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 = osd_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 = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_18_23, contains self-fid
 711  * \retval -v: other failure cases
 712  */
 713 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 714                  struct dentry *dentry, struct lu_fid *fid)
 715 {
 716         struct filter_fid *ff = &info->oti_ff;
 717         struct ost_id *ostid = &info->oti_ostid;
 718         int rc;
 719
 720         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 721         if (rc == sizeof(struct filter_fid_18_23)) {
 722                 struct filter_fid_18_23 *ff_old = (void *)ff;
 723
 724                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 725                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 726                 /*
 727                  * XXX: use 0 as the index for compatibility, the caller will
 728                  * handle index related issues when necessary.
 729                  */
 730                 if (!rc)
 731                         ostid_to_fid(fid, ostid, 0);
 732         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 733                 rc = 1;
 734         } else if (rc >= 0) {
 735                 rc = -EINVAL;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int osd_lma_self_repair(struct osd_thread_info *info,
 742                                struct osd_device *osd, struct inode *inode,
 743                                const struct lu_fid *fid, __u32 compat)
 744 {
 745         handle_t *jh;
 746         int rc;
 747
 748         LASSERT(current->journal_info == NULL);
 749
 750         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 751                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 752         if (IS_ERR(jh)) {
 753                 rc = PTR_ERR(jh);
 754                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 755                       osd_name(osd), rc);
 756                 return rc;
 757         }
 758
 759         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 760         if (rc != 0)
 761                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 762                       osd_name(osd), rc);
 763         ldiskfs_journal_stop(jh);
 764         return rc;
 765 }
 766
 767 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 768 {
 769         struct osd_thread_info *info = osd_oti_get(env);
 770         struct osd_device *osd = osd_obj2dev(obj);
 771         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 772         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 773         struct inode *inode = obj->oo_inode;
 774         struct dentry *dentry = &info->oti_obj_dentry;
 775         struct lu_fid *fid = NULL;
 776         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 777         int rc;
 778
 779         ENTRY;
 780
 781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 782                              (void *)loa, sizeof(*loa));
 783         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 784                 fid = &lma->lma_self_fid;
 785                 rc = osd_get_idif(info, inode, dentry, fid);
 786                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 787                         /*
 788                          * For the given OST-object, if it has neither LMA nor
 789                          * FID in XATTR_NAME_FID, then the given FID (which is
 790                          * contained in the @obj, from client RPC for locating
 791                          * the OST-object) is trusted. We use it to generate
 792                          * the LMA.
 793                          */
 794                         osd_lma_self_repair(info, osd, inode, rfid,
 795                                             LMAC_FID_ON_OST);
 796                         RETURN(0);
 797                 }
 798         }
 799
 800         if (rc < 0)
 801                 RETURN(rc);
 802
 803         if (rc > 0) {
 804                 rc = 0;
 805                 lustre_lma_swab(lma);
 806                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 807                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 808                               S_ISREG(inode->i_mode)))) {
 809                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 810                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 811                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 812                               PFID(rfid), inode->i_ino);
 813                         rc = -EOPNOTSUPP;
 814                 } else {
 815                         fid = &lma->lma_self_fid;
 816                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 817                             osd->od_is_ost)
 818                                 obj->oo_pfid_in_lma = 1;
 819                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 820                             !osd->od_is_ost)
 821                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 822                 }
 823         }
 824
 825         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 826                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 827                         struct ost_id   *oi   = &info->oti_ostid;
 828                         struct lu_fid   *fid1 = &info->oti_fid3;
 829                         __u32            idx  = fid_idif_ost_idx(rfid);
 830
 831                         /*
 832                          * For old IDIF, the OST index is not part of the IDIF,
 833                          * Means that different OSTs may have the same IDIFs.
 834                          * Under such case, we need to make some compatible
 835                          * check to make sure to trigger OI scrub properly.
 836                          */
 837                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 838                                 /* Given @rfid is new, LMA is old. */
 839                                 fid_to_ostid(fid, oi);
 840                                 ostid_to_fid(fid1, oi, idx);
 841                                 if (lu_fid_eq(fid1, rfid)) {
 842                                         if (osd->od_index_in_idif)
 843                                                 osd_lma_self_repair(info, osd,
 844                                                         inode, rfid,
 845                                                         LMAC_FID_ON_OST);
 846                                         RETURN(0);
 847                                 }
 848                         }
 849                 }
 850
 851                 rc = -EREMCHG;
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 struct osd_check_lmv_buf {
 858 #ifdef HAVE_DIR_CONTEXT
 859         /* please keep it as first member */
 860         struct dir_context ctx;
 861 #endif
 862         struct osd_thread_info *oclb_info;
 863         struct osd_device *oclb_dev;
 864         struct osd_idmap_cache *oclb_oic;
 865         int oclb_items;
 866         bool oclb_found;
 867 };
 868
 869 /**
 870  * It is called internally by ->readdir() to filter out the
 871  * local slave object's FID of the striped directory.
 872  *
 873  * \retval      1 found the local slave's FID
 874  * \retval      0 continue to check next item
 875  * \retval      -ve for failure
 876  */
 877 #ifdef HAVE_FILLDIR_USE_CTX
 878 static int osd_stripe_dir_filldir(struct dir_context *buf,
 879 #else
 880 static int osd_stripe_dir_filldir(void *buf,
 881 #endif
 882                                   const char *name, int namelen,
 883                                   loff_t offset, __u64 ino, unsigned int d_type)
 884 {
 885         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 886         struct osd_thread_info *oti = oclb->oclb_info;
 887         struct lu_fid *fid = &oti->oti_fid3;
 888         struct osd_inode_id *id = &oti->oti_id3;
 889         struct osd_device *dev = oclb->oclb_dev;
 890         struct osd_idmap_cache *oic = oclb->oclb_oic;
 891         struct inode *inode;
 892
 893         oclb->oclb_items++;
 894
 895         if (name[0] == '.')
 896                 return 0;
 897
 898         fid_zero(fid);
 899         sscanf(name + 1, SFID, RFID(fid));
 900         if (!fid_is_sane(fid))
 901                 return 0;
 902
 903         if (osd_remote_fid(oti->oti_env, dev, fid))
 904                 return 0;
 905
 906         osd_id_gen(id, ino, OSD_OII_NOGEN);
 907         inode = osd_iget(oti, dev, id);
 908         if (IS_ERR(inode))
 909                 return PTR_ERR(inode);
 910
 911         iput(inode);
 912         osd_add_oi_cache(oti, dev, id, fid);
 913         oic->oic_fid = *fid;
 914         oic->oic_lid = *id;
 915         oic->oic_dev = dev;
 916         osd_oii_insert(dev, oic, true);
 917         oclb->oclb_found = true;
 918
 919         return 1;
 920 }
 921
 922 /*
 923  * When lookup item under striped directory, we need to locate the master
 924  * MDT-object of the striped directory firstly, then the client will send
 925  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 926  * and the item's name. If the system is restored from MDT file level backup,
 927  * then before the OI scrub completely built the OI files, the OI mappings of
 928  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 929  * not a problem for the master MDT-object. Because when locate the master
 930  * MDT-object, we will do name based lookup (for the striped directory itself)
 931  * firstly, during such process we can setup the correct OI mapping for the
 932  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 933  * the client will not trigger name based lookup on the MDT to locate the slave
 934  * MDT-object before locating item under the striped directory, then when
 935  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 936  * is invalid and does not know what the right OI mapping is, then the MDT has
 937  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 938  * the OI file, related OI mapping is unknown yet, please try again later. And
 939  * then client will re-try the RPC again and again until related OI mapping has
 940  * been updated. That is quite inefficient.
 941  *
 942  * To resolve above trouble, we will handle it as the following two cases:
 943  *
 944  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 945  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 946  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 947  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 948  *    directly. Please check osd_fid_lookup().
 949  *
 950  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 951  *    Under such case, during lookup the master MDT-object, we will lookup the
 952  *    slave MDT-object via readdir against the master MDT-object, because the
 953  *    slave MDT-objects information are stored as sub-directories with the name
 954  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 955  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 956  *    the correct OI mapping for the slave MDT-object.
 957  */
 958 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 959                          struct inode *inode, struct osd_idmap_cache *oic)
 960 {
 961         struct lu_buf *buf = &oti->oti_big_buf;
 962         struct dentry *dentry = &oti->oti_obj_dentry;
 963         struct file *filp = &oti->oti_file;
 964         const struct file_operations *fops;
 965         struct lmv_mds_md_v1 *lmv1;
 966         struct osd_check_lmv_buf oclb = {
 967 #ifdef HAVE_DIR_CONTEXT
 968                 .ctx.actor = osd_stripe_dir_filldir,
 969 #endif
 970                 .oclb_info = oti,
 971                 .oclb_dev = dev,
 972                 .oclb_oic = oic,
 973                 .oclb_found = false,
 974         };
 975         int rc = 0;
 976
 977         ENTRY;
 978
 979 again:
 980         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 981                              buf->lb_len);
 982         if (rc == -ERANGE) {
 983                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 984                 if (rc > 0) {
 985                         lu_buf_realloc(buf, rc);
 986                         if (buf->lb_buf == NULL)
 987                                 GOTO(out, rc = -ENOMEM);
 988
 989                         goto again;
 990                 }
 991         }
 992
 993         if (unlikely(rc == 0 || rc == -ENODATA))
 994                 GOTO(out, rc = 0);
 995
 996         if (rc < 0)
 997                 GOTO(out, rc);
 998
 999         if (unlikely(buf->lb_buf == NULL)) {
1000                 lu_buf_realloc(buf, rc);
1001                 if (buf->lb_buf == NULL)
1002                         GOTO(out, rc = -ENOMEM);
1003
1004                 goto again;
1005         }
1006
1007         lmv1 = buf->lb_buf;
1008         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1009                 GOTO(out, rc = 0);
1010
1011         fops = inode->i_fop;
1012         dentry->d_inode = inode;
1013         dentry->d_sb = inode->i_sb;
1014         filp->f_pos = 0;
1015         filp->f_path.dentry = dentry;
1016         filp->f_mode = FMODE_64BITHASH;
1017         filp->f_mapping = inode->i_mapping;
1018         filp->f_op = fops;
1019         filp->private_data = NULL;
1020         set_file_inode(filp, inode);
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024 #ifdef HAVE_DIR_CONTEXT
1025                 oclb.ctx.pos = filp->f_pos;
1026 #ifdef HAVE_ITERATE_SHARED
1027                 rc = fops->iterate_shared(filp, &oclb.ctx);
1028 #else
1029                 rc = fops->iterate(filp, &oclb.ctx);
1030 #endif
1031                 filp->f_pos = oclb.ctx.pos;
1032 #else
1033                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1034 #endif
1035         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1036                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1037         fops->release(inode, filp);
1038
1039 out:
1040         if (rc < 0)
1041                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1042                        DFID": rc = %d\n", osd_ino2name(inode),
1043                        inode->i_ino, inode->i_generation,
1044                        PFID(&oic->oic_fid), rc);
1045         else
1046                 rc = 0;
1047
1048         RETURN(rc);
1049 }
1050
1051 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1052                           const struct lu_fid *fid,
1053                           const struct lu_object_conf *conf)
1054 {
1055         struct osd_thread_info *info;
1056         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1057         struct osd_device *dev;
1058         struct osd_idmap_cache *oic;
1059         struct osd_inode_id *id;
1060         struct inode *inode = NULL;
1061         struct lustre_scrub *scrub;
1062         struct scrub_file *sf;
1063         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1064         __u32 saved_ino;
1065         __u32 saved_gen;
1066         int result = 0;
1067         int rc1 = 0;
1068         bool remote = false;
1069         bool trusted = true;
1070         bool updated = false;
1071         bool checked = false;
1072
1073         ENTRY;
1074
1075         LINVRNT(osd_invariant(obj));
1076         LASSERT(obj->oo_inode == NULL);
1077         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1078
1079         dev = osd_dev(ldev);
1080         scrub = &dev->od_scrub.os_scrub;
1081         sf = &scrub->os_file;
1082         info = osd_oti_get(env);
1083         LASSERT(info);
1084         oic = &info->oti_cache;
1085
1086         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1087                 RETURN(-ENOENT);
1088
1089         /*
1090          * For the object is created as locking anchor, or for the object to
1091          * be created on disk. No need to osd_oi_lookup() at here because FID
1092          * shouldn't never be re-used, if it's really a duplicate FID from
1093          * unexpected reason, we should be able to detect it later by calling
1094          * do_create->osd_oi_insert().
1095          */
1096         if (conf && conf->loc_flags & LOC_F_NEW)
1097                 GOTO(out, result = 0);
1098
1099         /* Search order: 1. per-thread cache. */
1100         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1101                 id = &oic->oic_lid;
1102                 goto iget;
1103         }
1104
1105         id = &info->oti_id;
1106         if (!list_empty(&scrub->os_inconsistent_items)) {
1107                 /* Search order: 2. OI scrub pending list. */
1108                 result = osd_oii_lookup(dev, fid, id);
1109                 if (!result)
1110                         goto iget;
1111         }
1112
1113         /*
1114          * The OI mapping in the OI file can be updated by the OI scrub
1115          * when we locate the inode via FID. So it may be not trustable.
1116          */
1117         trusted = false;
1118
1119         /* Search order: 3. OI files. */
1120         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1121         if (result == -ENOENT) {
1122                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1123                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1124                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1125                                       sf->sf_oi_bitmap))
1126                         GOTO(out, result = 0);
1127
1128                 goto trigger;
1129         }
1130
1131         /* -ESTALE is returned if inode of OST object doesn't exist */
1132         if (result == -ESTALE &&
1133             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1134                 GOTO(out, result = 0);
1135         }
1136
1137         if (result)
1138                 GOTO(out, result);
1139
1140 iget:
1141         obj->oo_inode = NULL;
1142         /* for later passes through checks, not true on first pass */
1143         if (!IS_ERR_OR_NULL(inode))
1144                 iput(inode);
1145
1146         inode = osd_iget_check(info, dev, fid, id, trusted);
1147         if (!IS_ERR(inode)) {
1148                 obj->oo_inode = inode;
1149                 result = 0;
1150                 if (remote)
1151                         goto trigger;
1152
1153                 goto check_lma;
1154         }
1155
1156         result = PTR_ERR(inode);
1157         if (result == -ENOENT || result == -ESTALE)
1158                 GOTO(out, result = 0);
1159
1160         if (result != -EREMCHG)
1161                 GOTO(out, result);
1162
1163 trigger:
1164         /*
1165          * We still have chance to get the valid inode: for the
1166          * object which is referenced by remote name entry, the
1167          * object on the local MDT will be linked under the dir
1168          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1169          *
1170          * We do not know whether the object for the given FID
1171          * is referenced by some remote name entry or not, and
1172          * especially for DNE II, a multiple-linked object may
1173          * have many name entries reside on many MDTs.
1174          *
1175          * To simplify the operation, OSD will not distinguish
1176          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1177          * only happened for the RPC from other MDT during the
1178          * OI scrub, or for the client side RPC with FID only,
1179          * such as FID to path, or from old connected client.
1180          */
1181         if (!remote) {
1182                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1183                 if (!rc1) {
1184                         remote = true;
1185                         trusted = true;
1186                         flags |= SS_AUTO_PARTIAL;
1187                         flags &= ~SS_AUTO_FULL;
1188                         goto iget;
1189                 }
1190         }
1191
1192         if (thread_is_running(&scrub->os_thread)) {
1193                 if (scrub->os_partial_scan && !scrub->os_in_join)
1194                         goto join;
1195
1196                 if (IS_ERR_OR_NULL(inode) || result)
1197                         GOTO(out, result = -EINPROGRESS);
1198
1199                 LASSERT(remote);
1200                 LASSERT(obj->oo_inode == inode);
1201
1202                 osd_add_oi_cache(info, dev, id, fid);
1203                 osd_oii_insert(dev, oic, true);
1204                 goto found;
1205         }
1206
1207         if (dev->od_auto_scrub_interval == AS_NEVER) {
1208                 if (!remote)
1209                         GOTO(out, result = -EREMCHG);
1210
1211                 LASSERT(!result);
1212                 LASSERT(obj->oo_inode == inode);
1213
1214                 osd_add_oi_cache(info, dev, id, fid);
1215                 goto found;
1216         }
1217
1218 join:
1219         rc1 = osd_scrub_start(env, dev, flags);
1220         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1221                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1222         if (rc1 && rc1 != -EALREADY)
1223                 GOTO(out, result = -EREMCHG);
1224
1225         if (IS_ERR_OR_NULL(inode) || result)
1226                 GOTO(out, result = -EINPROGRESS);
1227
1228         LASSERT(remote);
1229         LASSERT(obj->oo_inode == inode);
1230
1231         osd_add_oi_cache(info, dev, id, fid);
1232         osd_oii_insert(dev, oic, true);
1233         goto found;
1234
1235 check_lma:
1236         checked = true;
1237         if (unlikely(obj->oo_header))
1238                 goto found;
1239
1240         result = osd_check_lma(env, obj);
1241         if (!result)
1242                 goto found;
1243
1244         LASSERTF(id->oii_ino == inode->i_ino &&
1245                  id->oii_gen == inode->i_generation,
1246                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1247                  PFID(fid), id->oii_ino, id->oii_gen,
1248                  inode->i_ino, inode->i_generation);
1249
1250         saved_ino = inode->i_ino;
1251         saved_gen = inode->i_generation;
1252
1253         if (unlikely(result == -ENODATA)) {
1254                 /*
1255                  * If the OI scrub updated the OI mapping by race, it
1256                  * must be valid. Trust the inode that has no LMA EA.
1257                  */
1258                 if (updated)
1259                         goto found;
1260
1261                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1262                 if (!result) {
1263                         /*
1264                          * The OI mapping is still there, the inode is still
1265                          * valid. It is just becaues the inode has no LMA EA.
1266                          */
1267                         if (saved_ino == id->oii_ino &&
1268                             saved_gen == id->oii_gen)
1269                                 goto found;
1270
1271                         /*
1272                          * It is the OI scrub updated the OI mapping by race.
1273                          * The new OI mapping must be valid.
1274                          */
1275                         trusted = true;
1276                         updated = true;
1277                         goto iget;
1278                 }
1279
1280                 /*
1281                  * "result == -ENOENT" means that the OI mappinghas been
1282                  * removed by race, so the inode belongs to other object.
1283                  *
1284                  * Others error can be returned  directly.
1285                  */
1286                 if (result == -ENOENT) {
1287                         LASSERT(trusted);
1288
1289                         obj->oo_inode = NULL;
1290                         result = 0;
1291                 }
1292         }
1293
1294         if (result != -EREMCHG)
1295                 GOTO(out, result);
1296
1297         LASSERT(!updated);
1298
1299         /*
1300          * if two OST objects map to the same inode, and inode mode is
1301          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1302          * reserved by precreate, and not written yet, in this case, don't
1303          * set inode for the object whose FID mismatch, so that it can create
1304          * inode and not block precreate.
1305          */
1306         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1307             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1308                 obj->oo_inode = NULL;
1309                 GOTO(out, result = 0);
1310         }
1311
1312         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1313         /*
1314          * "result == -ENOENT" means the cached OI mapping has been removed
1315          * from the OI file by race, above inode belongs to other object.
1316          */
1317         if (result == -ENOENT) {
1318                 LASSERT(trusted);
1319
1320                 obj->oo_inode = NULL;
1321                 GOTO(out, result = 0);
1322         }
1323
1324         if (result)
1325                 GOTO(out, result);
1326
1327         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1328                 result = -EREMCHG;
1329                 goto trigger;
1330         }
1331
1332         /*
1333          * It is the OI scrub updated the OI mapping by race.
1334          * The new OI mapping must be valid.
