 const struct osd_fid_pack *rec;
6234                 struct lu_fid *fid = &info->oti_fid;
6235                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6236                 char *name;
6237                 int namelen;
6238                 __u64 hash;
6239                 int rc;
6240
6241                 name = (char *)iam_it_key_get(&it->oi_it);
6242                 if (IS_ERR(name))
6243                         RETURN(PTR_ERR(name));
6244
6245                 namelen = iam_it_key_size(&it->oi_it);
6246
6247                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6248                 if (IS_ERR(rec))
6249                         RETURN(PTR_ERR(rec));
6250
6251                 rc = osd_fid_unpack(fid, rec);
6252                 if (rc)
6253                         RETURN(rc);
6254
6255                 hash = iam_it_store(&it->oi_it);
6256
6257                 /* IAM does not store object type in IAM index (dir) */
6258                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6259                                    0, LUDA_FID);
6260         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6261                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6262                            (struct iam_rec *)dtrec);
6263                 osd_quota_unpack(it->oi_obj, dtrec);
6264         } else {
6265                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6266                            (struct iam_rec *)dtrec);
6267         }
6268
6269         RETURN(0);
6270 }
6271
6272 /**
6273  * Returns cookie for current Iterator position.
6274  */
6275 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6276 {
6277         struct osd_it_iam *it = (struct osd_it_iam *)di;
6278
6279         return iam_it_store(&it->oi_it);
6280 }
6281
6282 /**
6283  * Restore iterator from cookie.
6284  *
6285  * \param  di      osd iterator
6286  * \param  hash    Iterator location cookie
6287  *
6288  * \retval +ve  di points to record with least key not larger than key.
6289  * \retval  0   di points to exact matched key
6290  * \retval -ve  failure
6291  */
6292
6293 static int osd_it_iam_load(const struct lu_env *env,
6294                            const struct dt_it *di, __u64 hash)
6295 {
6296         struct osd_it_iam *it = (struct osd_it_iam *)di;
6297
6298         return iam_it_load(&it->oi_it, hash);
6299 }
6300
6301 static const struct dt_index_operations osd_index_iam_ops = {
6302         .dio_lookup         = osd_index_iam_lookup,
6303         .dio_declare_insert = osd_index_declare_iam_insert,
6304         .dio_insert         = osd_index_iam_insert,
6305         .dio_declare_delete = osd_index_declare_iam_delete,
6306         .dio_delete         = osd_index_iam_delete,
6307         .dio_it     = {
6308                 .init     = osd_it_iam_init,
6309                 .fini     = osd_it_iam_fini,
6310                 .get      = osd_it_iam_get,
6311                 .put      = osd_it_iam_put,
6312                 .next     = osd_it_iam_next,
6313                 .key      = osd_it_iam_key,
6314                 .key_size = osd_it_iam_key_size,
6315                 .rec      = osd_it_iam_rec,
6316                 .store    = osd_it_iam_store,
6317                 .load     = osd_it_iam_load
6318         }
6319 };
6320
6321
6322 /**
6323  * Creates or initializes iterator context.
6324  *
6325  * \retval struct osd_it_ea, iterator structure on success
6326  *
6327  */
6328 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6329                                     struct dt_object *dt,
6330                                     __u32 attr)
6331 {
6332         struct osd_object *obj = osd_dt_obj(dt);
6333         struct osd_thread_info *info = osd_oti_get(env);
6334         struct osd_it_ea *oie;
6335         struct file *file;
6336         struct lu_object *lo = &dt->do_lu;
6337         struct dentry *obj_dentry;
6338
6339         ENTRY;
6340
6341         if (!dt_object_exists(dt) || obj->oo_destroyed)
6342                 RETURN(ERR_PTR(-ENOENT));
6343
6344         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6345         if (oie == NULL)
6346                 RETURN(ERR_PTR(-ENOMEM));
6347         obj_dentry = &oie->oie_dentry;
6348
6349         obj_dentry->d_inode = obj->oo_inode;
6350         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6351         obj_dentry->d_name.hash = 0;
6352
6353         oie->oie_rd_dirent       = 0;
6354         oie->oie_it_dirent       = 0;
6355         oie->oie_dirent          = NULL;
6356         if (unlikely(!info->oti_it_ea_buf_used)) {
6357                 oie->oie_buf = info->oti_it_ea_buf;
6358                 info->oti_it_ea_buf_used = 1;
6359         } else {
6360                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6361                 if (oie->oie_buf == NULL)
6362                         RETURN(ERR_PTR(-ENOMEM));
6363         }
6364         oie->oie_obj = obj;
6365
6366         file = &oie->oie_file;
6367
6368         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6369         if (attr & LUDA_64BITHASH)
6370                 file->f_mode    = FMODE_64BITHASH;
6371         else
6372                 file->f_mode    = FMODE_32BITHASH;
6373         file->f_path.dentry     = obj_dentry;
6374         file->f_mapping         = obj->oo_inode->i_mapping;
6375         file->f_op              = obj->oo_inode->i_fop;
6376         set_file_inode(file, obj->oo_inode);
6377
6378         lu_object_get(lo);
6379         RETURN((struct dt_it *)oie);
6380 }
6381
6382 /**
6383  * Destroy or finishes iterator context.
6384  *
6385  * \param di iterator structure to be destroyed
6386  */
6387 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6388 {
6389         struct osd_thread_info *info = osd_oti_get(env);
6390         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6391         struct osd_object *obj = oie->oie_obj;
6392         struct inode *inode = obj->oo_inode;
6393
6394         ENTRY;
6395         oie->oie_file.f_op->release(inode, &oie->oie_file);
6396         osd_object_put(env, obj);
6397         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6398                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6399         else
6400                 info->oti_it_ea_buf_used = 0;
6401         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6402         EXIT;
6403 }
6404
6405 /**
6406  * It position the iterator at given key, so that next lookup continues from
6407  * that key Or it is similar to dio_it->load() but based on a key,
6408  * rather than file position.
6409  *
6410  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6411  * to the beginning.
6412  *
6413  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6414  */
6415 static int osd_it_ea_get(const struct lu_env *env,
6416                          struct dt_it *di, const struct dt_key *key)
6417 {
6418         struct osd_it_ea *it = (struct osd_it_ea *)di;
6419
6420         ENTRY;
6421         LASSERT(((const char *)key)[0] == '\0');
6422         it->oie_file.f_pos = 0;
6423         it->oie_rd_dirent = 0;
6424         it->oie_it_dirent = 0;
6425         it->oie_dirent = NULL;
6426
6427         RETURN(+1);
6428 }
6429
6430 /**
6431  * Does nothing
6432  */
6433 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6434 {
6435 }
6436
6437 struct osd_filldir_cbs {
6438 #ifdef HAVE_DIR_CONTEXT
6439         struct dir_context ctx;
6440 #endif
6441         struct osd_it_ea  *it;
6442 };
6443 /**
6444  * It is called internally by ->readdir(). It fills the
6445  * iterator's in-memory data structure with required
6446  * information i.e. name, namelen, rec_size etc.
6447  *
6448  * \param buf in which information to be filled in.
