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
2344         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2345
2346         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2347
2348         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2349                rc);
2350
2351         return rc;
2352 }
2353
2354 /**
2355  * Start commit for OSD device.
2356  *
2357  * An implementation of dt_commit_async method for OSD device.
2358  * Asychronously starts underlayng fs sync and thereby a transaction
2359  * commit.
2360  *
2361  * \param env environment
2362  * \param d dt device
2363  *
2364  * \see dt_device_operations
2365  */
2366 static int osd_commit_async(const struct lu_env *env,
2367                             struct dt_device *d)
2368 {
2369         struct super_block *s = osd_sb(osd_dt_dev(d));
2370
2371         ENTRY;
2372
2373         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2374         RETURN(s->s_op->sync_fs(s, 0));
2375 }
2376
2377 /* Our own copy of the set readonly functions if present, or NU if not. */
2378 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2379 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2380 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2381
2382 /*
2383  * Concurrency: shouldn't matter.
2384  */
2385 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2386 {
2387         struct super_block *sb = osd_sb(osd_dt_dev(d));
2388         struct block_device *dev = sb->s_bdev;
2389         int rc = -EOPNOTSUPP;
2390
2391         ENTRY;
2392
2393         if (priv_dev_set_rdonly) {
2394                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2395
2396                 rc = 0;
2397                 CERROR("*** setting %s read-only ***\n",
2398                        osd_dt_dev(d)->od_svname);
2399
2400                 if (sb->s_op->freeze_fs) {
2401                         rc = sb->s_op->freeze_fs(sb);
2402                         if (rc)
2403                                 goto out;
2404                 }
2405
2406                 if (jdev && (jdev != dev)) {
2407                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2408                                (long)jdev);
2409                         priv_dev_set_rdonly(jdev);
2410                 }
2411                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2412                 priv_dev_set_rdonly(dev);
2413
2414                 if (sb->s_op->unfreeze_fs)
2415                         sb->s_op->unfreeze_fs(sb);
2416         }
2417
2418 out:
2419         if (rc)
2420                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2421                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2422
2423         RETURN(rc);
2424 }
2425
2426 /**
2427  * Note: we do not count into QUOTA here.
2428  * If we mount with --data_journal we may need more.
2429  */
2430 const int osd_dto_credits_noquota[DTO_NR] = {
2431         /**
2432          * Insert.
2433          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2434          * SINGLEDATA_TRANS_BLOCKS(8)
2435          * XXX Note: maybe iam need more, since iam have more level than
2436          *           EXT3 htree.
2437          */
2438         [DTO_INDEX_INSERT]  = 16,
2439         /**
2440          * Delete
2441          * just modify a single entry, probably merge few within a block
2442          */
2443         [DTO_INDEX_DELETE]  = 1,
2444         /**
2445          * Used for OI scrub
2446          */
2447         [DTO_INDEX_UPDATE]  = 16,
2448         /**
2449          * 4(inode, inode bits, groups, GDT)
2450          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2451          */
2452         [DTO_OBJECT_CREATE] = 4,
2453         /**
2454          * 4(inode, inode bits, groups, GDT)
2455          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2456          */
2457         [DTO_OBJECT_DELETE] = 4,
2458         /**
2459          * Attr set credits (inode)
2460          */
2461         [DTO_ATTR_SET_BASE] = 1,
2462         /**
2463          * Xattr set. The same as xattr of EXT3.
2464          * DATA_TRANS_BLOCKS(14)
2465          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2466          * are also counted in. Do not know why?
2467          */
2468         [DTO_XATTR_SET]     = 14,
2469         /**
2470          * credits for inode change during write.
2471          */
2472         [DTO_WRITE_BASE]    = 3,
2473         /**
2474          * credits for single block write.
2475          */
2476         [DTO_WRITE_BLOCK]   = 14,
2477         /**
2478          * Attr set credits for chown.
2479          * This is extra credits for setattr, and it is null without quota
2480          */
2481         [DTO_ATTR_SET_CHOWN] = 0
2482 };
2483
2484 static const struct dt_device_operations osd_dt_ops = {
2485         .dt_root_get       = osd_root_get,
2486         .dt_statfs         = osd_statfs,
2487         .dt_trans_create   = osd_trans_create,
2488         .dt_trans_start    = osd_trans_start,
2489         .dt_trans_stop     = osd_trans_stop,
2490         .dt_trans_cb_add   = osd_trans_cb_add,
2491         .dt_conf_get       = osd_conf_get,
2492         .dt_sync           = osd_sync,
2493         .dt_ro             = osd_ro,
2494         .dt_commit_async   = osd_commit_async,
2495 };
2496
2497 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2498                           unsigned int role)
2499 {
2500         struct osd_object *obj = osd_dt_obj(dt);
2501         struct osd_thread_info *oti = osd_oti_get(env);
2502
2503         LINVRNT(osd_invariant(obj));
2504
2505         LASSERT(obj->oo_owner != env);
2506         down_read_nested(&obj->oo_sem, role);
2507
2508         LASSERT(obj->oo_owner == NULL);
2509         oti->oti_r_locks++;
2510 }
2511
2512 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2513                            unsigned int role)
2514 {
2515         struct osd_object *obj = osd_dt_obj(dt);
2516         struct osd_thread_info *oti = osd_oti_get(env);
2517
2518         LINVRNT(osd_invariant(obj));
2519
2520         LASSERT(obj->oo_owner != env);
2521         down_write_nested(&obj->oo_sem, role);
2522
2523         LASSERT(obj->oo_owner == NULL);
2524         obj->oo_owner = env;
2525         oti->oti_w_locks++;
2526 }
2527
2528 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2529 {
2530         struct osd_object *obj = osd_dt_obj(dt);
2531         struct osd_thread_info *oti = osd_oti_get(env);
2532
2533         LINVRNT(osd_invariant(obj));
2534
2535         LASSERT(oti->oti_r_locks > 0);
2536         oti->oti_r_locks--;
2537         up_read(&obj->oo_sem);
2538 }
2539
2540 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2541 {
2542         struct osd_object *obj = osd_dt_obj(dt);
2543         struct osd_thread_info *oti = osd_oti_get(env);
2544
2545         LINVRNT(osd_invariant(obj));
2546
2547         LASSERT(obj->oo_owner == env);
2548         LASSERT(oti->oti_w_locks > 0);
2549         oti->oti_w_locks--;
2550         obj->oo_owner = NULL;
2551         up_write(&obj->oo_sem);
2552 }
2553
2554 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2555 {
2556         struct osd_object *obj = osd_dt_obj(dt);
2557
2558         LINVRNT(osd_invariant(obj));
2559
2560         return obj->oo_owner == env;
2561 }
2562
2563 static struct timespec *osd_inode_time(const struct lu_env *env,
2564                                        struct inode *inode, __u64 seconds)
2565 {
2566         struct osd_thread_info *oti = osd_oti_get(env);
2567         struct timespec *t = &oti->oti_time;
2568
2569         t->tv_sec = seconds;
2570         t->tv_nsec = 0;
2571         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2572         return t;
2573 }
2574
2575 static void osd_inode_getattr(const struct lu_env *env,
2576                               struct inode *inode, struct lu_attr *attr)
2577 {
2578         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2579                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2580                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2581                            LA_BLKSIZE | LA_TYPE;
2582
2583         attr->la_atime   = LTIME_S(inode->i_atime);
2584         attr->la_mtime   = LTIME_S(inode->i_mtime);
2585         attr->la_ctime   = LTIME_S(inode->i_ctime);
2586         attr->la_mode    = inode->i_mode;
2587         attr->la_size    = i_size_read(inode);
2588         attr->la_blocks  = inode->i_blocks;
2589         attr->la_uid     = i_uid_read(inode);
2590         attr->la_gid     = i_gid_read(inode);
2591         attr->la_projid  = i_projid_read(inode);
2592         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2593         attr->la_nlink   = inode->i_nlink;
2594         attr->la_rdev    = inode->i_rdev;
2595         attr->la_blksize = 1 << inode->i_blkbits;
2596         attr->la_blkbits = inode->i_blkbits;
2597         /*
2598          * Ext4 did not transfer inherit flags from raw inode
2599          * to inode flags, and ext4 internally test raw inode
2600          * @i_flags directly. Instead of patching ext4, we do it here.
2601          */
2602         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2603                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2604 }
2605
2606 static int osd_attr_get(const struct lu_env *env,
2607                         struct dt_object *dt,
2608                         struct lu_attr *attr)
2609 {
2610         struct osd_object *obj = osd_dt_obj(dt);
2611
2612         if (unlikely(!dt_object_exists(dt)))
2613                 return -ENOENT;
2614         if (unlikely(obj->oo_destroyed))
2615                 return -ENOENT;
2616
2617         LASSERT(!dt_object_remote(dt));
2618         LINVRNT(osd_invariant(obj));
2619
2620         spin_lock(&obj->oo_guard);
2621         osd_inode_getattr(env, obj->oo_inode, attr);
2622         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2623                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2624         spin_unlock(&obj->oo_guard);
2625
2626         return 0;
2627 }
2628
2629 static int osd_declare_attr_qid(const struct lu_env *env,
2630                                 struct osd_object *obj,
2631                                 struct osd_thandle *oh, long long bspace,
2632                                 qid_t old_id, qid_t new_id, bool enforce,
2633                                 unsigned int type, bool ignore_edquot)
2634 {
2635         int rc;
2636         struct osd_thread_info *info = osd_oti_get(env);
2637         struct lquota_id_info  *qi = &info->oti_qi;
2638
2639         qi->lqi_type = type;
2640         /* inode accounting */
2641         qi->lqi_is_blk = false;
2642
2643         /* one more inode for the new id ... */
2644         qi->lqi_id.qid_uid = new_id;
2645         qi->lqi_space      = 1;
2646         /* Reserve credits for the new id */
2647         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2648         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2649                 rc = 0;
2650         if (rc)
2651                 RETURN(rc);
2652
2653         /* and one less inode for the current id */
2654         qi->lqi_id.qid_uid = old_id;
2655         qi->lqi_space = -1;
2656         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2657         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2658                 rc = 0;
2659         if (rc)
2660                 RETURN(rc);
2661
2662         /* block accounting */
2663         qi->lqi_is_blk = true;
2664
2665         /* more blocks for the new id ... */
2666         qi->lqi_id.qid_uid = new_id;
2667         qi->lqi_space      = bspace;
2668         /*
2669          * Credits for the new uid has been reserved, re-use "obj"
2670          * to save credit reservation.
2671          */
2672         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2673         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2674                 rc = 0;
2675         if (rc)
2676                 RETURN(rc);
2677
2678         /* and finally less blocks for the current uid */
2679         qi->lqi_id.qid_uid = old_id;
2680         qi->lqi_space      = -bspace;
2681         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2682         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2683                 rc = 0;
2684
2685         RETURN(rc);
2686 }
2687
2688 static int osd_declare_attr_set(const struct lu_env *env,
2689                                 struct dt_object *dt,
2690                                 const struct lu_attr *attr,
2691                                 struct thandle *handle)
2692 {
2693         struct osd_thandle *oh;
2694         struct osd_object *obj;
2695         qid_t uid;
2696         qid_t gid;
2697         long long bspace;
2698         int rc = 0;
2699         bool enforce;
2700
2701         ENTRY;
2702
2703         LASSERT(dt != NULL);
2704         LASSERT(handle != NULL);
2705
2706         obj = osd_dt_obj(dt);
2707         LASSERT(osd_invariant(obj));
2708
2709         oh = container_of0(handle, struct osd_thandle, ot_super);
2710         LASSERT(oh->ot_handle == NULL);
2711
2712         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2713                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2714
2715         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2716                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2717
2718         if (attr == NULL || obj->oo_inode == NULL)
2719                 RETURN(rc);
2720
2721         bspace   = obj->oo_inode->i_blocks << 9;
2722         bspace   = toqb(bspace);
2723
2724         /*
2725          * Changing ownership is always preformed by super user, it should not
2726          * fail with EDQUOT unless required explicitly.
2727          *
2728          * We still need to call the osd_declare_qid() to calculate the journal
2729          * credits for updating quota accounting files and to trigger quota
2730          * space adjustment once the operation is completed.
2731          */
2732         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2733                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2734
2735                 if (!ignore_edquot)
2736                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2737                                "(the quota space is %lld)\n",
2738                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2739                                attr->la_gid, bspace);
2740
2741                 /* USERQUOTA */
2742                 uid = i_uid_read(obj->oo_inode);
2743                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2744                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2745                                           attr->la_uid, enforce, USRQUOTA,
2746                                           true);
2747                 if (rc)
2748                         RETURN(rc);
2749
2750                 gid = i_gid_read(obj->oo_inode);
2751                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2752                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2753                                           i_gid_read(obj->oo_inode),
2754                                           attr->la_gid, enforce, GRPQUOTA,
2755                                           ignore_edquot);
2756                 if (rc)
2757                         RETURN(rc);
2758
2759         }
2760 #ifdef HAVE_PROJECT_QUOTA
2761         if (attr->la_valid & LA_PROJID) {
2762                 __u32 projid = i_projid_read(obj->oo_inode);
2763
2764                 enforce = (attr->la_valid & LA_PROJID) &&
2765                                         (attr->la_projid != projid);
2766                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2767                                           (qid_t)projid, (qid_t)attr->la_projid,
2768                                           enforce, PRJQUOTA, true);
2769                 if (rc)
2770                         RETURN(rc);
2771         }
2772 #endif
2773         RETURN(rc);
2774 }
2775
2776 static int osd_inode_setattr(const struct lu_env *env,
2777                              struct inode *inode, const struct lu_attr *attr)
2778 {
2779         __u64 bits = attr->la_valid;
2780
2781         /* Only allow set size for regular file */
2782         if (!S_ISREG(inode->i_mode))
2783                 bits &= ~(LA_SIZE | LA_BLOCKS);
2784
2785         if (bits == 0)
2786                 return 0;
2787
2788         if (bits & LA_ATIME)
2789                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2790         if (bits & LA_CTIME)
2791                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2792         if (bits & LA_MTIME)
2793                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2794         if (bits & LA_SIZE) {
2795                 spin_lock(&inode->i_lock);
2796                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2797                 i_size_write(inode, attr->la_size);
2798                 spin_unlock(&inode->i_lock);
2799         }
2800
2801         /*
2802          * OSD should not change "i_blocks" which is used by quota.
2803          * "i_blocks" should be changed by ldiskfs only.
2804          */
2805         if (bits & LA_MODE)
2806                 inode->i_mode = (inode->i_mode & S_IFMT) |
2807                                 (attr->la_mode & ~S_IFMT);
2808         if (bits & LA_UID)
2809                 i_uid_write(inode, attr->la_uid);
2810         if (bits & LA_GID)
2811                 i_gid_write(inode, attr->la_gid);
2812         if (bits & LA_PROJID)
2813                 i_projid_write(inode, attr->la_projid);
2814         if (bits & LA_NLINK)
2815                 set_nlink(inode, attr->la_nlink);
2816         if (bits & LA_RDEV)
2817                 inode->i_rdev = attr->la_rdev;
2818
2819         if (bits & LA_FLAGS) {
2820                 /* always keep S_NOCMTIME */
2821                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2822                                  S_NOCMTIME;
2823                 /*
2824                  * Ext4 did not transfer inherit flags from
2825                  * @inode->i_flags to raw inode i_flags when writing
2826                  * flags, we do it explictly here.
2827                  */
2828                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2829                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2830                 else
2831                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2832         }
2833         return 0;
2834 }
2835
2836 #ifdef HAVE_PROJECT_QUOTA
2837 static int osd_transfer_project(struct inode *inode, __u32 projid)
2838 {
2839         struct super_block *sb = inode->i_sb;
2840         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2841         int err;
2842         kprojid_t kprojid;
2843         struct ldiskfs_iloc iloc;
2844         struct ldiskfs_inode *raw_inode;
2845         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2846
2847         if (!ldiskfs_has_feature_project(sb)) {
2848                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2849                         == LDISKFS_DEF_PROJID);
2850                 if (projid != LDISKFS_DEF_PROJID)
2851                         return -EOPNOTSUPP;
2852                 else
2853                         return 0;
2854         }
2855
2856         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2857                 return -EOPNOTSUPP;
2858
2859         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2860         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2861                 return 0;
2862
2863         err = ldiskfs_get_inode_loc(inode, &iloc);
2864         if (err)
2865                 return err;
2866
2867         raw_inode = ldiskfs_raw_inode(&iloc);
2868         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2869                 err = -EOVERFLOW;
2870                 brelse(iloc.bh);
2871                 return err;
2872         }
2873         brelse(iloc.bh);
2874
2875         dquot_initialize(inode);
2876         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2877         if (transfer_to[PRJQUOTA]) {
2878                 err = __dquot_transfer(inode, transfer_to);
2879                 dqput(transfer_to[PRJQUOTA]);
2880                 if (err)
2881                         return err;
2882         }
2883
2884         return err;
2885 }
2886 #endif
2887
2888 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2889 {
2890         int rc;
2891
2892         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2893             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2894                 struct iattr iattr;
2895
2896                 ll_vfs_dq_init(inode);
2897                 iattr.ia_valid = 0;
2898                 if (attr->la_valid & LA_UID)
2899                         iattr.ia_valid |= ATTR_UID;
2900                 if (attr->la_valid & LA_GID)
2901                         iattr.ia_valid |= ATTR_GID;
2902                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2903                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2904
2905                 rc = ll_vfs_dq_transfer(inode, &iattr);
2906                 if (rc) {
2907                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2908                                "enforcement enabled on the ldiskfs "
2909                                "filesystem?\n", inode->i_sb->s_id, rc);
2910                         return rc;
2911                 }
2912         }
2913
2914         /* Handle project id transfer here properly */
2915         if (attr->la_valid & LA_PROJID &&
2916             attr->la_projid != i_projid_read(inode)) {
2917 #ifdef HAVE_PROJECT_QUOTA
2918                 rc = osd_transfer_project(inode, attr->la_projid);
2919 #else
2920                 rc = -ENOTSUPP;
2921 #endif
2922                 if (rc) {
2923                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2924                                "enforcement enabled on the ldiskfs "
2925                                "filesystem?\n", inode->i_sb->s_id, rc);
2926                         return rc;
2927                 }
2928         }
2929         return 0;
2930 }
2931
2932 static int osd_attr_set(const struct lu_env *env,
2933                         struct dt_object *dt,
2934                         const struct lu_attr *attr,
2935                         struct thandle *handle)
2936 {
2937         struct osd_object *obj = osd_dt_obj(dt);
2938         struct inode *inode;
2939         int rc;
2940
2941         if (!dt_object_exists(dt))
2942                 return -ENOENT;
2943
2944         LASSERT(handle != NULL);
2945         LASSERT(!dt_object_remote(dt));
2946         LASSERT(osd_invariant(obj));
2947
2948         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2949
2950         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2951             !osd_obj2dev(obj)->od_is_ost) {
2952                 struct osd_thread_info *oti = osd_oti_get(env);
2953                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2954                 struct lu_fid *fid1 = &oti->oti_fid;
2955                 struct osd_inode_id *id = &oti->oti_id;
2956                 struct iam_path_descr *ipd;
2957                 struct iam_container *bag;
2958                 struct osd_thandle *oh;
2959                 int rc;
2960
2961                 fid_cpu_to_be(fid1, fid0);
2962                 memset(id, 1, sizeof(*id));
2963                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2964                                   fid0)->oi_dir.od_container;
2965                 ipd = osd_idx_ipd_get(env, bag);
2966                 if (unlikely(ipd == NULL))
2967                         RETURN(-ENOMEM);
2968
2969                 oh = container_of0(handle, struct osd_thandle, ot_super);
2970                 rc = iam_update(oh->ot_handle, bag,
2971                                 (const struct iam_key *)fid1,
2972                                 (const struct iam_rec *)id, ipd);
2973                 osd_ipd_put(env, bag, ipd);
2974                 return(rc > 0 ? 0 : rc);
2975         }
2976
2977         inode = obj->oo_inode;
2978
2979         rc = osd_quota_transfer(inode, attr);
2980         if (rc)
2981                 return rc;
2982
2983         spin_lock(&obj->oo_guard);
2984         rc = osd_inode_setattr(env, inode, attr);
2985         spin_unlock(&obj->oo_guard);
2986         if (rc != 0)
2987                 GOTO(out, rc);
2988
2989         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2990
2991         if (!(attr->la_valid & LA_FLAGS))
2992                 GOTO(out, rc);
2993
2994         /* Let's check if there are extra flags need to be set into LMA */
2995         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2996                 struct osd_thread_info *info = osd_oti_get(env);
2997                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2998
2999                 LASSERT(!obj->oo_pfid_in_lma);
3000
3001                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3002                                  &info->oti_ost_attrs);
3003                 if (rc)
3004                         GOTO(out, rc);
3005
3006                 lma->lma_incompat |=
3007                         lustre_to_lma_flags(attr->la_flags);
3008                 lustre_lma_swab(lma);
3009                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3010                                      lma, sizeof(*lma), XATTR_REPLACE);
3011                 if (rc != 0) {
3012                         struct osd_device *osd = osd_obj2dev(obj);
3013
3014                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3015                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3016                               lma->lma_incompat, rc);
3017                 } else {
3018                         obj->oo_lma_flags =
3019                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3020                 }
3021                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3022         }
3023 out:
3024         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3025
3026         return rc;
3027 }
3028
3029 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3030                                            struct osd_object *obj,
3031                                            const char *name, const int namelen)
3032 {
3033         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3034 }
3035
3036 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3037                       umode_t mode, struct dt_allocation_hint *hint,
3038                       struct thandle *th)
3039 {
3040         int result;
3041         struct osd_device *osd = osd_obj2dev(obj);
3042         struct osd_thandle *oth;
3043         struct dt_object *parent = NULL;
3044         struct inode *inode;
3045
3046         LINVRNT(osd_invariant(obj));
3047         LASSERT(obj->oo_inode == NULL);
3048         LASSERT(obj->oo_hl_head == NULL);
3049
3050         if (S_ISDIR(mode) && ldiskfs_pdo) {
3051                 obj->oo_hl_head =
3052                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3053                 if (obj->oo_hl_head == NULL)
3054                         return -ENOMEM;
3055         }
3056
3057         oth = container_of(th, struct osd_thandle, ot_super);
3058         LASSERT(oth->ot_handle->h_transaction != NULL);
3059
3060         if (hint != NULL && hint->dah_parent != NULL &&
3061             !dt_object_remote(hint->dah_parent))
3062                 parent = hint->dah_parent;
3063
3064         inode = ldiskfs_create_inode(oth->ot_handle,
3065                                      parent ? osd_dt_obj(parent)->oo_inode :
3066                                               osd_sb(osd)->s_root->d_inode,
3067                                      mode);
3068         if (!IS_ERR(inode)) {
3069                 /* Do not update file c/mtime in ldiskfs. */
3070                 inode->i_flags |= S_NOCMTIME;
3071
3072                 /*
3073                  * For new created object, it must be consistent,
3074                  * and it is unnecessary to scrub against it.