1335          */
1336         trusted = true;
1337         updated = true;
1338         goto iget;
1339
1340 found:
1341         if (!checked) {
1342                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1343                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1344
1345                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1346                 if (!result) {
1347                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1348                             dev->od_is_ost)
1349                                 obj->oo_pfid_in_lma = 1;
1350                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1351                             !dev->od_is_ost)
1352                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1353                 } else if (result != -ENODATA) {
1354                         GOTO(out, result);
1355                 }
1356         }
1357
1358         obj->oo_compat_dot_created = 1;
1359         obj->oo_compat_dotdot_created = 1;
1360
1361         if (S_ISDIR(inode->i_mode) &&
1362             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1363                 osd_check_lmv(info, dev, inode, oic);
1364
1365         result = osd_attach_jinode(inode);
1366         if (result)
1367                 GOTO(out, result);
1368
1369         if (!ldiskfs_pdo)
1370                 GOTO(out, result = 0);
1371
1372         LASSERT(!obj->oo_hl_head);
1373         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1374
1375         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1376
1377 out:
1378         if (result || !obj->oo_inode) {
1379                 if (!IS_ERR_OR_NULL(inode))
1380                         iput(inode);
1381
1382                 obj->oo_inode = NULL;
1383                 if (trusted)
1384                         fid_zero(&oic->oic_fid);
1385         }
1386
1387         LINVRNT(osd_invariant(obj));
1388         return result;
1389 }
1390
1391 /*
1392  * Concurrency: shouldn't matter.
1393  */
1394 static void osd_object_init0(struct osd_object *obj)
1395 {
1396         LASSERT(obj->oo_inode != NULL);
1397         obj->oo_dt.do_body_ops = &osd_body_ops;
1398         obj->oo_dt.do_lu.lo_header->loh_attr |=
1399                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1400 }
1401
1402 /*
1403  * Concurrency: no concurrent access is possible that early in object
1404  * life-cycle.
1405  */
1406 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1407                            const struct lu_object_conf *conf)
1408 {
1409         struct osd_object *obj = osd_obj(l);
1410         int result;
1411
1412         LINVRNT(osd_invariant(obj));
1413
1414         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1415                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1416                 l->lo_header->loh_attr |= LOHA_EXISTS;
1417                 return 0;
1418         }
1419
1420         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1421         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1422         if (result == 0 && obj->oo_inode != NULL) {
1423                 struct osd_thread_info *oti = osd_oti_get(env);
1424                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1425
1426                 osd_object_init0(obj);
1427                 if (unlikely(obj->oo_header))
1428                         return 0;
1429
1430                 result = osd_get_lma(oti, obj->oo_inode,
1431                                      &oti->oti_obj_dentry, loa);
1432                 if (!result) {
1433                         /*
1434                          * Convert LMAI flags to lustre LMA flags
1435                          * and cache it to oo_lma_flags
1436                          */
1437                         obj->oo_lma_flags =
1438                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1439                 } else if (result == -ENODATA) {
1440                         result = 0;
1441                 }
1442         }
1443
1444         LINVRNT(osd_invariant(obj));
1445         return result;
1446 }
1447
1448 /*
1449  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1450  * The left part is for value, binary mode.
1451  */
1452 struct osd_xattr_entry {
1453         struct list_head        oxe_list;
1454         size_t                  oxe_len;
1455         size_t                  oxe_namelen;
1456         bool                    oxe_exist;
1457         struct rcu_head         oxe_rcu;
1458         char                    oxe_buf[0];
1459 };
1460
1461 static int osd_oxc_get(struct osd_object *obj, const char *name,
1462                        struct lu_buf *buf)
1463 {
1464         struct osd_xattr_entry *tmp;
1465         struct osd_xattr_entry *oxe = NULL;
1466         size_t namelen = strlen(name);
1467         int rc;
1468
1469         ENTRY;
1470
1471         rcu_read_lock();
1472         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1473                 if (namelen == tmp->oxe_namelen &&
1474                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1475                         oxe = tmp;
1476                         break;
1477                 }
1478         }
1479
1480         if (oxe == NULL)
1481                 GOTO(out, rc = -ENOENT);
1482
1483         if (!oxe->oxe_exist)
1484                 GOTO(out, rc = -ENODATA);
1485
1486         /* vallen */
1487         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1488         LASSERT(rc > 0);
1489
1490         if (buf->lb_buf == NULL)
1491                 GOTO(out, rc);
1492
1493         if (buf->lb_len < rc)
1494                 GOTO(out, rc = -ERANGE);
1495
1496         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1497         EXIT;
1498 out:
1499         rcu_read_unlock();
1500
1501         return rc;
1502 }
1503
1504 static void osd_oxc_free(struct rcu_head *head)
1505 {
1506         struct osd_xattr_entry *oxe;
1507
1508         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1509         OBD_FREE(oxe, oxe->oxe_len);
1510 }
1511
1512 static void osd_oxc_add(struct osd_object *obj, const char *name,
1513                         const char *buf, int buflen)
1514 {
1515         struct osd_xattr_entry *oxe;
1516         struct osd_xattr_entry *old = NULL;
1517         struct osd_xattr_entry *tmp;
1518         size_t namelen = strlen(name);
1519         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1520
1521         OBD_ALLOC(oxe, len);
1522         if (oxe == NULL)
1523                 return;
1524
1525         INIT_LIST_HEAD(&oxe->oxe_list);
1526         oxe->oxe_len = len;
1527         oxe->oxe_namelen = namelen;
1528         memcpy(oxe->oxe_buf, name, namelen);
1529         if (buflen > 0) {
1530                 LASSERT(buf != NULL);
1531                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1532                 oxe->oxe_exist = true;
1533         } else {
1534                 oxe->oxe_exist = false;
1535         }
1536
1537         /* this should be rarely called, just remove old and add new */
1538         spin_lock(&obj->oo_guard);
1539         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1540                 if (namelen == tmp->oxe_namelen &&
1541                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1542                         old = tmp;
1543                         break;
1544                 }
1545         }
1546         if (old != NULL) {
1547                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1548                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1549         } else {
1550                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1551         }
1552         spin_unlock(&obj->oo_guard);
1553 }
1554
1555 static void osd_oxc_del(struct osd_object *obj, const char *name)
1556 {
1557         struct osd_xattr_entry *oxe;
1558         size_t namelen = strlen(name);
1559
1560         spin_lock(&obj->oo_guard);
1561         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1562                 if (namelen == oxe->oxe_namelen &&
1563                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1564                         list_del_rcu(&oxe->oxe_list);
1565                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1566                         break;
1567                 }
1568         }
1569         spin_unlock(&obj->oo_guard);
1570 }
1571
1572 static void osd_oxc_fini(struct osd_object *obj)
1573 {
1574         struct osd_xattr_entry *oxe, *next;
1575
1576         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1577                 list_del(&oxe->oxe_list);
1578                 OBD_FREE(oxe, oxe->oxe_len);
1579         }
1580 }
1581
1582 /*
1583  * Concurrency: no concurrent access is possible that late in object
1584  * life-cycle.
1585  */
1586 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1587 {
1588         struct osd_object *obj = osd_obj(l);
1589         struct lu_object_header *h = obj->oo_header;
1590
1591         LINVRNT(osd_invariant(obj));
1592
1593         osd_oxc_fini(obj);
1594         dt_object_fini(&obj->oo_dt);
1595         if (obj->oo_hl_head != NULL)
1596                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1597         OBD_FREE_PTR(obj);
1598         if (unlikely(h)) {
1599                 lu_object_header_fini(h);
1600                 OBD_FREE_PTR(h);
1601         }
1602 }
1603
1604 /*
1605  * Concurrency: no concurrent access is possible that late in object
1606  * life-cycle.
1607  */
1608 static void osd_index_fini(struct osd_object *o)
1609 {
1610         struct iam_container *bag;
1611
1612         if (o->oo_dir != NULL) {
1613                 bag = &o->oo_dir->od_container;
1614                 if (o->oo_inode != NULL) {
1615                         if (bag->ic_object == o->oo_inode)
1616                                 iam_container_fini(bag);
1617                 }
1618                 OBD_FREE_PTR(o->oo_dir);
1619                 o->oo_dir = NULL;
1620         }
1621 }
1622
1623 /*
1624  * Concurrency: no concurrent access is possible that late in object
1625  * life-cycle (for all existing callers, that is. New callers have to provide
1626  * their own locking.)
1627  */
1628 static int osd_inode_unlinked(const struct inode *inode)
1629 {
1630         return inode->i_nlink == 0;
1631 }
1632
1633 enum {
1634         OSD_TXN_OI_DELETE_CREDITS    = 20,
1635         OSD_TXN_INODE_DELETE_CREDITS = 20
1636 };
1637
1638 /*
1639  * Journal
1640  */
1641
1642 #if OSD_THANDLE_STATS
1643 /**
1644  * Set time when the handle is allocated
1645  */
1646 static void osd_th_alloced(struct osd_thandle *oth)
1647 {
1648         oth->oth_alloced = ktime_get();
1649 }
1650
1651 /**
1652  * Set time when the handle started
1653  */
1654 static void osd_th_started(struct osd_thandle *oth)
1655 {
1656         oth->oth_started = ktime_get();
1657 }
1658
1659 /**
1660  * Check whether the we deal with this handle for too long.
1661  */
1662 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1663                                 ktime_t alloced, ktime_t started,
1664                                 ktime_t closed)
1665 {
1666         ktime_t now = ktime_get();
1667
1668         LASSERT(dev != NULL);
1669
1670         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1671                             ktime_us_delta(started, alloced));
1672         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1673                             ktime_us_delta(closed, started));
1674         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1675                             ktime_us_delta(now, closed));
1676
1677         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1678                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1679                                 oth, now, alloced, started, closed);
1680                 libcfs_debug_dumpstack(NULL);
1681         }
1682 }
1683
1684 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1685 {                                                                       \
1686         ktime_t __closed = ktime_get();                                 \
1687         ktime_t __alloced = oth->oth_alloced;                           \
1688         ktime_t __started = oth->oth_started;                           \
1689                                                                         \
1690         expr;                                                           \
1691         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1692 }
1693
1694 #else /* OSD_THANDLE_STATS */
1695
1696 #define osd_th_alloced(h)                  do {} while(0)
1697 #define osd_th_started(h)                  do {} while(0)
1698 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1699
1700 #endif /* OSD_THANDLE_STATS */
1701
1702 /*
1703  * Concurrency: doesn't access mutable data.
1704  */
1705 static int osd_param_is_not_sane(const struct osd_device *dev,
1706                                  const struct thandle *th)
1707 {
1708         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1709
1710         return oh->ot_credits > osd_transaction_size(dev);
1711 }
1712
1713 /*
1714  * Concurrency: shouldn't matter.
1715  */
1716 static void osd_trans_commit_cb(struct super_block *sb,
1717                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1718 {
1719         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1720         struct thandle *th = &oh->ot_super;
1721         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1722         struct dt_txn_commit_cb *dcb, *tmp;
1723
1724         LASSERT(oh->ot_handle == NULL);
1725
1726         if (error)
1727                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1728
1729         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1730         /* call per-transaction callbacks if any */
1731         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1732                                  dcb_linkage) {
1733                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1734                          "commit callback entry: magic=%x name='%s'\n",
1735                          dcb->dcb_magic, dcb->dcb_name);
1736                 list_del_init(&dcb->dcb_linkage);
1737                 dcb->dcb_func(NULL, th, dcb, error);
1738         }
1739
1740         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1741         lu_device_put(lud);
1742         th->th_dev = NULL;
1743
1744         OBD_FREE_PTR(oh);
1745 }
1746
1747 #ifndef HAVE_SB_START_WRITE
1748 # define sb_start_write(sb) do {} while (0)
1749 # define sb_end_write(sb) do {} while (0)
1750 #endif
1751
1752 static struct thandle *osd_trans_create(const struct lu_env *env,
1753                                         struct dt_device *d)
1754 {
1755         struct osd_thread_info *oti = osd_oti_get(env);
1756         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1757         struct osd_thandle *oh;
1758         struct thandle *th;
1759
1760         ENTRY;
1761
1762         if (d->dd_rdonly) {
1763                 CERROR("%s: someone try to start transaction under "
1764                        "readonly mode, should be disabled.\n",
1765                        osd_name(osd_dt_dev(d)));
1766                 dump_stack();
1767                 RETURN(ERR_PTR(-EROFS));
1768         }
1769
1770         /* on pending IO in this thread should left from prev. request */
1771         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1772
1773         sb_start_write(osd_sb(osd_dt_dev(d)));
1774
1775         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1776         if (!oh) {
1777                 sb_end_write(osd_sb(osd_dt_dev(d)));
1778                 RETURN(ERR_PTR(-ENOMEM));
1779         }
1780
1781         oh->ot_quota_trans = &oti->oti_quota_trans;
1782         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1783         th = &oh->ot_super;
1784         th->th_dev = d;
1785         th->th_result = 0;
1786         oh->ot_credits = 0;
1787         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1788         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1789         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1790         osd_th_alloced(oh);
1791
1792         memset(oti->oti_declare_ops, 0,
1793                sizeof(oti->oti_declare_ops));
1794         memset(oti->oti_declare_ops_cred, 0,
1795                sizeof(oti->oti_declare_ops_cred));
1796         memset(oti->oti_declare_ops_used, 0,
1797                sizeof(oti->oti_declare_ops_used));
1798
1799         oti->oti_ins_cache_depth++;
1800
1801         RETURN(th);
1802 }
1803
1804 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1805 {
1806         struct osd_thread_info *oti = osd_oti_get(env);
1807         struct osd_thandle *oh;
1808
1809         oh = container_of0(th, struct osd_thandle, ot_super);
1810         LASSERT(oh != NULL);
1811
1812         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1813               oti->oti_declare_ops[OSD_OT_CREATE],
1814               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1815               oti->oti_declare_ops_used[OSD_OT_CREATE],
1816               oti->oti_declare_ops[OSD_OT_DESTROY],
1817               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1818               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1819         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1820               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1821               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1822               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1823               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1824               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1825               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1826         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1827               oti->oti_declare_ops[OSD_OT_WRITE],
1828               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1829               oti->oti_declare_ops_used[OSD_OT_WRITE],
1830               oti->oti_declare_ops[OSD_OT_PUNCH],
1831               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1832               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1833               oti->oti_declare_ops[OSD_OT_QUOTA],
1834               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1835               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1836         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1837               oti->oti_declare_ops[OSD_OT_INSERT],
1838               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1839               oti->oti_declare_ops_used[OSD_OT_INSERT],
1840               oti->oti_declare_ops[OSD_OT_DELETE],
1841               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1842               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1843         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1844               oti->oti_declare_ops[OSD_OT_REF_ADD],
1845               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1846               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1847               oti->oti_declare_ops[OSD_OT_REF_DEL],
1848               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1849               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1850 }
1851
1852 /*
1853  * Concurrency: shouldn't matter.
1854  */
1855 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1856                            struct thandle *th)
1857 {
1858         struct osd_thread_info *oti = osd_oti_get(env);
1859         struct osd_device *dev = osd_dt_dev(d);
1860         handle_t *jh;
1861         struct osd_thandle *oh;
1862         int rc;
1863
1864         ENTRY;
1865
1866         LASSERT(current->journal_info == NULL);
1867
1868         oh = container_of0(th, struct osd_thandle, ot_super);
1869         LASSERT(oh != NULL);
1870         LASSERT(oh->ot_handle == NULL);
1871
1872         rc = dt_txn_hook_start(env, d, th);
1873         if (rc != 0)
1874                 GOTO(out, rc);
1875
1876         if (unlikely(osd_param_is_not_sane(dev, th))) {
1877                 static unsigned long last_printed;
1878                 static int last_credits;
1879
1880                 /*
1881                  * don't make noise on a tiny testing systems
1882                  * actual credits misuse will be caught anyway
1883                  */
1884                 if (last_credits != oh->ot_credits &&
1885                     time_after(jiffies, last_printed +
1886                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1887                     osd_transaction_size(dev) > 512) {
1888                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1889                               oh->ot_credits, osd_transaction_size(dev));
1890                         osd_trans_dump_creds(env, th);
1891                         libcfs_debug_dumpstack(NULL);
1892                         last_credits = oh->ot_credits;
1893                         last_printed = jiffies;
1894                 }
1895                 /*
1896                  * XXX Limit the credits to 'max_transaction_buffers', and
1897                  *     let the underlying filesystem to catch the error if
1898                  *     we really need so many credits.
1899                  *
1900                  *     This should be removed when we can calculate the
1901                  *     credits precisely.
1902                  */
1903                 oh->ot_credits = osd_transaction_size(dev);
1904         } else if (ldiskfs_track_declares_assert != 0) {
1905                 /*
1906                  * reserve few credits to prevent an assertion in JBD
1907                  * our debugging mechanism will be able to detected
1908                  * overuse. this can help to debug single-update
1909                  * transactions
1910                  */
1911                 oh->ot_credits += 10;
1912                 if (unlikely(osd_param_is_not_sane(dev, th)))
1913                         oh->ot_credits = osd_transaction_size(dev);
1914         }
1915
1916         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1917                 GOTO(out, rc = -EIO);
1918
1919         /*
1920          * XXX temporary stuff. Some abstraction layer should
1921          * be used.
1922          */
1923         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1924         osd_th_started(oh);
1925         if (!IS_ERR(jh)) {
1926                 oh->ot_handle = jh;
1927                 LASSERT(oti->oti_txns == 0);
1928
1929                 lu_device_get(&d->dd_lu_dev);
1930                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1931                               "osd-tx", th);
1932                 oti->oti_txns++;
1933                 rc = 0;
1934         } else {
1935                 rc = PTR_ERR(jh);
1936         }
1937 out:
1938         RETURN(rc);
1939 }
1940
1941 static int osd_seq_exists(const struct lu_env *env,
1942                           struct osd_device *osd, u64 seq)
1943 {
1944         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1945         struct seq_server_site *ss = osd_seq_site(osd);
1946         int rc;
1947
1948         ENTRY;
1949
1950         LASSERT(ss != NULL);
1951         LASSERT(ss->ss_server_fld != NULL);
1952
1953         rc = osd_fld_lookup(env, osd, seq, range);
1954         if (rc != 0) {
1955                 if (rc != -ENOENT)
1956                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1957                                osd_name(osd), seq, rc);
1958                 RETURN(0);
1959         }
1960
1961         RETURN(ss->ss_node_id == range->lsr_index);
1962 }
1963
1964 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1965 {
1966         struct dt_txn_commit_cb *dcb;
1967         struct dt_txn_commit_cb *tmp;
1968
1969         /* call per-transaction stop callbacks if any */
1970         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1971                                  dcb_linkage) {
1972                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1973                          "commit callback entry: magic=%x name='%s'\n",
1974                          dcb->dcb_magic, dcb->dcb_name);
1975                 list_del_init(&dcb->dcb_linkage);
1976                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1977         }
1978 }
1979
1980 /*
1981  * Concurrency: shouldn't matter.
1982  */
1983 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1984                           struct thandle *th)
1985 {
1986         struct osd_thread_info *oti = osd_oti_get(env);
1987         struct osd_thandle *oh;
1988         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1989         struct osd_device *osd = osd_dt_dev(th->th_dev);
1990         struct qsd_instance *qsd = osd_def_qsd(osd);
1991         struct lquota_trans *qtrans;
1992         struct list_head truncates = LIST_HEAD_INIT(truncates);
1993         int rc = 0, remove_agents = 0;
1994
1995         ENTRY;
1996
1997         oh = container_of0(th, struct osd_thandle, ot_super);
1998
1999         remove_agents = oh->ot_remove_agents;
2000
2001         qtrans = oh->ot_quota_trans;
2002         oh->ot_quota_trans = NULL;
2003
2004         /* move locks to local list, stop tx, execute truncates */
2005         list_splice(&oh->ot_trunc_locks, &truncates);
2006
2007         if (oh->ot_handle != NULL) {
2008                 int rc2;
2009
2010                 handle_t *hdl = oh->ot_handle;
2011
2012                 /*
2013                  * add commit callback
2014                  * notice we don't do this in osd_trans_start()
2015                  * as underlying transaction can change during truncate
2016                  */
2017                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2018                                              &oh->ot_jcb);
2019
2020                 LASSERT(oti->oti_txns == 1);
2021                 oti->oti_txns--;
2022
2023                 rc = dt_txn_hook_stop(env, th);
2024                 if (rc != 0)
2025                         CERROR("%s: failed in transaction hook: rc = %d\n",
2026                                osd_name(osd), rc);
2027
2028                 osd_trans_stop_cb(oh, rc);
2029                 /* hook functions might modify th_sync */
2030                 hdl->h_sync = th->th_sync;
2031
2032                 oh->ot_handle = NULL;
2033                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2034                 if (rc2 != 0)
2035                         CERROR("%s: failed to stop transaction: rc = %d\n",
2036                                osd_name(osd), rc2);
2037                 if (!rc)
2038                         rc = rc2;
2039
2040                 osd_process_truncates(&truncates);
2041         } else {
2042                 osd_trans_stop_cb(oh, th->th_result);
2043                 OBD_FREE_PTR(oh);
2044         }
2045
2046         osd_trunc_unlock_all(&truncates);
2047
2048         /* inform the quota slave device that the transaction is stopping */
2049         qsd_op_end(env, qsd, qtrans);
2050
2051         /*
2052          * as we want IO to journal and data IO be concurrent, we don't block
2053          * awaiting data IO completion in osd_do_bio(), instead we wait here
2054          * once transaction is submitted to the journal. all reqular requests
2055          * don't do direct IO (except read/write), thus this wait_event becomes
2056          * no-op for them.