6449  * \param name name of the file in given dir
6450  *
6451  * \retval 0 on success
6452  * \retval 1 on buffer full
6453  */
6454 #ifdef HAVE_FILLDIR_USE_CTX
6455 static int osd_ldiskfs_filldir(struct dir_context *buf,
6456 #else
6457 static int osd_ldiskfs_filldir(void *buf,
6458 #endif
6459                                const char *name, int namelen,
6460                                loff_t offset, __u64 ino, unsigned int d_type)
6461 {
6462         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6463         struct osd_object *obj = it->oie_obj;
6464         struct osd_it_ea_dirent *ent = it->oie_dirent;
6465         struct lu_fid *fid = &ent->oied_fid;
6466         struct osd_fid_pack *rec;
6467
6468         ENTRY;
6469
6470 /* this should never happen */
6471         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6472                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6473                 RETURN(-EIO);
6474         }
6475
6476         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6477             OSD_IT_EA_BUFSIZE)
6478                 RETURN(1);
6479
6480         /* "." is just the object itself. */
6481         if (namelen == 1 && name[0] == '.') {
6482                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6483         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6484                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6485                 if (osd_fid_unpack(fid, rec) != 0)
6486                         fid_zero(fid);
6487         } else {
6488                 fid_zero(fid);
6489         }
6490         d_type &= ~LDISKFS_DIRENT_LUFID;
6491
6492         /* NOT export local root. */
6493         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6494                 ino = obj->oo_inode->i_ino;
6495                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6496         }
6497
6498         ent->oied_ino     = ino;
6499         ent->oied_off     = offset;
6500         ent->oied_namelen = namelen;
6501         ent->oied_type    = d_type;
6502
6503         memcpy(ent->oied_name, name, namelen);
6504
6505         it->oie_rd_dirent++;
6506         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6507         RETURN(0);
6508 }
6509
6510 /**
6511  * Calls ->readdir() to load a directory entry at a time
6512  * and stored it in iterator's in-memory data structure.
6513  *
6514  * \param di iterator's in memory structure
6515  *
6516  * \retval   0 on success
6517  * \retval -ve on error
6518  * \retval +1 reach the end of entry
6519  */
6520 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6521                                const struct dt_it *di)
6522 {
6523         struct osd_it_ea *it = (struct osd_it_ea *)di;
6524         struct osd_object *obj = it->oie_obj;
6525         struct inode *inode = obj->oo_inode;
6526         struct htree_lock *hlock = NULL;
6527         struct file *filp = &it->oie_file;
6528         int rc = 0;
6529         struct osd_filldir_cbs buf = {
6530 #ifdef HAVE_DIR_CONTEXT
6531                 .ctx.actor = osd_ldiskfs_filldir,
6532 #endif
6533                 .it = it
6534         };
6535
6536         ENTRY;
6537         it->oie_dirent = it->oie_buf;
6538         it->oie_rd_dirent = 0;
6539
6540         if (obj->oo_hl_head != NULL) {
6541                 hlock = osd_oti_get(env)->oti_hlock;
6542                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6543                                    inode, LDISKFS_HLOCK_READDIR);
6544         } else {
6545                 down_read(&obj->oo_ext_idx_sem);
6546         }
6547
6548 #ifdef HAVE_DIR_CONTEXT
6549         buf.ctx.pos = filp->f_pos;
6550 #ifdef HAVE_ITERATE_SHARED
6551         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6552 #else
6553         rc = inode->i_fop->iterate(filp, &buf.ctx);
6554 #endif
6555         filp->f_pos = buf.ctx.pos;
6556 #else
6557         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6558 #endif
6559
6560         if (hlock != NULL)
6561                 ldiskfs_htree_unlock(hlock);
6562         else
6563                 up_read(&obj->oo_ext_idx_sem);
6564
6565         if (it->oie_rd_dirent == 0) {
6566                 /*
6567                  * If it does not get any dirent, it means it has been reached
6568                  * to the end of the dir
6569                  */
6570                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6571                 if (rc == 0)
6572                         rc = 1;
6573         } else {
6574                 it->oie_dirent = it->oie_buf;
6575                 it->oie_it_dirent = 1;
6576         }
6577
6578         RETURN(rc);
6579 }
6580
6581 /**
6582  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6583  * to load a directory entry at a time and stored it in
6584  * iterator's in-memory data structure.
6585  *
6586  * \param di iterator's in memory structure
6587  *
6588  * \retval +ve iterator reached to end
6589  * \retval   0 iterator not reached to end
6590  * \retval -ve on error
6591  */
6592 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6593 {
6594         struct osd_it_ea *it = (struct osd_it_ea *)di;
6595         int rc;
6596
6597         ENTRY;
6598
6599         if (it->oie_it_dirent < it->oie_rd_dirent) {
6600                 it->oie_dirent =
6601                         (void *)it->oie_dirent +
6602                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6603                                        it->oie_dirent->oied_namelen);
6604                 it->oie_it_dirent++;
6605                 rc = 0;
6606         } else {
6607                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6608                         rc = 1;
6609                 else
6610                         rc = osd_ldiskfs_it_fill(env, di);
6611         }
6612
6613         RETURN(rc);
6614 }
6615
6616 /**
6617  * Returns the key at current position from iterator's in memory structure.
6618  *
6619  * \param di iterator's in memory structure
6620  *
6621  * \retval key i.e. struct dt_key on success
6622  */
6623 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6624                                     const struct dt_it *di)
6625 {
6626         struct osd_it_ea *it = (struct osd_it_ea *)di;
6627
6628         return (struct dt_key *)it->oie_dirent->oied_name;
6629 }
6630
6631 /**
6632  * Returns key's size at current position from iterator's in memory structure.
6633  *
6634  * \param di iterator's in memory structure
6635  *
6636  * \retval key_size i.e. struct dt_key on success
6637  */
6638 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6639 {
6640         struct osd_it_ea *it = (struct osd_it_ea *)di;
6641
6642         return it->oie_dirent->oied_namelen;
6643 }
6644
6645 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6646 {
6647         if (LDISKFS_DIR_REC_LEN(de) >=
6648             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6649                 return true;
6650
6651         return false;
6652 }
6653
6654 static inline bool
6655 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6656                      unsigned int blocksize, bool dotdot)
6657 {
6658         if (dotdot)
6659                 return osd_dotdot_has_space(de);
6660
6661         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6662             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6663                 return true;
6664
6665         return false;
6666 }
6667
6668 static int
6669 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6670                     handle_t *jh, struct dentry *dentry,
6671                     const struct lu_fid *fid, struct buffer_head *bh,
6672                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6673                     bool dotdot)
6674 {
6675         struct inode *dir = dentry->d_parent->d_inode;
6676         struct inode *inode = dentry->d_inode;
6677         struct osd_fid_pack *rec;
6678         struct ldiskfs_dentry_param *ldp;
6679         int namelen = dentry->d_name.len;
6680         int rc;
6681         struct osd_thread_info *info = osd_oti_get(env);
6682
6683         ENTRY;
6684
6685         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6686                 RETURN(0);
6687
6688         /* There is enough space to hold the FID-in-dirent. */
6689         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6690                 rc = ldiskfs_journal_get_write_access(jh, bh);
6691                 if (rc != 0)
6692                         RETURN(rc);
6693
6694                 de->name[namelen] = 0;
6695                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6696                 rec->fp_len = sizeof(struct lu_fid) + 1;
6697                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6698                 de->file_type |= LDISKFS_DIRENT_LUFID;
6699                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6700
6701                 RETURN(rc);
6702         }
6703
6704         LASSERT(!dotdot);
6705
6706         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6707         if (rc != 0)
6708                 RETURN(rc);
6709
6710         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6711         osd_get_ldiskfs_dirent_param(ldp, fid);
6712         dentry->d_fsdata = (void *)ldp;
6713         ll_vfs_dq_init(dir);
6714         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6715         /*
6716          * It is too bad, we cannot reinsert the name entry back.