3075                  */
3076                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3077
3078                 obj->oo_inode = inode;
3079                 result = 0;
3080         } else {
3081                 if (obj->oo_hl_head != NULL) {
3082                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3083                         obj->oo_hl_head = NULL;
3084                 }
3085                 result = PTR_ERR(inode);
3086         }
3087         LINVRNT(osd_invariant(obj));
3088         return result;
3089 }
3090
3091 enum {
3092         OSD_NAME_LEN = 255
3093 };
3094
3095 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3096                      struct lu_attr *attr,
3097                      struct dt_allocation_hint *hint,
3098                      struct dt_object_format *dof,
3099                      struct thandle *th)
3100 {
3101         int result;
3102         struct osd_thandle *oth;
3103         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3104
3105         LASSERT(S_ISDIR(attr->la_mode));
3106
3107         oth = container_of(th, struct osd_thandle, ot_super);
3108         LASSERT(oth->ot_handle->h_transaction != NULL);
3109         result = osd_mkfile(info, obj, mode, hint, th);
3110
3111         return result;
3112 }
3113
3114 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3115                         struct lu_attr *attr,
3116                         struct dt_allocation_hint *hint,
3117                         struct dt_object_format *dof,
3118                         struct thandle *th)
3119 {
3120         int result;
3121         struct osd_thandle *oth;
3122         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3123
3124         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3125
3126         LASSERT(S_ISREG(attr->la_mode));
3127
3128         oth = container_of(th, struct osd_thandle, ot_super);
3129         LASSERT(oth->ot_handle->h_transaction != NULL);
3130
3131         result = osd_mkfile(info, obj, mode, hint, th);
3132         if (result == 0) {
3133                 LASSERT(obj->oo_inode != NULL);
3134                 if (feat->dif_flags & DT_IND_VARKEY)
3135                         result = iam_lvar_create(obj->oo_inode,
3136                                                  feat->dif_keysize_max,
3137                                                  feat->dif_ptrsize,
3138                                                  feat->dif_recsize_max,
3139                                                  oth->ot_handle);
3140                 else
3141                         result = iam_lfix_create(obj->oo_inode,
3142                                                  feat->dif_keysize_max,
3143                                                  feat->dif_ptrsize,
3144                                                  feat->dif_recsize_max,
3145                                                  oth->ot_handle);
3146         }
3147         return result;
3148 }
3149
3150 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3151                      struct lu_attr *attr,
3152                      struct dt_allocation_hint *hint,
3153                      struct dt_object_format *dof,
3154                      struct thandle *th)
3155 {
3156         LASSERT(S_ISREG(attr->la_mode));
3157         return osd_mkfile(info, obj, (attr->la_mode &
3158                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3159 }
3160
3161 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3162                      struct lu_attr *attr,
3163                      struct dt_allocation_hint *hint,
3164                      struct dt_object_format *dof,
3165                      struct thandle *th)
3166 {
3167         LASSERT(S_ISLNK(attr->la_mode));
3168         return osd_mkfile(info, obj, (attr->la_mode &
3169                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3170 }
3171
3172 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3173                      struct lu_attr *attr,
3174                      struct dt_allocation_hint *hint,
3175                      struct dt_object_format *dof,
3176                      struct thandle *th)
3177 {
3178         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3179         int result;
3180
3181         LINVRNT(osd_invariant(obj));
3182         LASSERT(obj->oo_inode == NULL);
3183         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3184                 S_ISFIFO(mode) || S_ISSOCK(mode));
3185
3186         result = osd_mkfile(info, obj, mode, hint, th);
3187         if (result == 0) {
3188                 LASSERT(obj->oo_inode != NULL);
3189                 /*
3190                  * This inode should be marked dirty for i_rdev.  Currently
3191                  * that is done in the osd_attr_init().
3192                  */
3193                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3194                                    attr->la_rdev);
3195         }
3196         LINVRNT(osd_invariant(obj));
3197         return result;
3198 }
3199
3200 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3201                               struct lu_attr *,
3202                               struct dt_allocation_hint *hint,
3203                               struct dt_object_format *dof,
3204                               struct thandle *);
3205
3206 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3207 {
3208         osd_obj_type_f result;
3209
3210         switch (type) {
3211         case DFT_DIR:
3212                 result = osd_mkdir;
3213                 break;
3214         case DFT_REGULAR:
3215                 result = osd_mkreg;
3216                 break;
3217         case DFT_SYM:
3218                 result = osd_mksym;
3219                 break;
3220         case DFT_NODE:
3221                 result = osd_mknod;
3222                 break;
3223         case DFT_INDEX:
3224                 result = osd_mk_index;
3225                 break;
3226
3227         default:
3228                 LBUG();
3229                 break;
3230         }
3231         return result;
3232 }
3233
3234
3235 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3236                         struct dt_object *parent, struct dt_object *child,
3237                         umode_t child_mode)
3238 {
3239         LASSERT(ah);
3240
3241         ah->dah_parent = parent;
3242         ah->dah_mode = child_mode;
3243
3244         if (parent != NULL && !dt_object_remote(parent)) {
3245                 /* will help to find FID->ino at dt_insert("..") */
3246                 struct osd_object *pobj = osd_dt_obj(parent);
3247
3248                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3249         }
3250 }
3251
3252 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3253                           struct lu_attr *attr, struct dt_object_format *dof)
3254 {
3255         struct inode *inode = obj->oo_inode;
3256         __u64 valid = attr->la_valid;
3257         int result;
3258
3259         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3260
3261         if (dof->dof_type != DFT_NODE)
3262                 attr->la_valid &= ~LA_RDEV;
3263         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3264                 attr->la_valid &= ~LA_ATIME;
3265         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3266                 attr->la_valid &= ~LA_CTIME;
3267         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3268                 attr->la_valid &= ~LA_MTIME;
3269
3270         result = osd_quota_transfer(inode, attr);
3271         if (result)
3272                 return;
3273
3274         if (attr->la_valid != 0) {
3275                 result = osd_inode_setattr(info->oti_env, inode, attr);
3276                 /*
3277                  * The osd_inode_setattr() should always succeed here.  The
3278                  * only error that could be returned is EDQUOT when we are
3279                  * trying to change the UID or GID of the inode. However, this
3280                  * should not happen since quota enforcement is no longer
3281                  * enabled on ldiskfs (lquota takes care of it).
3282                  */
3283                 LASSERTF(result == 0, "%d\n", result);
3284                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3285         }
3286
3287         attr->la_valid = valid;
3288 }
3289
3290 /**
3291  * Helper function for osd_create()
3292  *
3293  * \retval 0, on success
3294  */
3295 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3296                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3297                         struct dt_object_format *dof, struct thandle *th)
3298 {
3299         int result;
3300         __u32 umask;
3301
3302         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3303
3304         /* we drop umask so that permissions we pass are not affected */
3305         umask = current->fs->umask;
3306         current->fs->umask = 0;
3307
3308         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3309                                                   th);
3310         if (likely(obj->oo_inode != NULL)) {
3311                 LASSERT(obj->oo_inode->i_state & I_NEW);
3312
3313                 /*
3314                  * Unlock the inode before attr initialization to avoid
3315                  * unnecessary dqget operations. LU-6378
3316                  */
3317                 unlock_new_inode(obj->oo_inode);
3318         }
3319
3320         if (likely(result == 0)) {
3321                 osd_attr_init(info, obj, attr, dof);
3322                 osd_object_init0(obj);
3323         }
3324
3325         /* restore previous umask value */
3326         current->fs->umask = umask;
3327
3328         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3329
3330         return result;
3331 }
3332
3333 /**
3334  * Helper function for osd_create()
3335  *
3336  * \retval 0, on success
3337  */
3338 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3339                            const struct lu_fid *fid, struct thandle *th)
3340 {
3341         struct osd_thread_info *info = osd_oti_get(env);
3342         struct osd_inode_id    *id   = &info->oti_id;
3343         struct osd_device      *osd  = osd_obj2dev(obj);
3344         struct osd_thandle     *oh;
3345         int rc;
3346
3347         LASSERT(obj->oo_inode != NULL);
3348
3349         oh = container_of0(th, struct osd_thandle, ot_super);
3350         LASSERT(oh->ot_handle);
3351         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3352
3353         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3354         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3355                            OI_CHECK_FLD, NULL);
3356         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3357
3358         return rc;
3359 }
3360
3361 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3362                    u64 seq, struct lu_seq_range *range)
3363 {
3364         struct seq_server_site *ss = osd_seq_site(osd);
3365
3366         if (fid_seq_is_idif(seq)) {
3367                 fld_range_set_ost(range);
3368                 range->lsr_index = idif_ost_idx(seq);
3369                 return 0;
3370         }
3371
3372         if (!fid_seq_in_fldb(seq)) {
3373                 fld_range_set_mdt(range);
3374                 if (ss != NULL)
3375                         /*
3376                          * FIXME: If ss is NULL, it suppose not get lsr_index
3377                          * at all
3378                          */
3379                         range->lsr_index = ss->ss_node_id;
3380                 return 0;
3381         }
3382
3383         LASSERT(ss != NULL);
3384         fld_range_set_any(range);
3385         /* OSD will only do local fld lookup */
3386         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3387 }
3388
3389 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3390                               struct lu_attr *attr,
3391                               struct dt_allocation_hint *hint,
3392                               struct dt_object_format *dof,
3393                               struct thandle *handle)
3394 {
3395         struct osd_thandle *oh;
3396         int rc;
3397
3398         ENTRY;
3399
3400         LASSERT(handle != NULL);
3401
3402         oh = container_of0(handle, struct osd_thandle, ot_super);
3403         LASSERT(oh->ot_handle == NULL);
3404
3405         /*
3406          * EA object consumes more credits than regular object: osd_mk_index
3407          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3408          * leaf_node, could involves the block, block bitmap, groups, GDT
3409          * change for each block, so add 4 * 2 credits in that case.
3410          */
3411         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3412                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3413                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3414         /*
3415          * Reuse idle OI block may cause additional one OI block
3416          * to be changed.
3417          */
3418         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3419                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3420
3421         if (!attr)
3422                 RETURN(0);
3423
3424         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3425                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3426                                    NULL, OSD_QID_INODE);
3427         if (rc != 0)
3428                 RETURN(rc);
3429
3430         /* will help to find FID->ino mapping at dt_insert() */
3431         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3432                                    osd_dt_obj(dt));
3433
3434         RETURN(rc);
3435 }
3436
3437 /**
3438  * Called to destroy on-disk representation of the object
3439  *
3440  * Concurrency: must be locked
3441  */
3442 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3443                                struct thandle *th)
3444 {
3445         struct osd_object *obj = osd_dt_obj(dt);
3446         struct inode *inode = obj->oo_inode;
3447         struct osd_thandle *oh;
3448         int rc;
3449
3450         ENTRY;
3451
3452         if (inode == NULL)
3453                 RETURN(-ENOENT);
3454
3455         oh = container_of0(th, struct osd_thandle, ot_super);
3456         LASSERT(oh->ot_handle == NULL);
3457
3458         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3459                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3460
3461         /* For removing agent entry */
3462         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3463                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3464
3465         /*
3466          * Recycle idle OI leaf may cause additional three OI blocks
3467          * to be changed.
3468          */
3469         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3470                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3471                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3472         /* one less inode */
3473         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3474                                    i_projid_read(inode), -1, oh, obj, NULL,
3475                                    OSD_QID_INODE);
3476         if (rc)
3477                 RETURN(rc);
3478         /* data to be truncated */
3479         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3480                                    i_projid_read(inode), 0, oh, obj, NULL,
3481                                    OSD_QID_BLK);
3482         if (rc)
3483                 RETURN(rc);
3484
3485         /*
3486          * will help to find FID->ino when this object is being
3487          * added to PENDING
3488          */
3489         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3490
3491         RETURN(rc);
3492 }
3493
3494 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3495                        struct thandle *th)
3496 {
3497         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3498         struct osd_object *obj = osd_dt_obj(dt);
3499         struct inode *inode = obj->oo_inode;
3500         struct osd_device *osd = osd_obj2dev(obj);
3501         struct osd_thandle *oh;
3502         int result;
3503
3504         ENTRY;
3505
3506         oh = container_of0(th, struct osd_thandle, ot_super);
3507         LASSERT(oh->ot_handle);
3508         LASSERT(inode);
3509         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3510
3511         if (unlikely(fid_is_acct(fid)))
3512                 RETURN(-EPERM);
3513
3514         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3515                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3516                 if (result != 0)
3517                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3518                                osd_name(osd), PFID(fid), result);
3519         }
3520
3521         if (S_ISDIR(inode->i_mode)) {
3522                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3523                         inode->i_nlink == 2);
3524
3525                 spin_lock(&obj->oo_guard);
3526                 clear_nlink(inode);
3527                 spin_unlock(&obj->oo_guard);
3528                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3529         }
3530
3531         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3532
3533         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3534
3535         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3536                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3537                                        oh->ot_handle, OI_CHECK_FLD);
3538
3539         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3540         /* XXX: add to ext3 orphan list */
3541         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3542
3543         /* not needed in the cache anymore */
3544         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3545         obj->oo_destroyed = 1;
3546
3547         RETURN(0);
3548 }
3549
3550 /**
3551  * Put the fid into lustre_mdt_attrs, and then place the structure
3552  * inode's ea. This fid should not be altered during the life time
3553  * of the inode.
3554  *
3555  * \retval +ve, on success
3556  * \retval -ve, on error
3557  *
3558  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3559  */
3560 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3561                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3562 {
3563         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3564         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3565         int rc;
3566
3567         ENTRY;
3568
3569         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3570                 RETURN(0);
3571
3572         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3573                 rc = -ENOMEM;
3574
3575         lustre_loa_init(loa, fid, compat, incompat);
3576         lustre_loa_swab(loa, false);
3577
3578         /*
3579          * For the OST device with 256 bytes inode size by default,
3580          * the PFID EA will be stored together with LMA EA to avoid
3581          * performance trouble. Otherwise the PFID EA can be stored
3582          * independently. LU-8998
3583          */
3584         if ((compat & LMAC_FID_ON_OST) &&
3585             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3586                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3587                                      sizeof(*loa), XATTR_CREATE);
3588         else
3589                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3590                                      sizeof(*lma), XATTR_CREATE);
3591         /*
3592          * LMA may already exist, but we need to check that all the
3593          * desired compat/incompat flags have been added.
3594          */
3595         if (unlikely(rc == -EEXIST)) {
3596                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3597                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3598                 if (rc < 0)
3599                         RETURN(rc);
3600
3601                 if (rc < sizeof(*lma))
3602                         RETURN(-EINVAL);
3603
3604                 lustre_loa_swab(loa, true);
3605                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3606                     ((compat == 0 && incompat == 0) ||
3607                      (!(~lma->lma_compat & compat) &&
3608                       !(~lma->lma_incompat & incompat))))
3609                         RETURN(0);
3610
3611                 lma->lma_self_fid = *fid;
3612                 lma->lma_compat |= compat;
3613                 lma->lma_incompat |= incompat;
3614                 if (rc == sizeof(*lma)) {
3615                         lustre_lma_swab(lma);
3616                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3617                                              sizeof(*lma), XATTR_REPLACE);
3618                 } else {
3619                         lustre_loa_swab(loa, false);
3620                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3621                                              sizeof(*loa), XATTR_REPLACE);
3622                 }
3623         }
3624
3625         RETURN(rc);
3626 }
3627
3628 /**
3629  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3630  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3631  * To have compatilibility with 1.8 ldiskfs driver we need to have
3632  * magic number at start of fid data.
3633  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3634  * its inmemory API.
3635  */
3636 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3637                                          const struct lu_fid *fid)
3638 {
3639         if (!fid_is_namespace_visible(fid) ||
3640             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3641                 param->edp_magic = 0;
3642                 return;
3643         }
3644
3645         param->edp_magic = LDISKFS_LUFID_MAGIC;
3646         param->edp_len =  sizeof(struct lu_fid) + 1;
3647         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3648 }
3649
3650 /**
3651  * Try to read the fid from inode ea into dt_rec.
3652  *
3653  * \param fid object fid.
3654  *
3655  * \retval 0 on success
3656  */
3657 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3658                           __u32 ino, struct lu_fid *fid,
3659                           struct osd_inode_id *id)
3660 {
3661         struct osd_thread_info *info  = osd_oti_get(env);
3662         struct inode *inode;
3663
3664         ENTRY;
3665
3666         osd_id_gen(id, ino, OSD_OII_NOGEN);
3667         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3668         if (IS_ERR(inode))
3669                 RETURN(PTR_ERR(inode));
3670
3671         iput(inode);
3672         RETURN(0);
3673 }
3674
3675 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3676                                         struct inode *dir,
3677                                         struct inode *parent_dir,
3678                                         const struct lu_fid *dot_fid,
3679                                         const struct lu_fid *dot_dot_fid,
3680                                         struct osd_thandle *oth)
3681 {
3682         struct ldiskfs_dentry_param *dot_ldp;
3683         struct ldiskfs_dentry_param *dot_dot_ldp;
3684         __u32 saved_nlink = dir->i_nlink;
3685         int rc;
3686
3687         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3688         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3689
3690         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3691         dot_ldp->edp_magic = 0;
3692
3693         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3694                                     dir, dot_ldp, dot_dot_ldp);
3695         /*
3696          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3697          * the subseqent ref_add() will increase the dir->i_nlink
3698          * as 3. That is incorrect for new created directory.
3699          *
3700          * It looks like hack, because we want to make the OSD API
3701          * to be order-independent for new created directory object
3702          * between dt_insert(..) and ref_add() operations.
3703          *
3704          * Here, we only restore the in-RAM dir-inode's nlink attr,
3705          * becuase if the nlink attr is not 2, then there will be
3706          * ref_add() called following the dt_insert(..), such call
3707          * will make both the in-RAM and on-disk dir-inode's nlink
3708          * attr to be set as 2. LU-7447
3709          */
3710         set_nlink(dir, saved_nlink);
3711         return rc;
3712 }
3713
3714 /**
3715  * Create an local agent inode for remote entry
3716  */
3717 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3718                                                   struct osd_device *osd,
3719                                                   struct osd_object *pobj,
3720                                                   const struct lu_fid *fid,
3721                                                   __u32 type,
3722                                                   struct thandle *th)
3723 {
3724         struct osd_thread_info *info = osd_oti_get(env);
3725         struct inode *local;
3726         struct osd_thandle *oh;
3727         int rc;
3728
3729         ENTRY;
3730
3731         LASSERT(th);
3732         oh = container_of(th, struct osd_thandle, ot_super);
3733         LASSERT(oh->ot_handle->h_transaction != NULL);
3734
3735         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3736         if (IS_ERR(local)) {
3737                 CERROR("%s: create local error %d\n", osd_name(osd),
3738                        (int)PTR_ERR(local));
3739                 RETURN(local);
3740         }
3741
3742         /*
3743          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3744          * correct gid on remote file, not agent here
3745          */
3746         local->i_gid = current_fsgid();
3747         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3748         unlock_new_inode(local);
3749
3750         /* Agent inode should not have project ID */
3751 #ifdef  HAVE_PROJECT_QUOTA
3752         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3753             i_projid_read(pobj->oo_inode) != 0) {
3754                 rc = osd_transfer_project(local, 0);
3755                 if (rc) {
3756                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3757                                "quota enforcement enabled on the ldiskfs "
3758                                "filesystem?\n", local->i_sb->s_id, rc);
3759                         RETURN(ERR_PTR(rc));
3760                 }
3761         }
3762 #endif
3763         /* Set special LMA flag for local agent inode */
3764         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3765         if (rc != 0) {
3766                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3767                        osd_name(osd), PFID(fid), rc);
3768                 RETURN(ERR_PTR(rc));
3769         }
3770
3771         if (!S_ISDIR(type))
3772                 RETURN(local);
3773
3774         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3775                                          lu_object_fid(&pobj->oo_dt.do_lu),
3776                                          fid, oh);
3777         if (rc != 0) {
3778                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3779                         osd_name(osd), PFID(fid), rc);
3780                 RETURN(ERR_PTR(rc));
3781         }
3782
3783         RETURN(local);
3784 }
3785
3786 /**
3787  * when direntry is deleted, we have to take care of possible agent inode
3788  * referenced by that. unfortunately we can't do this at that point:
3789  * iget() within a running transaction leads to deadlock and we better do
3790  * not call that every delete declaration to save performance. so we put
3791  * a potention agent inode on a list and process that once the transaction
3792  * is over. Notice it's not any worse in terms of real orphans as regular
3793  * object destroy doesn't put inodes on the on-disk orphan list. this should
3794  * be addressed separately
3795  */
3796 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3797                                             struct osd_thandle *oh,
3798                                             __u32 ino)
3799 {
3800         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3801         struct osd_obj_orphan *oor;
3802
3803         OBD_ALLOC_PTR(oor);
3804         if (oor == NULL)
3805                 return -ENOMEM;
3806
3807         oor->oor_ino = ino;
3808         oor->oor_env = (struct lu_env *)env;
3809         spin_lock(&osd->od_osfs_lock);
3810         list_add(&oor->oor_list, &osd->od_orphan_list);
3811         spin_unlock(&osd->od_osfs_lock);
3812
3813         oh->ot_remove_agents = 1;
3814
3815         return 0;
3816
3817 }
3818
3819 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3820                                                 struct osd_device *osd)
3821 {
3822         struct osd_thread_info *info = osd_oti_get(env);
3823         struct osd_obj_orphan *oor, *tmp;
3824         struct osd_inode_id id;
3825         struct list_head list;
3826         struct inode *inode;
3827         struct lu_fid fid;
3828         handle_t *jh;
3829         __u32 ino;
3830
3831         INIT_LIST_HEAD(&list);
3832
3833         spin_lock(&osd->od_osfs_lock);
3834         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3835                 if (oor->oor_env == env) {
3836                         list_del(&oor->oor_list);
3837                         list_add(&oor->oor_list, &list);
3838                 }
3839         }
3840         spin_unlock(&osd->od_osfs_lock);
3841
3842         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3843
3844                 ino = oor->oor_ino;
3845
3846                 list_del(&oor->oor_list);
3847                 OBD_FREE_PTR(oor);
3848
3849                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3850                 inode = osd_iget_fid(info, osd, &id, &fid);
3851                 if (IS_ERR(inode))
3852                         continue;
3853
3854                 if (!osd_remote_fid(env, osd, &fid)) {
3855                         iput(inode);
3856                         continue;
3857                 }
3858
3859                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3860                 clear_nlink(inode);
3861                 mark_inode_dirty(inode);
3862                 ldiskfs_journal_stop(jh);
3863                 iput(inode);
3864         }
3865
3866         return 0;
3867 }
3868
3869 /**
3870  * OSD layer object create function for OST objects (b=11826).