2057          *
2058          * IMPORTANT: we have to wait till any IO submited by the thread is
2059          * completed otherwise iobuf may be corrupted by different request
2060          */
2061         wait_event(iobuf->dr_wait,
2062                        atomic_read(&iobuf->dr_numreqs) == 0);
2063         osd_fini_iobuf(osd, iobuf);
2064         if (!rc)
2065                 rc = iobuf->dr_error;
2066
2067         if (unlikely(remove_agents != 0))
2068                 osd_process_scheduled_agent_removals(env, osd);
2069
2070         oti->oti_ins_cache_depth--;
2071         /* reset OI cache for safety */
2072         if (oti->oti_ins_cache_depth == 0)
2073                 oti->oti_ins_cache_used = 0;
2074
2075         sb_end_write(osd_sb(osd));
2076
2077         RETURN(rc);
2078 }
2079
2080 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2081 {
2082         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2083                                                ot_super);
2084
2085         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2086         LASSERT(&dcb->dcb_func != NULL);
2087         if (dcb->dcb_flags & DCB_TRANS_STOP)
2088                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2089         else
2090                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2091
2092         return 0;
2093 }
2094
2095 /*
2096  * Called just before object is freed. Releases all resources except for
2097  * object itself (that is released by osd_object_free()).
2098  *
2099  * Concurrency: no concurrent access is possible that late in object
2100  * life-cycle.
2101  */
2102 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2103 {
2104         struct osd_object *obj = osd_obj(l);
2105         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2106         struct inode *inode = obj->oo_inode;
2107         __u64 projid;
2108         qid_t uid;
2109         qid_t gid;
2110
2111         LINVRNT(osd_invariant(obj));
2112
2113         /*
2114          * If object is unlinked remove fid->ino mapping from object index.
2115          */
2116
2117         osd_index_fini(obj);
2118
2119         if (!inode)
2120                 return;
2121
2122         uid = i_uid_read(inode);
2123         gid = i_gid_read(inode);
2124         projid = i_projid_read(inode);
2125
2126         obj->oo_inode = NULL;
2127         iput(inode);
2128
2129         /* do not rebalance quota if the caller needs to release memory
2130          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2131          * transaction and risk to deadlock - LU-12178 */
2132         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2133                 return;
2134
2135         if (!obj->oo_header && qsd) {
2136                 struct osd_thread_info *info = osd_oti_get(env);
2137                 struct lquota_id_info *qi = &info->oti_qi;
2138
2139                 /* Release granted quota to master if necessary */
2140                 qi->lqi_id.qid_uid = uid;
2141                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2142
2143                 qi->lqi_id.qid_uid = gid;
2144                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2145
2146                 qi->lqi_id.qid_uid = projid;
2147                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2148         }
2149 }
2150
2151 /*
2152  * Concurrency: ->loo_object_release() is called under site spin-lock.
2153  */
2154 static void osd_object_release(const struct lu_env *env,
2155                                struct lu_object *l)
2156 {
2157         struct osd_object *o = osd_obj(l);
2158
2159         /*
2160          * nobody should be releasing a non-destroyed object with nlink=0
2161          * the API allows this, but ldiskfs doesn't like and then report
2162          * this inode as deleted
2163          */
2164         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2165                   o->oo_inode->i_nlink == 0));
2166 }
2167
2168 /*
2169  * Concurrency: shouldn't matter.
2170  */
2171 static int osd_object_print(const struct lu_env *env, void *cookie,
2172                             lu_printer_t p, const struct lu_object *l)
2173 {
2174         struct osd_object *o = osd_obj(l);
2175         struct iam_descr *d;
2176
2177         if (o->oo_dir != NULL)
2178                 d = o->oo_dir->od_container.ic_descr;
2179         else
2180                 d = NULL;
2181         return (*p)(env, cookie,
2182                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2183                     o, o->oo_inode,
2184                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2185                     o->oo_inode ? o->oo_inode->i_generation : 0,
2186                     d ? d->id_ops->id_name : "plain");
2187 }
2188
2189 /*
2190  * Concurrency: shouldn't matter.
2191  */
2192 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2193                 struct obd_statfs *sfs)
2194 {
2195         struct osd_device *osd = osd_dt_dev(d);
2196         struct super_block *sb = osd_sb(osd);
2197         struct kstatfs *ksfs;
2198         __u64 reserved;
2199         int result = 0;
2200
2201         if (unlikely(osd->od_mnt == NULL))
2202                 return -EINPROGRESS;
2203
2204         /* osd_lproc.c call this without env, allocate ksfs for that case */
2205         if (unlikely(env == NULL)) {
2206                 OBD_ALLOC_PTR(ksfs);
2207                 if (ksfs == NULL)
2208                         return -ENOMEM;
2209         } else {
2210                 ksfs = &osd_oti_get(env)->oti_ksfs;
2211         }
2212
2213         result = sb->s_op->statfs(sb->s_root, ksfs);
2214         if (result)
2215                 goto out;
2216
2217         statfs_pack(sfs, ksfs);
2218         if (unlikely(sb->s_flags & MS_RDONLY))
2219                 sfs->os_state |= OS_STATE_READONLY;
2220
2221         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2222
2223         if (ldiskfs_has_feature_extents(sb))
2224                 sfs->os_maxbytes = sb->s_maxbytes;
2225         else
2226                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2227
2228         /*
2229          * Reserve some space so to avoid fragmenting the filesystem too much.
2230          * Fragmentation not only impacts performance, but can also increase
2231          * metadata overhead significantly, causing grant calculation to be
2232          * wrong.
2233          *
2234          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2235          */
2236         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2237         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2238         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2239                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2240
2241         sfs->os_blocks -= reserved;
2242         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2243         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2244
2245 out:
2246         if (unlikely(env == NULL))
2247                 OBD_FREE_PTR(ksfs);
2248         return result;
2249 }
2250
2251 /**
2252  * Estimate space needed for file creations. We assume the largest filename
2253  * which is 2^64 - 1, hence a filename of 20 chars.
2254  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2255  */
2256 #ifdef __LDISKFS_DIR_REC_LEN
2257 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2258 #else
2259 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2260 #endif
2261
2262 /*
2263  * Concurrency: doesn't access mutable data.
2264  */
2265 static void osd_conf_get(const struct lu_env *env,
2266                          const struct dt_device *dev,
2267                          struct dt_device_param *param)
2268 {
2269         struct osd_device *d = osd_dt_dev(dev);
2270         struct super_block *sb = osd_sb(d);
2271         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2272         const char *name;
2273         int ea_overhead;
2274
2275         /*
2276          * XXX should be taken from not-yet-existing fs abstraction layer.
2277          */
2278         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2279         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2280         param->ddp_symlink_max  = sb->s_blocksize;
2281         param->ddp_mount_type   = LDD_MT_LDISKFS;
2282         if (ldiskfs_has_feature_extents(sb))
2283                 param->ddp_maxbytes = sb->s_maxbytes;
2284         else
2285                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2286         /*
2287          * inode are statically allocated, so per-inode space consumption
2288          * is the space consumed by the directory entry
2289          */
2290         param->ddp_inodespace     = PER_OBJ_USAGE;
2291         /*
2292          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2293          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2294          * maximum length to not under-count the actual number of extents
2295          * needed for writing a file if there are sub-optimal block allocations.
2296          */
2297         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2298         /* worst-case extent insertion metadata overhead */
2299         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2300         param->ddp_mntopts = 0;
2301         if (test_opt(sb, XATTR_USER))
2302                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2303         if (test_opt(sb, POSIX_ACL))
2304                 param->ddp_mntopts |= MNTOPT_ACL;
2305
2306         /*
2307          * LOD might calculate the max stripe count based on max_ea_size,
2308          * so we need take account in the overhead as well,
2309          * xattr_header + magic + xattr_entry_head
2310          */
2311         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2312                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2313
2314 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2315         if (ldiskfs_has_feature_ea_inode(sb))
2316                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2317                                                                 ea_overhead;
2318         else
2319 #endif
2320                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2321
2322         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE - ea_overhead)
2323                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE - ea_overhead;
2324
2325         /*
2326          * Preferred RPC size for efficient disk IO.  4MB shows good
2327          * all-around performance for ldiskfs, but use bigalloc chunk size
2328          * by default if larger.
2329          */
2330 #if defined(LDISKFS_CLUSTER_SIZE)
2331         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2332                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2333         else
2334 #endif
2335                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2336
2337         param->ddp_t10_cksum_type = 0;
2338         if (bi) {
2339                 unsigned short interval = blk_integrity_interval(bi);
2340                 name = blk_integrity_name(bi);
2341                 /*
2342                  * Expected values:
2343                  * T10-DIF-TYPE1-CRC
2344                  * T10-DIF-TYPE3-CRC
2345                  * T10-DIF-TYPE1-IP
2346                  * T10-DIF-TYPE3-IP
2347                  */
2348                 if (strncmp(name, "T10-DIF-TYPE",
2349                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2350                         /* also skip "1/3-" at end */
2351                         const int type_off = sizeof("T10-DIF-TYPE.");
2352                         char type_number = name[type_off - 2];
2353
2354                         if (interval != 512 && interval != 4096) {
2355                                 CERROR("%s: unsupported T10PI sector size %u\n",
2356                                        d->od_svname, interval);
2357                         } else if (type_number != '1' && type_number != '3') {
2358                                 CERROR("%s: unsupported T10PI type %s\n",
2359                                        d->od_svname, name);
2360                         } else if (strcmp(name + type_off, "CRC") == 0) {
2361                                 d->od_t10_type = type_number == '1' ?
2362                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2363                                 param->ddp_t10_cksum_type = interval == 512 ?
2364                                         OBD_CKSUM_T10CRC512 :
2365                                         OBD_CKSUM_T10CRC4K;
2366                         } else if (strcmp(name + type_off, "IP") == 0) {
2367                                 d->od_t10_type = type_number == '1' ?
2368                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2369                                 param->ddp_t10_cksum_type = interval == 512 ?
2370                                         OBD_CKSUM_T10IP512 :
2371                                         OBD_CKSUM_T10IP4K;
2372                         } else {
2373                                 CERROR("%s: unsupported checksum type of "
2374                                        "T10PI type '%s'",
2375                                        d->od_svname, name);
2376                         }
2377
2378                 } else {
2379                         CERROR("%s: unsupported T10PI type '%s'",
2380                                d->od_svname, name);
2381                 }
2382         }
2383 }
2384
2385 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2386 {
2387         return osd_sb(osd_dt_dev(d));
2388 }
2389
2390 /*
2391  * Concurrency: shouldn't matter.
2392  */
2393 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2394 {
2395         int rc;
2396         struct super_block *s = osd_sb(osd_dt_dev(d));
2397         ENTRY;
2398
2399         down_read(&s->s_umount);
2400         rc = s->s_op->sync_fs(s, 1);
2401         up_read(&s->s_umount);
2402
2403         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2404                rc);
2405
2406         return rc;
2407 }
2408
2409 /**
2410  * Start commit for OSD device.
2411  *
2412  * An implementation of dt_commit_async method for OSD device.
2413  * Asychronously starts underlayng fs sync and thereby a transaction
2414  * commit.
2415  *
2416  * \param env environment
2417  * \param d dt device
2418  *
2419  * \see dt_device_operations
2420  */
2421 static int osd_commit_async(const struct lu_env *env,
2422                             struct dt_device *d)
2423 {
2424         struct super_block *s = osd_sb(osd_dt_dev(d));
2425         int rc;
2426
2427         ENTRY;
2428
2429         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2430         down_read(&s->s_umount);
2431         rc = s->s_op->sync_fs(s, 0);
2432         up_read(&s->s_umount);
2433
2434         RETURN(rc);
2435 }
2436
2437 /* Our own copy of the set readonly functions if present, or NU if not. */
2438 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2439 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2440 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2441
2442 /*
2443  * Concurrency: shouldn't matter.
2444  */
2445 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2446 {
2447         struct super_block *sb = osd_sb(osd_dt_dev(d));
2448         struct block_device *dev = sb->s_bdev;
2449         int rc = -EOPNOTSUPP;
2450
2451         ENTRY;
2452
2453         if (priv_dev_set_rdonly) {
2454                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2455
2456                 rc = 0;
2457                 CERROR("*** setting %s read-only ***\n",
2458                        osd_dt_dev(d)->od_svname);
2459
2460                 if (sb->s_op->freeze_fs) {
2461                         rc = sb->s_op->freeze_fs(sb);
2462                         if (rc)
2463                                 goto out;
2464                 }
2465
2466                 if (jdev && (jdev != dev)) {
2467                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2468                                (long)jdev);
2469                         priv_dev_set_rdonly(jdev);
2470                 }
2471                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2472                 priv_dev_set_rdonly(dev);
2473
2474                 if (sb->s_op->unfreeze_fs)
2475                         sb->s_op->unfreeze_fs(sb);
2476         }
2477
2478 out:
2479         if (rc)
2480                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2481                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2482
2483         RETURN(rc);
2484 }
2485
2486 /**
2487  * Note: we do not count into QUOTA here.
2488  * If we mount with --data_journal we may need more.
2489  */
2490 const int osd_dto_credits_noquota[DTO_NR] = {
2491         /**
2492          * Insert.
2493          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2494          * SINGLEDATA_TRANS_BLOCKS(8)
2495          * XXX Note: maybe iam need more, since iam have more level than
2496          *           EXT3 htree.
2497          */
2498         [DTO_INDEX_INSERT]  = 16,
2499         /**
2500          * Delete
2501          * just modify a single entry, probably merge few within a block
2502          */
2503         [DTO_INDEX_DELETE]  = 1,
2504         /**
2505          * Used for OI scrub
2506          */
2507         [DTO_INDEX_UPDATE]  = 16,
2508         /**
2509          * 4(inode, inode bits, groups, GDT)
2510          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2511          */
2512         [DTO_OBJECT_CREATE] = 4,
2513         /**
2514          * 4(inode, inode bits, groups, GDT)
2515          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2516          */
2517         [DTO_OBJECT_DELETE] = 4,
2518         /**
2519          * Attr set credits (inode)
2520          */
2521         [DTO_ATTR_SET_BASE] = 1,
2522         /**
2523          * Xattr set. The same as xattr of EXT3.
2524          * DATA_TRANS_BLOCKS(14)
2525          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2526          * are also counted in. Do not know why?
2527          */
2528         [DTO_XATTR_SET]     = 14,
2529         /**
2530          * credits for inode change during write.
2531          */
2532         [DTO_WRITE_BASE]    = 3,
2533         /**
2534          * credits for single block write.
2535          */
2536         [DTO_WRITE_BLOCK]   = 14,
2537         /**
2538          * Attr set credits for chown.
2539          * This is extra credits for setattr, and it is null without quota
2540          */
2541         [DTO_ATTR_SET_CHOWN] = 0
2542 };
2543
2544 static const struct dt_device_operations osd_dt_ops = {
2545         .dt_root_get       = osd_root_get,
2546         .dt_statfs         = osd_statfs,
2547         .dt_trans_create   = osd_trans_create,
2548         .dt_trans_start    = osd_trans_start,
2549         .dt_trans_stop     = osd_trans_stop,
2550         .dt_trans_cb_add   = osd_trans_cb_add,
2551         .dt_conf_get       = osd_conf_get,
2552         .dt_mnt_sb_get     = osd_mnt_sb_get,
2553         .dt_sync           = osd_sync,
2554         .dt_ro             = osd_ro,
2555         .dt_commit_async   = osd_commit_async,
2556 };
2557
2558 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2559                           unsigned int role)
2560 {
2561         struct osd_object *obj = osd_dt_obj(dt);
2562         struct osd_thread_info *oti = osd_oti_get(env);
2563
2564         LINVRNT(osd_invariant(obj));
2565
2566         LASSERT(obj->oo_owner != env);
2567         down_read_nested(&obj->oo_sem, role);
2568
2569         LASSERT(obj->oo_owner == NULL);
2570         oti->oti_r_locks++;
2571 }
2572
2573 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2574                            unsigned int role)
2575 {
2576         struct osd_object *obj = osd_dt_obj(dt);
2577         struct osd_thread_info *oti = osd_oti_get(env);
2578
2579         LINVRNT(osd_invariant(obj));
2580
2581         LASSERT(obj->oo_owner != env);
2582         down_write_nested(&obj->oo_sem, role);
2583
2584         LASSERT(obj->oo_owner == NULL);
2585         obj->oo_owner = env;
2586         oti->oti_w_locks++;
2587 }
2588
2589 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2590 {
2591         struct osd_object *obj = osd_dt_obj(dt);
2592         struct osd_thread_info *oti = osd_oti_get(env);
2593
2594         LINVRNT(osd_invariant(obj));
2595
2596         LASSERT(oti->oti_r_locks > 0);
2597         oti->oti_r_locks--;
2598         up_read(&obj->oo_sem);
2599 }
2600
2601 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2602 {
2603         struct osd_object *obj = osd_dt_obj(dt);
2604         struct osd_thread_info *oti = osd_oti_get(env);
2605
2606         LINVRNT(osd_invariant(obj));
2607
2608         LASSERT(obj->oo_owner == env);
2609         LASSERT(oti->oti_w_locks > 0);
2610         oti->oti_w_locks--;
2611         obj->oo_owner = NULL;
2612         up_write(&obj->oo_sem);
2613 }
2614
2615 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2616 {
2617         struct osd_object *obj = osd_dt_obj(dt);
2618
2619         LINVRNT(osd_invariant(obj));
2620
2621         return obj->oo_owner == env;
2622 }
2623
2624 static void osd_inode_getattr(const struct lu_env *env,
2625                               struct inode *inode, struct lu_attr *attr)
2626 {
2627         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2628                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2629                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2630                            LA_BLKSIZE | LA_TYPE;
2631
2632         attr->la_atime = inode->i_atime.tv_sec;
2633         attr->la_mtime = inode->i_mtime.tv_sec;
2634         attr->la_ctime = inode->i_ctime.tv_sec;
2635         attr->la_mode    = inode->i_mode;
2636         attr->la_size    = i_size_read(inode);
2637         attr->la_blocks  = inode->i_blocks;
2638         attr->la_uid     = i_uid_read(inode);
2639         attr->la_gid     = i_gid_read(inode);
2640         attr->la_projid  = i_projid_read(inode);
2641         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2642         attr->la_nlink   = inode->i_nlink;
2643         attr->la_rdev    = inode->i_rdev;
2644         attr->la_blksize = 1 << inode->i_blkbits;
2645         attr->la_blkbits = inode->i_blkbits;
2646         /*
2647          * Ext4 did not transfer inherit flags from raw inode
2648          * to inode flags, and ext4 internally test raw inode
2649          * @i_flags directly. Instead of patching ext4, we do it here.
2650          */
2651         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2652                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2653 }
2654
2655 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2656                         struct lu_attr *attr)
2657 {
2658         struct osd_object *obj = osd_dt_obj(dt);
2659
2660         if (unlikely(!dt_object_exists(dt)))
2661                 return -ENOENT;
2662         if (unlikely(obj->oo_destroyed))
2663                 return -ENOENT;
2664
2665         LASSERT(!dt_object_remote(dt));
2666         LINVRNT(osd_invariant(obj));
2667
2668         spin_lock(&obj->oo_guard);
2669         osd_inode_getattr(env, obj->oo_inode, attr);
2670         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2671                 attr->la_valid |= LA_FLAGS;
2672                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2673         }
2674         spin_unlock(&obj->oo_guard);
2675
2676         return 0;
2677 }
2678
2679 static int osd_declare_attr_qid(const struct lu_env *env,
2680                                 struct osd_object *obj,
2681                                 struct osd_thandle *oh, long long bspace,
2682                                 qid_t old_id, qid_t new_id, bool enforce,
2683                                 unsigned int type, bool ignore_edquot)
2684 {
2685         int rc;
2686         struct osd_thread_info *info = osd_oti_get(env);
2687         struct lquota_id_info  *qi = &info->oti_qi;
2688
2689         qi->lqi_type = type;
2690         /* inode accounting */
2691         qi->lqi_is_blk = false;
2692
2693         /* one more inode for the new id ... */
2694         qi->lqi_id.qid_uid = new_id;
2695         qi->lqi_space      = 1;
2696         /* Reserve credits for the new id */
2697         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2698         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2699                 rc = 0;
2700         if (rc)
2701                 RETURN(rc);
2702
2703         /* and one less inode for the current id */
2704         qi->lqi_id.qid_uid = old_id;
2705         qi->lqi_space = -1;
2706         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2707         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2708                 rc = 0;
2709         if (rc)
2710                 RETURN(rc);
2711
2712         /* block accounting */
2713         qi->lqi_is_blk = true;
2714
2715         /* more blocks for the new id ... */
2716         qi->lqi_id.qid_uid = new_id;
2717         qi->lqi_space      = bspace;
2718         /*
2719          * Credits for the new uid has been reserved, re-use "obj"
2720          * to save credit reservation.