6717          * That means we lose it!
6718          */
6719         if (rc != 0)
6720                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6721                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6722                        osd_ino2name(inode),
6723                        dir->i_ino, dir->i_generation, namelen,
6724                        dentry->d_name.name, PFID(fid), rc);
6725
6726         RETURN(rc);
6727 }
6728
6729 static int
6730 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6731                         struct osd_it_ea *it, struct lu_fid *fid,
6732                         struct osd_inode_id *id, __u32 *attr)
6733 {
6734         struct osd_thread_info *info = osd_oti_get(env);
6735         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6736         struct osd_device *dev = osd_obj2dev(obj);
6737         struct super_block *sb = osd_sb(dev);
6738         const char *devname = osd_name(dev);
6739         struct osd_it_ea_dirent *ent = it->oie_dirent;
6740         struct inode *dir = obj->oo_inode;
6741         struct htree_lock *hlock = NULL;
6742         struct buffer_head *bh = NULL;
6743         handle_t *jh = NULL;
6744         struct ldiskfs_dir_entry_2 *de;
6745         struct dentry *dentry;
6746         struct inode *inode;
6747         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6748         int credits;
6749         int rc;
6750         bool dotdot = false;
6751         bool dirty = false;
6752
6753         ENTRY;
6754
6755         if (ent->oied_name[0] == '.') {
6756                 if (ent->oied_namelen == 1)
6757                         RETURN(0);
6758
6759                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6760                         dotdot = true;
6761         }
6762
6763         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6764         inode = osd_iget(info, dev, id);
6765         if (IS_ERR(inode)) {
6766                 rc = PTR_ERR(inode);
6767                 if (rc == -ENOENT || rc == -ESTALE) {
6768                         /*
6769                          * Maybe dangling name entry, or
6770                          * corrupted directory entry.
6771                          */
6772                         *attr |= LUDA_UNKNOWN;
6773                         rc = 0;
6774                 } else {
6775                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6776                                "check_repair, dir = %lu/%u, name = %.*s, "
6777                                "ino = %llu, rc = %d\n",
6778                                devname, dir->i_ino, dir->i_generation,
6779                                ent->oied_namelen, ent->oied_name,
6780                                ent->oied_ino, rc);
6781                 }
6782
6783                 RETURN(rc);
6784         }
6785
6786         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6787                                            ent->oied_namelen);
6788         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6789         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6790                 lma = NULL;
6791         else if (rc != 0)
6792                 GOTO(out, rc);
6793
6794         /*
6795          * We need to ensure that the name entry is still valid.
6796          * Because it may be removed or renamed by other already.
6797          *
6798          * The unlink or rename operation will start journal before PDO lock,
6799          * so to avoid deadlock, here we need to start journal handle before
6800          * related PDO lock also. But because we do not know whether there
6801          * will be something to be repaired before PDO lock, we just start
6802          * journal without conditions.
6803          *
6804          * We may need to remove the name entry firstly, then insert back.
6805          * One credit is for user quota file update.
6806          * One credit is for group quota file update.
6807          * Two credits are for dirty inode.
6808          */
6809         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6810                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6811
6812         if (dev->od_dirent_journal != 0) {
6813
6814 again:
6815                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6816                 if (IS_ERR(jh)) {
6817                         rc = PTR_ERR(jh);
6818                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6819                                "check_repair, dir = %lu/%u, credits = %d, "
6820                                "name = %.*s, ino = %llu: rc = %d\n",
6821                                devname, dir->i_ino, dir->i_generation, credits,
6822                                ent->oied_namelen, ent->oied_name,
6823                                ent->oied_ino, rc);
6824
6825                         GOTO(out_inode, rc);
6826                 }
6827
6828                 if (obj->oo_hl_head != NULL) {
6829                         hlock = osd_oti_get(env)->oti_hlock;
6830                         /*
6831                          * "0" means exclusive lock for the whole directory.
6832                          * We need to prevent others access such name entry
6833                          * during the delete + insert. Neither HLOCK_ADD nor
6834                          * HLOCK_DEL cannot guarantee the atomicity.
6835                          */
6836                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6837                 } else {
6838                         down_write(&obj->oo_ext_idx_sem);
6839                 }
6840         } else {
6841                 if (obj->oo_hl_head != NULL) {
6842                         hlock = osd_oti_get(env)->oti_hlock;
6843                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6844                                            LDISKFS_HLOCK_LOOKUP);
6845                 } else {
6846                         down_read(&obj->oo_ext_idx_sem);
6847                 }
6848         }
6849
6850         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6851         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6852                 *attr |= LUDA_IGNORE;
6853
6854                 GOTO(out, rc = 0);
6855         }
6856
6857         /*
6858          * For dotdot entry, if there is not enough space to hold the
6859          * FID-in-dirent, just keep them there. It only happens when the
6860          * device upgraded from 1.8 or restored from MDT file-level backup.
6861          * For the whole directory, only dotdot entry have no FID-in-dirent
6862          * and needs to get FID from LMA when readdir, it will not affect the
6863          * performance much.
6864          */
6865         if (dotdot && !osd_dotdot_has_space(de)) {
6866                 *attr |= LUDA_UNKNOWN;
6867
6868                 GOTO(out, rc = 0);
6869         }
6870
6871         if (lma != NULL) {
6872                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6873                         GOTO(out, rc = 0);
6874
6875                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6876                         struct lu_fid *tfid = &lma->lma_self_fid;
6877
6878                         if (likely(dotdot &&
6879                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6880                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6881                                 /*
6882                                  * It must be REMOTE_PARENT_DIR and as the
6883                                  * 'dotdot' entry of remote directory
6884                                  */
6885                                 *attr |= LUDA_IGNORE;
6886                         } else {
6887                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6888                                        "parent directory, but got %.*s under "
6889                                        "dir = %lu/%u with the FID "DFID"\n",
6890                                        devname, ent->oied_namelen,
6891                                        ent->oied_name, dir->i_ino,
6892                                        dir->i_generation, PFID(tfid));
6893
6894                                 *attr |= LUDA_UNKNOWN;
6895                         }
6896
6897                         GOTO(out, rc = 0);
6898                 }
6899         }
6900
6901         if (!fid_is_zero(fid)) {
6902                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6903                                               ent->oied_namelen);
6904                 if (rc == -ENOENT ||
6905                     (rc == -ENODATA &&
6906                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6907                         /*
6908                          * linkEA does not recognize the dirent entry,
6909                          * it may because the dirent entry corruption
6910                          * and points to other's inode.