3871  *
3872  * The FID is inserted into inode xattr here.
3873  *
3874  * \retval   0, on success
3875  * \retval -ve, on error
3876  */
3877 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3878                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3879                       struct dt_object_format *dof, struct thandle *th)
3880 {
3881         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3882         struct osd_object *obj = osd_dt_obj(dt);
3883         struct osd_thread_info *info = osd_oti_get(env);
3884         int result, on_ost = 0;
3885
3886         ENTRY;
3887
3888         if (dt_object_exists(dt))
3889                 RETURN(-EEXIST);
3890
3891         LINVRNT(osd_invariant(obj));
3892         LASSERT(!dt_object_remote(dt));
3893         LASSERT(osd_is_write_locked(env, obj));
3894         LASSERT(th != NULL);
3895
3896         if (unlikely(fid_is_acct(fid)))
3897                 /*
3898                  * Quota files can't be created from the kernel any more,
3899                  * 'tune2fs -O quota' will take care of creating them
3900                  */
3901                 RETURN(-EPERM);
3902
3903         result = __osd_create(info, obj, attr, hint, dof, th);
3904         if (result == 0) {
3905                 if (fid_is_idif(fid) &&
3906                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3907                         struct lu_fid *tfid = &info->oti_fid;
3908                         struct ost_id *oi   = &info->oti_ostid;
3909
3910                         fid_to_ostid(fid, oi);
3911                         ostid_to_fid(tfid, oi, 0);
3912                         on_ost = 1;
3913                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3914                                                 LMAC_FID_ON_OST, 0);
3915                 } else {
3916                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3917                                                fid, OI_CHECK_FLD);
3918                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3919                                                 on_ost ? LMAC_FID_ON_OST : 0,
3920                                                 0);
3921                 }
3922                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3923                         obj->oo_dt.do_body_ops = &osd_body_ops;
3924         }
3925
3926         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3927                 result = __osd_oi_insert(env, obj, fid, th);
3928
3929         /*
3930          * a small optimization - dt_insert() isn't usually applied
3931          * to OST objects, so we don't need to cache OI mapping for
3932          * OST objects
3933          */
3934         if (result == 0 && on_ost == 0) {
3935                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3936
3937                 result = osd_idc_find_and_init(env, osd, obj);
3938                 LASSERT(result == 0);
3939         }
3940
3941         LASSERT(ergo(result == 0,
3942                      dt_object_exists(dt) && !dt_object_remote(dt)));
3943         LINVRNT(osd_invariant(obj));
3944         RETURN(result);
3945 }
3946
3947 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3948                                struct thandle *handle)
3949 {
3950         struct osd_thandle *oh;
3951
3952         /* it's possible that object doesn't exist yet */
3953         LASSERT(handle != NULL);
3954
3955         oh = container_of0(handle, struct osd_thandle, ot_super);
3956         LASSERT(oh->ot_handle == NULL);
3957
3958         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3959                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3960
3961         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3962
3963         return 0;
3964 }
3965
3966 /*
3967  * Concurrency: @dt is write locked.
3968  */
3969 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3970                        struct thandle *th)
3971 {
3972         struct osd_object *obj = osd_dt_obj(dt);
3973         struct inode *inode = obj->oo_inode;
3974         struct osd_thandle *oh;
3975         int rc = 0;
3976
3977         if (!dt_object_exists(dt) || obj->oo_destroyed)
3978                 return -ENOENT;
3979
3980         LINVRNT(osd_invariant(obj));
3981         LASSERT(!dt_object_remote(dt));
3982         LASSERT(osd_is_write_locked(env, obj));
3983         LASSERT(th != NULL);
3984
3985         oh = container_of0(th, struct osd_thandle, ot_super);
3986         LASSERT(oh->ot_handle != NULL);
3987
3988         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3989
3990         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3991                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3992         /*
3993          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3994          * (65000) subdirectories by storing "1" in i_nlink if the link count
3995          * would otherwise overflow. Directory tranversal tools understand
3996          * that (st_nlink == 1) indicates that the filesystem dose not track
3997          * hard links count on the directory, and will not abort subdirectory
3998          * scanning early once (st_nlink - 2) subdirs have been found.
3999          *
4000          * This also has to properly handle the case of inodes with nlink == 0
4001          * in case they are being linked into the PENDING directory
4002          */
4003         spin_lock(&obj->oo_guard);
4004         if (unlikely(inode->i_nlink == 0))
4005                 /* inc_nlink from 0 may cause WARN_ON */
4006                 set_nlink(inode, 1);
4007         else {
4008                 ldiskfs_inc_count(oh->ot_handle, inode);
4009                 if (!S_ISDIR(inode->i_mode))
4010                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4011         }
4012         spin_unlock(&obj->oo_guard);
4013
4014         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4015         LINVRNT(osd_invariant(obj));
4016
4017         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4018
4019         return rc;
4020 }
4021
4022 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4023                                struct thandle *handle)
4024 {
4025         struct osd_thandle *oh;
4026
4027         if (!dt_object_exists(dt))
4028                 return -ENOENT;
4029
4030         LASSERT(!dt_object_remote(dt));
4031         LASSERT(handle != NULL);
4032
4033         oh = container_of0(handle, struct osd_thandle, ot_super);
4034         LASSERT(oh->ot_handle == NULL);
4035
4036         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4037                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4038
4039         return 0;
4040 }
4041
4042 /*
4043  * Concurrency: @dt is write locked.
4044  */
4045 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4046                        struct thandle *th)
4047 {
4048         struct osd_object *obj = osd_dt_obj(dt);
4049         struct inode *inode = obj->oo_inode;
4050         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4051         struct osd_thandle *oh;
4052
4053         if (!dt_object_exists(dt))
4054                 return -ENOENT;
4055
4056         LINVRNT(osd_invariant(obj));
4057         LASSERT(!dt_object_remote(dt));
4058         LASSERT(osd_is_write_locked(env, obj));
4059         LASSERT(th != NULL);
4060
4061         oh = container_of0(th, struct osd_thandle, ot_super);
4062         LASSERT(oh->ot_handle != NULL);
4063
4064         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4065
4066         spin_lock(&obj->oo_guard);
4067         /*
4068          * That can be result of upgrade from old Lustre version and
4069          * applied only to local files.  Just skip this ref_del call.
4070          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4071          */
4072         if (inode->i_nlink == 0) {
4073                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4074                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4075                              ", maybe an upgraded file? (LU-3915)\n",
4076                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4077                 spin_unlock(&obj->oo_guard);
4078                 return 0;
4079         }
4080
4081         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4082                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4083
4084         ldiskfs_dec_count(oh->ot_handle, inode);
4085         spin_unlock(&obj->oo_guard);
4086
4087         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4088         LINVRNT(osd_invariant(obj));
4089
4090         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4091
4092         return 0;
4093 }
4094
4095 /*
4096  * Concurrency: @dt is read locked.
4097  */
4098 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4099                          struct lu_buf *buf, const char *name)
4100 {
4101         struct osd_object      *obj    = osd_dt_obj(dt);
4102         struct inode           *inode  = obj->oo_inode;
4103         struct osd_thread_info *info   = osd_oti_get(env);
4104         struct dentry          *dentry = &info->oti_obj_dentry;
4105         bool cache_xattr = false;
4106         int rc;
4107
4108         LASSERT(buf);
4109
4110         /* version get is not real XATTR but uses xattr API */
4111         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4112                 dt_obj_version_t *ver = buf->lb_buf;
4113
4114                 /*
4115                  * for version we are just using xattr API but change inode
4116                  * field instead
4117                  */
4118                 if (buf->lb_len == 0)
4119                         return sizeof(dt_obj_version_t);
4120
4121                 if (buf->lb_len < sizeof(dt_obj_version_t))
4122                         return -ERANGE;
4123
4124                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4125                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4126
4127                 *ver = LDISKFS_I(inode)->i_fs_version;
4128
4129                 return sizeof(dt_obj_version_t);
4130         }
4131
4132         if (!dt_object_exists(dt))
4133                 return -ENOENT;
4134
4135         LASSERT(!dt_object_remote(dt));
4136         LASSERT(inode->i_op != NULL);
4137 #ifdef HAVE_IOP_XATTR
4138         LASSERT(inode->i_op->getxattr != NULL);
4139 #endif
4140
4141         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4142             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4143                 cache_xattr = true;
4144
4145         if (cache_xattr) {
4146                 rc = osd_oxc_get(obj, name, buf);
4147                 if (rc != -ENOENT)
4148                         return rc;
4149         }
4150
4151         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4152                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4153                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4154                 struct filter_fid *ff;
4155                 struct ost_layout *ol;
4156
4157                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4158                 if (rc)
4159                         return rc;
4160
4161                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4162
4163                 rc = sizeof(*ff);
4164                 if (buf->lb_len == 0 || !buf->lb_buf)
4165                         return rc;
4166
4167                 if (buf->lb_len < rc)
4168                         return -ERANGE;
4169
4170                 ff = buf->lb_buf;
4171                 ol = &ff->ff_layout;
4172                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4173                                                   PFID_STRIPE_IDX_BITS);
4174                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4175                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4176                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4177                 if (lma->lma_compat & LMAC_COMP_INFO) {
4178                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4179                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4180                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4181                 } else {
4182                         ol->ol_comp_start = 0;
4183                         ol->ol_comp_end = 0;
4184                         ol->ol_comp_id = 0;
4185                 }
4186         } else {
4187                 rc = __osd_xattr_get(inode, dentry, name,
4188                                      buf->lb_buf, buf->lb_len);
4189         }
4190
4191         if (cache_xattr) {
4192                 if (rc == -ENOENT || rc == -ENODATA)
4193                         osd_oxc_add(obj, name, NULL, 0);
4194                 else if (rc > 0 && buf->lb_buf != NULL)
4195                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4196         }
4197
4198         return rc;
4199 }
4200
4201 static int osd_declare_xattr_set(const struct lu_env *env,
4202                                  struct dt_object *dt,
4203                                  const struct lu_buf *buf, const char *name,
4204                                  int fl, struct thandle *handle)
4205 {
4206         struct osd_thandle *oh;
4207         int credits = 0;
4208         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4209
4210         LASSERT(handle != NULL);
4211
4212         oh = container_of0(handle, struct osd_thandle, ot_super);
4213         LASSERT(oh->ot_handle == NULL);
4214
4215         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4216                 /*
4217                  * For non-upgrading case, the LMA is set first and
4218                  * usually fit inode. But for upgrade case, the LMA
4219                  * may be in another separated EA block.
4220                  */
4221                 if (dt_object_exists(dt)) {
4222                         if (fl == LU_XATTR_REPLACE)
4223                                 credits = 1;
4224                         else
4225                                 goto upgrade;
4226                 }
4227         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4228                 credits = 1;
4229         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4230                 /* We may need to delete the old PFID EA. */
4231                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4232                 if (fl == LU_XATTR_REPLACE)
4233                         credits += 1;
4234                 else
4235                         goto upgrade;
4236         } else {
4237                 /*
4238                  * If some name entry resides on remote MDT, then will create
4239                  * agent entry under remote parent. On the other hand, if the
4240                  * remote entry will be removed, then related agent entry may
4241                  * need to be removed from the remote parent. So there may be
4242                  * kinds of cases, let's declare enough credits. The credits
4243                  * for create agent entry is enough for remove case.
4244                  */
4245                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4246                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4247                         if (dt_object_exists(dt))
4248                                 credits += 1; /* For updating LMA */
4249                 }
4250
4251 upgrade:
4252                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4253
4254                 if (buf != NULL) {
4255                         ssize_t buflen;
4256
4257                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4258                                 /*
4259                                  * learn xattr size from osd_xattr_get if
4260                                  * attribute has not been read yet
4261                                  */
4262                                 buflen = __osd_xattr_get(
4263                                     osd_dt_obj(dt)->oo_inode,
4264                                     &osd_oti_get(env)->oti_obj_dentry,
4265                                     name, NULL, 0);
4266                                 if (buflen < 0)
4267                                         buflen = 0;
4268                         } else {
4269                                 buflen = buf->lb_len;
4270                         }
4271
4272                         if (buflen > sb->s_blocksize) {
4273                                 credits += osd_calc_bkmap_credits(
4274                                     sb, NULL, 0, -1,
4275                                     (buflen + sb->s_blocksize - 1) >>
4276                                     sb->s_blocksize_bits);
4277                         }
4278                 }
4279                 /*
4280                  * xattr set may involve inode quota change, reserve credits for
4281                  * dquot_initialize()
4282                  */
4283                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4284         }
4285
4286         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4287
4288         return 0;
4289 }
4290
4291 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4292                               const struct lu_buf *buf, int fl,
4293                               struct thandle *handle)
4294 {
4295         struct osd_thread_info *info = osd_oti_get(env);
4296         struct dentry *dentry = &info->oti_obj_dentry;
4297         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4298         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4299         struct inode *inode = obj->oo_inode;
4300         struct filter_fid *ff = buf->lb_buf;
4301         struct ost_layout *ol = &ff->ff_layout;
4302         int flags = XATTR_REPLACE;
4303         int rc;
4304
4305         ENTRY;
4306
4307         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4308                 RETURN(-EINVAL);
4309
4310         rc = osd_get_lma(info, inode, dentry, loa);
4311         if (rc == -ENODATA) {
4312                 /* Usually for upgarding from old device */
4313                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4314                                 LMAC_FID_ON_OST, 0);
4315                 flags = XATTR_CREATE;
4316         } else if (rc) {
4317                 RETURN(rc);
4318         }
4319
4320         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4321                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4322                         RETURN(-EEXIST);
4323
4324                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4325                         /* Separate PFID EA from LMA */
4326                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4327                         lustre_lma_swab(lma);
4328                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4329                                              sizeof(*lma), XATTR_REPLACE);
4330                         if (!rc) {
4331                                 obj->oo_pfid_in_lma = 0;
4332                                 rc = LU_XATTR_CREATE;
4333                         }
4334
4335                         RETURN(rc);
4336                 }
4337         } else {
4338                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4339                         RETURN(fl);
4340
4341                 /*
4342                  * Old client does not send stripe information,
4343                  * then store the PFID EA on disk separatedly.
4344                  */
4345                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4346                              ol->ol_stripe_size == 0))
4347                         RETURN(fl);
4348
4349                 /* Remove old PFID EA entry firstly. */
4350                 ll_vfs_dq_init(inode);
4351                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4352                 if (rc == -ENODATA) {
4353                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4354                                 RETURN(rc);
4355                 } else if (rc) {
4356                         RETURN(rc);
4357                 }
4358         }
4359
4360         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4361         if (likely(ol->ol_stripe_size != 0)) {
4362                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4363                                              PFID_STRIPE_IDX_BITS;
4364                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4365                 lma->lma_compat |= LMAC_STRIPE_INFO;
4366                 if (ol->ol_comp_id != 0) {
4367                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4368                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4369                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4370                         lma->lma_compat |= LMAC_COMP_INFO;
4371                 }
4372         }
4373
4374         lustre_loa_swab(loa, false);
4375
4376         /* Store the PFID EA inside LMA. */
4377         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4378                              flags);
4379         if (!rc)
4380                 obj->oo_pfid_in_lma = 1;
4381
4382         RETURN(rc);
4383 }
4384
4385 /*
4386  * In DNE environment, the object (in spite of regular file or directory)
4387  * and its name entry may reside on different MDTs. Under such case, we will
4388  * create an agent entry on the MDT where the object resides. The agent entry
4389  * references the object locally, that makes the object to be visible to the
4390  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4391  * such as 'tar' can handle the object properly.
4392  *
4393  * We handle the agent entry during set linkEA that is the common interface
4394  * for both regular file and directroy, can handle kinds of cases, such as
4395  * create/link/unlink/rename, and so on.
4396  *
4397  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4398  *
4399  * XXX: There are two known issues:
4400  * 1. For one object, we will create at most one agent entry even if there
4401  *    may be more than one cross-MDTs hard links on the object. So the local
4402  *    e2fsck may claim that the object's nlink is larger than the name entries
4403  *    that reference such inode. And in further, the e2fsck will fix the nlink
4404  *    attribute to match the local references. Then it will cause the object's
4405  *    nlink attribute to be inconsistent with the global references. it is bad
4406  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4407  *    other hand, the global namespace LFSCK can repair the object's attribute
4408  *    according to the linkEA.
4409  * 2. There may be too many hard links on the object as to its linkEA overflow,
4410  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4411  *    case happened, then at this point, we do not know whether there are some
4412  *    cross-MDTs reference. But there are local references, it guarantees that
4413  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4414  */
4415 static int osd_xattr_handle_linkea(const struct lu_env *env,
4416                                    struct osd_device *osd,
4417                                    struct osd_object *obj,
4418                                    const struct lu_buf *buf,
4419                                    struct thandle *handle)
4420 {
4421         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4422         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4423         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4424         struct lu_name tmpname;
4425         struct osd_thandle *oh;
4426         int rc;
4427         bool remote = false;
4428
4429         ENTRY;
4430
4431         oh = container_of0(handle, struct osd_thandle, ot_super);
4432         LASSERT(oh->ot_handle != NULL);
4433
4434         rc = linkea_init_with_rec(&ldata);
4435         if (!rc) {
4436                 linkea_first_entry(&ldata);
4437                 while (ldata.ld_lee != NULL && !remote) {
4438                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4439                                             &tmpname, tfid);
4440                         if (osd_remote_fid(env, osd, tfid) > 0)
4441                                 remote = true;
4442                         else
4443                                 linkea_next_entry(&ldata);
4444                 }
4445         } else if (rc == -ENODATA) {
4446                 rc = 0;
4447         } else {
4448                 RETURN(rc);
4449         }
4450
4451         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4452                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4453                 if (rc)
4454                         CERROR("%s: failed to remove agent entry for "DFID
4455                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4456         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4457                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4458                 if (rc)
4459                         CERROR("%s: failed to create agent entry for "DFID
4460                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4461         }
4462
4463         RETURN(rc);
4464 }
4465
4466 /*
4467  * Concurrency: @dt is write locked.
4468  */
4469 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4470                          const struct lu_buf *buf, const char *name, int fl,
4471                          struct thandle *handle)
4472 {
4473         struct osd_object *obj = osd_dt_obj(dt);
4474         struct osd_device *osd = osd_obj2dev(obj);
4475         struct inode *inode = obj->oo_inode;
4476         struct osd_thread_info *info = osd_oti_get(env);
4477         int fs_flags = 0;
4478         int len;
4479         int rc;
4480
4481         ENTRY;
4482
4483         LASSERT(handle);
4484         LASSERT(buf);
4485
4486         /* version set is not real XATTR */
4487         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4488                 dt_obj_version_t *version = buf->lb_buf;
4489
4490                 /*
4491                  * for version we are just using xattr API but change inode
4492                  * field instead
4493                  */
4494                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4495
4496                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4497                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4498
4499                 LDISKFS_I(inode)->i_fs_version = *version;
4500                 /*
4501                  * Version is set after all inode operations are finished,
4502                  * so we should mark it dirty here
4503                  */
4504                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4505
4506                 RETURN(0);
4507         }
4508
4509         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4510                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4511
4512         len = buf->lb_len;
4513         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4514
4515         /*
4516          * For the OST device with 256 bytes inode size by default,
4517          * the PFID EA will be stored together with LMA EA to avoid
4518          * performance trouble. Otherwise the PFID EA can be stored
4519          * independently. LU-8998
4520          */
4521         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4522             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4523                 LASSERT(buf->lb_buf);
4524
4525                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4526                 if (fl <= 0)
4527                         RETURN(fl);
4528         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4529                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4530                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4531
4532                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4533                 if (rc)
4534                         RETURN(rc);
4535
4536                 lma->lma_incompat |= LMAI_STRIPED;
4537                 lustre_lma_swab(lma);
4538                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4539                                      sizeof(*lma), XATTR_REPLACE);
4540                 if (rc != 0)
4541                         RETURN(rc);
4542         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4543                 LASSERT(!osd->od_is_ost);
4544
4545                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4546                 if (rc)
4547                         RETURN(rc);
4548         }
4549
4550         if (fl & LU_XATTR_REPLACE)
4551                 fs_flags |= XATTR_REPLACE;
4552
4553         if (fl & LU_XATTR_CREATE)
4554                 fs_flags |= XATTR_CREATE;
4555
4556         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4557         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4558
4559         if (rc == 0 &&
4560             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4561              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4562                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4563
4564         return rc;
4565 }
4566
4567 /*
4568  * Concurrency: @dt is read locked.