2721          */
2722         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2723         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2724                 rc = 0;
2725         if (rc)
2726                 RETURN(rc);
2727
2728         /* and finally less blocks for the current uid */
2729         qi->lqi_id.qid_uid = old_id;
2730         qi->lqi_space      = -bspace;
2731         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2732         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2733                 rc = 0;
2734
2735         RETURN(rc);
2736 }
2737
2738 static int osd_declare_attr_set(const struct lu_env *env,
2739                                 struct dt_object *dt,
2740                                 const struct lu_attr *attr,
2741                                 struct thandle *handle)
2742 {
2743         struct osd_thandle *oh;
2744         struct osd_object *obj;
2745         qid_t uid;
2746         qid_t gid;
2747         long long bspace;
2748         int rc = 0;
2749         bool enforce;
2750
2751         ENTRY;
2752
2753         LASSERT(dt != NULL);
2754         LASSERT(handle != NULL);
2755
2756         obj = osd_dt_obj(dt);
2757         LASSERT(osd_invariant(obj));
2758
2759         oh = container_of0(handle, struct osd_thandle, ot_super);
2760         LASSERT(oh->ot_handle == NULL);
2761
2762         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2763                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2764
2765         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2766                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2767
2768         if (attr == NULL || obj->oo_inode == NULL)
2769                 RETURN(rc);
2770
2771         bspace   = obj->oo_inode->i_blocks << 9;
2772         bspace   = toqb(bspace);
2773
2774         /*
2775          * Changing ownership is always preformed by super user, it should not
2776          * fail with EDQUOT unless required explicitly.
2777          *
2778          * We still need to call the osd_declare_qid() to calculate the journal
2779          * credits for updating quota accounting files and to trigger quota
2780          * space adjustment once the operation is completed.
2781          */
2782         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2783                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2784
2785                 if (!ignore_edquot)
2786                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2787                                "(the quota space is %lld)\n",
2788                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2789                                attr->la_gid, bspace);
2790
2791                 /* USERQUOTA */
2792                 uid = i_uid_read(obj->oo_inode);
2793                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2794                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2795                                           attr->la_uid, enforce, USRQUOTA,
2796                                           true);
2797                 if (rc)
2798                         RETURN(rc);
2799
2800                 gid = i_gid_read(obj->oo_inode);
2801                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2802                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2803                                           i_gid_read(obj->oo_inode),
2804                                           attr->la_gid, enforce, GRPQUOTA,
2805                                           ignore_edquot);
2806                 if (rc)
2807                         RETURN(rc);
2808
2809         }
2810 #ifdef HAVE_PROJECT_QUOTA
2811         if (attr->la_valid & LA_PROJID) {
2812                 __u32 projid = i_projid_read(obj->oo_inode);
2813
2814                 enforce = (attr->la_valid & LA_PROJID) &&
2815                                         (attr->la_projid != projid);
2816                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2817                                           (qid_t)projid, (qid_t)attr->la_projid,
2818                                           enforce, PRJQUOTA, true);
2819                 if (rc)
2820                         RETURN(rc);
2821         }
2822 #endif
2823         RETURN(rc);
2824 }
2825
2826 static int osd_inode_setattr(const struct lu_env *env,
2827                              struct inode *inode, const struct lu_attr *attr)
2828 {
2829         __u64 bits = attr->la_valid;
2830
2831         /* Only allow set size for regular file */
2832         if (!S_ISREG(inode->i_mode))
2833                 bits &= ~(LA_SIZE | LA_BLOCKS);
2834
2835         if (bits == 0)
2836                 return 0;
2837
2838         if (bits & LA_ATIME)
2839                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2840         if (bits & LA_CTIME)
2841                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2842         if (bits & LA_MTIME)
2843                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2844         if (bits & LA_SIZE) {
2845                 spin_lock(&inode->i_lock);
2846                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2847                 i_size_write(inode, attr->la_size);
2848                 spin_unlock(&inode->i_lock);
2849         }
2850
2851         /*
2852          * OSD should not change "i_blocks" which is used by quota.
2853          * "i_blocks" should be changed by ldiskfs only.
2854          */
2855         if (bits & LA_MODE)
2856                 inode->i_mode = (inode->i_mode & S_IFMT) |
2857                                 (attr->la_mode & ~S_IFMT);
2858         if (bits & LA_UID)
2859                 i_uid_write(inode, attr->la_uid);
2860         if (bits & LA_GID)
2861                 i_gid_write(inode, attr->la_gid);
2862         if (bits & LA_PROJID)
2863                 i_projid_write(inode, attr->la_projid);
2864         if (bits & LA_NLINK)
2865                 set_nlink(inode, attr->la_nlink);
2866         if (bits & LA_RDEV)
2867                 inode->i_rdev = attr->la_rdev;
2868
2869         if (bits & LA_FLAGS) {
2870                 /* always keep S_NOCMTIME */
2871                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2872                                  S_NOCMTIME;
2873                 /*
2874                  * Ext4 did not transfer inherit flags from
2875                  * @inode->i_flags to raw inode i_flags when writing
2876                  * flags, we do it explictly here.
2877                  */
2878                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2879                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2880                 else
2881                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2882         }
2883         return 0;
2884 }
2885
2886 #ifdef HAVE_PROJECT_QUOTA
2887 static int osd_transfer_project(struct inode *inode, __u32 projid)
2888 {
2889         struct super_block *sb = inode->i_sb;
2890         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2891         int err;
2892         kprojid_t kprojid;
2893         struct ldiskfs_iloc iloc;
2894         struct ldiskfs_inode *raw_inode;
2895         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2896
2897         if (!ldiskfs_has_feature_project(sb)) {
2898                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2899                         == LDISKFS_DEF_PROJID);
2900                 if (projid != LDISKFS_DEF_PROJID)
2901                         return -EOPNOTSUPP;
2902                 else
2903                         return 0;
2904         }
2905
2906         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2907                 return -EOPNOTSUPP;
2908
2909         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2910         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2911                 return 0;
2912
2913         err = ldiskfs_get_inode_loc(inode, &iloc);
2914         if (err)
2915                 return err;
2916
2917         raw_inode = ldiskfs_raw_inode(&iloc);
2918         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2919                 err = -EOVERFLOW;
2920                 brelse(iloc.bh);
2921                 return err;
2922         }
2923         brelse(iloc.bh);
2924
2925         dquot_initialize(inode);
2926         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2927         if (transfer_to[PRJQUOTA]) {
2928                 err = __dquot_transfer(inode, transfer_to);
2929                 dqput(transfer_to[PRJQUOTA]);
2930                 if (err)
2931                         return err;
2932         }
2933
2934         return err;
2935 }
2936 #endif
2937
2938 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2939 {
2940         int rc;
2941
2942         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2943             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2944                 struct iattr iattr;
2945
2946                 ll_vfs_dq_init(inode);
2947                 iattr.ia_valid = 0;
2948                 if (attr->la_valid & LA_UID)
2949                         iattr.ia_valid |= ATTR_UID;
2950                 if (attr->la_valid & LA_GID)
2951                         iattr.ia_valid |= ATTR_GID;
2952                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2953                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2954
2955                 rc = ll_vfs_dq_transfer(inode, &iattr);
2956                 if (rc) {
2957                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2958                                "enforcement enabled on the ldiskfs "
2959                                "filesystem?\n", inode->i_sb->s_id, rc);
2960                         return rc;
2961                 }
2962         }
2963
2964         /* Handle project id transfer here properly */
2965         if (attr->la_valid & LA_PROJID &&
2966             attr->la_projid != i_projid_read(inode)) {
2967 #ifdef HAVE_PROJECT_QUOTA
2968                 rc = osd_transfer_project(inode, attr->la_projid);
2969 #else
2970                 rc = -ENOTSUPP;
2971 #endif
2972                 if (rc) {
2973                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2974                                "enforcement enabled on the ldiskfs "
2975                                "filesystem?\n", inode->i_sb->s_id, rc);
2976                         return rc;
2977                 }
2978         }
2979         return 0;
2980 }
2981
2982 static int osd_attr_set(const struct lu_env *env,
2983                         struct dt_object *dt,
2984                         const struct lu_attr *attr,
2985                         struct thandle *handle)
2986 {
2987         struct osd_object *obj = osd_dt_obj(dt);
2988         struct inode *inode;
2989         int rc;
2990
2991         if (!dt_object_exists(dt))
2992                 return -ENOENT;
2993
2994         LASSERT(handle != NULL);
2995         LASSERT(!dt_object_remote(dt));
2996         LASSERT(osd_invariant(obj));
2997
2998         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2999
3000         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3001             !osd_obj2dev(obj)->od_is_ost) {
3002                 struct osd_thread_info *oti = osd_oti_get(env);
3003                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3004                 struct lu_fid *fid1 = &oti->oti_fid;
3005                 struct osd_inode_id *id = &oti->oti_id;
3006                 struct iam_path_descr *ipd;
3007                 struct iam_container *bag;
3008                 struct osd_thandle *oh;
3009                 int rc;
3010
3011                 fid_cpu_to_be(fid1, fid0);
3012                 memset(id, 1, sizeof(*id));
3013                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3014                                   fid0)->oi_dir.od_container;
3015                 ipd = osd_idx_ipd_get(env, bag);
3016                 if (unlikely(ipd == NULL))
3017                         RETURN(-ENOMEM);
3018
3019                 oh = container_of0(handle, struct osd_thandle, ot_super);
3020                 rc = iam_update(oh->ot_handle, bag,
3021                                 (const struct iam_key *)fid1,
3022                                 (const struct iam_rec *)id, ipd);
3023                 osd_ipd_put(env, bag, ipd);
3024                 return(rc > 0 ? 0 : rc);
3025         }
3026
3027         inode = obj->oo_inode;
3028
3029         rc = osd_quota_transfer(inode, attr);
3030         if (rc)
3031                 return rc;
3032
3033         spin_lock(&obj->oo_guard);
3034         rc = osd_inode_setattr(env, inode, attr);
3035         spin_unlock(&obj->oo_guard);
3036         if (rc != 0)
3037                 GOTO(out, rc);
3038
3039         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3040
3041         if (!(attr->la_valid & LA_FLAGS))
3042                 GOTO(out, rc);
3043
3044         /* Let's check if there are extra flags need to be set into LMA */
3045         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3046                 struct osd_thread_info *info = osd_oti_get(env);
3047                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3048
3049                 LASSERT(!obj->oo_pfid_in_lma);
3050
3051                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3052                                  &info->oti_ost_attrs);
3053                 if (rc)
3054                         GOTO(out, rc);
3055
3056                 lma->lma_incompat |=
3057                         lustre_to_lma_flags(attr->la_flags);
3058                 lustre_lma_swab(lma);
3059                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3060                                      lma, sizeof(*lma), XATTR_REPLACE);
3061                 if (rc != 0) {
3062                         struct osd_device *osd = osd_obj2dev(obj);
3063
3064                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3065                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3066                               lma->lma_incompat, rc);
3067                 } else {
3068                         obj->oo_lma_flags =
3069                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3070                 }
3071                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3072         }
3073 out:
3074         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3075
3076         return rc;
3077 }
3078
3079 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3080                                            struct osd_object *obj,
3081                                            const char *name, const int namelen)
3082 {
3083         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3084 }
3085
3086 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3087                       umode_t mode, struct dt_allocation_hint *hint,
3088                       struct thandle *th, struct lu_attr *attr)
3089 {
3090         int result;
3091         struct osd_device *osd = osd_obj2dev(obj);
3092         struct osd_thandle *oth;
3093         struct dt_object *parent = NULL;
3094         struct inode *inode;
3095         uid_t owner[2] = {0, 0};
3096
3097         if (attr->la_valid & LA_UID)
3098                 owner[0] = attr->la_uid;
3099         if (attr->la_valid & LA_GID)
3100                 owner[1] = attr->la_gid;
3101
3102         LINVRNT(osd_invariant(obj));
3103         LASSERT(obj->oo_inode == NULL);
3104         LASSERT(obj->oo_hl_head == NULL);
3105
3106         if (S_ISDIR(mode) && ldiskfs_pdo) {
3107                 obj->oo_hl_head =
3108                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3109                 if (obj->oo_hl_head == NULL)
3110                         return -ENOMEM;
3111         }
3112
3113         oth = container_of(th, struct osd_thandle, ot_super);
3114         LASSERT(oth->ot_handle->h_transaction != NULL);
3115
3116         if (hint != NULL && hint->dah_parent != NULL &&
3117             !dt_object_remote(hint->dah_parent))
3118                 parent = hint->dah_parent;
3119
3120         inode = ldiskfs_create_inode(oth->ot_handle,
3121                                      parent ? osd_dt_obj(parent)->oo_inode :
3122                                               osd_sb(osd)->s_root->d_inode,
3123                                      mode, owner);
3124         if (!IS_ERR(inode)) {
3125                 /* Do not update file c/mtime in ldiskfs. */
3126                 inode->i_flags |= S_NOCMTIME;
3127
3128                 /*
3129                  * For new created object, it must be consistent,
3130                  * and it is unnecessary to scrub against it.
3131                  */
3132                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3133
3134                 obj->oo_inode = inode;
3135                 result = 0;
3136         } else {
3137                 if (obj->oo_hl_head != NULL) {
3138                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3139                         obj->oo_hl_head = NULL;
3140                 }
3141                 result = PTR_ERR(inode);
3142         }
3143         LINVRNT(osd_invariant(obj));
3144         return result;
3145 }
3146
3147 enum {
3148         OSD_NAME_LEN = 255
3149 };
3150
3151 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3152                      struct lu_attr *attr,
3153                      struct dt_allocation_hint *hint,
3154                      struct dt_object_format *dof,
3155                      struct thandle *th)
3156 {
3157         int result;
3158         struct osd_thandle *oth;
3159         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3160
3161         LASSERT(S_ISDIR(attr->la_mode));
3162
3163         oth = container_of(th, struct osd_thandle, ot_super);
3164         LASSERT(oth->ot_handle->h_transaction != NULL);
3165         result = osd_mkfile(info, obj, mode, hint, th, attr);
3166
3167         return result;
3168 }
3169
3170 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3171                         struct lu_attr *attr,
3172                         struct dt_allocation_hint *hint,
3173                         struct dt_object_format *dof,
3174                         struct thandle *th)
3175 {
3176         int result;
3177         struct osd_thandle *oth;
3178         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3179
3180         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3181
3182         LASSERT(S_ISREG(attr->la_mode));
3183
3184         oth = container_of(th, struct osd_thandle, ot_super);
3185         LASSERT(oth->ot_handle->h_transaction != NULL);
3186
3187         result = osd_mkfile(info, obj, mode, hint, th, attr);
3188         if (result == 0) {
3189                 LASSERT(obj->oo_inode != NULL);
3190                 if (feat->dif_flags & DT_IND_VARKEY)
3191                         result = iam_lvar_create(obj->oo_inode,
3192                                                  feat->dif_keysize_max,
3193                                                  feat->dif_ptrsize,
3194                                                  feat->dif_recsize_max,
3195                                                  oth->ot_handle);
3196                 else
3197                         result = iam_lfix_create(obj->oo_inode,
3198                                                  feat->dif_keysize_max,
3199                                                  feat->dif_ptrsize,
3200                                                  feat->dif_recsize_max,
3201                                                  oth->ot_handle);
3202         }
3203         return result;
3204 }
3205
3206 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3207                      struct lu_attr *attr,
3208                      struct dt_allocation_hint *hint,
3209                      struct dt_object_format *dof,
3210                      struct thandle *th)
3211 {
3212         LASSERT(S_ISREG(attr->la_mode));
3213         return osd_mkfile(info, obj, (attr->la_mode &
3214                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3215                           attr);
3216 }
3217
3218 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3219                      struct lu_attr *attr,
3220                      struct dt_allocation_hint *hint,
3221                      struct dt_object_format *dof,
3222                      struct thandle *th)
3223 {
3224         LASSERT(S_ISLNK(attr->la_mode));
3225         return osd_mkfile(info, obj, (attr->la_mode &
3226                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3227                           attr);
3228 }
3229
3230 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3231                      struct lu_attr *attr,
3232                      struct dt_allocation_hint *hint,
3233                      struct dt_object_format *dof,
3234                      struct thandle *th)
3235 {
3236         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3237         int result;
3238
3239         LINVRNT(osd_invariant(obj));
3240         LASSERT(obj->oo_inode == NULL);
3241         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3242                 S_ISFIFO(mode) || S_ISSOCK(mode));
3243
3244         result = osd_mkfile(info, obj, mode, hint, th, attr);
3245         if (result == 0) {
3246                 LASSERT(obj->oo_inode != NULL);
3247                 /*
3248                  * This inode should be marked dirty for i_rdev.  Currently
3249                  * that is done in the osd_attr_init().
3250                  */
3251                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3252                                    attr->la_rdev);
3253         }
3254         LINVRNT(osd_invariant(obj));
3255         return result;
3256 }
3257
3258 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3259                               struct lu_attr *,
3260                               struct dt_allocation_hint *hint,
3261                               struct dt_object_format *dof,
3262                               struct thandle *);
3263
3264 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3265 {
3266         osd_obj_type_f result;
3267
3268         switch (type) {
3269         case DFT_DIR:
3270                 result = osd_mkdir;
3271                 break;
3272         case DFT_REGULAR:
3273                 result = osd_mkreg;
3274                 break;
3275         case DFT_SYM:
3276                 result = osd_mksym;
3277                 break;
3278         case DFT_NODE:
3279                 result = osd_mknod;
3280                 break;
3281         case DFT_INDEX:
3282                 result = osd_mk_index;
3283                 break;
3284
3285         default:
3286                 LBUG();
3287                 break;
3288         }
3289         return result;
3290 }
3291
3292
3293 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3294                         struct dt_object *parent, struct dt_object *child,
3295                         umode_t child_mode)
3296 {
3297         LASSERT(ah);
3298
3299         ah->dah_parent = parent;
3300         ah->dah_mode = child_mode;
3301
3302         if (parent != NULL && !dt_object_remote(parent)) {
3303                 /* will help to find FID->ino at dt_insert("..") */
3304                 struct osd_object *pobj = osd_dt_obj(parent);
3305
3306                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3307         }
3308 }
3309
3310 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3311                           struct lu_attr *attr, struct dt_object_format *dof)
3312 {
3313         struct inode *inode = obj->oo_inode;
3314         __u64 valid = attr->la_valid;
3315         int result;
3316
3317         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3318
3319         if (dof->dof_type != DFT_NODE)
3320                 attr->la_valid &= ~LA_RDEV;
3321         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3322                 attr->la_valid &= ~LA_ATIME;
3323         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3324                 attr->la_valid &= ~LA_CTIME;
3325         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3326                 attr->la_valid &= ~LA_MTIME;
3327
3328         result = osd_quota_transfer(inode, attr);
3329         if (result)
3330                 return;
3331
3332         if (attr->la_valid != 0) {
3333                 result = osd_inode_setattr(info->oti_env, inode, attr);
3334                 /*
3335                  * The osd_inode_setattr() should always succeed here.  The
3336                  * only error that could be returned is EDQUOT when we are
3337                  * trying to change the UID or GID of the inode. However, this
3338                  * should not happen since quota enforcement is no longer
3339                  * enabled on ldiskfs (lquota takes care of it).