6911                          */
6912                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6913                                "recognize the dirent, dir = %lu/%u, "
6914                                " name = %.*s, ino = %llu, "
6915                                DFID": rc = %d\n", devname, dir->i_ino,
6916                                dir->i_generation, ent->oied_namelen,
6917                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6918                         *attr |= LUDA_UNKNOWN;
6919
6920                         GOTO(out, rc = 0);
6921                 }
6922
6923                 if (rc && rc != -ENODATA) {
6924                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6925                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6926                                DFID": rc = %d\n", devname, dir->i_ino,
6927                                dir->i_generation, ent->oied_namelen,
6928                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6929                         *attr |= LUDA_UNKNOWN;
6930
6931                         GOTO(out, rc = 0);
6932                 }
6933         }
6934
6935         if (lma != NULL) {
6936                 /*
6937                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6938                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6939                  */
6940                 if (*attr & LUDA_VERIFY_DRYRUN) {
6941                         *fid = lma->lma_self_fid;
6942                         *attr |= LUDA_REPAIR;
6943
6944                         GOTO(out, rc = 0);
6945                 }
6946
6947                 if (jh == NULL) {
6948                         brelse(bh);
6949                         dev->od_dirent_journal = 1;
6950                         if (hlock != NULL) {
6951                                 ldiskfs_htree_unlock(hlock);
6952                                 hlock = NULL;
6953                         } else {
6954                                 up_read(&obj->oo_ext_idx_sem);
6955                         }
6956
6957                         goto again;
6958                 }
6959
6960                 *fid = lma->lma_self_fid;
6961                 dirty = true;
6962                 /* Update or append the FID-in-dirent. */
6963                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6964                                          bh, de, hlock, dotdot);
6965                 if (rc == 0)
6966                         *attr |= LUDA_REPAIR;
6967                 else
6968                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6969                                "the dirent, dir = %lu/%u, name = %.*s, "
6970                                "ino = %llu, "DFID": rc = %d\n",
6971                                devname, dir->i_ino, dir->i_generation,
6972                                ent->oied_namelen, ent->oied_name,
6973                                ent->oied_ino, PFID(fid), rc);
6974         } else {
6975                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6976                 if (*attr & LUDA_VERIFY_DRYRUN) {
6977                         if (fid_is_sane(fid)) {
6978                                 *attr |= LUDA_REPAIR;
6979                         } else if (dev->od_index == 0) {
6980                                 lu_igif_build(fid, inode->i_ino,
6981                                               inode->i_generation);
6982                                 *attr |= LUDA_UPGRADE;
6983                         }
6984
6985                         GOTO(out, rc = 0);
6986                 }
6987
6988                 if (jh == NULL) {
6989                         brelse(bh);
6990                         dev->od_dirent_journal = 1;
6991                         if (hlock != NULL) {
6992                                 ldiskfs_htree_unlock(hlock);
6993                                 hlock = NULL;
6994                         } else {
6995                                 up_read(&obj->oo_ext_idx_sem);
6996                         }
6997
6998                         goto again;
6999                 }
7000
7001                 dirty = true;
7002                 if (unlikely(fid_is_sane(fid))) {
7003                         /*
7004                          * FID-in-dirent exists, but FID-in-LMA is lost.
7005                          * Trust the FID-in-dirent, and add FID-in-LMA.
7006                          */
7007                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7008                         if (rc == 0)
7009                                 *attr |= LUDA_REPAIR;
7010                         else
7011                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7012                                        "update dirent, dir = %lu/%u, "
7013                                        "name = %.*s, ino = %llu, "
7014                                        DFID": rc = %d\n",
7015                                        devname, dir->i_ino, dir->i_generation,
7016                                        ent->oied_namelen, ent->oied_name,
7017                                        ent->oied_ino, PFID(fid), rc);
7018                 } else if (dev->od_index == 0) {
7019                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7020                         /*
7021                          * It is probably IGIF object. Only aappend the
7022                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7023                          */
7024                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7025                                                  bh, de, hlock, dotdot);
7026                         if (rc == 0)
7027                                 *attr |= LUDA_UPGRADE;
7028                         else
7029                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7030                                        "after the dirent, dir = %lu/%u, "
7031                                        "name = %.*s, ino = %llu, "
7032                                        DFID": rc = %d\n",
7033                                        devname, dir->i_ino, dir->i_generation,
7034                                        ent->oied_namelen, ent->oied_name,
7035                                        ent->oied_ino, PFID(fid), rc);
7036                 }
7037         }
7038
7039         GOTO(out, rc);
7040
7041 out:
7042         if (!IS_ERR(bh))
7043                 brelse(bh);
7044         if (hlock != NULL) {
7045                 ldiskfs_htree_unlock(hlock);
7046         } else {
7047                 if (dev->od_dirent_journal != 0)
7048                         up_write(&obj->oo_ext_idx_sem);
7049                 else
7050                         up_read(&obj->oo_ext_idx_sem);
7051         }
7052
7053         if (jh != NULL)
7054                 ldiskfs_journal_stop(jh);
7055
7056 out_inode:
7057         iput(inode);
7058         if (rc >= 0 && !dirty)
7059                 dev->od_dirent_journal = 0;
7060
7061         return rc;
7062 }
7063
7064 /**
7065  * Returns the value at current position from iterator's in memory structure.
7066  *
7067  * \param di struct osd_it_ea, iterator's in memory structure
7068  * \param attr attr requested for dirent.
7069  * \param lde lustre dirent
7070  *
7071  * \retval   0 no error and \param lde has correct lustre dirent.
7072  * \retval -ve on error
7073  */
7074 static inline int osd_it_ea_rec(const struct lu_env *env,
7075                                 const struct dt_it *di,
7076                                 struct dt_rec *dtrec, __u32 attr)
7077 {
7078         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7079         struct osd_object      *obj   = it->oie_obj;
7080         struct osd_device      *dev   = osd_obj2dev(obj);
7081         struct osd_thread_info *oti   = osd_oti_get(env);
7082         struct osd_inode_id    *id    = &oti->oti_id;
7083         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7084         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7085         __u32 ino = it->oie_dirent->oied_ino;
7086         int rc = 0;
7087
7088         ENTRY;
7089
7090         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7091
7092         if (attr & LUDA_VERIFY) {
7093                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7094                         attr |= LUDA_IGNORE;
7095                         /*
7096                          * If the parent is on remote MDT, and there
7097                          * is no FID-in-dirent, then we have to get
7098                          * the parent FID from the linkEA.
7099                          */
7100                         if (!fid_is_sane(fid) &&
7101                             it->oie_dirent->oied_namelen == 2 &&
7102                             it->oie_dirent->oied_name[0] == '.' &&
7103                             it->oie_dirent->oied_name[1] == '.')
7104                                 osd_get_pfid_from_linkea(env, obj, fid);
7105                 } else {
7106                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7107                                                      &attr);
7108                 }
7109
7110                 if (!fid_is_sane(fid))
7111                         attr |= LUDA_UNKNOWN;
7112         } else {
7113                 attr &= ~LU_DIRENT_ATTRS_MASK;
7114                 if (!fid_is_sane(fid)) {
7115                         bool is_dotdot = false;
7116
7117                         if (it->oie_dirent->oied_namelen == 2 &&
7118                             it->oie_dirent->oied_name[0] == '.' &&
7119                             it->oie_dirent->oied_name[1] == '.')
7120                                 is_dotdot = true;
7121                         /*
7122                          * If the parent is on remote MDT, and there
7123                          * is no FID-in-dirent, then we have to get
7124                          * the parent FID from the linkEA.