4569  */
4570 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4571                           const struct lu_buf *buf)
4572 {
4573         struct osd_object *obj = osd_dt_obj(dt);
4574         struct inode *inode = obj->oo_inode;
4575         struct osd_thread_info *info = osd_oti_get(env);
4576         struct dentry *dentry = &info->oti_obj_dentry;
4577
4578         if (!dt_object_exists(dt))
4579                 return -ENOENT;
4580
4581         LASSERT(!dt_object_remote(dt));
4582         LASSERT(inode->i_op != NULL);
4583         LASSERT(inode->i_op->listxattr != NULL);
4584
4585         dentry->d_inode = inode;
4586         dentry->d_sb = inode->i_sb;
4587         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4588 }
4589
4590 static int osd_declare_xattr_del(const struct lu_env *env,
4591                                  struct dt_object *dt, const char *name,
4592                                  struct thandle *handle)
4593 {
4594         struct osd_thandle *oh;
4595         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4596
4597         LASSERT(!dt_object_remote(dt));
4598         LASSERT(handle != NULL);
4599
4600         oh = container_of0(handle, struct osd_thandle, ot_super);
4601         LASSERT(oh->ot_handle == NULL);
4602
4603         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4604                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4605         /*
4606          * xattr del may involve inode quota change, reserve credits for
4607          * dquot_initialize()
4608          */
4609         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4610
4611         return 0;
4612 }
4613
4614 /*
4615  * Concurrency: @dt is write locked.
4616  */
4617 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4618                          const char *name, struct thandle *handle)
4619 {
4620         struct osd_object *obj = osd_dt_obj(dt);
4621         struct inode *inode = obj->oo_inode;
4622         struct osd_thread_info *info = osd_oti_get(env);
4623         struct dentry *dentry = &info->oti_obj_dentry;
4624         int rc;
4625
4626         if (!dt_object_exists(dt))
4627                 return -ENOENT;
4628
4629         LASSERT(!dt_object_remote(dt));
4630         LASSERT(inode->i_op != NULL);
4631         LASSERT(handle != NULL);
4632 #ifdef HAVE_IOP_XATTR
4633         LASSERT(inode->i_op->removexattr != NULL);
4634 #endif
4635
4636         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4637
4638         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4639                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4640
4641                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4642                                  &info->oti_ost_attrs);
4643                 if (!rc) {
4644                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4645
4646                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4647                         lustre_lma_swab(lma);
4648                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4649                                              sizeof(*lma), XATTR_REPLACE);
4650                         if (!rc)
4651                                 obj->oo_pfid_in_lma = 0;
4652                 }
4653         } else {
4654                 ll_vfs_dq_init(inode);
4655                 dentry->d_inode = inode;
4656                 dentry->d_sb = inode->i_sb;
4657                 rc = osd_removexattr(dentry, inode, name);
4658         }
4659
4660         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4661
4662         if (rc == 0 &&
4663             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4664              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4665                 osd_oxc_del(obj, name);
4666
4667         return rc;
4668 }
4669
4670 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4671                            __u64 start, __u64 end)
4672 {
4673         struct osd_object *obj = osd_dt_obj(dt);
4674         struct inode *inode = obj->oo_inode;
4675         struct osd_thread_info *info = osd_oti_get(env);
4676         struct dentry *dentry = &info->oti_obj_dentry;
4677         struct file *file = &info->oti_file;
4678         int rc;
4679
4680         ENTRY;
4681
4682         dentry->d_inode = inode;
4683         dentry->d_sb = inode->i_sb;
4684         file->f_path.dentry = dentry;
4685         file->f_mapping = inode->i_mapping;
4686         file->f_op = inode->i_fop;
4687         set_file_inode(file, inode);
4688
4689         rc = ll_vfs_fsync_range(file, start, end, 0);
4690
4691         RETURN(rc);
4692 }
4693
4694 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4695 {
4696         return 0;
4697 }
4698
4699 /*
4700  * Index operations.
4701  */
4702
4703 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4704                                const struct dt_index_features *feat)
4705 {
4706         struct iam_descr *descr;
4707
4708         if (osd_object_is_root(o))
4709                 return feat == &dt_directory_features;
4710
4711         LASSERT(o->oo_dir != NULL);
4712
4713         descr = o->oo_dir->od_container.ic_descr;
4714         if (feat == &dt_directory_features) {
4715                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4716                         return 1;
4717                 else
4718                         return 0;
4719         } else {
4720                 return feat->dif_keysize_min <= descr->id_key_size &&
4721                        descr->id_key_size <= feat->dif_keysize_max &&
4722                        feat->dif_recsize_min <= descr->id_rec_size &&
4723                        descr->id_rec_size <= feat->dif_recsize_max &&
4724                        !(feat->dif_flags & (DT_IND_VARKEY |
4725                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4726                        ergo(feat->dif_flags & DT_IND_UPDATE,
4727                             1 /* XXX check that object (and fs) is writable */);
4728         }
4729 }
4730
4731 static int osd_iam_container_init(const struct lu_env *env,
4732                                   struct osd_object *obj,
4733                                   struct osd_directory *dir)
4734 {
4735         struct iam_container *bag = &dir->od_container;
4736         int result;
4737
4738         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4739         if (result != 0)
4740                 return result;
4741
4742         result = iam_container_setup(bag);
4743         if (result == 0)
4744                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4745         else
4746                 iam_container_fini(bag);
4747
4748         return result;
4749 }
4750
4751
4752 /*
4753  * Concurrency: no external locking is necessary.
4754  */
4755 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4756                          const struct dt_index_features *feat)
4757 {
4758         int result;
4759         int skip_iam = 0;
4760         struct osd_object *obj = osd_dt_obj(dt);
4761
4762         LINVRNT(osd_invariant(obj));
4763
4764         if (osd_object_is_root(obj)) {
4765                 dt->do_index_ops = &osd_index_ea_ops;
4766                 result = 0;
4767         } else if (feat == &dt_directory_features) {
4768                 dt->do_index_ops = &osd_index_ea_ops;
4769                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4770                         result = 0;
4771                 else
4772                         result = -ENOTDIR;
4773                 skip_iam = 1;
4774         } else if (unlikely(feat == &dt_otable_features)) {
4775                 dt->do_index_ops = &osd_otable_ops;
4776                 return 0;
4777         } else if (unlikely(feat == &dt_acct_features)) {
4778                 dt->do_index_ops = &osd_acct_index_ops;
4779                 result = 0;
4780                 skip_iam = 1;
4781         } else if (!osd_has_index(obj)) {
4782                 struct osd_directory *dir;
4783                 struct osd_device *osd = osd_obj2dev(obj);
4784                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4785
4786                 OBD_ALLOC_PTR(dir);
4787                 if (dir) {
4788
4789                         spin_lock(&obj->oo_guard);
4790                         if (obj->oo_dir == NULL)
4791                                 obj->oo_dir = dir;
4792                         else
4793                                 /*
4794                                  * Concurrent thread allocated container data.
4795                                  */
4796                                 OBD_FREE_PTR(dir);
4797                         spin_unlock(&obj->oo_guard);
4798                         /*
4799                          * Now, that we have container data, serialize its
4800                          * initialization.
4801                          */
4802                         down_write(&obj->oo_ext_idx_sem);
4803                         /*
4804                          * recheck under lock.
4805                          */
4806
4807                         if (osd_has_index(obj)) {
4808                                 result = 0;
4809                                 goto unlock;
4810                         }
4811
4812                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4813                         if (result || feat == &dt_lfsck_namespace_features ||
4814                             feat == &dt_lfsck_layout_orphan_features ||
4815                             feat == &dt_lfsck_layout_dangling_features)
4816                                 goto unlock;
4817
4818                         result = osd_index_register(osd, fid,
4819                                                     feat->dif_keysize_max,
4820                                                     feat->dif_recsize_max);
4821                         if (result < 0)
4822                                 CWARN("%s: failed to register index "
4823                                       DFID": rc = %d\n",
4824                                       osd_name(osd), PFID(fid), result);
4825                         else if (result > 0)
4826                                 result = 0;
4827                         else
4828                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4829                                        " (%d/%d) registered\n",
4830                                        osd_name(osd), PFID(fid),
4831                                        (int)feat->dif_keysize_max,
4832                                        (int)feat->dif_recsize_max);
4833
4834 unlock:
4835                         up_write(&obj->oo_ext_idx_sem);
4836                 } else {
4837                         result = -ENOMEM;
4838                 }
4839         } else {
4840                 result = 0;
4841         }
4842
4843         if (result == 0 && skip_iam == 0) {
4844                 if (!osd_iam_index_probe(env, obj, feat))
4845                         result = -ENOTDIR;
4846         }
4847         LINVRNT(osd_invariant(obj));
4848
4849         return result;
4850 }
4851
4852 static int osd_otable_it_attr_get(const struct lu_env *env,
4853                                  struct dt_object *dt,
4854                                  struct lu_attr *attr)
4855 {
4856         attr->la_valid = 0;
4857         return 0;
4858 }
4859
4860 static const struct dt_object_operations osd_obj_ops = {
4861         .do_read_lock           = osd_read_lock,
4862         .do_write_lock          = osd_write_lock,
4863         .do_read_unlock         = osd_read_unlock,
4864         .do_write_unlock        = osd_write_unlock,
4865         .do_write_locked        = osd_write_locked,
4866         .do_attr_get            = osd_attr_get,
4867         .do_declare_attr_set    = osd_declare_attr_set,
4868         .do_attr_set            = osd_attr_set,
4869         .do_ah_init             = osd_ah_init,
4870         .do_declare_create      = osd_declare_create,
4871         .do_create              = osd_create,
4872         .do_declare_destroy     = osd_declare_destroy,
4873         .do_destroy             = osd_destroy,
4874         .do_index_try           = osd_index_try,
4875         .do_declare_ref_add     = osd_declare_ref_add,
4876         .do_ref_add             = osd_ref_add,
4877         .do_declare_ref_del     = osd_declare_ref_del,
4878         .do_ref_del             = osd_ref_del,
4879         .do_xattr_get           = osd_xattr_get,
4880         .do_declare_xattr_set   = osd_declare_xattr_set,
4881         .do_xattr_set           = osd_xattr_set,
4882         .do_declare_xattr_del   = osd_declare_xattr_del,
4883         .do_xattr_del           = osd_xattr_del,
4884         .do_xattr_list          = osd_xattr_list,
4885         .do_object_sync         = osd_object_sync,
4886         .do_invalidate          = osd_invalidate,
4887 };
4888
4889 static const struct dt_object_operations osd_obj_otable_it_ops = {
4890         .do_attr_get    = osd_otable_it_attr_get,
4891         .do_index_try   = osd_index_try,
4892 };
4893
4894 static int osd_index_declare_iam_delete(const struct lu_env *env,
4895                                         struct dt_object *dt,
4896                                         const struct dt_key *key,
4897                                         struct thandle *handle)
4898 {
4899         struct osd_thandle *oh;
4900
4901         oh = container_of0(handle, struct osd_thandle, ot_super);
4902         LASSERT(oh->ot_handle == NULL);
4903
4904         /* Recycle  may cause additional three blocks to be changed. */
4905         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4906                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4907
4908         return 0;
4909 }
4910
4911 /**
4912  *      delete a (key, value) pair from index \a dt specified by \a key
4913  *
4914  *      \param  dt      osd index object
4915  *      \param  key     key for index
4916  *      \param  rec     record reference
4917  *      \param  handle  transaction handler
4918  *
4919  *      \retval  0  success
4920  *      \retval -ve   failure
4921  */
4922 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4923                                 const struct dt_key *key,
4924                                 struct thandle *handle)
4925 {
4926         struct osd_thread_info *oti = osd_oti_get(env);
4927         struct osd_object *obj = osd_dt_obj(dt);
4928         struct osd_thandle *oh;
4929         struct iam_path_descr *ipd;
4930         struct iam_container *bag = &obj->oo_dir->od_container;
4931         int rc;
4932
4933         ENTRY;
4934
4935         if (!dt_object_exists(dt))
4936                 RETURN(-ENOENT);
4937
4938         LINVRNT(osd_invariant(obj));
4939         LASSERT(!dt_object_remote(dt));
4940         LASSERT(bag->ic_object == obj->oo_inode);
4941         LASSERT(handle != NULL);
4942
4943         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4944
4945         ipd = osd_idx_ipd_get(env, bag);
4946         if (unlikely(ipd == NULL))
4947                 RETURN(-ENOMEM);
4948
4949         oh = container_of0(handle, struct osd_thandle, ot_super);
4950         LASSERT(oh->ot_handle != NULL);
4951         LASSERT(oh->ot_handle->h_transaction != NULL);
4952
4953         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4954                 /* swab quota uid/gid provided by caller */
4955                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4956                 key = (const struct dt_key *)&oti->oti_quota_id;
4957         }
4958
4959         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4960         osd_ipd_put(env, bag, ipd);
4961         LINVRNT(osd_invariant(obj));
4962         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4963         RETURN(rc);
4964 }
4965
4966 static int osd_index_declare_ea_delete(const struct lu_env *env,
4967                                        struct dt_object *dt,
4968                                        const struct dt_key *key,
4969                                        struct thandle *handle)
4970 {
4971         struct osd_thandle *oh;
4972         struct inode *inode;
4973         int rc, credits;
4974
4975         ENTRY;
4976
4977         LASSERT(!dt_object_remote(dt));
4978         LASSERT(handle != NULL);
4979
4980         oh = container_of0(handle, struct osd_thandle, ot_super);
4981         LASSERT(oh->ot_handle == NULL);
4982
4983         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4984         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4985
4986         inode = osd_dt_obj(dt)->oo_inode;
4987         if (inode == NULL)
4988                 RETURN(-ENOENT);
4989
4990         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4991                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4992                                    NULL, OSD_QID_BLK);
4993         RETURN(rc);
4994 }
4995
4996 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4997                                           struct dt_rec *fid)
4998 {
4999         struct osd_fid_pack *rec;
5000         int rc = -ENODATA;
5001
5002         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5003                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5004                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5005                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5006                         rc = -EINVAL;
5007         }
5008         return rc;
5009 }
5010
5011 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5012                           const struct lu_fid *fid)
5013 {
5014         struct seq_server_site *ss = osd_seq_site(osd);
5015
5016         ENTRY;
5017
5018         /* FID seqs not in FLDB, must be local seq */
5019         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5020                 RETURN(0);
5021
5022         /*
5023          * If FLD is not being initialized yet, it only happens during the
5024          * initialization, likely during mgs initialization, and we assume
5025          * this is local FID.
5026          */
5027         if (ss == NULL || ss->ss_server_fld == NULL)
5028                 RETURN(0);
5029
5030         /* Only check the local FLDB here */
5031         if (osd_seq_exists(env, osd, fid_seq(fid)))
5032                 RETURN(0);
5033
5034         RETURN(1);
5035 }
5036
5037 static void osd_take_care_of_agent(const struct lu_env *env,
5038                                    struct osd_device *osd,
5039                                    struct osd_thandle *oh,
5040                                    struct ldiskfs_dir_entry_2 *de)
5041 {
5042         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5043         struct osd_idmap_cache *idc;
5044         int rc, schedule = 0;
5045
5046         LASSERT(de != NULL);
5047
5048         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5049         if (likely(rc == 0)) {
5050                 idc = osd_idc_find_or_init(env, osd, fid);
5051                 if (IS_ERR(idc) || idc->oic_remote)
5052                         schedule = 1;
5053         } else if (rc == -ENODATA) {
5054                 /*
5055                  * can't get FID, postpone to the end of the
5056                  * transaction when iget() is safe
5057                  */
5058                 schedule = 1;
5059         } else {
5060                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5061         }
5062         if (schedule)
5063                 osd_schedule_agent_inode_removal(env, oh,
5064                                                  le32_to_cpu(de->inode));
5065 }
5066
5067 /**
5068  * Index delete function for interoperability mode (b11826).
5069  * It will remove the directory entry added by osd_index_ea_insert().
5070  * This entry is needed to maintain name->fid mapping.
5071  *
5072  * \param key,  key i.e. file entry to be deleted
5073  *
5074  * \retval   0, on success
5075  * \retval -ve, on error
5076  */
5077 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5078                                const struct dt_key *key, struct thandle *handle)
5079 {
5080         struct osd_object *obj = osd_dt_obj(dt);
5081         struct inode *dir = obj->oo_inode;
5082         struct dentry *dentry;
5083         struct osd_thandle *oh;
5084         struct ldiskfs_dir_entry_2 *de = NULL;
5085         struct buffer_head *bh;
5086         struct htree_lock *hlock = NULL;
5087         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5088         int rc;
5089
5090         ENTRY;
5091
5092         if (!dt_object_exists(dt))
5093                 RETURN(-ENOENT);
5094
5095         LINVRNT(osd_invariant(obj));
5096         LASSERT(!dt_object_remote(dt));
5097         LASSERT(handle != NULL);
5098
5099         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5100
5101         oh = container_of(handle, struct osd_thandle, ot_super);
5102         LASSERT(oh->ot_handle != NULL);
5103         LASSERT(oh->ot_handle->h_transaction != NULL);
5104
5105         ll_vfs_dq_init(dir);
5106         dentry = osd_child_dentry_get(env, obj,
5107                                       (char *)key, strlen((char *)key));
5108
5109         if (obj->oo_hl_head != NULL) {
5110                 hlock = osd_oti_get(env)->oti_hlock;
5111                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5112                                    dir, LDISKFS_HLOCK_DEL);
5113         } else {
5114                 down_write(&obj->oo_ext_idx_sem);
5115         }
5116
5117         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5118         if (!IS_ERR(bh)) {
5119                 /*
5120                  * If this is not the ".." entry, it might be a remote DNE
5121                  * entry and  we need to check if the FID is for a remote
5122                  * MDT.  If the FID is  not in the directory entry (e.g.
5123                  * upgraded 1.8 filesystem without dirdata enabled) then
5124                  * we need to get the FID from the LMA. For a remote directory
5125                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5126                  * case.
5127                  *
5128                  * Delete the entry before the agent inode in order to
5129                  * simplify error handling.  At worst an error after deleting
5130                  * the entry first might leak the agent inode afterward. The
5131                  * reverse would need filesystem abort in case of error deleting
5132                  * the entry after the agent had been removed, or leave a
5133                  * dangling entry pointing at a random inode.
5134                  */
5135                 if (strcmp((char *)key, dotdot) != 0)
5136                         osd_take_care_of_agent(env, osd, oh, de);
5137                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5138                 brelse(bh);
5139         } else {
5140                 rc = PTR_ERR(bh);
5141         }
5142         if (hlock != NULL)
5143                 ldiskfs_htree_unlock(hlock);
5144         else
5145                 up_write(&obj->oo_ext_idx_sem);
5146
5147         GOTO(out, rc);
5148 out:
5149         LASSERT(osd_invariant(obj));
5150         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5151         RETURN(rc);
5152 }
5153
5154 /**
5155  *      Lookup index for \a key and copy record to \a rec.
5156  *
5157  *      \param  dt      osd index object
5158  *      \param  key     key for index
5159  *      \param  rec     record reference
5160  *
5161  *      \retval  +ve  success : exact mach
5162  *      \retval  0    return record with key not greater than \a key
5163  *      \retval -ve   failure
5164  */
5165 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5166                                 struct dt_rec *rec, const struct dt_key *key)
5167 {
5168         struct osd_object *obj = osd_dt_obj(dt);
5169         struct iam_path_descr *ipd;
5170         struct iam_container *bag = &obj->oo_dir->od_container;
5171         struct osd_thread_info *oti = osd_oti_get(env);
5172         struct iam_iterator *it = &oti->oti_idx_it;
5173         struct iam_rec *iam_rec;
5174         int rc;
5175
5176         ENTRY;
5177
5178         if (!dt_object_exists(dt))
5179                 RETURN(-ENOENT);
5180
5181         LASSERT(osd_invariant(obj));
5182         LASSERT(!dt_object_remote(dt));
5183         LASSERT(bag->ic_object == obj->oo_inode);
5184
5185         ipd = osd_idx_ipd_get(env, bag);
5186         if (IS_ERR(ipd))
5187                 RETURN(-ENOMEM);
5188
5189         /* got ipd now we can start iterator. */
5190         iam_it_init(it, bag, 0, ipd);
5191
5192         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5193                 /* swab quota uid/gid provided by caller */
5194                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5195                 key = (const struct dt_key *)&oti->oti_quota_id;
5196         }
5197
5198         rc = iam_it_get(it, (struct iam_key *)key);
5199         if (rc >= 0) {
5200                 if (S_ISDIR(obj->oo_inode->i_mode))
5201                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5202                 else
5203                         iam_rec = (struct iam_rec *)rec;
5204
5205                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5206
5207                 if (S_ISDIR(obj->oo_inode->i_mode))
5208                         osd_fid_unpack((struct lu_fid *)rec,
5209                                        (struct osd_fid_pack *)iam_rec);
5210                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5211                         osd_quota_unpack(obj, rec);
5212         }
5213
5214         iam_it_put(it);
5215         iam_it_fini(it);
5216         osd_ipd_put(env, bag, ipd);
5217
5218         LINVRNT(osd_invariant(obj));
5219
5220         RETURN(rc);
5221 }
5222
5223 static int osd_index_declare_iam_insert(const struct lu_env *env,
5224                                         struct dt_object *dt,
5225                                         const struct dt_rec *rec,
5226                                         const struct dt_key *key,
5227                                         struct thandle *handle)
5228 {
5229         struct osd_thandle *oh;
5230
5231         LASSERT(handle != NULL);
5232
5233         oh = container_of0(handle, struct osd_thandle, ot_super);
5234         LASSERT(oh->ot_handle == NULL);
5235
5236         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5237                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5238
5239         return 0;
5240 }
5241
5242 /**
5243  *      Inserts (key, value) pair in \a dt index object.