3340                  */
3341                 LASSERTF(result == 0, "%d\n", result);
3342                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3343         }
3344
3345         attr->la_valid = valid;
3346 }
3347
3348 /**
3349  * Helper function for osd_create()
3350  *
3351  * \retval 0, on success
3352  */
3353 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3354                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3355                         struct dt_object_format *dof, struct thandle *th)
3356 {
3357         int result;
3358         __u32 umask;
3359
3360         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3361
3362         /* we drop umask so that permissions we pass are not affected */
3363         umask = current->fs->umask;
3364         current->fs->umask = 0;
3365
3366         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3367                                                   th);
3368         if (likely(obj->oo_inode != NULL)) {
3369                 LASSERT(obj->oo_inode->i_state & I_NEW);
3370
3371                 /*
3372                  * Unlock the inode before attr initialization to avoid
3373                  * unnecessary dqget operations. LU-6378
3374                  */
3375                 unlock_new_inode(obj->oo_inode);
3376         }
3377
3378         if (likely(result == 0)) {
3379                 osd_attr_init(info, obj, attr, dof);
3380                 osd_object_init0(obj);
3381         }
3382
3383         /* restore previous umask value */
3384         current->fs->umask = umask;
3385
3386         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3387
3388         return result;
3389 }
3390
3391 /**
3392  * Helper function for osd_create()
3393  *
3394  * \retval 0, on success
3395  */
3396 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3397                            const struct lu_fid *fid, struct thandle *th)
3398 {
3399         struct osd_thread_info *info = osd_oti_get(env);
3400         struct osd_inode_id    *id   = &info->oti_id;
3401         struct osd_device      *osd  = osd_obj2dev(obj);
3402         struct osd_thandle     *oh;
3403         int rc;
3404
3405         LASSERT(obj->oo_inode != NULL);
3406
3407         oh = container_of0(th, struct osd_thandle, ot_super);
3408         LASSERT(oh->ot_handle);
3409         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3410
3411         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3412         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3413                            OI_CHECK_FLD, NULL);
3414         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3415                 struct lu_fid next_fid = *fid;
3416
3417                 /* insert next object in advance, and map to the same inode */
3418                 next_fid.f_oid++;
3419                 if (next_fid.f_oid != 0) {
3420                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3421                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3422                                       OI_CHECK_FLD, NULL);
3423                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3424                 }
3425         }
3426
3427         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3428
3429         return rc;
3430 }
3431
3432 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3433                    u64 seq, struct lu_seq_range *range)
3434 {
3435         struct seq_server_site *ss = osd_seq_site(osd);
3436
3437         if (fid_seq_is_idif(seq)) {
3438                 fld_range_set_ost(range);
3439                 range->lsr_index = idif_ost_idx(seq);
3440                 return 0;
3441         }
3442
3443         if (!fid_seq_in_fldb(seq)) {
3444                 fld_range_set_mdt(range);
3445                 if (ss != NULL)
3446                         /*
3447                          * FIXME: If ss is NULL, it suppose not get lsr_index
3448                          * at all
3449                          */
3450                         range->lsr_index = ss->ss_node_id;
3451                 return 0;
3452         }
3453
3454         LASSERT(ss != NULL);
3455         fld_range_set_any(range);
3456         /* OSD will only do local fld lookup */
3457         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3458 }
3459
3460 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3461                               struct lu_attr *attr,
3462                               struct dt_allocation_hint *hint,
3463                               struct dt_object_format *dof,
3464                               struct thandle *handle)
3465 {
3466         struct osd_thandle *oh;
3467         int rc;
3468
3469         ENTRY;
3470
3471         LASSERT(handle != NULL);
3472
3473         oh = container_of0(handle, struct osd_thandle, ot_super);
3474         LASSERT(oh->ot_handle == NULL);
3475
3476         /*
3477          * EA object consumes more credits than regular object: osd_mk_index
3478          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3479          * leaf_node, could involves the block, block bitmap, groups, GDT
3480          * change for each block, so add 4 * 2 credits in that case.
3481          */
3482         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3483                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3484                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3485         /*
3486          * Reuse idle OI block may cause additional one OI block
3487          * to be changed.
3488          */
3489         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3490                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3491         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3492                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3493                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3494
3495         /* will help to find FID->ino mapping at dt_insert() */
3496         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3497                                    osd_dt_obj(dt));
3498         if (rc != 0)
3499                 RETURN(rc);
3500
3501         if (!attr)
3502                 RETURN(0);
3503
3504         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3505                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3506                                    NULL, OSD_QID_INODE);
3507         if (rc != 0)
3508                 RETURN(rc);
3509
3510         RETURN(rc);
3511 }
3512
3513 /**
3514  * Called to destroy on-disk representation of the object
3515  *
3516  * Concurrency: must be locked
3517  */
3518 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3519                                struct thandle *th)
3520 {
3521         struct osd_object *obj = osd_dt_obj(dt);
3522         struct inode *inode = obj->oo_inode;
3523         struct osd_thandle *oh;
3524         int rc;
3525
3526         ENTRY;
3527
3528         if (inode == NULL)
3529                 RETURN(-ENOENT);
3530
3531         oh = container_of0(th, struct osd_thandle, ot_super);
3532         LASSERT(oh->ot_handle == NULL);
3533
3534         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3535                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3536
3537         /* For removing agent entry */
3538         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3539                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3540
3541         /*
3542          * Recycle idle OI leaf may cause additional three OI blocks
3543          * to be changed.
3544          */
3545         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3546                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3547                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3548         /* one less inode */
3549         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3550                                    i_projid_read(inode), -1, oh, obj, NULL,
3551                                    OSD_QID_INODE);
3552         if (rc)
3553                 RETURN(rc);
3554         /* data to be truncated */
3555         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3556                                    i_projid_read(inode), 0, oh, obj, NULL,
3557                                    OSD_QID_BLK);
3558         if (rc)
3559                 RETURN(rc);
3560
3561         /*
3562          * will help to find FID->ino when this object is being
3563          * added to PENDING
3564          */
3565         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3566
3567         RETURN(rc);
3568 }
3569
3570 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3571                        struct thandle *th)
3572 {
3573         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3574         struct osd_object *obj = osd_dt_obj(dt);
3575         struct inode *inode = obj->oo_inode;
3576         struct osd_device *osd = osd_obj2dev(obj);
3577         struct osd_thandle *oh;
3578         int result;
3579
3580         ENTRY;
3581
3582         oh = container_of0(th, struct osd_thandle, ot_super);
3583         LASSERT(oh->ot_handle);
3584         LASSERT(inode);
3585         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3586
3587         if (unlikely(fid_is_acct(fid)))
3588                 RETURN(-EPERM);
3589
3590         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3591                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3592                 if (result != 0)
3593                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3594                                osd_name(osd), PFID(fid), result);
3595         }
3596
3597         if (S_ISDIR(inode->i_mode)) {
3598                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3599                         inode->i_nlink == 2);
3600
3601                 spin_lock(&obj->oo_guard);
3602                 clear_nlink(inode);
3603                 spin_unlock(&obj->oo_guard);
3604                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3605         }
3606
3607         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3608
3609         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3610
3611         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3612                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3613                                        oh->ot_handle, OI_CHECK_FLD);
3614
3615         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3616         /* XXX: add to ext3 orphan list */
3617         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3618
3619         /* not needed in the cache anymore */
3620         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3621         obj->oo_destroyed = 1;
3622
3623         RETURN(0);
3624 }
3625
3626 /**
3627  * Put the fid into lustre_mdt_attrs, and then place the structure
3628  * inode's ea. This fid should not be altered during the life time
3629  * of the inode.
3630  *
3631  * \retval +ve, on success
3632  * \retval -ve, on error
3633  *
3634  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3635  */
3636 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3637                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3638 {
3639         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3640         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3641         int rc;
3642
3643         ENTRY;
3644
3645         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3646                 RETURN(0);
3647
3648         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3649                 rc = -ENOMEM;
3650
3651         lustre_loa_init(loa, fid, compat, incompat);
3652         lustre_loa_swab(loa, false);
3653
3654         /*
3655          * For the OST device with 256 bytes inode size by default,
3656          * the PFID EA will be stored together with LMA EA to avoid
3657          * performance trouble. Otherwise the PFID EA can be stored
3658          * independently. LU-8998
3659          */
3660         if ((compat & LMAC_FID_ON_OST) &&
3661             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3662                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3663                                      sizeof(*loa), XATTR_CREATE);
3664         else
3665                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3666                                      sizeof(*lma), XATTR_CREATE);
3667         /*
3668          * LMA may already exist, but we need to check that all the
3669          * desired compat/incompat flags have been added.
3670          */
3671         if (unlikely(rc == -EEXIST)) {
3672                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3673                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3674                 if (rc < 0)
3675                         RETURN(rc);
3676
3677                 if (rc < sizeof(*lma))
3678                         RETURN(-EINVAL);
3679
3680                 lustre_loa_swab(loa, true);
3681                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3682                     ((compat == 0 && incompat == 0) ||
3683                      (!(~lma->lma_compat & compat) &&
3684                       !(~lma->lma_incompat & incompat))))
3685                         RETURN(0);
3686
3687                 lma->lma_self_fid = *fid;
3688                 lma->lma_compat |= compat;
3689                 lma->lma_incompat |= incompat;
3690                 if (rc == sizeof(*lma)) {
3691                         lustre_lma_swab(lma);
3692                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3693                                              sizeof(*lma), XATTR_REPLACE);
3694                 } else {
3695                         lustre_loa_swab(loa, false);
3696                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3697                                              sizeof(*loa), XATTR_REPLACE);
3698                 }
3699         }
3700
3701         RETURN(rc);
3702 }
3703
3704 /**
3705  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3706  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3707  * To have compatilibility with 1.8 ldiskfs driver we need to have
3708  * magic number at start of fid data.
3709  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3710  * its inmemory API.
3711  */
3712 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3713                                          const struct lu_fid *fid)
3714 {
3715         if (!fid_is_namespace_visible(fid) ||
3716             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3717                 param->edp_magic = 0;
3718                 return;
3719         }
3720
3721         param->edp_magic = LDISKFS_LUFID_MAGIC;
3722         param->edp_len =  sizeof(struct lu_fid) + 1;
3723         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3724 }
3725
3726 /**
3727  * Try to read the fid from inode ea into dt_rec.
3728  *
3729  * \param fid object fid.
3730  *
3731  * \retval 0 on success
3732  */
3733 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3734                           __u32 ino, struct lu_fid *fid,
3735                           struct osd_inode_id *id)
3736 {
3737         struct osd_thread_info *info  = osd_oti_get(env);
3738         struct inode *inode;
3739
3740         ENTRY;
3741
3742         osd_id_gen(id, ino, OSD_OII_NOGEN);
3743         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3744         if (IS_ERR(inode))
3745                 RETURN(PTR_ERR(inode));
3746
3747         iput(inode);
3748         RETURN(0);
3749 }
3750
3751 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3752                                         struct inode *dir,
3753                                         struct inode *parent_dir,
3754                                         const struct lu_fid *dot_fid,
3755                                         const struct lu_fid *dot_dot_fid,
3756                                         struct osd_thandle *oth)
3757 {
3758         struct ldiskfs_dentry_param *dot_ldp;
3759         struct ldiskfs_dentry_param *dot_dot_ldp;
3760         __u32 saved_nlink = dir->i_nlink;
3761         int rc;
3762
3763         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3764         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3765
3766         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3767         dot_ldp->edp_magic = 0;
3768
3769         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3770                                     dir, dot_ldp, dot_dot_ldp);
3771         /*
3772          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3773          * the subseqent ref_add() will increase the dir->i_nlink
3774          * as 3. That is incorrect for new created directory.
3775          *
3776          * It looks like hack, because we want to make the OSD API
3777          * to be order-independent for new created directory object
3778          * between dt_insert(..) and ref_add() operations.
3779          *
3780          * Here, we only restore the in-RAM dir-inode's nlink attr,
3781          * becuase if the nlink attr is not 2, then there will be
3782          * ref_add() called following the dt_insert(..), such call
3783          * will make both the in-RAM and on-disk dir-inode's nlink
3784          * attr to be set as 2. LU-7447
3785          */
3786         set_nlink(dir, saved_nlink);
3787         return rc;
3788 }
3789
3790 /**
3791  * Create an local agent inode for remote entry
3792  */
3793 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3794                                                   struct osd_device *osd,
3795                                                   struct osd_object *pobj,
3796                                                   const struct lu_fid *fid,
3797                                                   __u32 type,
3798                                                   struct thandle *th)
3799 {
3800         struct osd_thread_info *info = osd_oti_get(env);
3801         struct inode *local;
3802         struct osd_thandle *oh;
3803         int rc;
3804
3805         ENTRY;
3806
3807         LASSERT(th);
3808         oh = container_of(th, struct osd_thandle, ot_super);
3809         LASSERT(oh->ot_handle->h_transaction != NULL);
3810
3811         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3812                                      NULL);
3813         if (IS_ERR(local)) {
3814                 CERROR("%s: create local error %d\n", osd_name(osd),
3815                        (int)PTR_ERR(local));
3816                 RETURN(local);
3817         }
3818
3819         /*
3820          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3821          * correct gid on remote file, not agent here
3822          */
3823         local->i_gid = current_fsgid();
3824         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3825
3826         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3827          * That gives e2fsck something to look at and be happy, and allows
3828          * debugging if we need to determine where this symlink came from.
3829          */
3830         if (S_ISLNK(type)) {
3831                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3832                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3833                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3834
3835                 i_size_write(local, rc);
3836                 LDISKFS_I(local)->i_disksize = rc;
3837         }
3838         unlock_new_inode(local);
3839
3840         /* Agent inode should not have project ID */
3841 #ifdef  HAVE_PROJECT_QUOTA
3842         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3843             i_projid_read(pobj->oo_inode) != 0) {
3844                 rc = osd_transfer_project(local, 0);
3845                 if (rc) {
3846                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3847                                "quota enforcement enabled on the ldiskfs "
3848                                "filesystem?\n", local->i_sb->s_id, rc);
3849                         RETURN(ERR_PTR(rc));
3850                 }
3851         }
3852 #endif
3853         /* Set special LMA flag for local agent inode */
3854         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3855         if (rc != 0) {
3856                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3857                        osd_name(osd), PFID(fid), rc);
3858                 RETURN(ERR_PTR(rc));
3859         }
3860
3861         if (!S_ISDIR(type))
3862                 RETURN(local);
3863
3864         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3865                                          lu_object_fid(&pobj->oo_dt.do_lu),
3866                                          fid, oh);
3867         if (rc != 0) {
3868                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3869                         osd_name(osd), PFID(fid), rc);
3870                 RETURN(ERR_PTR(rc));
3871         }
3872
3873         RETURN(local);
3874 }
3875
3876 /**
3877  * when direntry is deleted, we have to take care of possible agent inode
3878  * referenced by that. unfortunately we can't do this at that point:
3879  * iget() within a running transaction leads to deadlock and we better do
3880  * not call that every delete declaration to save performance. so we put
3881  * a potention agent inode on a list and process that once the transaction
3882  * is over. Notice it's not any worse in terms of real orphans as regular
3883  * object destroy doesn't put inodes on the on-disk orphan list. this should
3884  * be addressed separately
3885  */
3886 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3887                                             struct osd_thandle *oh,
3888                                             __u32 ino)
3889 {
3890         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3891         struct osd_obj_orphan *oor;
3892
3893         OBD_ALLOC_PTR(oor);
3894         if (oor == NULL)
3895                 return -ENOMEM;
3896
3897         oor->oor_ino = ino;
3898         oor->oor_env = (struct lu_env *)env;
3899         spin_lock(&osd->od_osfs_lock);
3900         list_add(&oor->oor_list, &osd->od_orphan_list);
3901         spin_unlock(&osd->od_osfs_lock);
3902
3903         oh->ot_remove_agents = 1;
3904
3905         return 0;
3906
3907 }
3908
3909 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3910                                                 struct osd_device *osd)
3911 {
3912         struct osd_thread_info *info = osd_oti_get(env);
3913         struct osd_obj_orphan *oor, *tmp;
3914         struct osd_inode_id id;
3915         struct list_head list;
3916         struct inode *inode;
3917         struct lu_fid fid;
3918         handle_t *jh;
3919         __u32 ino;
3920
3921         INIT_LIST_HEAD(&list);
3922
3923         spin_lock(&osd->od_osfs_lock);
3924         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3925                 if (oor->oor_env == env) {
3926                         list_del(&oor->oor_list);
3927                         list_add(&oor->oor_list, &list);
3928                 }
3929         }
3930         spin_unlock(&osd->od_osfs_lock);
3931
3932         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3933
3934                 ino = oor->oor_ino;
3935
3936                 list_del(&oor->oor_list);
3937                 OBD_FREE_PTR(oor);
3938
3939                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3940                 inode = osd_iget_fid(info, osd, &id, &fid);
3941                 if (IS_ERR(inode))
3942                         continue;
3943
3944                 if (!osd_remote_fid(env, osd, &fid)) {
3945                         iput(inode);
3946                         continue;
3947                 }
3948
3949                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3950                 clear_nlink(inode);
3951                 mark_inode_dirty(inode);
3952                 ldiskfs_journal_stop(jh);
3953                 iput(inode);
3954         }
3955
3956         return 0;
3957 }
3958
3959 /**
3960  * OSD layer object create function for OST objects (b=11826).
3961  *
3962  * The FID is inserted into inode xattr here.
3963  *
3964  * \retval   0, on success
3965  * \retval -ve, on error
3966  */
3967 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3968                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3969                       struct dt_object_format *dof, struct thandle *th)
3970 {
3971         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3972         struct osd_object *obj = osd_dt_obj(dt);
3973         struct osd_thread_info *info = osd_oti_get(env);
3974         int result, on_ost = 0;
3975
3976         ENTRY;
3977
3978         if (dt_object_exists(dt))
3979                 RETURN(-EEXIST);
3980
3981         LINVRNT(osd_invariant(obj));
3982         LASSERT(!dt_object_remote(dt));
3983         LASSERT(osd_is_write_locked(env, obj));
3984         LASSERT(th != NULL);
3985
3986         if (unlikely(fid_is_acct(fid)))
3987                 /*
3988                  * Quota files can't be created from the kernel any more,
3989                  * 'tune2fs -O quota' will take care of creating them
3990                  */
3991                 RETURN(-EPERM);
3992
3993         result = __osd_create(info, obj, attr, hint, dof, th);
3994         if (result == 0) {
3995                 if (fid_is_idif(fid) &&
3996                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3997                         struct lu_fid *tfid = &info->oti_fid;
3998                         struct ost_id *oi   = &info->oti_ostid;
3999
4000                         fid_to_ostid(fid, oi);
4001                         ostid_to_fid(tfid, oi, 0);
4002                         on_ost = 1;
4003                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4004                                                 LMAC_FID_ON_OST, 0);
4005                 } else {
4006                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4007                                                fid, OI_CHECK_FLD);
4008                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4009                                                 on_ost ? LMAC_FID_ON_OST : 0,
4010                                                 0);
4011                 }
4012                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4013                         obj->oo_dt.do_body_ops = &osd_body_ops;
4014         }
4015
4016         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4017                 result = __osd_oi_insert(env, obj, fid, th);
4018
4019         /*
4020          * a small optimization - dt_insert() isn't usually applied
4021          * to OST objects, so we don't need to cache OI mapping for
4022          * OST objects
4023          */
4024         if (result == 0 && on_ost == 0) {
4025                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4026
4027                 result = osd_idc_find_and_init(env, osd, obj);
4028                 LASSERT(result == 0);
4029         }
4030
4031         LASSERT(ergo(result == 0,
4032                      dt_object_exists(dt) && !dt_object_remote(dt)));
4033         LINVRNT(osd_invariant(obj));
4034         RETURN(result);
4035 }
4036
4037 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4038                                struct thandle *handle)
4039 {
4040         struct osd_thandle *oh;
4041
4042         /* it's possible that object doesn't exist yet */
4043         LASSERT(handle != NULL);
4044
4045         oh = container_of0(handle, struct osd_thandle, ot_super);
4046         LASSERT(oh->ot_handle == NULL);
4047
4048         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4049                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4050
4051         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4052
4053         return 0;
4054 }
4055
4056 /*
4057  * Concurrency: @dt is write locked.