7125                          */
7126                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7127                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7128                         } else {
7129                                 if (is_dotdot == false &&
7130                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7131                                         RETURN(-ENOENT);
7132
7133                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7134                         }
7135                 } else {
7136                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7137                 }
7138         }
7139
7140         /* Pack the entry anyway, at least the offset is right. */
7141         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7142                            it->oie_dirent->oied_name,
7143                            it->oie_dirent->oied_namelen,
7144                            it->oie_dirent->oied_type, attr);
7145
7146         if (rc < 0)
7147                 RETURN(rc);
7148
7149         if (osd_remote_fid(env, dev, fid))
7150                 RETURN(0);
7151
7152         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7153                 osd_add_oi_cache(oti, dev, id, fid);
7154
7155         RETURN(rc > 0 ? 0 : rc);
7156 }
7157
7158 /**
7159  * Returns the record size size at current position.
7160  *
7161  * This function will return record(lu_dirent) size in bytes.
7162  *
7163  * \param[in] env       execution environment
7164  * \param[in] di        iterator's in memory structure
7165  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7166  *                      calculate the lu_dirent size.
7167  *
7168  * \retval      record size(in bytes & in memory) of the current lu_dirent
7169  *              entry.
7170  */
7171 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7172                               __u32 attr)
7173 {
7174         struct osd_it_ea *it = (struct osd_it_ea *)di;
7175
7176         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7177 }
7178
7179 /**
7180  * Returns a cookie for current position of the iterator head, so that
7181  * user can use this cookie to load/start the iterator next time.
7182  *
7183  * \param di iterator's in memory structure
7184  *
7185  * \retval cookie for current position, on success
7186  */
7187 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7188 {
7189         struct osd_it_ea *it = (struct osd_it_ea *)di;
7190
7191         return it->oie_dirent->oied_off;
7192 }
7193
7194 /**
7195  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7196  * to load a directory entry at a time and stored it i inn,
7197  * in iterator's in-memory data structure.
7198  *
7199  * \param di struct osd_it_ea, iterator's in memory structure
7200  *
7201  * \retval +ve on success
7202  * \retval -ve on error
7203  */
7204 static int osd_it_ea_load(const struct lu_env *env,
7205                           const struct dt_it *di, __u64 hash)
7206 {
7207         struct osd_it_ea *it = (struct osd_it_ea *)di;
7208         int rc;
7209
7210         ENTRY;
7211         it->oie_file.f_pos = hash;
7212
7213         rc =  osd_ldiskfs_it_fill(env, di);
7214         if (rc > 0)
7215                 rc = -ENODATA;
7216
7217         if (rc == 0)
7218                 rc = 1;
7219
7220         RETURN(rc);
7221 }
7222
7223 /**
7224  * Index lookup function for interoperability mode (b11826).
7225  *
7226  * \param key,  key i.e. file name to be searched
7227  *
7228  * \retval +ve, on success
7229  * \retval -ve, on error
7230  */
7231 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7232                                struct dt_rec *rec, const struct dt_key *key)
7233 {
7234         struct osd_object *obj = osd_dt_obj(dt);
7235         int rc = 0;
7236
7237         ENTRY;
7238
7239         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7240         LINVRNT(osd_invariant(obj));
7241
7242         rc = osd_ea_lookup_rec(env, obj, rec, key);
7243         if (rc == 0)
7244                 rc = 1;
7245         RETURN(rc);
7246 }
7247
7248 /**
7249  * Index and Iterator operations for interoperability
7250  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7251  */
7252 static const struct dt_index_operations osd_index_ea_ops = {
7253         .dio_lookup         = osd_index_ea_lookup,
7254         .dio_declare_insert = osd_index_declare_ea_insert,
7255         .dio_insert         = osd_index_ea_insert,
7256         .dio_declare_delete = osd_index_declare_ea_delete,
7257         .dio_delete         = osd_index_ea_delete,
7258         .dio_it     = {
7259                 .init     = osd_it_ea_init,
7260                 .fini     = osd_it_ea_fini,
7261                 .get      = osd_it_ea_get,
7262                 .put      = osd_it_ea_put,
7263                 .next     = osd_it_ea_next,
7264                 .key      = osd_it_ea_key,
7265                 .key_size = osd_it_ea_key_size,
7266                 .rec      = osd_it_ea_rec,
7267                 .rec_size = osd_it_ea_rec_size,
7268                 .store    = osd_it_ea_store,
7269                 .load     = osd_it_ea_load
7270         }
7271 };
7272
7273 static void *osd_key_init(const struct lu_context *ctx,
7274                           struct lu_context_key *key)
7275 {
7276         struct osd_thread_info *info;
7277
7278         OBD_ALLOC_PTR(info);
7279         if (info == NULL)
7280                 return ERR_PTR(-ENOMEM);
7281
7282         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7283         if (info->oti_it_ea_buf == NULL)
7284                 goto out_free_info;
7285
7286         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7287
7288         info->oti_hlock = ldiskfs_htree_lock_alloc();
7289         if (info->oti_hlock == NULL)
7290                 goto out_free_ea;
7291
7292         return info;
7293
7294 out_free_ea:
7295         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7296 out_free_info:
7297         OBD_FREE_PTR(info);
7298         return ERR_PTR(-ENOMEM);
7299 }
7300
7301 static void osd_key_fini(const struct lu_context *ctx,
7302                          struct lu_context_key *key, void *data)
7303 {
7304         struct osd_thread_info *info = data;
7305         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7306         struct osd_idmap_cache *idc = info->oti_ins_cache;
7307
7308         if (info->oti_inode != NULL)
7309                 OBD_FREE_PTR(lli);
7310         if (info->oti_hlock != NULL)
7311                 ldiskfs_htree_lock_free(info->oti_hlock);
7312         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7313         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7314         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7315         lu_buf_free(&info->oti_big_buf);
7316         if (idc != NULL) {
7317                 LASSERT(info->oti_ins_cache_size > 0);
7318                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7319                 info->oti_ins_cache = NULL;
7320                 info->oti_ins_cache_size = 0;
7321         }
7322         OBD_FREE_PTR(info);
7323 }
7324
7325 static void osd_key_exit(const struct lu_context *ctx,
7326                          struct lu_context_key *key, void *data)
7327 {
7328         struct osd_thread_info *info = data;
7329
7330         LASSERT(info->oti_r_locks == 0);
7331         LASSERT(info->oti_w_locks == 0);
7332         LASSERT(info->oti_txns    == 0);
7333 }
7334
7335 /* type constructor/destructor: osd_type_init, osd_type_fini */
7336 LU_TYPE_INIT_FINI(osd, &osd_key);
7337
7338 struct lu_context_key osd_key = {
7339         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7340         .lct_init = osd_key_init,
7341         .lct_fini = osd_key_fini,
7342         .lct_exit = osd_key_exit
7343 };
7344
7345
7346 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7347                            const char *name, struct lu_device *next)
7348 {
7349         struct osd_device *osd = osd_dev(d);
7350
7351         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7352             sizeof(osd->od_svname))
7353                 return -E2BIG;
7354         return osd_procfs_init(osd, name);
7355 }
7356
7357 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7358 {
7359         struct seq_server_site *ss = osd_seq_site(osd);
7360         int rc;
7361
7362         ENTRY;
7363
7364         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7365                 RETURN(0);
7366
7367         if (unlikely(ss == NULL))
7368                 RETURN(-ENODEV);
7369
7370         OBD_ALLOC_PTR(osd->od_cl_seq);
7371         if (osd->od_cl_seq == NULL)
7372                 RETURN(-ENOMEM);
7373
7374         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7375                              osd->od_svname, ss->ss_server_seq);
7376         if (rc != 0) {
7377                 OBD_FREE_PTR(osd->od_cl_seq);
7378                 osd->od_cl_seq = NULL;
7379                 RETURN(rc);
7380         }
7381
7382         if (ss->ss_node_id == 0) {
7383                 /*
7384                  * If the OSD on the sequence controller(MDT0), then allocate
7385                  * sequence here, otherwise allocate sequence after connected
7386                  * to MDT0 (see mdt_register_lwp_callback()).