5244  *
5245  *      \param  dt      osd index object
5246  *      \param  key     key for index
5247  *      \param  rec     record reference
5248  *      \param  th      transaction handler
5249  *
5250  *      \retval  0  success
5251  *      \retval -ve failure
5252  */
5253 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5254                                 const struct dt_rec *rec,
5255                                 const struct dt_key *key, struct thandle *th,
5256                                 int ignore_quota)
5257 {
5258         struct osd_object *obj = osd_dt_obj(dt);
5259         struct iam_path_descr *ipd;
5260         struct osd_thandle *oh;
5261         struct iam_container *bag;
5262         struct osd_thread_info *oti = osd_oti_get(env);
5263         struct iam_rec *iam_rec;
5264         int rc;
5265
5266         ENTRY;
5267
5268         if (!dt_object_exists(dt))
5269                 RETURN(-ENOENT);
5270
5271         LINVRNT(osd_invariant(obj));
5272         LASSERT(!dt_object_remote(dt));
5273
5274         bag = &obj->oo_dir->od_container;
5275         LASSERT(bag->ic_object == obj->oo_inode);
5276         LASSERT(th != NULL);
5277
5278         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5279
5280         ipd = osd_idx_ipd_get(env, bag);
5281         if (unlikely(ipd == NULL))
5282                 RETURN(-ENOMEM);
5283
5284         oh = container_of0(th, struct osd_thandle, ot_super);
5285         LASSERT(oh->ot_handle != NULL);
5286         LASSERT(oh->ot_handle->h_transaction != NULL);
5287         if (S_ISDIR(obj->oo_inode->i_mode)) {
5288                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5289                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5290                              &oti->oti_fid);
5291         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5292                 /* pack quota uid/gid */
5293                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5294                 key = (const struct dt_key *)&oti->oti_quota_id;
5295                 /* pack quota record */
5296                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5297                 iam_rec = (struct iam_rec *)rec;
5298         } else {
5299                 iam_rec = (struct iam_rec *)rec;
5300         }
5301
5302         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5303                         iam_rec, ipd);
5304         osd_ipd_put(env, bag, ipd);
5305         LINVRNT(osd_invariant(obj));
5306         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5307         RETURN(rc);
5308 }
5309
5310 /**
5311  * Calls ldiskfs_add_entry() to add directory entry
5312  * into the directory. This is required for
5313  * interoperability mode (b11826)
5314  *
5315  * \retval   0, on success
5316  * \retval -ve, on error
5317  */
5318 static int __osd_ea_add_rec(struct osd_thread_info *info,
5319                             struct osd_object *pobj, struct inode  *cinode,
5320                             const char *name, const struct lu_fid *fid,
5321                             struct htree_lock *hlock, struct thandle *th)
5322 {
5323         struct ldiskfs_dentry_param *ldp;
5324         struct dentry *child;
5325         struct osd_thandle *oth;
5326         int rc;
5327
5328         oth = container_of(th, struct osd_thandle, ot_super);
5329         LASSERT(oth->ot_handle != NULL);
5330         LASSERT(oth->ot_handle->h_transaction != NULL);
5331         LASSERT(pobj->oo_inode);
5332
5333         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5334         if (unlikely(pobj->oo_inode ==
5335                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5336                 ldp->edp_magic = 0;
5337         else
5338                 osd_get_ldiskfs_dirent_param(ldp, fid);
5339         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5340         child->d_fsdata = (void *)ldp;
5341         ll_vfs_dq_init(pobj->oo_inode);
5342         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5343                                    child, cinode, hlock);
5344         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5345                 struct ldiskfs_dir_entry_2      *de;
5346                 struct buffer_head              *bh;
5347                 int                              rc1;
5348
5349                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5350                                             NULL, hlock);
5351                 if (!IS_ERR(bh)) {
5352                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5353                                                                bh);
5354                         if (rc1 == 0) {
5355                                 if (S_ISDIR(cinode->i_mode))
5356                                         de->file_type = LDISKFS_DIRENT_LUFID |
5357                                                         LDISKFS_FT_REG_FILE;
5358                                 else
5359                                         de->file_type = LDISKFS_DIRENT_LUFID |
5360                                                         LDISKFS_FT_DIR;
5361                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5362                                                               NULL, bh);
5363                         }
5364                         brelse(bh);
5365                 }
5366         }
5367
5368         RETURN(rc);
5369 }
5370
5371 /**
5372  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5373  * into the directory.Also sets flags into osd object to
5374  * indicate dot and dotdot are created. This is required for
5375  * interoperability mode (b11826)
5376  *
5377  * \param dir   directory for dot and dotdot fixup.
5378  * \param obj   child object for linking
5379  *
5380  * \retval   0, on success
5381  * \retval -ve, on error
5382  */
5383 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5384                               struct osd_object *dir,
5385                               struct inode *parent_dir, const char *name,
5386                               const struct lu_fid *dot_fid,
5387                               const struct lu_fid *dot_dot_fid,
5388                               struct thandle *th)
5389 {
5390         struct inode *inode = dir->oo_inode;
5391         struct osd_thandle *oth;
5392         int result = 0;
5393
5394         oth = container_of(th, struct osd_thandle, ot_super);
5395         LASSERT(oth->ot_handle->h_transaction != NULL);
5396         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5397
5398         if (strcmp(name, dot) == 0) {
5399                 if (dir->oo_compat_dot_created) {
5400                         result = -EEXIST;
5401                 } else {
5402                         LASSERT(inode->i_ino == parent_dir->i_ino);
5403                         dir->oo_compat_dot_created = 1;
5404                         result = 0;
5405                 }
5406         } else if (strcmp(name, dotdot) == 0) {
5407                 if (!dir->oo_compat_dot_created)
5408                         return -EINVAL;
5409                 /* in case of rename, dotdot is already created */
5410                 if (dir->oo_compat_dotdot_created) {
5411                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5412                                                 dot_dot_fid, NULL, th);
5413                 }
5414
5415                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5416                         struct lu_fid tfid = *dot_dot_fid;
5417
5418                         tfid.f_oid--;
5419                         result = osd_add_dot_dotdot_internal(info,
5420                                         dir->oo_inode, parent_dir, dot_fid,
5421                                         &tfid, oth);
5422                 } else {
5423                         result = osd_add_dot_dotdot_internal(info,
5424                                         dir->oo_inode, parent_dir, dot_fid,
5425                                         dot_dot_fid, oth);
5426                 }
5427
5428                 if (result == 0)
5429                         dir->oo_compat_dotdot_created = 1;
5430         }
5431
5432         return result;
5433 }
5434
5435
5436 /**
5437  * It will call the appropriate osd_add* function and return the
5438  * value, return by respective functions.
5439  */
5440 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5441                           struct inode *cinode, const char *name,
5442                           const struct lu_fid *fid, struct thandle *th)
5443 {
5444         struct osd_thread_info *info = osd_oti_get(env);
5445         struct htree_lock *hlock;
5446         int rc;
5447
5448         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5449
5450         if (name[0] == '.' && (name[1] == '\0' ||
5451                                (name[1] == '.' && name[2] == '\0'))) {
5452                 if (hlock != NULL) {
5453                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5454                                            pobj->oo_inode, 0);
5455                 } else {
5456                         down_write(&pobj->oo_ext_idx_sem);
5457                 }
5458
5459                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5460                                         lu_object_fid(&pobj->oo_dt.do_lu),
5461                                         fid, th);
5462         } else {
5463                 if (hlock != NULL) {
5464                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5465                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5466                 } else {
5467                         down_write(&pobj->oo_ext_idx_sem);
5468                 }
5469
5470                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5471                         struct lu_fid *tfid = &info->oti_fid;
5472
5473                         *tfid = *fid;
5474                         tfid->f_ver = ~0;
5475                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5476                                               tfid, hlock, th);
5477                 } else {
5478                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5479                                               hlock, th);
5480                 }
5481         }
5482         if (hlock != NULL)
5483                 ldiskfs_htree_unlock(hlock);
5484         else
5485                 up_write(&pobj->oo_ext_idx_sem);
5486
5487         return rc;
5488 }
5489
5490 static int
5491 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5492                       struct osd_idmap_cache *oic)
5493 {
5494         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5495         struct lu_fid *fid = &oic->oic_fid;
5496         struct osd_inode_id *id = &oic->oic_lid;
5497         struct inode *inode = NULL;
5498         int once = 0;
5499         bool insert;
5500         int rc;
5501
5502         ENTRY;
5503
5504         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5505                 RETURN(0);
5506
5507         if (thread_is_running(&scrub->os_thread) &&
5508             scrub->os_pos_current > id->oii_ino)
5509                 RETURN(0);
5510
5511         if (dev->od_auto_scrub_interval == AS_NEVER ||
5512             ktime_get_real_seconds() <
5513             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5514                 RETURN(0);
5515
5516 again:
5517         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5518         if (rc == -ENOENT) {
5519                 __u32 gen = id->oii_gen;
5520
5521                 insert = true;
5522                 if (inode != NULL)
5523                         goto trigger;
5524
5525                 inode = osd_iget(oti, dev, id);
5526                 /* The inode has been removed (by race maybe). */
5527                 if (IS_ERR(inode)) {
5528                         rc = PTR_ERR(inode);
5529
5530                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5531                 }
5532
5533                 /* The OI mapping is lost. */
5534                 if (gen != OSD_OII_NOGEN)
5535                         goto trigger;
5536
5537                 /*
5538                  * The inode may has been reused by others, we do not know,
5539                  * leave it to be handled by subsequent osd_fid_lookup().
5540                  */
5541                 GOTO(out, rc = 0);
5542         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5543                 GOTO(out, rc);
5544         }
5545
5546         insert = false;
5547
5548 trigger:
5549         if (thread_is_running(&scrub->os_thread)) {
5550                 if (inode == NULL) {
5551                         inode = osd_iget(oti, dev, id);
5552                         /* The inode has been removed (by race maybe). */
5553                         if (IS_ERR(inode)) {
5554                                 rc = PTR_ERR(inode);
5555
5556                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5557                         }
5558                 }
5559
5560                 rc = osd_oii_insert(dev, oic, insert);
5561                 /*
5562                  * There is race condition between osd_oi_lookup and OI scrub.
5563                  * The OI scrub finished just after osd_oi_lookup() failure.
5564                  * Under such case, it is unnecessary to trigger OI scrub again,
5565                  * but try to call osd_oi_lookup() again.
5566                  */
5567                 if (unlikely(rc == -EAGAIN))
5568                         goto again;
5569
5570                 if (!S_ISDIR(inode->i_mode))
5571                         rc = 0;
5572                 else
5573                         rc = osd_check_lmv(oti, dev, inode, oic);
5574
5575                 GOTO(out, rc);
5576         }
5577
5578         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5579                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5580                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5581                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5582                        "%s: trigger partial OI scrub for RPC inconsistency "
5583                        "checking FID "DFID": rc = %d\n",
5584                        osd_dev2name(dev), PFID(fid), rc);
5585                 if (rc == 0 || rc == -EALREADY)
5586                         goto again;
5587         }
5588
5589         GOTO(out, rc);
5590
5591 out:
5592         if (inode)
5593                 iput(inode);
5594
5595         RETURN(rc);
5596 }
5597
5598 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5599                                struct osd_device *dev,
5600                                struct osd_idmap_cache *oic,
5601                                struct lu_fid *fid, __u32 ino)
5602 {
5603         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5604         struct inode *inode;
5605         int rc;
5606
5607         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5608         inode = osd_iget(oti, dev, &oic->oic_lid);
5609         if (IS_ERR(inode)) {
5610                 fid_zero(&oic->oic_fid);
5611                 return PTR_ERR(inode);
5612         }
5613
5614         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5615         iput(inode);
5616         if (rc != 0)
5617                 fid_zero(&oic->oic_fid);
5618         else
5619                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5620         return rc;
5621 }
5622
5623 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5624                       struct osd_inode_id *id, const struct lu_fid *fid)
5625 {
5626         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5627                id->oii_ino, id->oii_gen, info);
5628         info->oti_cache.oic_lid = *id;
5629         info->oti_cache.oic_fid = *fid;
5630         info->oti_cache.oic_dev = osd;
5631 }
5632
5633 /**
5634  * Get parent FID from the linkEA.
5635  *
5636  * For a directory which parent resides on remote MDT, to satisfy the
5637  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5638  * the other hand, to make the lookup(..) on the directory can return
5639  * the real parent FID, we append the real parent FID after its ".."
5640  * name entry in the /REMOTE_PARENT_DIR.
5641  *
5642  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5643  * backup. So after the restore, we cannot get the right parent FID from
5644  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5645  * we have stored the real parent FID in the directory object's linkEA,
5646  * we can parse the linkEA for the real parent FID.
5647  *
5648  * \param[in] env       pointer to the thread context
5649  * \param[in] obj       pointer to the object to be handled
5650  * \param[out]fid       pointer to the buffer to hold the parent FID
5651  *
5652  * \retval              0 for getting the real parent FID successfully
5653  * \retval              negative error number on failure
5654  */
5655 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5656                                     struct osd_object *obj,
5657                                     struct lu_fid *fid)
5658 {
5659         struct osd_thread_info *oti = osd_oti_get(env);
5660         struct lu_buf *buf = &oti->oti_big_buf;
5661         struct dentry *dentry = &oti->oti_obj_dentry;
5662         struct inode *inode = obj->oo_inode;
5663         struct linkea_data ldata = { NULL };
5664         int rc;
5665
5666         ENTRY;
5667
5668         fid_zero(fid);
5669         if (!S_ISDIR(inode->i_mode))
5670                 RETURN(-EIO);
5671
5672 again:
5673         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5674                              buf->lb_buf, buf->lb_len);
5675         if (rc == -ERANGE) {
5676                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5677                                      NULL, 0);
5678                 if (rc > 0) {
5679                         lu_buf_realloc(buf, rc);
5680                         if (buf->lb_buf == NULL)
5681                                 RETURN(-ENOMEM);
5682
5683                         goto again;
5684                 }
5685         }
5686
5687         if (unlikely(rc == 0))
5688                 RETURN(-ENODATA);
5689
5690         if (rc < 0)
5691                 RETURN(rc);
5692
5693         if (unlikely(buf->lb_buf == NULL)) {
5694                 lu_buf_realloc(buf, rc);
5695                 if (buf->lb_buf == NULL)
5696                         RETURN(-ENOMEM);
5697
5698                 goto again;
5699         }
5700
5701         ldata.ld_buf = buf;
5702         rc = linkea_init_with_rec(&ldata);
5703         if (!rc) {
5704                 linkea_first_entry(&ldata);
5705                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5706         }
5707
5708         RETURN(rc);
5709 }
5710
5711 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5712                                     struct inode *inode,
5713                                     const struct lu_fid *pfid,
5714                                     const char *name, const int namelen)
5715 {
5716         struct osd_thread_info *oti = osd_oti_get(env);
5717         struct lu_buf *buf = &oti->oti_big_buf;
5718         struct dentry *dentry = &oti->oti_obj_dentry;
5719         struct linkea_data ldata = { NULL };
5720         struct lu_name cname = { .ln_name = name,
5721                                  .ln_namelen = namelen };
5722         int rc;
5723
5724         ENTRY;
5725
5726 again:
5727         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5728                              buf->lb_buf, buf->lb_len);
5729         if (rc == -ERANGE)
5730                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5731
5732         if (rc < 0)
5733                 RETURN(rc);
5734
5735         if (unlikely(rc == 0))
5736                 RETURN(-ENODATA);
5737
5738         if (buf->lb_len < rc) {
5739                 lu_buf_realloc(buf, rc);
5740                 if (buf->lb_buf == NULL)
5741                         RETURN(-ENOMEM);
5742
5743                 goto again;
5744         }
5745
5746         ldata.ld_buf = buf;
5747         rc = linkea_init_with_rec(&ldata);
5748         if (!rc)
5749                 rc = linkea_links_find(&ldata, &cname, pfid);
5750
5751         RETURN(rc);
5752 }
5753
5754 /**
5755  * Calls ->lookup() to find dentry. From dentry get inode and
5756  * read inode's ea to get fid. This is required for  interoperability
5757  * mode (b11826)
5758  *
5759  * \retval   0, on success
5760  * \retval -ve, on error
5761  */
5762 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5763                              struct dt_rec *rec, const struct dt_key *key)
5764 {
5765         struct inode *dir = obj->oo_inode;
5766         struct dentry *dentry;
5767         struct ldiskfs_dir_entry_2 *de;
5768         struct buffer_head *bh;
5769         struct lu_fid *fid = (struct lu_fid *)rec;
5770         struct htree_lock *hlock = NULL;
5771         int ino;
5772         int rc;
5773
5774         ENTRY;
5775
5776         LASSERT(dir->i_op != NULL);
5777         LASSERT(dir->i_op->lookup != NULL);
5778
5779         dentry = osd_child_dentry_get(env, obj,
5780                                       (char *)key, strlen((char *)key));
5781
5782         if (obj->oo_hl_head != NULL) {
5783                 hlock = osd_oti_get(env)->oti_hlock;
5784                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5785                                    dir, LDISKFS_HLOCK_LOOKUP);
5786         } else {
5787                 down_read(&obj->oo_ext_idx_sem);
5788         }
5789
5790         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5791         if (!IS_ERR(bh)) {
5792                 struct osd_thread_info *oti = osd_oti_get(env);
5793                 struct osd_inode_id *id = &oti->oti_id;
5794                 struct osd_idmap_cache *oic = &oti->oti_cache;
5795                 struct osd_device *dev = osd_obj2dev(obj);
5796
5797                 ino = le32_to_cpu(de->inode);
5798                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5799                         brelse(bh);
5800                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5801                         GOTO(out, rc);
5802                 }
5803
5804                 rc = osd_get_fid_from_dentry(de, rec);
5805
5806                 /* done with de, release bh */
5807                 brelse(bh);
5808                 if (rc != 0) {
5809                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5810                                 const char *name = (const char *)key;
5811
5812                                 /*
5813                                  * If the parent is on remote MDT, and there
5814                                  * is no FID-in-dirent, then we have to get
5815                                  * the parent FID from the linkEA.
5816                                  */
5817                                 if (likely(strlen(name) == 2 &&
5818                                            name[0] == '.' && name[1] == '.'))
5819                                         rc = osd_get_pfid_from_linkea(env, obj,
5820                                                                       fid);
5821                         } else {
5822                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5823                         }
5824                 } else {
5825                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5826                 }
5827
5828                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5829                         fid_zero(&oic->oic_fid);
5830
5831                         GOTO(out, rc);
5832                 }
5833
5834                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5835                 rc = osd_consistency_check(oti, dev, oic);
5836                 if (rc == -ENOENT)
5837                         fid_zero(&oic->oic_fid);
5838                 else
5839                         /* Other error should not affect lookup result. */
5840                         rc = 0;
5841         } else {
5842                 rc = PTR_ERR(bh);
5843         }
5844
5845         GOTO(out, rc);
5846
5847 out:
5848         if (hlock != NULL)
5849                 ldiskfs_htree_unlock(hlock);
5850         else
5851                 up_read(&obj->oo_ext_idx_sem);
5852         return rc;
5853 }
5854
5855 static int osd_index_declare_ea_insert(const struct lu_env *env,
5856                                        struct dt_object *dt,
5857                                        const struct dt_rec *rec,
5858                                        const struct dt_key *key,
5859                                        struct thandle *handle)
5860 {
5861         struct osd_thandle *oh;
5862         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5863         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5864         const struct lu_fid *fid = rec1->rec_fid;
5865         int credits, rc = 0;
5866         struct osd_idmap_cache *idc;
5867
5868         ENTRY;
5869
5870         LASSERT(!dt_object_remote(dt));
5871         LASSERT(handle != NULL);
5872         LASSERT(fid != NULL);
5873         LASSERT(rec1->rec_type != 0);
5874
5875         oh = container_of0(handle, struct osd_thandle, ot_super);
5876         LASSERT(oh->ot_handle == NULL);
5877
5878         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5879
5880         /*
5881          * we can't call iget() while a transactions is running
5882          * (this can lead to a deadlock), but we need to know
5883          * inum and object type. so we find this information at
5884          * declaration and cache in per-thread info
5885          */
5886         idc = osd_idc_find_or_init(env, osd, fid);
5887         if (IS_ERR(idc))
5888                 RETURN(PTR_ERR(idc));
5889         if (idc->oic_remote) {
5890                 /*
5891                  * a reference to remote inode is represented by an
5892                  * agent inode which we have to create
5893                  */
5894                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5895                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5896         }
5897
5898         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5899
5900         if (osd_dt_obj(dt)->oo_inode != NULL) {
5901                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5902
5903                 /*
5904                  * We ignore block quota on meta pool (MDTs), so needn't
5905                  * calculate how many blocks will be consumed by this index
5906                  * insert
5907                  */
5908                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5909                                            i_gid_read(inode),
5910                                            i_projid_read(inode), 0,
5911                                            oh, osd_dt_obj(dt), NULL,
5912                                            OSD_QID_BLK);
5913                 if (rc)
5914                         RETURN(rc);
5915
5916 #ifdef HAVE_PROJECT_QUOTA
5917                 /*
5918                  * Reserve credits for local agent inode to transfer
5919                  * to 0, quota enforcement is ignored in this case.
5920                  */
5921                 if (idc->oic_remote &&
5922                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5923                     i_projid_read(inode) != 0)
5924                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5925                                                   0, i_projid_read(inode),
5926                                                   0, false, PRJQUOTA, true);
5927 #endif
5928         }
5929
5930         RETURN(rc);
5931 }
5932
5933 /**
5934  * Index add function for interoperability mode (b11826).
5935  * It will add the directory entry.This entry is needed to
5936  * maintain name->fid mapping.