4058  */
4059 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4060                        struct thandle *th)
4061 {
4062         struct osd_object *obj = osd_dt_obj(dt);
4063         struct inode *inode = obj->oo_inode;
4064         struct osd_thandle *oh;
4065         int rc = 0;
4066
4067         if (!dt_object_exists(dt) || obj->oo_destroyed)
4068                 return -ENOENT;
4069
4070         LINVRNT(osd_invariant(obj));
4071         LASSERT(!dt_object_remote(dt));
4072         LASSERT(osd_is_write_locked(env, obj));
4073         LASSERT(th != NULL);
4074
4075         oh = container_of0(th, struct osd_thandle, ot_super);
4076         LASSERT(oh->ot_handle != NULL);
4077
4078         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4079
4080         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4081                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4082         /*
4083          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4084          * (65000) subdirectories by storing "1" in i_nlink if the link count
4085          * would otherwise overflow. Directory tranversal tools understand
4086          * that (st_nlink == 1) indicates that the filesystem dose not track
4087          * hard links count on the directory, and will not abort subdirectory
4088          * scanning early once (st_nlink - 2) subdirs have been found.
4089          *
4090          * This also has to properly handle the case of inodes with nlink == 0
4091          * in case they are being linked into the PENDING directory
4092          */
4093         spin_lock(&obj->oo_guard);
4094         if (unlikely(inode->i_nlink == 0))
4095                 /* inc_nlink from 0 may cause WARN_ON */
4096                 set_nlink(inode, 1);
4097         else {
4098                 ldiskfs_inc_count(oh->ot_handle, inode);
4099                 if (!S_ISDIR(inode->i_mode))
4100                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4101         }
4102         spin_unlock(&obj->oo_guard);
4103
4104         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4105         LINVRNT(osd_invariant(obj));
4106
4107         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4108
4109         return rc;
4110 }
4111
4112 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4113                                struct thandle *handle)
4114 {
4115         struct osd_thandle *oh;
4116
4117         if (!dt_object_exists(dt))
4118                 return -ENOENT;
4119
4120         LASSERT(!dt_object_remote(dt));
4121         LASSERT(handle != NULL);
4122
4123         oh = container_of0(handle, struct osd_thandle, ot_super);
4124         LASSERT(oh->ot_handle == NULL);
4125
4126         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4127                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4128
4129         return 0;
4130 }
4131
4132 /*
4133  * Concurrency: @dt is write locked.
4134  */
4135 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4136                        struct thandle *th)
4137 {
4138         struct osd_object *obj = osd_dt_obj(dt);
4139         struct inode *inode = obj->oo_inode;
4140         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4141         struct osd_thandle *oh;
4142
4143         if (!dt_object_exists(dt))
4144                 return -ENOENT;
4145
4146         LINVRNT(osd_invariant(obj));
4147         LASSERT(!dt_object_remote(dt));
4148         LASSERT(osd_is_write_locked(env, obj));
4149         LASSERT(th != NULL);
4150
4151         oh = container_of0(th, struct osd_thandle, ot_super);
4152         LASSERT(oh->ot_handle != NULL);
4153
4154         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4155
4156         spin_lock(&obj->oo_guard);
4157         /*
4158          * That can be result of upgrade from old Lustre version and
4159          * applied only to local files.  Just skip this ref_del call.
4160          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4161          */
4162         if (inode->i_nlink == 0) {
4163                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4164                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4165                              ", maybe an upgraded file? (LU-3915)\n",
4166                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4167                 spin_unlock(&obj->oo_guard);
4168                 return 0;
4169         }
4170
4171         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4172                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4173
4174         ldiskfs_dec_count(oh->ot_handle, inode);
4175         spin_unlock(&obj->oo_guard);
4176
4177         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4178         LINVRNT(osd_invariant(obj));
4179
4180         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4181
4182         return 0;
4183 }
4184
4185 /*
4186  * Concurrency: @dt is read locked.
4187  */
4188 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4189                          struct lu_buf *buf, const char *name)
4190 {
4191         struct osd_object      *obj    = osd_dt_obj(dt);
4192         struct inode           *inode  = obj->oo_inode;
4193         struct osd_thread_info *info   = osd_oti_get(env);
4194         struct dentry          *dentry = &info->oti_obj_dentry;
4195         bool cache_xattr = false;
4196         int rc;
4197
4198         LASSERT(buf);
4199
4200         /* version get is not real XATTR but uses xattr API */
4201         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4202                 dt_obj_version_t *ver = buf->lb_buf;
4203
4204                 /*
4205                  * for version we are just using xattr API but change inode
4206                  * field instead
4207                  */
4208                 if (buf->lb_len == 0)
4209                         return sizeof(dt_obj_version_t);
4210
4211                 if (buf->lb_len < sizeof(dt_obj_version_t))
4212                         return -ERANGE;
4213
4214                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4215                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4216
4217                 *ver = LDISKFS_I(inode)->i_fs_version;
4218
4219                 return sizeof(dt_obj_version_t);
4220         }
4221
4222         if (!dt_object_exists(dt))
4223                 return -ENOENT;
4224
4225         LASSERT(!dt_object_remote(dt));
4226         LASSERT(inode->i_op != NULL);
4227 #ifdef HAVE_IOP_XATTR
4228         LASSERT(inode->i_op->getxattr != NULL);
4229 #endif
4230
4231         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4232             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4233                 cache_xattr = true;
4234
4235         if (cache_xattr) {
4236                 rc = osd_oxc_get(obj, name, buf);
4237                 if (rc != -ENOENT)
4238                         return rc;
4239         }
4240
4241         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4242                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4243                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4244                 struct filter_fid *ff;
4245                 struct ost_layout *ol;
4246
4247                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4248                 if (rc)
4249                         return rc;
4250
4251                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4252
4253                 rc = sizeof(*ff);
4254                 if (buf->lb_len == 0 || !buf->lb_buf)
4255                         return rc;
4256
4257                 if (buf->lb_len < rc)
4258                         return -ERANGE;
4259
4260                 ff = buf->lb_buf;
4261                 ol = &ff->ff_layout;
4262                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4263                                                   PFID_STRIPE_IDX_BITS);
4264                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4265                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4266                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4267                 if (lma->lma_compat & LMAC_COMP_INFO) {
4268                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4269                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4270                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4271                 } else {
4272                         ol->ol_comp_start = 0;
4273                         ol->ol_comp_end = 0;
4274                         ol->ol_comp_id = 0;
4275                 }
4276         } else {
4277                 rc = __osd_xattr_get(inode, dentry, name,
4278                                      buf->lb_buf, buf->lb_len);
4279         }
4280
4281         if (cache_xattr) {
4282                 if (rc == -ENOENT || rc == -ENODATA)
4283                         osd_oxc_add(obj, name, NULL, 0);
4284                 else if (rc > 0 && buf->lb_buf != NULL)
4285                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4286         }
4287
4288         return rc;
4289 }
4290
4291 static int osd_declare_xattr_set(const struct lu_env *env,
4292                                  struct dt_object *dt,
4293                                  const struct lu_buf *buf, const char *name,
4294                                  int fl, struct thandle *handle)
4295 {
4296         struct osd_thandle *oh;
4297         int credits = 0;
4298         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4299
4300         LASSERT(handle != NULL);
4301
4302         oh = container_of0(handle, struct osd_thandle, ot_super);
4303         LASSERT(oh->ot_handle == NULL);
4304
4305         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4306                 /*
4307                  * For non-upgrading case, the LMA is set first and
4308                  * usually fit inode. But for upgrade case, the LMA
4309                  * may be in another separated EA block.
4310                  */
4311                 if (dt_object_exists(dt)) {
4312                         if (fl == LU_XATTR_REPLACE)
4313                                 credits = 1;
4314                         else
4315                                 goto upgrade;
4316                 }
4317         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4318                 credits = 1;
4319         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4320                 /* We may need to delete the old PFID EA. */
4321                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4322                 if (fl == LU_XATTR_REPLACE)
4323                         credits += 1;
4324                 else
4325                         goto upgrade;
4326         } else {
4327                 /*
4328                  * If some name entry resides on remote MDT, then will create
4329                  * agent entry under remote parent. On the other hand, if the
4330                  * remote entry will be removed, then related agent entry may
4331                  * need to be removed from the remote parent. So there may be
4332                  * kinds of cases, let's declare enough credits. The credits
4333                  * for create agent entry is enough for remove case.
4334                  */
4335                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4336                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4337                         if (dt_object_exists(dt))
4338                                 credits += 1; /* For updating LMA */
4339                 }
4340
4341 upgrade:
4342                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4343
4344                 if (buf != NULL) {
4345                         ssize_t buflen;
4346
4347                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4348                                 /*
4349                                  * learn xattr size from osd_xattr_get if
4350                                  * attribute has not been read yet
4351                                  */
4352                                 buflen = __osd_xattr_get(
4353                                     osd_dt_obj(dt)->oo_inode,
4354                                     &osd_oti_get(env)->oti_obj_dentry,
4355                                     name, NULL, 0);
4356                                 if (buflen < 0)
4357                                         buflen = 0;
4358                         } else {
4359                                 buflen = buf->lb_len;
4360                         }
4361
4362                         if (buflen > sb->s_blocksize) {
4363                                 credits += osd_calc_bkmap_credits(
4364                                     sb, NULL, 0, -1,
4365                                     (buflen + sb->s_blocksize - 1) >>
4366                                     sb->s_blocksize_bits);
4367                         }
4368                 }
4369                 /*
4370                  * xattr set may involve inode quota change, reserve credits for
4371                  * dquot_initialize()
4372                  */
4373                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4374         }
4375
4376         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4377
4378         return 0;
4379 }
4380
4381 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4382                               const struct lu_buf *buf, int fl,
4383                               struct thandle *handle)
4384 {
4385         struct osd_thread_info *info = osd_oti_get(env);
4386         struct dentry *dentry = &info->oti_obj_dentry;
4387         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4388         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4389         struct inode *inode = obj->oo_inode;
4390         struct filter_fid *ff = buf->lb_buf;
4391         struct ost_layout *ol = &ff->ff_layout;
4392         int flags = XATTR_REPLACE;
4393         int rc;
4394
4395         ENTRY;
4396
4397         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4398                 RETURN(-EINVAL);
4399
4400         rc = osd_get_lma(info, inode, dentry, loa);
4401         if (rc == -ENODATA) {
4402                 /* Usually for upgarding from old device */
4403                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4404                                 LMAC_FID_ON_OST, 0);
4405                 flags = XATTR_CREATE;
4406         } else if (rc) {
4407                 RETURN(rc);
4408         }
4409
4410         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4411                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4412                         RETURN(-EEXIST);
4413
4414                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4415                         /* Separate PFID EA from LMA */
4416                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4417                         lustre_lma_swab(lma);
4418                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4419                                              sizeof(*lma), XATTR_REPLACE);
4420                         if (!rc) {
4421                                 obj->oo_pfid_in_lma = 0;
4422                                 rc = LU_XATTR_CREATE;
4423                         }
4424
4425                         RETURN(rc);
4426                 }
4427         } else {
4428                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4429                         RETURN(fl);
4430
4431                 /*
4432                  * Old client does not send stripe information,
4433                  * then store the PFID EA on disk separatedly.
4434                  */
4435                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4436                              ol->ol_stripe_size == 0))
4437                         RETURN(fl);
4438
4439                 /* Remove old PFID EA entry firstly. */
4440                 ll_vfs_dq_init(inode);
4441                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4442                 if (rc == -ENODATA) {
4443                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4444                                 RETURN(rc);
4445                 } else if (rc) {
4446                         RETURN(rc);
4447                 }
4448         }
4449
4450         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4451         if (likely(ol->ol_stripe_size != 0)) {
4452                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4453                                              PFID_STRIPE_IDX_BITS;
4454                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4455                 lma->lma_compat |= LMAC_STRIPE_INFO;
4456                 if (ol->ol_comp_id != 0) {
4457                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4458                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4459                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4460                         lma->lma_compat |= LMAC_COMP_INFO;
4461                 }
4462         }
4463
4464         lustre_loa_swab(loa, false);
4465
4466         /* Store the PFID EA inside LMA. */
4467         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4468                              flags);
4469         if (!rc)
4470                 obj->oo_pfid_in_lma = 1;
4471
4472         RETURN(rc);
4473 }
4474
4475 /*
4476  * In DNE environment, the object (in spite of regular file or directory)
4477  * and its name entry may reside on different MDTs. Under such case, we will
4478  * create an agent entry on the MDT where the object resides. The agent entry
4479  * references the object locally, that makes the object to be visible to the
4480  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4481  * such as 'tar' can handle the object properly.
4482  *
4483  * We handle the agent entry during set linkEA that is the common interface
4484  * for both regular file and directroy, can handle kinds of cases, such as
4485  * create/link/unlink/rename, and so on.
4486  *
4487  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4488  *
4489  * XXX: There are two known issues:
4490  * 1. For one object, we will create at most one agent entry even if there
4491  *    may be more than one cross-MDTs hard links on the object. So the local
4492  *    e2fsck may claim that the object's nlink is larger than the name entries
4493  *    that reference such inode. And in further, the e2fsck will fix the nlink
4494  *    attribute to match the local references. Then it will cause the object's
4495  *    nlink attribute to be inconsistent with the global references. it is bad
4496  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4497  *    other hand, the global namespace LFSCK can repair the object's attribute
4498  *    according to the linkEA.
4499  * 2. There may be too many hard links on the object as to its linkEA overflow,
4500  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4501  *    case happened, then at this point, we do not know whether there are some
4502  *    cross-MDTs reference. But there are local references, it guarantees that
4503  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4504  */
4505 static int osd_xattr_handle_linkea(const struct lu_env *env,
4506                                    struct osd_device *osd,
4507                                    struct osd_object *obj,
4508                                    const struct lu_buf *buf,
4509                                    struct thandle *handle)
4510 {
4511         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4512         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4513         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4514         struct lu_name tmpname;
4515         struct osd_thandle *oh;
4516         int rc;
4517         bool remote = false;
4518
4519         ENTRY;
4520
4521         oh = container_of0(handle, struct osd_thandle, ot_super);
4522         LASSERT(oh->ot_handle != NULL);
4523
4524         rc = linkea_init_with_rec(&ldata);
4525         if (!rc) {
4526                 linkea_first_entry(&ldata);
4527                 while (ldata.ld_lee != NULL && !remote) {
4528                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4529                                             &tmpname, tfid);
4530                         if (osd_remote_fid(env, osd, tfid) > 0)
4531                                 remote = true;
4532                         else
4533                                 linkea_next_entry(&ldata);
4534                 }
4535         } else if (rc == -ENODATA) {
4536                 rc = 0;
4537         } else {
4538                 RETURN(rc);
4539         }
4540
4541         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4542                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4543                 if (rc)
4544                         CERROR("%s: failed to remove agent entry for "DFID
4545                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4546         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4547                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4548                 if (rc)
4549                         CERROR("%s: failed to create agent entry for "DFID
4550                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4551         }
4552
4553         RETURN(rc);
4554 }
4555
4556 /*
4557  * Concurrency: @dt is write locked.
4558  */
4559 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4560                          const struct lu_buf *buf, const char *name, int fl,
4561                          struct thandle *handle)
4562 {
4563         struct osd_object *obj = osd_dt_obj(dt);
4564         struct osd_device *osd = osd_obj2dev(obj);
4565         struct inode *inode = obj->oo_inode;
4566         struct osd_thread_info *info = osd_oti_get(env);
4567         int fs_flags = 0;
4568         int len;
4569         int rc;
4570
4571         ENTRY;
4572
4573         LASSERT(handle);
4574         LASSERT(buf);
4575
4576         /* version set is not real XATTR */
4577         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4578                 dt_obj_version_t *version = buf->lb_buf;
4579
4580                 /*
4581                  * for version we are just using xattr API but change inode
4582                  * field instead
4583                  */
4584                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4585
4586                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4587                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4588
4589                 LDISKFS_I(inode)->i_fs_version = *version;
4590                 /*
4591                  * Version is set after all inode operations are finished,
4592                  * so we should mark it dirty here
4593                  */
4594                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4595
4596                 RETURN(0);
4597         }
4598
4599         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4600                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4601
4602         len = buf->lb_len;
4603         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4604
4605         /*
4606          * For the OST device with 256 bytes inode size by default,
4607          * the PFID EA will be stored together with LMA EA to avoid
4608          * performance trouble. Otherwise the PFID EA can be stored
4609          * independently. LU-8998
4610          */
4611         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4612             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4613                 LASSERT(buf->lb_buf);
4614
4615                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4616                 if (fl <= 0)
4617                         RETURN(fl);
4618         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4619                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4620                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4621
4622                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4623                 if (rc)
4624                         RETURN(rc);
4625
4626                 lma->lma_incompat |= LMAI_STRIPED;
4627                 lustre_lma_swab(lma);
4628                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4629                                      sizeof(*lma), XATTR_REPLACE);
4630                 if (rc != 0)
4631                         RETURN(rc);
4632         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4633                 LASSERT(!osd->od_is_ost);
4634
4635                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4636                 if (rc)
4637                         RETURN(rc);
4638         }
4639
4640         if (fl & LU_XATTR_REPLACE)
4641                 fs_flags |= XATTR_REPLACE;
4642
4643         if (fl & LU_XATTR_CREATE)
4644                 fs_flags |= XATTR_CREATE;
4645
4646         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4647         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4648
4649         if (rc == 0 &&
4650             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4651              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4652                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4653
4654         return rc;
4655 }
4656
4657 /*
4658  * Concurrency: @dt is read locked.
4659  */
4660 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4661                           const struct lu_buf *buf)
4662 {
4663         struct osd_object *obj = osd_dt_obj(dt);
4664         struct inode *inode = obj->oo_inode;
4665         struct osd_thread_info *info = osd_oti_get(env);
4666         struct dentry *dentry = &info->oti_obj_dentry;
4667
4668         if (!dt_object_exists(dt))
4669                 return -ENOENT;
4670
4671         LASSERT(!dt_object_remote(dt));
4672         LASSERT(inode->i_op != NULL);
4673         LASSERT(inode->i_op->listxattr != NULL);
4674
4675         dentry->d_inode = inode;
4676         dentry->d_sb = inode->i_sb;
4677         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4678 }
4679
4680 static int osd_declare_xattr_del(const struct lu_env *env,
4681                                  struct dt_object *dt, const char *name,
4682                                  struct thandle *handle)
4683 {
4684         struct osd_thandle *oh;
4685         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4686
4687         LASSERT(!dt_object_remote(dt));
4688         LASSERT(handle != NULL);
4689
4690         oh = container_of0(handle, struct osd_thandle, ot_super);
4691         LASSERT(oh->ot_handle == NULL);
4692
4693         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4694                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4695         /*
4696          * xattr del may involve inode quota change, reserve credits for
4697          * dquot_initialize()
4698          */
4699         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4700
4701         return 0;
4702 }
4703
4704 /*
4705  * Concurrency: @dt is write locked.
4706  */
4707 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4708                          const char *name, struct thandle *handle)
4709 {
4710         struct osd_object *obj = osd_dt_obj(dt);
4711         struct inode *inode = obj->oo_inode;
4712         struct osd_thread_info *info = osd_oti_get(env);
4713         struct dentry *dentry = &info->oti_obj_dentry;
4714         int rc;
4715
4716         if (!dt_object_exists(dt))
4717                 return -ENOENT;
4718
4719         LASSERT(!dt_object_remote(dt));
4720         LASSERT(inode->i_op != NULL);
4721         LASSERT(handle != NULL);
4722 #ifdef HAVE_IOP_XATTR
4723         LASSERT(inode->i_op->removexattr != NULL);
4724 #endif
4725
4726         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4727
4728         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4729                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4730
4731                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4732                                  &info->oti_ost_attrs);
4733                 if (!rc) {
4734                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4735
4736                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4737                         lustre_lma_swab(lma);
4738                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4739                                              sizeof(*lma), XATTR_REPLACE);
4740                         if (!rc)
4741                                 obj->oo_pfid_in_lma = 0;
4742                 }
4743         } else {
4744                 ll_vfs_dq_init(inode);
4745                 dentry->d_inode = inode;
4746                 dentry->d_sb = inode->i_sb;
4747                 rc = osd_removexattr(dentry, inode, name);
4748         }
4749
4750         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4751
4752         if (rc == 0 &&
4753             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4754              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4755                 osd_oxc_del(obj, name);
4756
4757         return rc;
4758 }
4759
4760 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4761                            __u64 start, __u64 end)
4762 {
4763         struct osd_object *obj = osd_dt_obj(dt);
4764         struct inode *inode = obj->oo_inode;
4765         struct osd_thread_info *info = osd_oti_get(env);
4766         struct dentry *dentry = &info->oti_obj_dentry;
4767         struct file *file = &info->oti_file;
4768         int rc;
4769
4770         ENTRY;
4771
4772         dentry->d_inode = inode;
4773         dentry->d_sb = inode->i_sb;
4774         file->f_path.dentry = dentry;
4775         file->f_mapping = inode->i_mapping;
4776         file->f_op = inode->i_fop;
4777         set_file_inode(file, inode);
4778
4779         rc = ll_vfs_fsync_range(file, start, end, 0);
4780
4781         RETURN(rc);
4782 }
4783
4784 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4785 {
4786         return 0;
4787 }
4788
4789 /*
4790  * Index operations.