7387                  */
7388                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7389                                    &osd->od_cl_seq->lcs_space, env);
7390         }
7391
7392         RETURN(rc);
7393 }
7394
7395 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7396 {
7397         if (osd->od_cl_seq == NULL)
7398                 return;
7399
7400         seq_client_fini(osd->od_cl_seq);
7401         OBD_FREE_PTR(osd->od_cl_seq);
7402         osd->od_cl_seq = NULL;
7403 }
7404
7405 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7406 {
7407         ENTRY;
7408
7409         /* shutdown quota slave instance associated with the device */
7410         if (o->od_quota_slave != NULL) {
7411                 struct qsd_instance *qsd = o->od_quota_slave;
7412
7413                 o->od_quota_slave = NULL;
7414                 qsd_fini(env, qsd);
7415         }
7416
7417         osd_fid_fini(env, o);
7418         osd_scrub_cleanup(env, o);
7419
7420         RETURN(0);
7421 }
7422
7423 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7424 {
7425         ENTRY;
7426
7427         if (o->od_mnt != NULL) {
7428                 shrink_dcache_sb(osd_sb(o));
7429                 osd_sync(env, &o->od_dt_dev);
7430
7431                 mntput(o->od_mnt);
7432                 o->od_mnt = NULL;
7433         }
7434
7435         EXIT;
7436 }
7437
7438 static int osd_mount(const struct lu_env *env,
7439                      struct osd_device *o, struct lustre_cfg *cfg)
7440 {
7441         const char *name = lustre_cfg_string(cfg, 0);
7442         const char *dev = lustre_cfg_string(cfg, 1);
7443         const char *opts;
7444         unsigned long page, s_flags, lmd_flags = 0;
7445         struct page *__page;
7446         struct file_system_type *type;
7447         char *options = NULL;
7448         char *str;
7449         struct osd_thread_info *info = osd_oti_get(env);
7450         struct lu_fid *fid = &info->oti_fid;
7451         struct inode *inode;
7452         int rc = 0, force_over_512tb = 0;
7453
7454         ENTRY;
7455
7456         if (o->od_mnt != NULL)
7457                 RETURN(0);
7458
7459         if (strlen(dev) >= sizeof(o->od_mntdev))
7460                 RETURN(-E2BIG);
7461         strcpy(o->od_mntdev, dev);
7462
7463         str = lustre_cfg_string(cfg, 2);
7464         s_flags = simple_strtoul(str, NULL, 0);
7465         str = strstr(str, ":");
7466         if (str)
7467                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7468         opts = lustre_cfg_string(cfg, 3);
7469 #ifdef __BIG_ENDIAN
7470         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7471                 CERROR("%s: device %s extents feature is not guaranteed to "
7472                        "work on big-endian systems. Use \"bigendian_extents\" "
7473                        "mount option to override.\n", name, dev);
7474                 RETURN(-EINVAL);
7475         }
7476 #endif
7477 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7478         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7479                 CWARN("force_over_128tb option is deprecated. "
7480                       "Filesystems less than 512TB can be created without any "
7481                       "force options. Use force_over_512tb option for "
7482                       "filesystems greater than 512TB.\n");
7483         }
7484 #endif
7485 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7486         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7487                 CWARN("force_over_256tb option is deprecated. "
7488                       "Filesystems less than 512TB can be created without any "
7489                       "force options. Use force_over_512tb option for "
7490                       "filesystems greater than 512TB.\n");
7491         }
7492 #endif
7493
7494         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7495                 force_over_512tb = 1;
7496
7497         __page = alloc_page(GFP_KERNEL);
7498         if (__page == NULL)
7499                 GOTO(out, rc = -ENOMEM);
7500         page = (unsigned long)page_address(__page);
7501         options = (char *)page;
7502         *options = '\0';
7503         if (opts != NULL) {
7504                 /* strip out the options for back compatiblity */
7505                 static char *sout[] = {
7506                         "mballoc",
7507                         "iopen",
7508                         "noiopen",
7509                         "iopen_nopriv",
7510                         "extents",
7511                         "noextents",
7512                         /* strip out option we processed in osd */
7513                         "bigendian_extents",
7514                         "force_over_128tb",
7515                         "force_over_256tb",
7516                         "force_over_512tb",
7517                         NULL
7518                 };
7519                 strcat(options, opts);
7520                 for (rc = 0, str = options; sout[rc]; ) {
7521                         char *op = strstr(str, sout[rc]);
7522
7523                         if (op == NULL) {
7524                                 rc++;
7525                                 str = options;
7526                                 continue;
7527                         }
7528                         if (op == options || *(op - 1) == ',') {
7529                                 str = op + strlen(sout[rc]);
7530                                 if (*str == ',' || *str == '\0') {
7531                                         *str == ',' ? str++ : str;
7532                                         memmove(op, str, strlen(str) + 1);
7533                                 }
7534                         }
7535                         for (str = op; *str != ',' && *str != '\0'; str++)
7536                                 ;
7537                 }
7538         } else {
7539                 strncat(options, "user_xattr,acl", 14);
7540         }
7541
7542         /* Glom up mount options */
7543         if (*options != '\0')
7544                 strcat(options, ",");
7545         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7546
7547         type = get_fs_type("ldiskfs");
7548         if (!type) {
7549                 CERROR("%s: cannot find ldiskfs module\n", name);
7550                 GOTO(out, rc = -ENODEV);
7551         }
7552
7553         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7554         module_put(type->owner);
7555
7556         if (IS_ERR(o->od_mnt)) {
7557                 rc = PTR_ERR(o->od_mnt);
7558                 o->od_mnt = NULL;
7559                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7560                 GOTO(out, rc);
7561         }
7562
7563         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7564                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7565                                  force_over_512tb == 0) {
7566                 CERROR("%s: device %s LDISKFS does not support filesystems "
7567                        "greater than 512TB and can cause data corruption. "
7568                        "Use \"force_over_512tb\" mount option to override.\n",
7569                        name, dev);
7570                 GOTO(out_mnt, rc = -EINVAL);
7571         }
7572
7573         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7574                 if (priv_dev_set_rdonly) {
7575                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7576                         o->od_dt_dev.dd_rdonly = 1;
7577                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7578                                       name);
7579                 } else {
7580                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7581                                       name);
7582
7583                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7584                 }
7585         } else if (priv_dev_check_rdonly &&
7586                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7587                 CERROR("%s: underlying device %s is marked as "