5937  *
5938  * \param key it is key i.e. file entry to be inserted
5939  * \param rec it is value of given key i.e. fid
5940  *
5941  * \retval   0, on success
5942  * \retval -ve, on error
5943  */
5944 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5945                                const struct dt_rec *rec,
5946                                const struct dt_key *key, struct thandle *th,
5947                                int ignore_quota)
5948 {
5949         struct osd_object *obj = osd_dt_obj(dt);
5950         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5951         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5952         const struct lu_fid *fid = rec1->rec_fid;
5953         const char *name = (const char *)key;
5954         struct osd_thread_info *oti = osd_oti_get(env);
5955         struct inode *child_inode = NULL;
5956         struct osd_idmap_cache *idc;
5957         int rc;
5958
5959         ENTRY;
5960
5961         if (!dt_object_exists(dt))
5962                 RETURN(-ENOENT);
5963
5964         LASSERT(osd_invariant(obj));
5965         LASSERT(!dt_object_remote(dt));
5966         LASSERT(th != NULL);
5967
5968         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5969
5970         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5971
5972         idc = osd_idc_find(env, osd, fid);
5973         if (unlikely(idc == NULL)) {
5974                 /*
5975                  * this dt_insert() wasn't declared properly, so
5976                  * FID is missing in OI cache. we better do not
5977                  * lookup FID in FLDB/OI and don't risk to deadlock,
5978                  * but in some special cases (lfsck testing, etc)
5979                  * it's much simpler than fixing a caller
5980                  */
5981                 CERROR("%s: "DFID" wasn't declared for insert\n",
5982                        osd_name(osd), PFID(fid));
5983                 dump_stack();
5984                 idc = osd_idc_find_or_init(env, osd, fid);
5985                 if (IS_ERR(idc))
5986                         RETURN(PTR_ERR(idc));
5987         }
5988
5989         if (idc->oic_remote) {
5990                 /* Insert remote entry */
5991                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5992                         child_inode =
5993                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
5994                 } else {
5995                         child_inode = osd_create_local_agent_inode(env, osd,
5996                                         obj, fid, rec1->rec_type & S_IFMT, th);
5997                         if (IS_ERR(child_inode))
5998                                 RETURN(PTR_ERR(child_inode));
5999                 }
6000         } else {
6001                 /* Insert local entry */
6002                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6003                         /* for a reason OI cache wasn't filled properly */
6004                         CERROR("%s: OIC for "DFID" isn't filled\n",
6005                                osd_name(osd), PFID(fid));
6006                         RETURN(-EINVAL);
6007                 }
6008                 child_inode = oti->oti_inode;
6009                 if (unlikely(child_inode == NULL)) {
6010                         struct ldiskfs_inode_info *lii;
6011
6012                         OBD_ALLOC_PTR(lii);
6013                         if (lii == NULL)
6014                                 RETURN(-ENOMEM);
6015                         child_inode = oti->oti_inode = &lii->vfs_inode;
6016                 }
6017                 child_inode->i_sb = osd_sb(osd);
6018                 child_inode->i_ino = idc->oic_lid.oii_ino;
6019                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6020         }
6021
6022         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6023
6024         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6025                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6026
6027         if (child_inode && child_inode != oti->oti_inode)
6028                 iput(child_inode);
6029         LASSERT(osd_invariant(obj));
6030         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6031         RETURN(rc);
6032 }
6033
6034 /**
6035  *  Initialize osd Iterator for given osd index object.
6036  *
6037  *  \param  dt      osd index object
6038  */
6039
6040 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6041                                      struct dt_object *dt,
6042                                      __u32 unused)
6043 {
6044         struct osd_it_iam *it;
6045         struct osd_object *obj = osd_dt_obj(dt);
6046         struct lu_object *lo = &dt->do_lu;
6047         struct iam_path_descr *ipd;
6048         struct iam_container *bag = &obj->oo_dir->od_container;
6049
6050         if (!dt_object_exists(dt))
6051                 return ERR_PTR(-ENOENT);
6052
6053         OBD_ALLOC_PTR(it);
6054         if (it == NULL)
6055                 return ERR_PTR(-ENOMEM);
6056
6057         ipd = osd_it_ipd_get(env, bag);
6058         if (likely(ipd != NULL)) {
6059                 it->oi_obj = obj;
6060                 it->oi_ipd = ipd;
6061                 lu_object_get(lo);
6062                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6063                 return (struct dt_it *)it;
6064         } else {
6065                 OBD_FREE_PTR(it);
6066                 return ERR_PTR(-ENOMEM);
6067         }
6068 }
6069
6070 /**
6071  * free given Iterator.
6072  */
6073 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6074 {
6075         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6076         struct osd_object *obj = it->oi_obj;
6077
6078         iam_it_fini(&it->oi_it);
6079         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6080         osd_object_put(env, obj);
6081         OBD_FREE_PTR(it);
6082 }
6083
6084 /**
6085  *  Move Iterator to record specified by \a key
6086  *
6087  *  \param  di      osd iterator
6088  *  \param  key     key for index
6089  *
6090  *  \retval +ve  di points to record with least key not larger than key
6091  *  \retval  0   di points to exact matched key
6092  *  \retval -ve  failure
6093  */
6094
6095 static int osd_it_iam_get(const struct lu_env *env,
6096                           struct dt_it *di, const struct dt_key *key)
6097 {
6098         struct osd_thread_info *oti = osd_oti_get(env);
6099         struct osd_it_iam *it = (struct osd_it_iam *)di;
6100
6101         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6102                 /* swab quota uid/gid */
6103                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6104                 key = (struct dt_key *)&oti->oti_quota_id;
6105         }
6106
6107         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6108 }
6109
6110 /**
6111  *  Release Iterator
6112  *
6113  *  \param  di      osd iterator
6114  */
6115 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6116 {
6117         struct osd_it_iam *it = (struct osd_it_iam *)di;
6118
6119         iam_it_put(&it->oi_it);
6120 }
6121
6122 /**
6123  *  Move iterator by one record
6124  *
6125  *  \param  di      osd iterator
6126  *
6127  *  \retval +1   end of container reached
6128  *  \retval  0   success
6129  *  \retval -ve  failure
6130  */
6131
6132 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6133 {
6134         struct osd_it_iam *it = (struct osd_it_iam *)di;
6135
6136         return iam_it_next(&it->oi_it);
6137 }
6138
6139 /**
6140  * Return pointer to the key under iterator.
6141  */
6142
6143 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6144                                      const struct dt_it *di)
6145 {
6146         struct osd_thread_info *oti = osd_oti_get(env);
6147         struct osd_it_iam *it = (struct osd_it_iam *)di;
6148         struct osd_object *obj = it->oi_obj;
6149         struct dt_key *key;
6150
6151         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6152
6153         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6154                 /* swab quota uid/gid */
6155                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6156                 key = (struct dt_key *)&oti->oti_quota_id;
6157         }
6158
6159         return key;
6160 }
6161
6162 /**
6163  * Return size of key under iterator (in bytes)
6164  */
6165
6166 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6167 {
6168         struct osd_it_iam *it = (struct osd_it_iam *)di;
6169
6170         return iam_it_key_size(&it->oi_it);
6171 }
6172
6173 static inline void
6174 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6175 {
6176         /* check if file type is required */
6177         if (ent->lde_attrs & LUDA_TYPE) {
6178                 struct luda_type *lt;
6179                 int align = sizeof(*lt) - 1;
6180
6181                 len = (len + align) & ~align;
6182                 lt = (struct luda_type *)(ent->lde_name + len);
6183                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6184         }
6185
6186         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6187 }
6188
6189 /**
6190  * build lu direct from backend fs dirent.
6191  */
6192
6193 static inline void
6194 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6195                    char *name, __u16 namelen, __u16 type, __u32 attr)
6196 {
6197         ent->lde_attrs = attr | LUDA_FID;
6198         fid_cpu_to_le(&ent->lde_fid, fid);
6199
6200         ent->lde_hash = cpu_to_le64(offset);
6201         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6202
6203         strncpy(ent->lde_name, name, namelen);
6204         ent->lde_name[namelen] = '\0';
6205         ent->lde_namelen = cpu_to_le16(namelen);
6206
6207         /* append lustre attributes */
6208         osd_it_append_attrs(ent, namelen, type);
6209 }
6210
6211 /**
6212  * Return pointer to the record under iterator.
6213  */
6214 static int osd_it_iam_rec(const struct lu_env *env,
6215                           const struct dt_it *di,
6216                           struct dt_rec *dtrec, __u32 attr)
6217 {
6218         struct osd_it_iam *it = (struct osd_it_iam *)di;
6219         struct osd_thread_info *info = osd_oti_get(env);
6220
6221         ENTRY;
6222
6223         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6224                 const struct osd_fid_pack *rec;
6225                 struct lu_fid *fid = &info->oti_fid;
6226                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6227                 char *name;
6228                 int namelen;
6229                 __u64 hash;
6230                 int rc;
6231
6232                 name = (char *)iam_it_key_get(&it->oi_it);
6233                 if (IS_ERR(name))
6234                         RETURN(PTR_ERR(name));
6235
6236                 namelen = iam_it_key_size(&it->oi_it);
6237
6238                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6239                 if (IS_ERR(rec))
6240                         RETURN(PTR_ERR(rec));
6241
6242                 rc = osd_fid_unpack(fid, rec);
6243                 if (rc)
6244                         RETURN(rc);
6245
6246                 hash = iam_it_store(&it->oi_it);
6247
6248                 /* IAM does not store object type in IAM index (dir) */
6249                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6250                                    0, LUDA_FID);
6251         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6252                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6253                            (struct iam_rec *)dtrec);
6254                 osd_quota_unpack(it->oi_obj, dtrec);
6255         } else {
6256                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6257                            (struct iam_rec *)dtrec);
6258         }
6259
6260         RETURN(0);
6261 }
6262
6263 /**
6264  * Returns cookie for current Iterator position.
6265  */
6266 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6267 {
6268         struct osd_it_iam *it = (struct osd_it_iam *)di;
6269
6270         return iam_it_store(&it->oi_it);
6271 }
6272
6273 /**
6274  * Restore iterator from cookie.
6275  *
6276  * \param  di      osd iterator
6277  * \param  hash    Iterator location cookie
6278  *
6279  * \retval +ve  di points to record with least key not larger than key.
6280  * \retval  0   di points to exact matched key
6281  * \retval -ve  failure
6282  */
6283
6284 static int osd_it_iam_load(const struct lu_env *env,
6285                            const struct dt_it *di, __u64 hash)
6286 {
6287         struct osd_it_iam *it = (struct osd_it_iam *)di;
6288
6289         return iam_it_load(&it->oi_it, hash);
6290 }
6291
6292 static const struct dt_index_operations osd_index_iam_ops = {
6293         .dio_lookup         = osd_index_iam_lookup,
6294         .dio_declare_insert = osd_index_declare_iam_insert,
6295         .dio_insert         = osd_index_iam_insert,
6296         .dio_declare_delete = osd_index_declare_iam_delete,
6297         .dio_delete         = osd_index_iam_delete,
6298         .dio_it     = {
6299                 .init     = osd_it_iam_init,
6300                 .fini     = osd_it_iam_fini,
6301                 .get      = osd_it_iam_get,
6302                 .put      = osd_it_iam_put,
6303                 .next     = osd_it_iam_next,
6304                 .key      = osd_it_iam_key,
6305                 .key_size = osd_it_iam_key_size,
6306                 .rec      = osd_it_iam_rec,
6307                 .store    = osd_it_iam_store,
6308                 .load     = osd_it_iam_load
6309         }
6310 };
6311
6312
6313 /**
6314  * Creates or initializes iterator context.
6315  *
6316  * \retval struct osd_it_ea, iterator structure on success
6317  *
6318  */
6319 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6320                                     struct dt_object *dt,
6321                                     __u32 attr)
6322 {
6323         struct osd_object *obj = osd_dt_obj(dt);
6324         struct osd_thread_info *info = osd_oti_get(env);
6325         struct osd_it_ea *oie;
6326         struct file *file;
6327         struct lu_object *lo = &dt->do_lu;
6328         struct dentry *obj_dentry;
6329
6330         ENTRY;
6331
6332         if (!dt_object_exists(dt) || obj->oo_destroyed)
6333                 RETURN(ERR_PTR(-ENOENT));
6334
6335         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6336         if (oie == NULL)
6337                 RETURN(ERR_PTR(-ENOMEM));
6338         obj_dentry = &oie->oie_dentry;
6339
6340         obj_dentry->d_inode = obj->oo_inode;
6341         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6342         obj_dentry->d_name.hash = 0;
6343
6344         oie->oie_rd_dirent       = 0;
6345         oie->oie_it_dirent       = 0;
6346         oie->oie_dirent          = NULL;
6347         if (unlikely(!info->oti_it_ea_buf_used)) {
6348                 oie->oie_buf = info->oti_it_ea_buf;
6349                 info->oti_it_ea_buf_used = 1;
6350         } else {
6351                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6352                 if (oie->oie_buf == NULL)
6353                         RETURN(ERR_PTR(-ENOMEM));
6354         }
6355         oie->oie_obj = obj;
6356
6357         file = &oie->oie_file;
6358
6359         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6360         if (attr & LUDA_64BITHASH)
6361                 file->f_mode    = FMODE_64BITHASH;
6362         else
6363                 file->f_mode    = FMODE_32BITHASH;
6364         file->f_path.dentry     = obj_dentry;
6365         file->f_mapping         = obj->oo_inode->i_mapping;
6366         file->f_op              = obj->oo_inode->i_fop;
6367         set_file_inode(file, obj->oo_inode);
6368
6369         lu_object_get(lo);
6370         RETURN((struct dt_it *)oie);
6371 }
6372
6373 /**
6374  * Destroy or finishes iterator context.
6375  *
6376  * \param di iterator structure to be destroyed
6377  */
6378 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6379 {
6380         struct osd_thread_info *info = osd_oti_get(env);
6381         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6382         struct osd_object *obj = oie->oie_obj;
6383         struct inode *inode = obj->oo_inode;
6384
6385         ENTRY;
6386         oie->oie_file.f_op->release(inode, &oie->oie_file);
6387         osd_object_put(env, obj);
6388         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6389                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6390         else
6391                 info->oti_it_ea_buf_used = 0;
6392         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6393         EXIT;
6394 }
6395
6396 /**
6397  * It position the iterator at given key, so that next lookup continues from
6398  * that key Or it is similar to dio_it->load() but based on a key,
6399  * rather than file position.
6400  *
6401  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6402  * to the beginning.
6403  *
6404  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6405  */
6406 static int osd_it_ea_get(const struct lu_env *env,
6407                          struct dt_it *di, const struct dt_key *key)
6408 {
6409         struct osd_it_ea *it = (struct osd_it_ea *)di;
6410
6411         ENTRY;
6412         LASSERT(((const char *)key)[0] == '\0');
6413         it->oie_file.f_pos = 0;
6414         it->oie_rd_dirent = 0;
6415         it->oie_it_dirent = 0;
6416         it->oie_dirent = NULL;
6417
6418         RETURN(+1);
6419 }
6420
6421 /**
6422  * Does nothing
6423  */
6424 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6425 {
6426 }
6427
6428 struct osd_filldir_cbs {
6429 #ifdef HAVE_DIR_CONTEXT
6430         struct dir_context ctx;
6431 #endif
6432         struct osd_it_ea  *it;
6433 };
6434 /**
6435  * It is called internally by ->readdir(). It fills the
6436  * iterator's in-memory data structure with required
6437  * information i.e. name, namelen, rec_size etc.
6438  *
6439  * \param buf in which information to be filled in.
6440  * \param name name of the file in given dir
6441  *
6442  * \retval 0 on success
6443  * \retval 1 on buffer full
6444  */
6445 #ifdef HAVE_FILLDIR_USE_CTX
6446 static int osd_ldiskfs_filldir(struct dir_context *buf,
6447 #else
6448 static int osd_ldiskfs_filldir(void *buf,
6449 #endif
6450                                const char *name, int namelen,
6451                                loff_t offset, __u64 ino, unsigned int d_type)
6452 {
6453         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6454         struct osd_object *obj = it->oie_obj;
6455         struct osd_it_ea_dirent *ent = it->oie_dirent;
6456         struct lu_fid *fid = &ent->oied_fid;
6457         struct osd_fid_pack *rec;
6458
6459         ENTRY;
6460
6461 /* this should never happen */
6462         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6463                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6464                 RETURN(-EIO);
6465         }
6466
6467         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6468             OSD_IT_EA_BUFSIZE)
6469                 RETURN(1);
6470
6471         /* "." is just the object itself. */
6472         if (namelen == 1 && name[0] == '.') {
6473                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6474         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6475                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6476                 if (osd_fid_unpack(fid, rec) != 0)
6477                         fid_zero(fid);
6478         } else {
6479                 fid_zero(fid);
6480         }
6481         d_type &= ~LDISKFS_DIRENT_LUFID;
6482
6483         /* NOT export local root. */
6484         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6485                 ino = obj->oo_inode->i_ino;
6486                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6487         }
6488
6489         ent->oied_ino     = ino;
6490         ent->oied_off     = offset;
6491         ent->oied_namelen = namelen;
6492         ent->oied_type    = d_type;
6493
6494         memcpy(ent->oied_name, name, namelen);
6495
6496         it->oie_rd_dirent++;
6497         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6498         RETURN(0);
6499 }
6500
6501 /**
6502  * Calls ->readdir() to load a directory entry at a time
6503  * and stored it in iterator's in-memory data structure.
6504  *
6505  * \param di iterator's in memory structure
6506  *
6507  * \retval   0 on success
6508  * \retval -ve on error
6509  * \retval +1 reach the end of entry
6510  */
6511 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6512                                const struct dt_it *di)
6513 {
6514         struct osd_it_ea *it = (struct osd_it_ea *)di;
6515         struct osd_object *obj = it->oie_obj;
6516         struct inode *inode = obj->oo_inode;
6517         struct htree_lock *hlock = NULL;
6518         struct file *filp = &it->oie_file;
6519         int rc = 0;
6520         struct osd_filldir_cbs buf = {
6521 #ifdef HAVE_DIR_CONTEXT
6522                 .ctx.actor = osd_ldiskfs_filldir,
6523 #endif
6524                 .it = it
6525         };
6526
6527         ENTRY;
6528         it->oie_dirent = it->oie_buf;
6529         it->oie_rd_dirent = 0;
6530
6531         if (obj->oo_hl_head != NULL) {
6532                 hlock = osd_oti_get(env)->oti_hlock;
6533                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6534                                    inode, LDISKFS_HLOCK_READDIR);
6535         } else {
6536                 down_read(&obj->oo_ext_idx_sem);
6537         }
6538
6539 #ifdef HAVE_DIR_CONTEXT
6540         buf.ctx.pos = filp->f_pos;
6541 #ifdef HAVE_ITERATE_SHARED
6542         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6543 #else
6544         rc = inode->i_fop->iterate(filp, &buf.ctx);
6545 #endif
6546         filp->f_pos = buf.ctx.pos;
6547 #else
6548         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6549 #endif
6550
6551         if (hlock != NULL)
6552                 ldiskfs_htree_unlock(hlock);
6553         else
6554                 up_read(&obj->oo_ext_idx_sem);
6555
6556         if (it->oie_rd_dirent == 0) {
6557                 /*
6558                  * If it does not get any dirent, it means it has been reached
6559                  * to the end of the dir
6560                  */
6561                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6562                 if (rc == 0)
6563                         rc = 1;
6564         } else {
6565                 it->oie_dirent = it->oie_buf;
6566                 it->oie_it_dirent = 1;
6567         }
6568
6569         RETURN(rc);
6570 }
6571
6572 /**
6573  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6574  * to load a directory entry at a time and stored it in
6575  * iterator's in-memory data structure.
6576  *
6577  * \param di iterator's in memory structure
6578  *
6579  * \retval +ve iterator reached to end
6580  * \retval   0 iterator not reached to end
6581  * \retval -ve on error
6582  */
6583 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6584 {
6585         struct osd_it_ea *it = (struct osd_it_ea *)di;
6586         int rc;
6587
6588         ENTRY;
6589
6590         if (it->oie_it_dirent < it->oie_rd_dirent) {
6591                 it->oie_dirent =
6592                         (void *)it->oie_dirent +
6593                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6594                                        it->oie_dirent->oied_namelen);
6595                 it->oie_it_dirent++;
6596                 rc = 0;
6597         } else {
6598                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6599                         rc = 1;
6600                 else
6601                         rc = osd_ldiskfs_it_fill(env, di);
6602         }
6603
6604         RETURN(rc);
6605 }
6606
6607 /**
6608  * Returns the key at current position from iterator's in memory structure.
6609  *
6610  * \param di iterator's in memory structure
6611  *
6612  * \retval key i.e. struct dt_key on success
6613  */
6614 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6615                                     const struct dt_it *di)
6616 {
6617         struct osd_it_ea *it = (struct osd_it_ea *)di;
6618
6619         return (struct dt_key *)it->oie_dirent->oied_name;
6620 }
6621
6622 /**
6623  * Returns key's size at current position from iterator's in memory structure.