4791  */
4792
4793 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4794                                const struct dt_index_features *feat)
4795 {
4796         struct iam_descr *descr;
4797
4798         if (osd_object_is_root(o))
4799                 return feat == &dt_directory_features;
4800
4801         LASSERT(o->oo_dir != NULL);
4802
4803         descr = o->oo_dir->od_container.ic_descr;
4804         if (feat == &dt_directory_features) {
4805                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4806                         return 1;
4807                 else
4808                         return 0;
4809         } else {
4810                 return feat->dif_keysize_min <= descr->id_key_size &&
4811                        descr->id_key_size <= feat->dif_keysize_max &&
4812                        feat->dif_recsize_min <= descr->id_rec_size &&
4813                        descr->id_rec_size <= feat->dif_recsize_max &&
4814                        !(feat->dif_flags & (DT_IND_VARKEY |
4815                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4816                        ergo(feat->dif_flags & DT_IND_UPDATE,
4817                             1 /* XXX check that object (and fs) is writable */);
4818         }
4819 }
4820
4821 static int osd_iam_container_init(const struct lu_env *env,
4822                                   struct osd_object *obj,
4823                                   struct osd_directory *dir)
4824 {
4825         struct iam_container *bag = &dir->od_container;
4826         int result;
4827
4828         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4829         if (result != 0)
4830                 return result;
4831
4832         result = iam_container_setup(bag);
4833         if (result == 0)
4834                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4835         else
4836                 iam_container_fini(bag);
4837
4838         return result;
4839 }
4840
4841
4842 /*
4843  * Concurrency: no external locking is necessary.
4844  */
4845 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4846                          const struct dt_index_features *feat)
4847 {
4848         int result;
4849         int skip_iam = 0;
4850         struct osd_object *obj = osd_dt_obj(dt);
4851
4852         LINVRNT(osd_invariant(obj));
4853
4854         if (osd_object_is_root(obj)) {
4855                 dt->do_index_ops = &osd_index_ea_ops;
4856                 result = 0;
4857         } else if (feat == &dt_directory_features) {
4858                 dt->do_index_ops = &osd_index_ea_ops;
4859                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4860                         result = 0;
4861                 else
4862                         result = -ENOTDIR;
4863                 skip_iam = 1;
4864         } else if (unlikely(feat == &dt_otable_features)) {
4865                 dt->do_index_ops = &osd_otable_ops;
4866                 return 0;
4867         } else if (unlikely(feat == &dt_acct_features)) {
4868                 dt->do_index_ops = &osd_acct_index_ops;
4869                 result = 0;
4870                 skip_iam = 1;
4871         } else if (!osd_has_index(obj)) {
4872                 struct osd_directory *dir;
4873                 struct osd_device *osd = osd_obj2dev(obj);
4874                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4875
4876                 OBD_ALLOC_PTR(dir);
4877                 if (dir) {
4878
4879                         spin_lock(&obj->oo_guard);
4880                         if (obj->oo_dir == NULL)
4881                                 obj->oo_dir = dir;
4882                         else
4883                                 /*
4884                                  * Concurrent thread allocated container data.
4885                                  */
4886                                 OBD_FREE_PTR(dir);
4887                         spin_unlock(&obj->oo_guard);
4888                         /*
4889                          * Now, that we have container data, serialize its
4890                          * initialization.
4891                          */
4892                         down_write(&obj->oo_ext_idx_sem);
4893                         /*
4894                          * recheck under lock.
4895                          */
4896
4897                         if (osd_has_index(obj)) {
4898                                 result = 0;
4899                                 goto unlock;
4900                         }
4901
4902                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4903                         if (result || feat == &dt_lfsck_namespace_features ||
4904                             feat == &dt_lfsck_layout_orphan_features ||
4905                             feat == &dt_lfsck_layout_dangling_features)
4906                                 goto unlock;
4907
4908                         result = osd_index_register(osd, fid,
4909                                                     feat->dif_keysize_max,
4910                                                     feat->dif_recsize_max);
4911                         if (result < 0)
4912                                 CWARN("%s: failed to register index "
4913                                       DFID": rc = %d\n",
4914                                       osd_name(osd), PFID(fid), result);
4915                         else if (result > 0)
4916                                 result = 0;
4917                         else
4918                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4919                                        " (%d/%d) registered\n",
4920                                        osd_name(osd), PFID(fid),
4921                                        (int)feat->dif_keysize_max,
4922                                        (int)feat->dif_recsize_max);
4923
4924 unlock:
4925                         up_write(&obj->oo_ext_idx_sem);
4926                 } else {
4927                         result = -ENOMEM;
4928                 }
4929         } else {
4930                 result = 0;
4931         }
4932
4933         if (result == 0 && skip_iam == 0) {
4934                 if (!osd_iam_index_probe(env, obj, feat))
4935                         result = -ENOTDIR;
4936         }
4937         LINVRNT(osd_invariant(obj));
4938
4939         return result;
4940 }
4941
4942 static int osd_otable_it_attr_get(const struct lu_env *env,
4943                                  struct dt_object *dt,
4944                                  struct lu_attr *attr)
4945 {
4946         attr->la_valid = 0;
4947         return 0;
4948 }
4949
4950 static const struct dt_object_operations osd_obj_ops = {
4951         .do_read_lock           = osd_read_lock,
4952         .do_write_lock          = osd_write_lock,
4953         .do_read_unlock         = osd_read_unlock,
4954         .do_write_unlock        = osd_write_unlock,
4955         .do_write_locked        = osd_write_locked,
4956         .do_attr_get            = osd_attr_get,
4957         .do_declare_attr_set    = osd_declare_attr_set,
4958         .do_attr_set            = osd_attr_set,
4959         .do_ah_init             = osd_ah_init,
4960         .do_declare_create      = osd_declare_create,
4961         .do_create              = osd_create,
4962         .do_declare_destroy     = osd_declare_destroy,
4963         .do_destroy             = osd_destroy,
4964         .do_index_try           = osd_index_try,
4965         .do_declare_ref_add     = osd_declare_ref_add,
4966         .do_ref_add             = osd_ref_add,
4967         .do_declare_ref_del     = osd_declare_ref_del,
4968         .do_ref_del             = osd_ref_del,
4969         .do_xattr_get           = osd_xattr_get,
4970         .do_declare_xattr_set   = osd_declare_xattr_set,
4971         .do_xattr_set           = osd_xattr_set,
4972         .do_declare_xattr_del   = osd_declare_xattr_del,
4973         .do_xattr_del           = osd_xattr_del,
4974         .do_xattr_list          = osd_xattr_list,
4975         .do_object_sync         = osd_object_sync,
4976         .do_invalidate          = osd_invalidate,
4977 };
4978
4979 static const struct dt_object_operations osd_obj_otable_it_ops = {
4980         .do_attr_get    = osd_otable_it_attr_get,
4981         .do_index_try   = osd_index_try,
4982 };
4983
4984 static int osd_index_declare_iam_delete(const struct lu_env *env,
4985                                         struct dt_object *dt,
4986                                         const struct dt_key *key,
4987                                         struct thandle *handle)
4988 {
4989         struct osd_thandle *oh;
4990
4991         oh = container_of0(handle, struct osd_thandle, ot_super);
4992         LASSERT(oh->ot_handle == NULL);
4993
4994         /* Recycle  may cause additional three blocks to be changed. */
4995         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4996                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4997
4998         return 0;
4999 }
5000
5001 /**
5002  *      delete a (key, value) pair from index \a dt specified by \a key
5003  *
5004  *      \param  dt      osd index object
5005  *      \param  key     key for index
5006  *      \param  rec     record reference
5007  *      \param  handle  transaction handler
5008  *
5009  *      \retval  0  success
5010  *      \retval -ve   failure
5011  */
5012 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5013                                 const struct dt_key *key,
5014                                 struct thandle *handle)
5015 {
5016         struct osd_thread_info *oti = osd_oti_get(env);
5017         struct osd_object *obj = osd_dt_obj(dt);
5018         struct osd_thandle *oh;
5019         struct iam_path_descr *ipd;
5020         struct iam_container *bag = &obj->oo_dir->od_container;
5021         int rc;
5022
5023         ENTRY;
5024
5025         if (!dt_object_exists(dt))
5026                 RETURN(-ENOENT);
5027
5028         LINVRNT(osd_invariant(obj));
5029         LASSERT(!dt_object_remote(dt));
5030         LASSERT(bag->ic_object == obj->oo_inode);
5031         LASSERT(handle != NULL);
5032
5033         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5034
5035         ipd = osd_idx_ipd_get(env, bag);
5036         if (unlikely(ipd == NULL))
5037                 RETURN(-ENOMEM);
5038
5039         oh = container_of0(handle, struct osd_thandle, ot_super);
5040         LASSERT(oh->ot_handle != NULL);
5041         LASSERT(oh->ot_handle->h_transaction != NULL);
5042
5043         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5044                 /* swab quota uid/gid provided by caller */
5045                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5046                 key = (const struct dt_key *)&oti->oti_quota_id;
5047         }
5048
5049         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5050         osd_ipd_put(env, bag, ipd);
5051         LINVRNT(osd_invariant(obj));
5052         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5053         RETURN(rc);
5054 }
5055
5056 static int osd_index_declare_ea_delete(const struct lu_env *env,
5057                                        struct dt_object *dt,
5058                                        const struct dt_key *key,
5059                                        struct thandle *handle)
5060 {
5061         struct osd_thandle *oh;
5062         struct inode *inode;
5063         int rc, credits;
5064
5065         ENTRY;
5066
5067         LASSERT(!dt_object_remote(dt));
5068         LASSERT(handle != NULL);
5069
5070         oh = container_of0(handle, struct osd_thandle, ot_super);
5071         LASSERT(oh->ot_handle == NULL);
5072
5073         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5074         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5075
5076         inode = osd_dt_obj(dt)->oo_inode;
5077         if (inode == NULL)
5078                 RETURN(-ENOENT);
5079
5080         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5081                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5082                                    NULL, OSD_QID_BLK);
5083         RETURN(rc);
5084 }
5085
5086 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5087                                           struct dt_rec *fid)
5088 {
5089         struct osd_fid_pack *rec;
5090         int rc = -ENODATA;
5091
5092         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5093                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5094                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5095                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5096                         rc = -EINVAL;
5097         }
5098         return rc;
5099 }
5100
5101 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5102                           const struct lu_fid *fid)
5103 {
5104         struct seq_server_site *ss = osd_seq_site(osd);
5105
5106         ENTRY;
5107
5108         /* FID seqs not in FLDB, must be local seq */
5109         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5110                 RETURN(0);
5111
5112         /*
5113          * If FLD is not being initialized yet, it only happens during the
5114          * initialization, likely during mgs initialization, and we assume
5115          * this is local FID.
5116          */
5117         if (ss == NULL || ss->ss_server_fld == NULL)
5118                 RETURN(0);
5119
5120         /* Only check the local FLDB here */
5121         if (osd_seq_exists(env, osd, fid_seq(fid)))
5122                 RETURN(0);
5123
5124         RETURN(1);
5125 }
5126
5127 static void osd_take_care_of_agent(const struct lu_env *env,
5128                                    struct osd_device *osd,
5129                                    struct osd_thandle *oh,
5130                                    struct ldiskfs_dir_entry_2 *de)
5131 {
5132         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5133         struct osd_idmap_cache *idc;
5134         int rc, schedule = 0;
5135
5136         LASSERT(de != NULL);
5137
5138         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5139         if (likely(rc == 0)) {
5140                 idc = osd_idc_find_or_init(env, osd, fid);
5141                 if (IS_ERR(idc) || idc->oic_remote)
5142                         schedule = 1;
5143         } else if (rc == -ENODATA) {
5144                 /*
5145                  * can't get FID, postpone to the end of the
5146                  * transaction when iget() is safe
5147                  */
5148                 schedule = 1;
5149         } else {
5150                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5151         }
5152         if (schedule)
5153                 osd_schedule_agent_inode_removal(env, oh,
5154                                                  le32_to_cpu(de->inode));
5155 }
5156
5157 /**
5158  * Index delete function for interoperability mode (b11826).
5159  * It will remove the directory entry added by osd_index_ea_insert().
5160  * This entry is needed to maintain name->fid mapping.
5161  *
5162  * \param key,  key i.e. file entry to be deleted
5163  *
5164  * \retval   0, on success
5165  * \retval -ve, on error
5166  */
5167 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5168                                const struct dt_key *key, struct thandle *handle)
5169 {
5170         struct osd_object *obj = osd_dt_obj(dt);
5171         struct inode *dir = obj->oo_inode;
5172         struct dentry *dentry;
5173         struct osd_thandle *oh;
5174         struct ldiskfs_dir_entry_2 *de = NULL;
5175         struct buffer_head *bh;
5176         struct htree_lock *hlock = NULL;
5177         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5178         int rc;
5179
5180         ENTRY;
5181
5182         if (!dt_object_exists(dt))
5183                 RETURN(-ENOENT);
5184
5185         LINVRNT(osd_invariant(obj));
5186         LASSERT(!dt_object_remote(dt));
5187         LASSERT(handle != NULL);
5188
5189         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5190
5191         oh = container_of(handle, struct osd_thandle, ot_super);
5192         LASSERT(oh->ot_handle != NULL);
5193         LASSERT(oh->ot_handle->h_transaction != NULL);
5194
5195         ll_vfs_dq_init(dir);
5196         dentry = osd_child_dentry_get(env, obj,
5197                                       (char *)key, strlen((char *)key));
5198
5199         if (obj->oo_hl_head != NULL) {
5200                 hlock = osd_oti_get(env)->oti_hlock;
5201                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5202                                    dir, LDISKFS_HLOCK_DEL);
5203         } else {
5204                 down_write(&obj->oo_ext_idx_sem);
5205         }
5206
5207         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5208         if (!IS_ERR(bh)) {
5209                 /*
5210                  * If this is not the ".." entry, it might be a remote DNE
5211                  * entry and  we need to check if the FID is for a remote
5212                  * MDT.  If the FID is  not in the directory entry (e.g.
5213                  * upgraded 1.8 filesystem without dirdata enabled) then
5214                  * we need to get the FID from the LMA. For a remote directory
5215                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5216                  * case.
5217                  *
5218                  * Delete the entry before the agent inode in order to
5219                  * simplify error handling.  At worst an error after deleting
5220                  * the entry first might leak the agent inode afterward. The
5221                  * reverse would need filesystem abort in case of error deleting
5222                  * the entry after the agent had been removed, or leave a
5223                  * dangling entry pointing at a random inode.
5224                  */
5225                 if (strcmp((char *)key, dotdot) != 0)
5226                         osd_take_care_of_agent(env, osd, oh, de);
5227                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5228                 brelse(bh);
5229         } else {
5230                 rc = PTR_ERR(bh);
5231         }
5232         if (hlock != NULL)
5233                 ldiskfs_htree_unlock(hlock);
5234         else
5235                 up_write(&obj->oo_ext_idx_sem);
5236
5237         GOTO(out, rc);
5238 out:
5239         LASSERT(osd_invariant(obj));
5240         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5241         RETURN(rc);
5242 }
5243
5244 /**
5245  *      Lookup index for \a key and copy record to \a rec.
5246  *
5247  *      \param  dt      osd index object
5248  *      \param  key     key for index
5249  *      \param  rec     record reference
5250  *
5251  *      \retval  +ve  success : exact mach
5252  *      \retval  0    return record with key not greater than \a key
5253  *      \retval -ve   failure
5254  */
5255 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5256                                 struct dt_rec *rec, const struct dt_key *key)
5257 {
5258         struct osd_object *obj = osd_dt_obj(dt);
5259         struct iam_path_descr *ipd;
5260         struct iam_container *bag = &obj->oo_dir->od_container;
5261         struct osd_thread_info *oti = osd_oti_get(env);
5262         struct iam_iterator *it = &oti->oti_idx_it;
5263         struct iam_rec *iam_rec;
5264         int rc;
5265
5266         ENTRY;
5267
5268         if (!dt_object_exists(dt))
5269                 RETURN(-ENOENT);
5270
5271         LASSERT(osd_invariant(obj));
5272         LASSERT(!dt_object_remote(dt));
5273         LASSERT(bag->ic_object == obj->oo_inode);
5274
5275         ipd = osd_idx_ipd_get(env, bag);
5276         if (IS_ERR(ipd))
5277                 RETURN(-ENOMEM);
5278
5279         /* got ipd now we can start iterator. */
5280         iam_it_init(it, bag, 0, ipd);
5281
5282         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5283                 /* swab quota uid/gid provided by caller */
5284                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5285                 key = (const struct dt_key *)&oti->oti_quota_id;
5286         }
5287
5288         rc = iam_it_get(it, (struct iam_key *)key);
5289         if (rc >= 0) {
5290                 if (S_ISDIR(obj->oo_inode->i_mode))
5291                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5292                 else
5293                         iam_rec = (struct iam_rec *)rec;
5294
5295                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5296
5297                 if (S_ISDIR(obj->oo_inode->i_mode))
5298                         osd_fid_unpack((struct lu_fid *)rec,
5299                                        (struct osd_fid_pack *)iam_rec);
5300                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5301                         osd_quota_unpack(obj, rec);
5302         }
5303
5304         iam_it_put(it);
5305         iam_it_fini(it);
5306         osd_ipd_put(env, bag, ipd);
5307
5308         LINVRNT(osd_invariant(obj));
5309
5310         RETURN(rc);
5311 }
5312
5313 static int osd_index_declare_iam_insert(const struct lu_env *env,
5314                                         struct dt_object *dt,
5315                                         const struct dt_rec *rec,
5316                                         const struct dt_key *key,
5317                                         struct thandle *handle)
5318 {
5319         struct osd_thandle *oh;
5320
5321         LASSERT(handle != NULL);
5322
5323         oh = container_of0(handle, struct osd_thandle, ot_super);
5324         LASSERT(oh->ot_handle == NULL);
5325
5326         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5327                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5328
5329         return 0;
5330 }
5331
5332 /**
5333  *      Inserts (key, value) pair in \a dt index object.