7588                        "read-only. Setup failed\n", name, dev);
7589
7590                 GOTO(out_mnt, rc = -EROFS);
7591         }
7592
7593         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7594                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7595                 GOTO(out_mnt, rc = -EINVAL);
7596         }
7597
7598 #ifdef LDISKFS_MOUNT_DIRDATA
7599         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7600                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7601         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7602                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7603                       "enabling the dirdata feature. If you do not want to "
7604                       "downgrade to Lustre-1.x again, you can enable it via "
7605                       "'tune2fs -O dirdata device'\n", name, dev);
7606 #endif
7607         inode = osd_sb(o)->s_root->d_inode;
7608         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7609         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7610         if (rc != 0) {
7611                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7612                 GOTO(out_mnt, rc);
7613         }
7614
7615         if (lmd_flags & LMD_FLG_NOSCRUB)
7616                 o->od_auto_scrub_interval = AS_NEVER;
7617
7618         GOTO(out, rc = 0);
7619
7620 out_mnt:
7621         mntput(o->od_mnt);
7622         o->od_mnt = NULL;
7623
7624 out:
7625         if (__page)
7626                 __free_page(__page);
7627
7628         return rc;
7629 }
7630
7631 static struct lu_device *osd_device_fini(const struct lu_env *env,
7632                                          struct lu_device *d)
7633 {
7634         struct osd_device *o = osd_dev(d);
7635
7636         ENTRY;
7637
7638         osd_index_backup(env, o, false);
7639         osd_shutdown(env, o);
7640         osd_procfs_fini(o);
7641         osd_obj_map_fini(o);
7642         osd_umount(env, o);
7643
7644         RETURN(NULL);
7645 }
7646
7647 static int osd_device_init0(const struct lu_env *env,
7648                             struct osd_device *o,
7649                             struct lustre_cfg *cfg)
7650 {
7651         struct lu_device *l = osd2lu_dev(o);
7652         struct osd_thread_info *info;
7653         int rc;
7654         int cplen = 0;
7655
7656         /* if the module was re-loaded, env can loose its keys */
7657         rc = lu_env_refill((struct lu_env *)env);
7658         if (rc)
7659                 GOTO(out, rc);
7660         info = osd_oti_get(env);
7661         LASSERT(info);
7662
7663         l->ld_ops = &osd_lu_ops;
7664         o->od_dt_dev.dd_ops = &osd_dt_ops;
7665
7666         spin_lock_init(&o->od_osfs_lock);
7667         mutex_init(&o->od_otable_mutex);
7668         INIT_LIST_HEAD(&o->od_orphan_list);
7669         INIT_LIST_HEAD(&o->od_index_backup_list);
7670         INIT_LIST_HEAD(&o->od_index_restore_list);
7671         spin_lock_init(&o->od_lock);
7672         o->od_index_backup_policy = LIBP_NONE;
7673
7674         o->od_read_cache = 1;
7675         o->od_writethrough_cache = 1;
7676         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7677         o->od_auto_scrub_interval = AS_DEFAULT;
7678
7679         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7680                         sizeof(o->od_svname));
7681         if (cplen >= sizeof(o->od_svname)) {
7682                 rc = -E2BIG;
7683                 GOTO(out, rc);
7684         }
7685
7686         o->od_index_backup_stop = 0;
7687         o->od_index = -1; /* -1 means index is invalid */
7688         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7689         if (rc == LDD_F_SV_TYPE_OST)
7690                 o->od_is_ost = 1;
7691
7692         o->od_full_scrub_ratio = OFSR_DEFAULT;
7693         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7694         rc = osd_mount(env, o, cfg);
7695         if (rc != 0)
7696                 GOTO(out, rc);
7697
7698         rc = osd_obj_map_init(env, o);
7699         if (rc != 0)
7700                 GOTO(out_mnt, rc);
7701
7702         rc = lu_site_init(&o->od_site, l);
7703         if (rc != 0)
7704                 GOTO(out_compat, rc);
7705         o->od_site.ls_bottom_dev = l;
7706
7707         rc = lu_site_init_finish(&o->od_site);
7708         if (rc != 0)
7709                 GOTO(out_site, rc);
7710
7711         INIT_LIST_HEAD(&o->od_ios_list);
7712         /* setup scrub, including OI files initialization */
7713         o->od_in_init = 1;
7714         rc = osd_scrub_setup(env, o);
7715         o->od_in_init = 0;
7716         if (rc < 0)
7717                 GOTO(out_site, rc);
7718
7719         rc = osd_procfs_init(o, o->od_svname);
7720         if (rc != 0) {
7721                 CERROR("%s: can't initialize procfs: rc = %d\n",
7722                        o->od_svname, rc);
7723                 GOTO(out_scrub, rc);
7724         }
7725
7726         LASSERT(l->ld_site->ls_linkage.next != NULL);
7727         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7728
7729         /* initialize quota slave instance */
7730         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7731                                      o->od_proc_entry);
7732         if (IS_ERR(o->od_quota_slave)) {
7733                 rc = PTR_ERR(o->od_quota_slave);
7734                 o->od_quota_slave = NULL;
7735                 GOTO(out_procfs, rc);
7736         }
7737
7738         RETURN(0);
7739
7740 out_procfs:
7741         osd_procfs_fini(o);
7742 out_scrub:
7743         osd_scrub_cleanup(env, o);
7744 out_site:
7745         lu_site_fini(&o->od_site);
7746 out_compat:
7747         osd_obj_map_fini(o);
7748 out_mnt:
7749         osd_umount(env, o);
7750 out:
7751         return rc;
7752 }
7753
7754 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7755                                           struct lu_device_type *t,
7756                                           struct lustre_cfg *cfg)
7757 {
7758         struct osd_device *o;
7759         int rc;
7760
7761         OBD_ALLOC_PTR(o);
7762         if (o == NULL)
7763                 return ERR_PTR(-ENOMEM);
7764
7765         rc = dt_device_init(&o->od_dt_dev, t);
7766         if (rc == 0) {
7767                 /*
7768                  * Because the ctx might be revived in dt_device_init,
7769                  * refill the env here
7770                  */
7771                 lu_env_refill((struct lu_env *)env);
7772                 rc = osd_device_init0(env, o, cfg);
7773                 if (rc)
7774                         dt_device_fini(&o->od_dt_dev);
7775         }
7776
7777         if (unlikely(rc != 0))
7778                 OBD_FREE_PTR(o);
7779
7780         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7781 }
7782
7783 static struct lu_device *osd_device_free(const struct lu_env *env,
7784                                          struct lu_device *d)
7785 {
7786         struct osd_device *o = osd_dev(d);
7787
7788         ENTRY;
7789
7790         /* XXX: make osd top device in order to release reference */
7791         d->ld_site->ls_top_dev = d;
7792         lu_site_purge(env, d->ld_site, -1);
7793         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7794                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7795                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7796         }
7797         lu_site_fini(&o->od_site);
7798         dt_device_fini(&o->od_dt_dev);
7799         OBD_FREE_PTR(o);
7800         RETURN(NULL);
7801 }
7802
7803 static int osd_process_config(const struct lu_env *env,
7804                               struct lu_device *d, struct lustre_cfg *cfg)
7805 {
7806         struct osd_device *o = osd_dev(d);
7807         int rc;
7808
7809         ENTRY;
7810
7811         switch (cfg->lcfg_command) {
7812         case LCFG_SETUP:
7813                 rc = osd_mount(env, o, cfg);
7814                 break;
7815         case LCFG_CLEANUP:
7816                 /*
7817                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7818                  * that may happend if hit failure during mount process.