6624  *
6625  * \param di iterator's in memory structure
6626  *
6627  * \retval key_size i.e. struct dt_key on success
6628  */
6629 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6630 {
6631         struct osd_it_ea *it = (struct osd_it_ea *)di;
6632
6633         return it->oie_dirent->oied_namelen;
6634 }
6635
6636 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6637 {
6638         if (LDISKFS_DIR_REC_LEN(de) >=
6639             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6640                 return true;
6641
6642         return false;
6643 }
6644
6645 static inline bool
6646 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6647                      unsigned int blocksize, bool dotdot)
6648 {
6649         if (dotdot)
6650                 return osd_dotdot_has_space(de);
6651
6652         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6653             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6654                 return true;
6655
6656         return false;
6657 }
6658
6659 static int
6660 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6661                     handle_t *jh, struct dentry *dentry,
6662                     const struct lu_fid *fid, struct buffer_head *bh,
6663                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6664                     bool dotdot)
6665 {
6666         struct inode *dir = dentry->d_parent->d_inode;
6667         struct inode *inode = dentry->d_inode;
6668         struct osd_fid_pack *rec;
6669         struct ldiskfs_dentry_param *ldp;
6670         int namelen = dentry->d_name.len;
6671         int rc;
6672         struct osd_thread_info *info = osd_oti_get(env);
6673
6674         ENTRY;
6675
6676         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6677                 RETURN(0);
6678
6679         /* There is enough space to hold the FID-in-dirent. */
6680         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6681                 rc = ldiskfs_journal_get_write_access(jh, bh);
6682                 if (rc != 0)
6683                         RETURN(rc);
6684
6685                 de->name[namelen] = 0;
6686                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6687                 rec->fp_len = sizeof(struct lu_fid) + 1;
6688                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6689                 de->file_type |= LDISKFS_DIRENT_LUFID;
6690                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6691
6692                 RETURN(rc);
6693         }
6694
6695         LASSERT(!dotdot);
6696
6697         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6698         if (rc != 0)
6699                 RETURN(rc);
6700
6701         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6702         osd_get_ldiskfs_dirent_param(ldp, fid);
6703         dentry->d_fsdata = (void *)ldp;
6704         ll_vfs_dq_init(dir);
6705         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6706         /*
6707          * It is too bad, we cannot reinsert the name entry back.
6708          * That means we lose it!
6709          */
6710         if (rc != 0)
6711                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6712                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6713                        osd_ino2name(inode),
6714                        dir->i_ino, dir->i_generation, namelen,
6715                        dentry->d_name.name, PFID(fid), rc);
6716
6717         RETURN(rc);
6718 }
6719
6720 static int
6721 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6722                         struct osd_it_ea *it, struct lu_fid *fid,
6723                         struct osd_inode_id *id, __u32 *attr)
6724 {
6725         struct osd_thread_info *info = osd_oti_get(env);
6726         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6727         struct osd_device *dev = osd_obj2dev(obj);
6728         struct super_block *sb = osd_sb(dev);
6729         const char *devname = osd_name(dev);
6730         struct osd_it_ea_dirent *ent = it->oie_dirent;
6731         struct inode *dir = obj->oo_inode;
6732         struct htree_lock *hlock = NULL;
6733         struct buffer_head *bh = NULL;
6734         handle_t *jh = NULL;
6735         struct ldiskfs_dir_entry_2 *de;
6736         struct dentry *dentry;
6737         struct inode *inode;
6738         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6739         int credits;
6740         int rc;
6741         bool dotdot = false;
6742         bool dirty = false;
6743
6744         ENTRY;
6745
6746         if (ent->oied_name[0] == '.') {
6747                 if (ent->oied_namelen == 1)
6748                         RETURN(0);
6749
6750                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6751                         dotdot = true;
6752         }
6753
6754         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6755         inode = osd_iget(info, dev, id);
6756         if (IS_ERR(inode)) {
6757                 rc = PTR_ERR(inode);
6758                 if (rc == -ENOENT || rc == -ESTALE) {
6759                         /*
6760                          * Maybe dangling name entry, or
6761                          * corrupted directory entry.
6762                          */
6763                         *attr |= LUDA_UNKNOWN;
6764                         rc = 0;
6765                 } else {
6766                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6767                                "check_repair, dir = %lu/%u, name = %.*s, "
6768                                "ino = %llu, rc = %d\n",
6769                                devname, dir->i_ino, dir->i_generation,
6770                                ent->oied_namelen, ent->oied_name,
6771                                ent->oied_ino, rc);
6772                 }
6773
6774                 RETURN(rc);
6775         }
6776
6777         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6778                                            ent->oied_namelen);
6779         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6780         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6781                 lma = NULL;
6782         else if (rc != 0)
6783                 GOTO(out, rc);
6784
6785         /*
6786          * We need to ensure that the name entry is still valid.
6787          * Because it may be removed or renamed by other already.
6788          *
6789          * The unlink or rename operation will start journal before PDO lock,
6790          * so to avoid deadlock, here we need to start journal handle before
6791          * related PDO lock also. But because we do not know whether there
6792          * will be something to be repaired before PDO lock, we just start
6793          * journal without conditions.
6794          *
6795          * We may need to remove the name entry firstly, then insert back.
6796          * One credit is for user quota file update.
6797          * One credit is for group quota file update.
6798          * Two credits are for dirty inode.
6799          */
6800         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6801                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6802
6803         if (dev->od_dirent_journal != 0) {
6804
6805 again:
6806                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6807                 if (IS_ERR(jh)) {
6808                         rc = PTR_ERR(jh);
6809                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6810                                "check_repair, dir = %lu/%u, credits = %d, "
6811                                "name = %.*s, ino = %llu: rc = %d\n",
6812                                devname, dir->i_ino, dir->i_generation, credits,
6813                                ent->oied_namelen, ent->oied_name,
6814                                ent->oied_ino, rc);
6815
6816                         GOTO(out_inode, rc);
6817                 }
6818
6819                 if (obj->oo_hl_head != NULL) {
6820                         hlock = osd_oti_get(env)->oti_hlock;
6821                         /*
6822                          * "0" means exclusive lock for the whole directory.
6823                          * We need to prevent others access such name entry
6824                          * during the delete + insert. Neither HLOCK_ADD nor
6825                          * HLOCK_DEL cannot guarantee the atomicity.
6826                          */
6827                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6828                 } else {
6829                         down_write(&obj->oo_ext_idx_sem);
6830                 }
6831         } else {
6832                 if (obj->oo_hl_head != NULL) {
6833                         hlock = osd_oti_get(env)->oti_hlock;
6834                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6835                                            LDISKFS_HLOCK_LOOKUP);
6836                 } else {
6837                         down_read(&obj->oo_ext_idx_sem);
6838                 }
6839         }
6840
6841         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6842         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6843                 *attr |= LUDA_IGNORE;
6844
6845                 GOTO(out, rc = 0);
6846         }
6847
6848         /*
6849          * For dotdot entry, if there is not enough space to hold the
6850          * FID-in-dirent, just keep them there. It only happens when the
6851          * device upgraded from 1.8 or restored from MDT file-level backup.
6852          * For the whole directory, only dotdot entry have no FID-in-dirent
6853          * and needs to get FID from LMA when readdir, it will not affect the
6854          * performance much.
6855          */
6856         if (dotdot && !osd_dotdot_has_space(de)) {
6857                 *attr |= LUDA_UNKNOWN;
6858
6859                 GOTO(out, rc = 0);
6860         }
6861
6862         if (lma != NULL) {
6863                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6864                         GOTO(out, rc = 0);
6865
6866                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6867                         struct lu_fid *tfid = &lma->lma_self_fid;
6868
6869                         if (likely(dotdot &&
6870                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6871                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6872                                 /*
6873                                  * It must be REMOTE_PARENT_DIR and as the
6874                                  * 'dotdot' entry of remote directory
6875                                  */
6876                                 *attr |= LUDA_IGNORE;
6877                         } else {
6878                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6879                                        "parent directory, but got %.*s under "
6880                                        "dir = %lu/%u with the FID "DFID"\n",
6881                                        devname, ent->oied_namelen,
6882                                        ent->oied_name, dir->i_ino,
6883                                        dir->i_generation, PFID(tfid));
6884
6885                                 *attr |= LUDA_UNKNOWN;
6886                         }
6887
6888                         GOTO(out, rc = 0);
6889                 }
6890         }
6891
6892         if (!fid_is_zero(fid)) {
6893                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6894                                               ent->oied_namelen);
6895                 if (rc == -ENOENT ||
6896                     (rc == -ENODATA &&
6897                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6898                         /*
6899                          * linkEA does not recognize the dirent entry,
6900                          * it may because the dirent entry corruption
6901                          * and points to other's inode.
6902                          */
6903                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6904                                "recognize the dirent, dir = %lu/%u, "
6905                                " name = %.*s, ino = %llu, "
6906                                DFID": rc = %d\n", devname, dir->i_ino,
6907                                dir->i_generation, ent->oied_namelen,
6908                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6909                         *attr |= LUDA_UNKNOWN;
6910
6911                         GOTO(out, rc = 0);
6912                 }
6913
6914                 if (rc && rc != -ENODATA) {
6915                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6916                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6917                                DFID": rc = %d\n", devname, dir->i_ino,
6918                                dir->i_generation, ent->oied_namelen,
6919                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6920                         *attr |= LUDA_UNKNOWN;
6921
6922                         GOTO(out, rc = 0);
6923                 }
6924         }
6925
6926         if (lma != NULL) {
6927                 /*
6928                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6929                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6930                  */
6931                 if (*attr & LUDA_VERIFY_DRYRUN) {
6932                         *fid = lma->lma_self_fid;
6933                         *attr |= LUDA_REPAIR;
6934
6935                         GOTO(out, rc = 0);
6936                 }
6937
6938                 if (jh == NULL) {
6939                         brelse(bh);
6940                         dev->od_dirent_journal = 1;
6941                         if (hlock != NULL) {
6942                                 ldiskfs_htree_unlock(hlock);
6943                                 hlock = NULL;
6944                         } else {
6945                                 up_read(&obj->oo_ext_idx_sem);
6946                         }
6947
6948                         goto again;
6949                 }
6950
6951                 *fid = lma->lma_self_fid;
6952                 dirty = true;
6953                 /* Update or append the FID-in-dirent. */
6954                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6955                                          bh, de, hlock, dotdot);
6956                 if (rc == 0)
6957                         *attr |= LUDA_REPAIR;
6958                 else
6959                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6960                                "the dirent, dir = %lu/%u, name = %.*s, "
6961                                "ino = %llu, "DFID": rc = %d\n",
6962                                devname, dir->i_ino, dir->i_generation,
6963                                ent->oied_namelen, ent->oied_name,
6964                                ent->oied_ino, PFID(fid), rc);
6965         } else {
6966                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6967                 if (*attr & LUDA_VERIFY_DRYRUN) {
6968                         if (fid_is_sane(fid)) {
6969                                 *attr |= LUDA_REPAIR;
6970                         } else if (dev->od_index == 0) {
6971                                 lu_igif_build(fid, inode->i_ino,
6972                                               inode->i_generation);
6973                                 *attr |= LUDA_UPGRADE;
6974                         }
6975
6976                         GOTO(out, rc = 0);
6977                 }
6978
6979                 if (jh == NULL) {
6980                         brelse(bh);
6981                         dev->od_dirent_journal = 1;
6982                         if (hlock != NULL) {
6983                                 ldiskfs_htree_unlock(hlock);
6984                                 hlock = NULL;
6985                         } else {
6986                                 up_read(&obj->oo_ext_idx_sem);
6987                         }
6988
6989                         goto again;
6990                 }
6991
6992                 dirty = true;
6993                 if (unlikely(fid_is_sane(fid))) {
6994                         /*
6995                          * FID-in-dirent exists, but FID-in-LMA is lost.
6996                          * Trust the FID-in-dirent, and add FID-in-LMA.
6997                          */
6998                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6999                         if (rc == 0)
7000                                 *attr |= LUDA_REPAIR;
7001                         else
7002                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7003                                        "update dirent, dir = %lu/%u, "
7004                                        "name = %.*s, ino = %llu, "
7005                                        DFID": rc = %d\n",
7006                                        devname, dir->i_ino, dir->i_generation,
7007                                        ent->oied_namelen, ent->oied_name,
7008                                        ent->oied_ino, PFID(fid), rc);
7009                 } else if (dev->od_index == 0) {
7010                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7011                         /*
7012                          * It is probably IGIF object. Only aappend the
7013                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7014                          */
7015                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7016                                                  bh, de, hlock, dotdot);
7017                         if (rc == 0)
7018                                 *attr |= LUDA_UPGRADE;
7019                         else
7020                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7021                                        "after the dirent, dir = %lu/%u, "
7022                                        "name = %.*s, ino = %llu, "
7023                                        DFID": rc = %d\n",
7024                                        devname, dir->i_ino, dir->i_generation,
7025                                        ent->oied_namelen, ent->oied_name,
7026                                        ent->oied_ino, PFID(fid), rc);
7027                 }
7028         }
7029
7030         GOTO(out, rc);
7031
7032 out:
7033         if (!IS_ERR(bh))
7034                 brelse(bh);
7035         if (hlock != NULL) {
7036                 ldiskfs_htree_unlock(hlock);
7037         } else {
7038                 if (dev->od_dirent_journal != 0)
7039                         up_write(&obj->oo_ext_idx_sem);
7040                 else
7041                         up_read(&obj->oo_ext_idx_sem);
7042         }
7043
7044         if (jh != NULL)
7045                 ldiskfs_journal_stop(jh);
7046
7047 out_inode:
7048         iput(inode);
7049         if (rc >= 0 && !dirty)
7050                 dev->od_dirent_journal = 0;
7051
7052         return rc;
7053 }
7054
7055 /**
7056  * Returns the value at current position from iterator's in memory structure.
7057  *
7058  * \param di struct osd_it_ea, iterator's in memory structure
7059  * \param attr attr requested for dirent.
7060  * \param lde lustre dirent
7061  *
7062  * \retval   0 no error and \param lde has correct lustre dirent.
7063  * \retval -ve on error
7064  */
7065 static inline int osd_it_ea_rec(const struct lu_env *env,
7066                                 const struct dt_it *di,
7067                                 struct dt_rec *dtrec, __u32 attr)
7068 {
7069         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7070         struct osd_object      *obj   = it->oie_obj;
7071         struct osd_device      *dev   = osd_obj2dev(obj);
7072         struct osd_thread_info *oti   = osd_oti_get(env);
7073         struct osd_inode_id    *id    = &oti->oti_id;
7074         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7075         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7076         __u32 ino = it->oie_dirent->oied_ino;
7077         int rc = 0;
7078
7079         ENTRY;
7080
7081         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7082
7083         if (attr & LUDA_VERIFY) {
7084                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7085                         attr |= LUDA_IGNORE;
7086                         /*
7087                          * If the parent is on remote MDT, and there
7088                          * is no FID-in-dirent, then we have to get
7089                          * the parent FID from the linkEA.
7090                          */
7091                         if (!fid_is_sane(fid) &&
7092                             it->oie_dirent->oied_namelen == 2 &&
7093                             it->oie_dirent->oied_name[0] == '.' &&
7094                             it->oie_dirent->oied_name[1] == '.')
7095                                 osd_get_pfid_from_linkea(env, obj, fid);
7096                 } else {
7097                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7098                                                      &attr);
7099                 }
7100
7101                 if (!fid_is_sane(fid))
7102                         attr |= LUDA_UNKNOWN;
7103         } else {
7104                 attr &= ~LU_DIRENT_ATTRS_MASK;
7105                 if (!fid_is_sane(fid)) {
7106                         bool is_dotdot = false;
7107
7108                         if (it->oie_dirent->oied_namelen == 2 &&
7109                             it->oie_dirent->oied_name[0] == '.' &&
7110                             it->oie_dirent->oied_name[1] == '.')
7111                                 is_dotdot = true;
7112                         /*
7113                          * If the parent is on remote MDT, and there
7114                          * is no FID-in-dirent, then we have to get
7115                          * the parent FID from the linkEA.
7116                          */
7117                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7118                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7119                         } else {
7120                                 if (is_dotdot == false &&
7121                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7122                                         RETURN(-ENOENT);
7123
7124                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7125                         }
7126                 } else {
7127                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7128                 }
7129         }
7130
7131         /* Pack the entry anyway, at least the offset is right. */
7132         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7133                            it->oie_dirent->oied_name,
7134                            it->oie_dirent->oied_namelen,
7135                            it->oie_dirent->oied_type, attr);
7136
7137         if (rc < 0)
7138                 RETURN(rc);
7139
7140         if (osd_remote_fid(env, dev, fid))
7141                 RETURN(0);
7142
7143         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7144                 osd_add_oi_cache(oti, dev, id, fid);
7145
7146         RETURN(rc > 0 ? 0 : rc);
7147 }
7148
7149 /**
7150  * Returns the record size size at current position.
7151  *
7152  * This function will return record(lu_dirent) size in bytes.
7153  *
7154  * \param[in] env       execution environment
7155  * \param[in] di        iterator's in memory structure
7156  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7157  *                      calculate the lu_dirent size.
7158  *
7159  * \retval      record size(in bytes & in memory) of the current lu_dirent
7160  *              entry.
7161  */
7162 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7163                               __u32 attr)
7164 {
7165         struct osd_it_ea *it = (struct osd_it_ea *)di;
7166
7167         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7168 }
7169
7170 /**
7171  * Returns a cookie for current position of the iterator head, so that
7172  * user can use this cookie to load/start the iterator next time.
7173  *
7174  * \param di iterator's in memory structure
7175  *
7176  * \retval cookie for current position, on success
7177  */
7178 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7179 {
7180         struct osd_it_ea *it = (struct osd_it_ea *)di;
7181
7182         return it->oie_dirent->oied_off;
7183 }
7184
7185 /**
7186  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7187  * to load a directory entry at a time and stored it i inn,
7188  * in iterator's in-memory data structure.
7189  *
7190  * \param di struct osd_it_ea, iterator's in memory structure
7191  *
7192  * \retval +ve on success
7193  * \retval -ve on error
7194  */
7195 static int osd_it_ea_load(const struct lu_env *env,
7196                           const struct dt_it *di, __u64 hash)
7197 {
7198         struct osd_it_ea *it = (struct osd_it_ea *)di;
7199         int rc;
7200
7201         ENTRY;
7202         it->oie_file.f_pos = hash;
7203
7204         rc =  osd_ldiskfs_it_fill(env, di);
7205         if (rc > 0)
7206                 rc = -ENODATA;
7207
7208         if (rc == 0)
7209                 rc = 1;
7210
7211         RETURN(rc);
7212 }
7213
7214 /**
7215  * Index lookup function for interoperability mode (b11826).
7216  *
7217  * \param key,  key i.e. file name to be searched
7218  *
7219  * \retval +ve, on success
7220  * \retval -ve, on error
7221  */
7222 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7223                                struct dt_rec *rec, const struct dt_key *key)
7224 {
7225         struct osd_object *obj = osd_dt_obj(dt);
7226         int rc = 0;
7227
7228         ENTRY;
7229
7230         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7231         LINVRNT(osd_invariant(obj));
7232
7233         rc = osd_ea_lookup_rec(env, obj, rec, key);
7234         if (rc == 0)
7235                 rc = 1;
7236         RETURN(rc);
7237 }
7238
7239 /**
7240  * Index and Iterator operations for interoperability
7241  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7242  */
7243 static const struct dt_index_operations osd_index_ea_ops = {
7244         .dio_lookup         = osd_index_ea_lookup,
7245         .dio_declare_insert = osd_index_declare_ea_insert,
7246         .dio_insert         = osd_index_ea_insert,
7247         .dio_declare_delete = osd_index_declare_ea_delete,
7248         .dio_delete         = osd_index_ea_delete,
7249         .dio_it     = {
7250                 .init     = osd_it_ea_init,
7251                 .fini     = osd_it_ea_fini,
7252                 .get      = osd_it_ea_get,
7253                 .put      = osd_it_ea_put,
7254                 .next     = osd_it_ea_next,
7255                 .key      = osd_it_ea_key,
7256                 .key_size = osd_it_ea_key_size,
7257                 .rec      = osd_it_ea_rec,
7258                 .rec_size = osd_it_ea_rec_size,
7259                 .store    = osd_it_ea_store,
7260                 .load     = osd_it_ea_load
7261         }
7262 };
7263
7264 static void *osd_key_init(const struct lu_context *ctx,
7265                           struct lu_context_key *key)
7266 {
7267         struct osd_thread_info *info;
7268
7269         OBD_ALLOC_PTR(info);
7270         if (info == NULL)
7271                 return ERR_PTR(-ENOMEM);
7272
7273         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7274         if (info->oti_it_ea_buf == NULL)
7275                 goto out_free_info;
7276
7277         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7278
7279         info->oti_hlock = ldiskfs_htree_lock_alloc();
7280         if (info->oti_hlock == NULL)
7281                 goto out_free_ea;
7282
7283         return info;
7284
7285 out_free_ea:
7286         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7287 out_free_info:
7288         OBD_FREE_PTR(info);
7289         return ERR_PTR(-ENOMEM);
7290 }
7291
7292 static void osd_key_fini(const struct lu_context *ctx,
7293                          struct lu_context_key *key, void *data)
7294 {
7295         struct osd_thread_info *info = data;
7296         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7297         struct osd_idmap_cache *idc = info->oti_ins_cache;
7298
7299         if (info->oti_inode != NULL)
7300                 OBD_FREE_PTR(lli);
7301         if (info->oti_hlock != NULL)
7302                 ldiskfs_htree_lock_free(info->oti_hlock);
7303         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7304         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7305         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7306         lu_buf_free(&info->oti_big_buf);
7307         if (idc != NULL) {
7308                 LASSERT(info->oti_ins_cache_size > 0);
7309                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7310                 info->oti_ins_cache = NULL;
7311                 info->oti_ins_cache_size = 0;
7312         }
7313         OBD_FREE_PTR(info);
7314 }
7315
7316 static void osd_key_exit(const struct lu_context *ctx,
7317                          struct lu_context_key *key, void *data)
7318 {
7319         struct osd_thread_info *info = data;
7320
7321         LASSERT(info->oti_r_locks == 0);
7322         LASSERT(info->oti_w_locks == 0);
7323         LASSERT(info->oti_txns    == 0);
7324 }
7325
7326 /* type constructor/destructor: osd_type_init, osd_type_fini */
7327 LU_TYPE_INIT_FINI(osd, &osd_key);
7328
7329 struct lu_context_key osd_key = {
7330         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7331         .lct_init = osd_key_init,
7332         .lct_fini = osd_key_fini,
7333         .lct_exit = osd_key_exit
7334 };
7335
7336
7337 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7338                            const char *name, struct lu_device *next)
7339 {
7340         struct osd_device *osd = osd_dev(d);
7341
7342         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7343             sizeof(osd->od_svname))
7344                 return -E2BIG;
7345         return osd_procfs_init(osd, name);
7346 }
7347
7348 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7349 {
7350         struct seq_server_site *ss = osd_seq_site(osd);
7351         int rc;
7352
7353         ENTRY;
7354
7355         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7356                 RETURN(0);
7357
7358         if (unlikely(ss == NULL))
7359                 RETURN(-ENODEV);
7360
7361         OBD_ALLOC_PTR(osd->od_cl_seq);
7362         if (osd->od_cl_seq == NULL)
7363                 RETURN(-ENOMEM);
7364
7365         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7366                              osd->od_svname, ss->ss_server_seq);
7367         if (rc != 0) {
7368                 OBD_FREE_PTR(osd->od_cl_seq);
7369                 osd->od_cl_seq = NULL;
7370                 RETURN(rc);
7371         }
7372
7373         if (ss->ss_node_id == 0) {
7374                 /*
7375                  * If the OSD on the sequence controller(MDT0), then allocate
7376                  * sequence here, otherwise allocate sequence after connected
7377                  * to MDT0 (see mdt_register_lwp_callback()).