7819                  */
7820                 osd_index_backup(env, o, false);
7821                 lu_dev_del_linkage(d->ld_site, d);
7822                 rc = osd_shutdown(env, o);
7823                 break;
7824         case LCFG_PARAM:
7825                 LASSERT(&o->od_dt_dev);
7826                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7827                                               cfg, &o->od_dt_dev);
7828                 if (rc > 0 || rc == -ENOSYS) {
7829                         rc = class_process_proc_param(PARAM_OST,
7830                                                       lprocfs_osd_obd_vars,
7831                                                       cfg, &o->od_dt_dev);
7832                         if (rc > 0)
7833                                 rc = 0;
7834                 }
7835                 break;
7836         case LCFG_PRE_CLEANUP:
7837                 osd_scrub_stop(o);
7838                 osd_index_backup(env, o,
7839                                  o->od_index_backup_policy != LIBP_NONE);
7840                 rc = 0;
7841                 break;
7842         default:
7843                 rc = -ENOSYS;
7844         }
7845
7846         RETURN(rc);
7847 }
7848
7849 static int osd_recovery_complete(const struct lu_env *env,
7850                                  struct lu_device *d)
7851 {
7852         struct osd_device *osd = osd_dev(d);
7853         int rc = 0;
7854
7855         ENTRY;
7856
7857         if (osd->od_quota_slave == NULL)
7858                 RETURN(0);
7859
7860         /*
7861          * start qsd instance on recovery completion, this notifies the quota
7862          * slave code that we are about to process new requests now
7863          */
7864         rc = qsd_start(env, osd->od_quota_slave);
7865         RETURN(rc);
7866 }
7867
7868 /*
7869  * we use exports to track all osd users
7870  */
7871 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7872                            struct obd_device *obd, struct obd_uuid *cluuid,
7873                            struct obd_connect_data *data, void *localdata)
7874 {
7875         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7876         struct lustre_handle conn;
7877         int rc;
7878
7879         ENTRY;
7880
7881         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7882
7883         rc = class_connect(&conn, obd, cluuid);
7884         if (rc)
7885                 RETURN(rc);
7886
7887         *exp = class_conn2export(&conn);
7888
7889         spin_lock(&osd->od_osfs_lock);
7890         osd->od_connects++;
7891         spin_unlock(&osd->od_osfs_lock);
7892
7893         RETURN(0);
7894 }
7895
7896 /*
7897  * once last export (we don't count self-export) disappeared
7898  * osd can be released
7899  */
7900 static int osd_obd_disconnect(struct obd_export *exp)
7901 {
7902         struct obd_device *obd = exp->exp_obd;
7903         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7904         int rc, release = 0;
7905
7906         ENTRY;
7907
7908         /* Only disconnect the underlying layers on the final disconnect. */
7909         spin_lock(&osd->od_osfs_lock);
7910         osd->od_connects--;
7911         if (osd->od_connects == 0)
7912                 release = 1;
7913         spin_unlock(&osd->od_osfs_lock);
7914
7915         rc = class_disconnect(exp); /* bz 9811 */
7916
7917         if (rc == 0 && release)
7918                 class_manual_cleanup(obd);
7919         RETURN(rc);
7920 }
7921
7922 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7923                        struct lu_device *dev)
7924 {
7925         struct osd_device *osd = osd_dev(dev);
7926         struct lr_server_data *lsd =
7927                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7928         int result = 0;
7929
7930         ENTRY;
7931
7932         if (osd->od_quota_slave != NULL) {
7933                 /* set up quota slave objects */
7934                 result = qsd_prepare(env, osd->od_quota_slave);
7935                 if (result != 0)
7936                         RETURN(result);
7937         }
7938
7939         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7940 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7941                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7942                         osd->od_index_in_idif = 1;
7943                 } else {
7944                         osd->od_index_in_idif = 0;
7945                         result = osd_register_proc_index_in_idif(osd);
7946                         if (result != 0)
7947                                 RETURN(result);
7948                 }
7949 #else
7950                 osd->od_index_in_idif = 1;
7951 #endif
7952         }
7953
7954         result = osd_fid_init(env, osd);
7955
7956         RETURN(result);
7957 }
7958
7959 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7960                          struct lu_fid *fid, struct md_op_data *op_data)
7961 {
7962         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7963
7964         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7965 }
7966
7967 static const struct lu_object_operations osd_lu_obj_ops = {
7968         .loo_object_init      = osd_object_init,
7969         .loo_object_delete    = osd_object_delete,
7970         .loo_object_release   = osd_object_release,
7971         .loo_object_free      = osd_object_free,
7972         .loo_object_print     = osd_object_print,
7973         .loo_object_invariant = osd_object_invariant
7974 };
7975
7976 const struct lu_device_operations osd_lu_ops = {
7977         .ldo_object_alloc      = osd_object_alloc,
7978         .ldo_process_config    = osd_process_config,
7979         .ldo_recovery_complete = osd_recovery_complete,
7980         .ldo_prepare           = osd_prepare,
7981 };
7982
7983 static const struct lu_device_type_operations osd_device_type_ops = {
7984         .ldto_init = osd_type_init,
7985         .ldto_fini = osd_type_fini,
7986
7987         .ldto_start = osd_type_start,
7988         .ldto_stop  = osd_type_stop,
7989
7990         .ldto_device_alloc = osd_device_alloc,
7991         .ldto_device_free  = osd_device_free,
7992
7993         .ldto_device_init = osd_device_init,
7994         .ldto_device_fini = osd_device_fini
7995 };
7996
7997 static struct lu_device_type osd_device_type = {
7998         .ldt_tags     = LU_DEVICE_DT,
7999         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8000         .ldt_ops      = &osd_device_type_ops,
8001         .ldt_ctx_tags = LCT_LOCAL,
8002 };
8003
8004 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8005 {
8006         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8007         struct super_block *sb = osd_sb(osd);
8008
8009         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8010 }
8011
8012 /*
8013  * lprocfs legacy support.
8014  */
8015 static struct obd_ops osd_obd_device_ops = {
8016         .o_owner = THIS_MODULE,
8017         .o_connect      = osd_obd_connect,
8018         .o_disconnect   = osd_obd_disconnect,
8019         .o_fid_alloc    = osd_fid_alloc,
8020         .o_health_check = osd_health_check,
8021 };
8022
8023 static int __init osd_init(void)
8024 {
8025         int rc;
8026
8027         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8028 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8029         /* please, try to keep osd_thread_info smaller than a page */
8030         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8031 #endif
8032
8033         osd_oi_mod_init();
8034
8035         rc = lu_kmem_init(ldiskfs_caches);
8036         if (rc)
8037                 return rc;
8038
8039 #ifdef CONFIG_KALLSYMS
8040         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8041         priv_dev_check_rdonly =
8042                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8043 #endif
8044
8045         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8046                                  lprocfs_osd_module_vars,
8047                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8048         if (rc)
8049                 lu_kmem_fini(ldiskfs_caches);
8050         return rc;
8051 }
8052
8053 static void __exit osd_exit(void)
8054 {
8055         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8056         lu_kmem_fini(ldiskfs_caches);
8057 }
8058
8059 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8060 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8061 MODULE_VERSION(LUSTRE_VERSION_STRING);
8062 MODULE_LICENSE("GPL");
8063
8064 module_init(osd_init);
8065 module_exit(osd_exit);