7378                  */
7379                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7380                                    &osd->od_cl_seq->lcs_space, env);
7381         }
7382
7383         RETURN(rc);
7384 }
7385
7386 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7387 {
7388         if (osd->od_cl_seq == NULL)
7389                 return;
7390
7391         seq_client_fini(osd->od_cl_seq);
7392         OBD_FREE_PTR(osd->od_cl_seq);
7393         osd->od_cl_seq = NULL;
7394 }
7395
7396 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7397 {
7398         ENTRY;
7399
7400         /* shutdown quota slave instance associated with the device */
7401         if (o->od_quota_slave != NULL) {
7402                 struct qsd_instance *qsd = o->od_quota_slave;
7403
7404                 o->od_quota_slave = NULL;
7405                 qsd_fini(env, qsd);
7406         }
7407
7408         osd_fid_fini(env, o);
7409         osd_scrub_cleanup(env, o);
7410
7411         RETURN(0);
7412 }
7413
7414 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7415 {
7416         ENTRY;
7417
7418         if (o->od_mnt != NULL) {
7419                 shrink_dcache_sb(osd_sb(o));
7420                 osd_sync(env, &o->od_dt_dev);
7421
7422                 mntput(o->od_mnt);
7423                 o->od_mnt = NULL;
7424         }
7425
7426         EXIT;
7427 }
7428
7429 static int osd_mount(const struct lu_env *env,
7430                      struct osd_device *o, struct lustre_cfg *cfg)
7431 {
7432         const char *name = lustre_cfg_string(cfg, 0);
7433         const char *dev = lustre_cfg_string(cfg, 1);
7434         const char *opts;
7435         unsigned long page, s_flags, lmd_flags = 0;
7436         struct page *__page;
7437         struct file_system_type *type;
7438         char *options = NULL;
7439         char *str;
7440         struct osd_thread_info *info = osd_oti_get(env);
7441         struct lu_fid *fid = &info->oti_fid;
7442         struct inode *inode;
7443         int rc = 0, force_over_512tb = 0;
7444
7445         ENTRY;
7446
7447         if (o->od_mnt != NULL)
7448                 RETURN(0);
7449
7450         if (strlen(dev) >= sizeof(o->od_mntdev))
7451                 RETURN(-E2BIG);
7452         strcpy(o->od_mntdev, dev);
7453
7454         str = lustre_cfg_string(cfg, 2);
7455         s_flags = simple_strtoul(str, NULL, 0);
7456         str = strstr(str, ":");
7457         if (str)
7458                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7459         opts = lustre_cfg_string(cfg, 3);
7460 #ifdef __BIG_ENDIAN
7461         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7462                 CERROR("%s: device %s extents feature is not guaranteed to "
7463                        "work on big-endian systems. Use \"bigendian_extents\" "
7464                        "mount option to override.\n", name, dev);
7465                 RETURN(-EINVAL);
7466         }
7467 #endif
7468 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7469         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7470                 CWARN("force_over_128tb option is deprecated. "
7471                       "Filesystems less than 512TB can be created without any "
7472                       "force options. Use force_over_512tb option for "
7473                       "filesystems greater than 512TB.\n");
7474         }
7475 #endif
7476 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7477         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7478                 CWARN("force_over_256tb option is deprecated. "
7479                       "Filesystems less than 512TB can be created without any "
7480                       "force options. Use force_over_512tb option for "
7481                       "filesystems greater than 512TB.\n");
7482         }
7483 #endif
7484
7485         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7486                 force_over_512tb = 1;
7487
7488         __page = alloc_page(GFP_KERNEL);
7489         if (__page == NULL)
7490                 GOTO(out, rc = -ENOMEM);
7491         page = (unsigned long)page_address(__page);
7492         options = (char *)page;
7493         *options = '\0';
7494         if (opts != NULL) {
7495                 /* strip out the options for back compatiblity */
7496                 static char *sout[] = {
7497                         "mballoc",
7498                         "iopen",
7499                         "noiopen",
7500                         "iopen_nopriv",
7501                         "extents",
7502                         "noextents",
7503                         /* strip out option we processed in osd */
7504                         "bigendian_extents",
7505                         "force_over_128tb",
7506                         "force_over_256tb",
7507                         "force_over_512tb",
7508                         NULL
7509                 };
7510                 strcat(options, opts);
7511                 for (rc = 0, str = options; sout[rc]; ) {
7512                         char *op = strstr(str, sout[rc]);
7513
7514                         if (op == NULL) {
7515                                 rc++;
7516                                 str = options;
7517                                 continue;
7518                         }
7519                         if (op == options || *(op - 1) == ',') {
7520                                 str = op + strlen(sout[rc]);
7521                                 if (*str == ',' || *str == '\0') {
7522                                         *str == ',' ? str++ : str;
7523                                         memmove(op, str, strlen(str) + 1);
7524                                 }
7525                         }
7526                         for (str = op; *str != ',' && *str != '\0'; str++)
7527                                 ;
7528                 }
7529         } else {
7530                 strncat(options, "user_xattr,acl", 14);
7531         }
7532
7533         /* Glom up mount options */
7534         if (*options != '\0')
7535                 strcat(options, ",");
7536         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7537
7538         type = get_fs_type("ldiskfs");
7539         if (!type) {
7540                 CERROR("%s: cannot find ldiskfs module\n", name);
7541                 GOTO(out, rc = -ENODEV);
7542         }
7543
7544         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7545         module_put(type->owner);
7546
7547         if (IS_ERR(o->od_mnt)) {
7548                 rc = PTR_ERR(o->od_mnt);
7549                 o->od_mnt = NULL;
7550                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7551                 GOTO(out, rc);
7552         }
7553
7554         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7555                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7556                                  force_over_512tb == 0) {
7557                 CERROR("%s: device %s LDISKFS does not support filesystems "
7558                        "greater than 512TB and can cause data corruption. "
7559                        "Use \"force_over_512tb\" mount option to override.\n",
7560                        name, dev);
7561                 GOTO(out_mnt, rc = -EINVAL);
7562         }
7563
7564         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7565                 if (priv_dev_set_rdonly) {
7566                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7567                         o->od_dt_dev.dd_rdonly = 1;
7568                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7569                                       name);
7570                 } else {
7571                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7572                                       name);
7573
7574                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7575                 }
7576         } else if (priv_dev_check_rdonly &&
7577                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7578                 CERROR("%s: underlying device %s is marked as "
7579                        "read-only. Setup failed\n", name, dev);
7580
7581                 GOTO(out_mnt, rc = -EROFS);
7582         }
7583
7584         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7585                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7586                 GOTO(out_mnt, rc = -EINVAL);
7587         }
7588
7589 #ifdef LDISKFS_MOUNT_DIRDATA
7590         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7591                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7592         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7593                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7594                       "enabling the dirdata feature. If you do not want to "
7595                       "downgrade to Lustre-1.x again, you can enable it via "
7596                       "'tune2fs -O dirdata device'\n", name, dev);
7597 #endif
7598         inode = osd_sb(o)->s_root->d_inode;
7599         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7600         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7601         if (rc != 0) {
7602                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7603                 GOTO(out_mnt, rc);
7604         }
7605
7606         if (lmd_flags & LMD_FLG_NOSCRUB)
7607                 o->od_auto_scrub_interval = AS_NEVER;
7608
7609         GOTO(out, rc = 0);
7610
7611 out_mnt:
7612         mntput(o->od_mnt);
7613         o->od_mnt = NULL;
7614
7615 out:
7616         if (__page)
7617                 __free_page(__page);
7618
7619         return rc;
7620 }
7621
7622 static struct lu_device *osd_device_fini(const struct lu_env *env,
7623                                          struct lu_device *d)
7624 {
7625         struct osd_device *o = osd_dev(d);
7626
7627         ENTRY;
7628
7629         osd_index_backup(env, o, false);
7630         osd_shutdown(env, o);
7631         osd_procfs_fini(o);
7632         osd_obj_map_fini(o);
7633         osd_umount(env, o);
7634
7635         RETURN(NULL);
7636 }
7637
7638 static int osd_device_init0(const struct lu_env *env,
7639                             struct osd_device *o,
7640                             struct lustre_cfg *cfg)
7641 {
7642         struct lu_device *l = osd2lu_dev(o);
7643         struct osd_thread_info *info;
7644         int rc;
7645         int cplen = 0;
7646
7647         /* if the module was re-loaded, env can loose its keys */
7648         rc = lu_env_refill((struct lu_env *)env);
7649         if (rc)
7650                 GOTO(out, rc);
7651         info = osd_oti_get(env);
7652         LASSERT(info);
7653
7654         l->ld_ops = &osd_lu_ops;
7655         o->od_dt_dev.dd_ops = &osd_dt_ops;
7656
7657         spin_lock_init(&o->od_osfs_lock);
7658         mutex_init(&o->od_otable_mutex);
7659         INIT_LIST_HEAD(&o->od_orphan_list);
7660         INIT_LIST_HEAD(&o->od_index_backup_list);
7661         INIT_LIST_HEAD(&o->od_index_restore_list);
7662         spin_lock_init(&o->od_lock);
7663         o->od_index_backup_policy = LIBP_NONE;
7664
7665         o->od_read_cache = 1;
7666         o->od_writethrough_cache = 1;
7667         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7668         o->od_auto_scrub_interval = AS_DEFAULT;
7669
7670         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7671                         sizeof(o->od_svname));
7672         if (cplen >= sizeof(o->od_svname)) {
7673                 rc = -E2BIG;
7674                 GOTO(out, rc);
7675         }
7676
7677         o->od_index_backup_stop = 0;
7678         o->od_index = -1; /* -1 means index is invalid */
7679         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7680         if (rc == LDD_F_SV_TYPE_OST)
7681                 o->od_is_ost = 1;
7682
7683         o->od_full_scrub_ratio = OFSR_DEFAULT;
7684         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7685         rc = osd_mount(env, o, cfg);
7686         if (rc != 0)
7687                 GOTO(out, rc);
7688
7689         rc = osd_obj_map_init(env, o);
7690         if (rc != 0)
7691                 GOTO(out_mnt, rc);
7692
7693         rc = lu_site_init(&o->od_site, l);
7694         if (rc != 0)
7695                 GOTO(out_compat, rc);
7696         o->od_site.ls_bottom_dev = l;
7697
7698         rc = lu_site_init_finish(&o->od_site);
7699         if (rc != 0)
7700                 GOTO(out_site, rc);
7701
7702         INIT_LIST_HEAD(&o->od_ios_list);
7703         /* setup scrub, including OI files initialization */
7704         o->od_in_init = 1;
7705         rc = osd_scrub_setup(env, o);
7706         o->od_in_init = 0;
7707         if (rc < 0)
7708                 GOTO(out_site, rc);
7709
7710         rc = osd_procfs_init(o, o->od_svname);
7711         if (rc != 0) {
7712                 CERROR("%s: can't initialize procfs: rc = %d\n",
7713                        o->od_svname, rc);
7714                 GOTO(out_scrub, rc);
7715         }
7716
7717         LASSERT(l->ld_site->ls_linkage.next != NULL);
7718         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7719
7720         /* initialize quota slave instance */
7721         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7722                                      o->od_proc_entry);
7723         if (IS_ERR(o->od_quota_slave)) {
7724                 rc = PTR_ERR(o->od_quota_slave);
7725                 o->od_quota_slave = NULL;
7726                 GOTO(out_procfs, rc);
7727         }
7728
7729         RETURN(0);
7730
7731 out_procfs:
7732         osd_procfs_fini(o);
7733 out_scrub:
7734         osd_scrub_cleanup(env, o);
7735 out_site:
7736         lu_site_fini(&o->od_site);
7737 out_compat:
7738         osd_obj_map_fini(o);
7739 out_mnt:
7740         osd_umount(env, o);
7741 out:
7742         return rc;
7743 }
7744
7745 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7746                                           struct lu_device_type *t,
7747                                           struct lustre_cfg *cfg)
7748 {
7749         struct osd_device *o;
7750         int rc;
7751
7752         OBD_ALLOC_PTR(o);
7753         if (o == NULL)
7754                 return ERR_PTR(-ENOMEM);
7755
7756         rc = dt_device_init(&o->od_dt_dev, t);
7757         if (rc == 0) {
7758                 /*
7759                  * Because the ctx might be revived in dt_device_init,
7760                  * refill the env here
7761                  */
7762                 lu_env_refill((struct lu_env *)env);
7763                 rc = osd_device_init0(env, o, cfg);
7764                 if (rc)
7765                         dt_device_fini(&o->od_dt_dev);
7766         }
7767
7768         if (unlikely(rc != 0))
7769                 OBD_FREE_PTR(o);
7770
7771         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7772 }
7773
7774 static struct lu_device *osd_device_free(const struct lu_env *env,
7775                                          struct lu_device *d)
7776 {
7777         struct osd_device *o = osd_dev(d);
7778
7779         ENTRY;
7780
7781         /* XXX: make osd top device in order to release reference */
7782         d->ld_site->ls_top_dev = d;
7783         lu_site_purge(env, d->ld_site, -1);
7784         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7785                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7786                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7787         }
7788         lu_site_fini(&o->od_site);
7789         dt_device_fini(&o->od_dt_dev);
7790         OBD_FREE_PTR(o);
7791         RETURN(NULL);
7792 }
7793
7794 static int osd_process_config(const struct lu_env *env,
7795                               struct lu_device *d, struct lustre_cfg *cfg)
7796 {
7797         struct osd_device *o = osd_dev(d);
7798         int rc;
7799
7800         ENTRY;
7801
7802         switch (cfg->lcfg_command) {
7803         case LCFG_SETUP:
7804                 rc = osd_mount(env, o, cfg);
7805                 break;
7806         case LCFG_CLEANUP:
7807                 /*
7808                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7809                  * that may happend if hit failure during mount process.
7810                  */
7811                 osd_index_backup(env, o, false);
7812                 lu_dev_del_linkage(d->ld_site, d);
7813                 rc = osd_shutdown(env, o);
7814                 break;
7815         case LCFG_PARAM:
7816                 LASSERT(&o->od_dt_dev);
7817                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7818                                               cfg, &o->od_dt_dev);
7819                 if (rc > 0 || rc == -ENOSYS) {
7820                         rc = class_process_proc_param(PARAM_OST,
7821                                                       lprocfs_osd_obd_vars,
7822                                                       cfg, &o->od_dt_dev);
7823                         if (rc > 0)
7824                                 rc = 0;
7825                 }
7826                 break;
7827         case LCFG_PRE_CLEANUP:
7828                 osd_scrub_stop(o);
7829                 osd_index_backup(env, o,
7830                                  o->od_index_backup_policy != LIBP_NONE);
7831                 rc = 0;
7832                 break;
7833         default:
7834                 rc = -ENOSYS;
7835         }
7836
7837         RETURN(rc);
7838 }
7839
7840 static int osd_recovery_complete(const struct lu_env *env,
7841                                  struct lu_device *d)
7842 {
7843         struct osd_device *osd = osd_dev(d);
7844         int rc = 0;
7845
7846         ENTRY;
7847
7848         if (osd->od_quota_slave == NULL)
7849                 RETURN(0);
7850
7851         /*
7852          * start qsd instance on recovery completion, this notifies the quota
7853          * slave code that we are about to process new requests now
7854          */
7855         rc = qsd_start(env, osd->od_quota_slave);
7856         RETURN(rc);
7857 }
7858
7859 /*
7860  * we use exports to track all osd users
7861  */
7862 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7863                            struct obd_device *obd, struct obd_uuid *cluuid,
7864                            struct obd_connect_data *data, void *localdata)
7865 {
7866         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7867         struct lustre_handle conn;
7868         int rc;
7869
7870         ENTRY;
7871
7872         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7873
7874         rc = class_connect(&conn, obd, cluuid);
7875         if (rc)
7876                 RETURN(rc);
7877
7878         *exp = class_conn2export(&conn);
7879
7880         spin_lock(&osd->od_osfs_lock);
7881         osd->od_connects++;
7882         spin_unlock(&osd->od_osfs_lock);
7883
7884         RETURN(0);
7885 }
7886
7887 /*
7888  * once last export (we don't count self-export) disappeared
7889  * osd can be released
7890  */
7891 static int osd_obd_disconnect(struct obd_export *exp)
7892 {
7893         struct obd_device *obd = exp->exp_obd;
7894         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7895         int rc, release = 0;
7896
7897         ENTRY;
7898
7899         /* Only disconnect the underlying layers on the final disconnect. */
7900         spin_lock(&osd->od_osfs_lock);
7901         osd->od_connects--;
7902         if (osd->od_connects == 0)
7903                 release = 1;
7904         spin_unlock(&osd->od_osfs_lock);
7905
7906         rc = class_disconnect(exp); /* bz 9811 */
7907
7908         if (rc == 0 && release)
7909                 class_manual_cleanup(obd);
7910         RETURN(rc);
7911 }
7912
7913 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7914                        struct lu_device *dev)
7915 {
7916         struct osd_device *osd = osd_dev(dev);
7917         struct lr_server_data *lsd =
7918                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7919         int result = 0;
7920
7921         ENTRY;
7922
7923         if (osd->od_quota_slave != NULL) {
7924                 /* set up quota slave objects */
7925                 result = qsd_prepare(env, osd->od_quota_slave);
7926                 if (result != 0)
7927                         RETURN(result);
7928         }
7929
7930         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7931 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7932                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7933                         osd->od_index_in_idif = 1;
7934                 } else {
7935                         osd->od_index_in_idif = 0;
7936                         result = osd_register_proc_index_in_idif(osd);
7937                         if (result != 0)
7938                                 RETURN(result);
7939                 }
7940 #else
7941                 osd->od_index_in_idif = 1;
7942 #endif
7943         }
7944
7945         result = osd_fid_init(env, osd);
7946
7947         RETURN(result);
7948 }
7949
7950 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7951                          struct lu_fid *fid, struct md_op_data *op_data)
7952 {
7953         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7954
7955         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7956 }
7957
7958 static const struct lu_object_operations osd_lu_obj_ops = {
7959         .loo_object_init      = osd_object_init,
7960         .loo_object_delete    = osd_object_delete,
7961         .loo_object_release   = osd_object_release,
7962         .loo_object_free      = osd_object_free,
7963         .loo_object_print     = osd_object_print,
7964         .loo_object_invariant = osd_object_invariant
7965 };
7966
7967 const struct lu_device_operations osd_lu_ops = {
7968         .ldo_object_alloc      = osd_object_alloc,
7969         .ldo_process_config    = osd_process_config,
7970         .ldo_recovery_complete = osd_recovery_complete,
7971         .ldo_prepare           = osd_prepare,
7972 };
7973
7974 static const struct lu_device_type_operations osd_device_type_ops = {
7975         .ldto_init = osd_type_init,
7976         .ldto_fini = osd_type_fini,
7977
7978         .ldto_start = osd_type_start,
7979         .ldto_stop  = osd_type_stop,
7980
7981         .ldto_device_alloc = osd_device_alloc,
7982         .ldto_device_free  = osd_device_free,
7983
7984         .ldto_device_init = osd_device_init,
7985         .ldto_device_fini = osd_device_fini
7986 };
7987
7988 static struct lu_device_type osd_device_type = {
7989         .ldt_tags     = LU_DEVICE_DT,
7990         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7991         .ldt_ops      = &osd_device_type_ops,
7992         .ldt_ctx_tags = LCT_LOCAL,
7993 };
7994
7995 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7996 {
7997         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7998         struct super_block *sb = osd_sb(osd);
7999
8000         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8001 }
8002
8003 /*
8004  * lprocfs legacy support.
8005  */
8006 static struct obd_ops osd_obd_device_ops = {
8007         .o_owner = THIS_MODULE,
8008         .o_connect      = osd_obd_connect,
8009         .o_disconnect   = osd_obd_disconnect,
8010         .o_fid_alloc    = osd_fid_alloc,
8011         .o_health_check = osd_health_check,
8012 };
8013
8014 static int __init osd_init(void)
8015 {
8016         int rc;
8017
8018         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8019 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8020         /* please, try to keep osd_thread_info smaller than a page */
8021         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8022 #endif
8023
8024         osd_oi_mod_init();
8025
8026         rc = lu_kmem_init(ldiskfs_caches);
8027         if (rc)
8028                 return rc;
8029
8030 #ifdef CONFIG_KALLSYMS
8031         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8032         priv_dev_check_rdonly =
8033                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8034 #endif
8035
8036         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8037                                  lprocfs_osd_module_vars,
8038                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8039         if (rc)
8040                 lu_kmem_fini(ldiskfs_caches);
8041         return rc;
8042 }
8043
8044 static void __exit osd_exit(void)
8045 {
8046         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8047         lu_kmem_fini(ldiskfs_caches);
8048 }
8049
8050 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8051 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8052 MODULE_VERSION(LUSTRE_VERSION_STRING);
8053 MODULE_LICENSE("GPL");
8054
8055 module_init(osd_init);
8056 module_exit(osd_exit);