5334  *
5335  *      \param  dt      osd index object
5336  *      \param  key     key for index
5337  *      \param  rec     record reference
5338  *      \param  th      transaction handler
5339  *
5340  *      \retval  0  success
5341  *      \retval -ve failure
5342  */
5343 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5344                                 const struct dt_rec *rec,
5345                                 const struct dt_key *key, struct thandle *th)
5346 {
5347         struct osd_object *obj = osd_dt_obj(dt);
5348         struct iam_path_descr *ipd;
5349         struct osd_thandle *oh;
5350         struct iam_container *bag;
5351         struct osd_thread_info *oti = osd_oti_get(env);
5352         struct iam_rec *iam_rec;
5353         int rc;
5354
5355         ENTRY;
5356
5357         if (!dt_object_exists(dt))
5358                 RETURN(-ENOENT);
5359
5360         LINVRNT(osd_invariant(obj));
5361         LASSERT(!dt_object_remote(dt));
5362
5363         bag = &obj->oo_dir->od_container;
5364         LASSERT(bag->ic_object == obj->oo_inode);
5365         LASSERT(th != NULL);
5366
5367         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5368
5369         ipd = osd_idx_ipd_get(env, bag);
5370         if (unlikely(ipd == NULL))
5371                 RETURN(-ENOMEM);
5372
5373         oh = container_of0(th, struct osd_thandle, ot_super);
5374         LASSERT(oh->ot_handle != NULL);
5375         LASSERT(oh->ot_handle->h_transaction != NULL);
5376         if (S_ISDIR(obj->oo_inode->i_mode)) {
5377                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5378                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5379                              &oti->oti_fid);
5380         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5381                 /* pack quota uid/gid */
5382                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5383                 key = (const struct dt_key *)&oti->oti_quota_id;
5384                 /* pack quota record */
5385                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5386                 iam_rec = (struct iam_rec *)rec;
5387         } else {
5388                 iam_rec = (struct iam_rec *)rec;
5389         }
5390
5391         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5392                         iam_rec, ipd);
5393         osd_ipd_put(env, bag, ipd);
5394         LINVRNT(osd_invariant(obj));
5395         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5396         RETURN(rc);
5397 }
5398
5399 /**
5400  * Calls ldiskfs_add_entry() to add directory entry
5401  * into the directory. This is required for
5402  * interoperability mode (b11826)
5403  *
5404  * \retval   0, on success
5405  * \retval -ve, on error
5406  */
5407 static int __osd_ea_add_rec(struct osd_thread_info *info,
5408                             struct osd_object *pobj, struct inode  *cinode,
5409                             const char *name, const struct lu_fid *fid,
5410                             struct htree_lock *hlock, struct thandle *th)
5411 {
5412         struct ldiskfs_dentry_param *ldp;
5413         struct dentry *child;
5414         struct osd_thandle *oth;
5415         int rc;
5416
5417         oth = container_of(th, struct osd_thandle, ot_super);
5418         LASSERT(oth->ot_handle != NULL);
5419         LASSERT(oth->ot_handle->h_transaction != NULL);
5420         LASSERT(pobj->oo_inode);
5421
5422         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5423         if (unlikely(pobj->oo_inode ==
5424                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5425                 ldp->edp_magic = 0;
5426         else
5427                 osd_get_ldiskfs_dirent_param(ldp, fid);
5428         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5429         child->d_fsdata = (void *)ldp;
5430         ll_vfs_dq_init(pobj->oo_inode);
5431         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5432                                    child, cinode, hlock);
5433         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5434                 struct ldiskfs_dir_entry_2      *de;
5435                 struct buffer_head              *bh;
5436                 int                              rc1;
5437
5438                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5439                                             NULL, hlock);
5440                 if (!IS_ERR(bh)) {
5441                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5442                                                                bh);
5443                         if (rc1 == 0) {
5444                                 if (S_ISDIR(cinode->i_mode))
5445                                         de->file_type = LDISKFS_DIRENT_LUFID |
5446                                                         LDISKFS_FT_REG_FILE;
5447                                 else
5448                                         de->file_type = LDISKFS_DIRENT_LUFID |
5449                                                         LDISKFS_FT_DIR;
5450                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5451                                                               NULL, bh);
5452                         }
5453                         brelse(bh);
5454                 }
5455         }
5456
5457         RETURN(rc);
5458 }
5459
5460 /**
5461  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5462  * into the directory.Also sets flags into osd object to
5463  * indicate dot and dotdot are created. This is required for
5464  * interoperability mode (b11826)
5465  *
5466  * \param dir   directory for dot and dotdot fixup.
5467  * \param obj   child object for linking
5468  *
5469  * \retval   0, on success
5470  * \retval -ve, on error
5471  */
5472 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5473                               struct osd_object *dir,
5474                               struct inode *parent_dir, const char *name,
5475                               const struct lu_fid *dot_fid,
5476                               const struct lu_fid *dot_dot_fid,
5477                               struct thandle *th)
5478 {
5479         struct inode *inode = dir->oo_inode;
5480         struct osd_thandle *oth;
5481         int result = 0;
5482
5483         oth = container_of(th, struct osd_thandle, ot_super);
5484         LASSERT(oth->ot_handle->h_transaction != NULL);
5485         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5486
5487         if (strcmp(name, dot) == 0) {
5488                 if (dir->oo_compat_dot_created) {
5489                         result = -EEXIST;
5490                 } else {
5491                         LASSERT(inode->i_ino == parent_dir->i_ino);
5492                         dir->oo_compat_dot_created = 1;
5493                         result = 0;
5494                 }
5495         } else if (strcmp(name, dotdot) == 0) {
5496                 if (!dir->oo_compat_dot_created)
5497                         return -EINVAL;
5498                 /* in case of rename, dotdot is already created */
5499                 if (dir->oo_compat_dotdot_created) {
5500                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5501                                                 dot_dot_fid, NULL, th);
5502                 }
5503
5504                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5505                         struct lu_fid tfid = *dot_dot_fid;
5506
5507                         tfid.f_oid--;
5508                         result = osd_add_dot_dotdot_internal(info,
5509                                         dir->oo_inode, parent_dir, dot_fid,
5510                                         &tfid, oth);
5511                 } else {
5512                         result = osd_add_dot_dotdot_internal(info,
5513                                         dir->oo_inode, parent_dir, dot_fid,
5514                                         dot_dot_fid, oth);
5515                 }
5516
5517                 if (result == 0)
5518                         dir->oo_compat_dotdot_created = 1;
5519         }
5520
5521         return result;
5522 }
5523
5524
5525 /**
5526  * It will call the appropriate osd_add* function and return the
5527  * value, return by respective functions.
5528  */
5529 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5530                           struct inode *cinode, const char *name,
5531                           const struct lu_fid *fid, struct thandle *th)
5532 {
5533         struct osd_thread_info *info = osd_oti_get(env);
5534         struct htree_lock *hlock;
5535         int rc;
5536
5537         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5538
5539         if (name[0] == '.' && (name[1] == '\0' ||
5540                                (name[1] == '.' && name[2] == '\0'))) {
5541                 if (hlock != NULL) {
5542                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5543                                            pobj->oo_inode, 0);
5544                 } else {
5545                         down_write(&pobj->oo_ext_idx_sem);
5546                 }
5547
5548                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5549                                         lu_object_fid(&pobj->oo_dt.do_lu),
5550                                         fid, th);
5551         } else {
5552                 if (hlock != NULL) {
5553                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5554                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5555                 } else {
5556                         down_write(&pobj->oo_ext_idx_sem);
5557                 }
5558
5559                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5560                         struct lu_fid *tfid = &info->oti_fid;
5561
5562                         *tfid = *fid;
5563                         tfid->f_ver = ~0;
5564                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5565                                               tfid, hlock, th);
5566                 } else {
5567                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5568                                               hlock, th);
5569                 }
5570         }
5571         if (hlock != NULL)
5572                 ldiskfs_htree_unlock(hlock);
5573         else
5574                 up_write(&pobj->oo_ext_idx_sem);
5575
5576         return rc;
5577 }
5578
5579 static int
5580 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5581                       struct osd_idmap_cache *oic)
5582 {
5583         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5584         struct lu_fid *fid = &oic->oic_fid;
5585         struct osd_inode_id *id = &oic->oic_lid;
5586         struct inode *inode = NULL;
5587         int once = 0;
5588         bool insert;
5589         int rc;
5590
5591         ENTRY;
5592
5593         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5594                 RETURN(0);
5595
5596         if (thread_is_running(&scrub->os_thread) &&
5597             scrub->os_pos_current > id->oii_ino)
5598                 RETURN(0);
5599
5600         if (dev->od_auto_scrub_interval == AS_NEVER ||
5601             ktime_get_real_seconds() <
5602             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5603                 RETURN(0);
5604
5605 again:
5606         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5607         if (rc == -ENOENT) {
5608                 __u32 gen = id->oii_gen;
5609
5610                 insert = true;
5611                 if (inode != NULL)
5612                         goto trigger;
5613
5614                 inode = osd_iget(oti, dev, id);
5615                 /* The inode has been removed (by race maybe). */
5616                 if (IS_ERR(inode)) {
5617                         rc = PTR_ERR(inode);
5618
5619                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5620                 }
5621
5622                 /* The OI mapping is lost. */
5623                 if (gen != OSD_OII_NOGEN)
5624                         goto trigger;
5625
5626                 /*
5627                  * The inode may has been reused by others, we do not know,
5628                  * leave it to be handled by subsequent osd_fid_lookup().
5629                  */
5630                 GOTO(out, rc = 0);
5631         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5632                 GOTO(out, rc);
5633         }
5634
5635         insert = false;
5636
5637 trigger:
5638         if (thread_is_running(&scrub->os_thread)) {
5639                 if (inode == NULL) {
5640                         inode = osd_iget(oti, dev, id);
5641                         /* The inode has been removed (by race maybe). */
5642                         if (IS_ERR(inode)) {
5643                                 rc = PTR_ERR(inode);
5644
5645                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5646                         }
5647                 }
5648
5649                 rc = osd_oii_insert(dev, oic, insert);
5650                 /*
5651                  * There is race condition between osd_oi_lookup and OI scrub.
5652                  * The OI scrub finished just after osd_oi_lookup() failure.
5653                  * Under such case, it is unnecessary to trigger OI scrub again,
5654                  * but try to call osd_oi_lookup() again.
5655                  */
5656                 if (unlikely(rc == -EAGAIN))
5657                         goto again;
5658
5659                 if (!S_ISDIR(inode->i_mode))
5660                         rc = 0;
5661                 else
5662                         rc = osd_check_lmv(oti, dev, inode, oic);
5663
5664                 GOTO(out, rc);
5665         }
5666
5667         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5668                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5669                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5670                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5671                        "%s: trigger partial OI scrub for RPC inconsistency "
5672                        "checking FID "DFID": rc = %d\n",
5673                        osd_dev2name(dev), PFID(fid), rc);
5674                 if (rc == 0 || rc == -EALREADY)
5675                         goto again;
5676         }
5677
5678         GOTO(out, rc);
5679
5680 out:
5681         if (inode)
5682                 iput(inode);
5683
5684         RETURN(rc);
5685 }
5686
5687 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5688                                struct osd_device *dev,
5689                                struct osd_idmap_cache *oic,
5690                                struct lu_fid *fid, __u32 ino)
5691 {
5692         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5693         struct inode *inode;
5694         int rc;
5695
5696         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5697         inode = osd_iget(oti, dev, &oic->oic_lid);
5698         if (IS_ERR(inode)) {
5699                 fid_zero(&oic->oic_fid);
5700                 return PTR_ERR(inode);
5701         }
5702
5703         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5704         iput(inode);
5705         if (rc != 0)
5706                 fid_zero(&oic->oic_fid);
5707         else
5708                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5709         return rc;
5710 }
5711
5712 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5713                       struct osd_inode_id *id, const struct lu_fid *fid)
5714 {
5715         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5716                id->oii_ino, id->oii_gen, info);
5717         info->oti_cache.oic_lid = *id;
5718         info->oti_cache.oic_fid = *fid;
5719         info->oti_cache.oic_dev = osd;
5720 }
5721
5722 /**
5723  * Get parent FID from the linkEA.
5724  *
5725  * For a directory which parent resides on remote MDT, to satisfy the
5726  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5727  * the other hand, to make the lookup(..) on the directory can return
5728  * the real parent FID, we append the real parent FID after its ".."
5729  * name entry in the /REMOTE_PARENT_DIR.
5730  *
5731  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5732  * backup. So after the restore, we cannot get the right parent FID from
5733  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5734  * we have stored the real parent FID in the directory object's linkEA,
5735  * we can parse the linkEA for the real parent FID.
5736  *
5737  * \param[in] env       pointer to the thread context
5738  * \param[in] obj       pointer to the object to be handled
5739  * \param[out]fid       pointer to the buffer to hold the parent FID
5740  *
5741  * \retval              0 for getting the real parent FID successfully
5742  * \retval              negative error number on failure
5743  */
5744 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5745                                     struct osd_object *obj,
5746                                     struct lu_fid *fid)
5747 {
5748         struct osd_thread_info *oti = osd_oti_get(env);
5749         struct lu_buf *buf = &oti->oti_big_buf;
5750         struct dentry *dentry = &oti->oti_obj_dentry;
5751         struct inode *inode = obj->oo_inode;
5752         struct linkea_data ldata = { NULL };
5753         int rc;
5754
5755         ENTRY;
5756
5757         fid_zero(fid);
5758         if (!S_ISDIR(inode->i_mode))
5759                 RETURN(-EIO);
5760
5761 again:
5762         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5763                              buf->lb_buf, buf->lb_len);
5764         if (rc == -ERANGE) {
5765                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5766                                      NULL, 0);
5767                 if (rc > 0) {
5768                         lu_buf_realloc(buf, rc);
5769                         if (buf->lb_buf == NULL)
5770                                 RETURN(-ENOMEM);
5771
5772                         goto again;
5773                 }
5774         }
5775
5776         if (unlikely(rc == 0))
5777                 RETURN(-ENODATA);
5778
5779         if (rc < 0)
5780                 RETURN(rc);
5781
5782         if (unlikely(buf->lb_buf == NULL)) {
5783                 lu_buf_realloc(buf, rc);
5784                 if (buf->lb_buf == NULL)
5785                         RETURN(-ENOMEM);
5786
5787                 goto again;
5788         }
5789
5790         ldata.ld_buf = buf;
5791         rc = linkea_init_with_rec(&ldata);
5792         if (!rc) {
5793                 linkea_first_entry(&ldata);
5794                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5795         }
5796
5797         RETURN(rc);
5798 }
5799
5800 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5801                                     struct inode *inode,
5802                                     const struct lu_fid *pfid,
5803                                     const char *name, const int namelen)
5804 {
5805         struct osd_thread_info *oti = osd_oti_get(env);
5806         struct lu_buf *buf = &oti->oti_big_buf;
5807         struct dentry *dentry = &oti->oti_obj_dentry;
5808         struct linkea_data ldata = { NULL };
5809         struct lu_name cname = { .ln_name = name,
5810                                  .ln_namelen = namelen };
5811         int rc;
5812
5813         ENTRY;
5814
5815 again:
5816         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5817                              buf->lb_buf, buf->lb_len);
5818         if (rc == -ERANGE)
5819                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5820
5821         if (rc < 0)
5822                 RETURN(rc);
5823
5824         if (unlikely(rc == 0))
5825                 RETURN(-ENODATA);
5826
5827         if (buf->lb_len < rc) {
5828                 lu_buf_realloc(buf, rc);
5829                 if (buf->lb_buf == NULL)
5830                         RETURN(-ENOMEM);
5831
5832                 goto again;
5833         }
5834
5835         ldata.ld_buf = buf;
5836         rc = linkea_init_with_rec(&ldata);
5837         if (!rc)
5838                 rc = linkea_links_find(&ldata, &cname, pfid);
5839
5840         RETURN(rc);
5841 }
5842
5843 /**
5844  * Calls ->lookup() to find dentry. From dentry get inode and
5845  * read inode's ea to get fid. This is required for  interoperability
5846  * mode (b11826)
5847  *
5848  * \retval   0, on success
5849  * \retval -ve, on error
5850  */
5851 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5852                              struct dt_rec *rec, const struct dt_key *key)
5853 {
5854         struct inode *dir = obj->oo_inode;
5855         struct dentry *dentry;
5856         struct ldiskfs_dir_entry_2 *de;
5857         struct buffer_head *bh;
5858         struct lu_fid *fid = (struct lu_fid *)rec;
5859         struct htree_lock *hlock = NULL;
5860         int ino;
5861         int rc;
5862
5863         ENTRY;
5864
5865         LASSERT(dir->i_op != NULL);
5866         LASSERT(dir->i_op->lookup != NULL);
5867
5868         dentry = osd_child_dentry_get(env, obj,
5869                                       (char *)key, strlen((char *)key));
5870
5871         if (obj->oo_hl_head != NULL) {
5872                 hlock = osd_oti_get(env)->oti_hlock;
5873                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5874                                    dir, LDISKFS_HLOCK_LOOKUP);
5875         } else {
5876                 down_read(&obj->oo_ext_idx_sem);
5877         }
5878
5879         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5880         if (!IS_ERR(bh)) {
5881                 struct osd_thread_info *oti = osd_oti_get(env);
5882                 struct osd_inode_id *id = &oti->oti_id;
5883                 struct osd_idmap_cache *oic = &oti->oti_cache;
5884                 struct osd_device *dev = osd_obj2dev(obj);
5885
5886                 ino = le32_to_cpu(de->inode);
5887                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5888                         brelse(bh);
5889                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5890                         GOTO(out, rc);
5891                 }
5892
5893                 rc = osd_get_fid_from_dentry(de, rec);
5894
5895                 /* done with de, release bh */
5896                 brelse(bh);
5897                 if (rc != 0) {
5898                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5899                                 const char *name = (const char *)key;
5900
5901                                 /*
5902                                  * If the parent is on remote MDT, and there
5903                                  * is no FID-in-dirent, then we have to get
5904                                  * the parent FID from the linkEA.
5905                                  */
5906                                 if (likely(strlen(name) == 2 &&
5907                                            name[0] == '.' && name[1] == '.'))
5908                                         rc = osd_get_pfid_from_linkea(env, obj,
5909                                                                       fid);
5910                         } else {
5911                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5912                         }
5913                 } else {
5914                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5915                 }
5916
5917                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5918                         fid_zero(&oic->oic_fid);
5919
5920                         GOTO(out, rc);
5921                 }
5922
5923                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5924                 rc = osd_consistency_check(oti, dev, oic);
5925                 if (rc == -ENOENT)
5926                         fid_zero(&oic->oic_fid);
5927                 else
5928                         /* Other error should not affect lookup result. */
5929                         rc = 0;
5930         } else {
5931                 rc = PTR_ERR(bh);
5932         }
5933
5934         GOTO(out, rc);
5935
5936 out:
5937         if (hlock != NULL)
5938                 ldiskfs_htree_unlock(hlock);
5939         else
5940                 up_read(&obj->oo_ext_idx_sem);
5941         return rc;
5942 }
5943
5944 static int osd_index_declare_ea_insert(const struct lu_env *env,
5945                                        struct dt_object *dt,
5946                                        const struct dt_rec *rec,
5947                                        const struct dt_key *key,
5948                                        struct thandle *handle)
5949 {
5950         struct osd_thandle *oh;
5951         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5952         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5953         const struct lu_fid *fid = rec1->rec_fid;
5954         int credits, rc = 0;
5955         struct osd_idmap_cache *idc;
5956
5957         ENTRY;
5958
5959         LASSERT(!dt_object_remote(dt));
5960         LASSERT(handle != NULL);
5961         LASSERT(fid != NULL);
5962         LASSERT(rec1->rec_type != 0);
5963
5964         oh = container_of0(handle, struct osd_thandle, ot_super);
5965         LASSERT(oh->ot_handle == NULL);
5966
5967         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5968
5969         /*
5970          * we can't call iget() while a transactions is running
5971          * (this can lead to a deadlock), but we need to know
5972          * inum and object type. so we find this information at
5973          * declaration and cache in per-thread info
5974          */
5975         idc = osd_idc_find_or_init(env, osd, fid);
5976         if (IS_ERR(idc))
5977                 RETURN(PTR_ERR(idc));
5978         if (idc->oic_remote) {
5979                 /*
5980                  * a reference to remote inode is represented by an
5981                  * agent inode which we have to create
5982                  */
5983                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5984                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5985         }
5986
5987         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5988
5989         if (osd_dt_obj(dt)->oo_inode != NULL) {
5990                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5991
5992                 /*
5993                  * We ignore block quota on meta pool (MDTs), so needn't
5994                  * calculate how many blocks will be consumed by this index
5995                  * insert
5996                  */
5997                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5998                                            i_gid_read(inode),
5999                                            i_projid_read(inode), 0,
6000                                            oh, osd_dt_obj(dt), NULL,
6001                                            OSD_QID_BLK);
6002                 if (rc)
6003                         RETURN(rc);
6004
6005 #ifdef HAVE_PROJECT_QUOTA
6006                 /*
6007                  * Reserve credits for local agent inode to transfer
6008                  * to 0, quota enforcement is ignored in this case.
6009                  */
6010                 if (idc->oic_remote &&
6011                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6012                     i_projid_read(inode) != 0)
6013                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6014                                                   0, i_projid_read(inode),
6015                                                   0, false, PRJQUOTA, true);
6016 #endif
6017         }
6018
6019         RETURN(rc);
6020 }
6021
6022 /**
6023  * Index add function for interoperability mode (b11826).
6024  * It will add the directory entry.This entry is needed to
6025  * maintain name->fid mapping.
6026  *
6027  * \param key it is key i.e. file entry to be inserted
6028  * \param rec it is value of given key i.e. fid
6029  *
6030  * \retval   0, on success
6031  * \retval -ve, on error
6032  */
6033 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6034                                const struct dt_rec *rec,
6035                                const struct dt_key *key, struct thandle *th)
6036 {
6037         struct osd_object *obj = osd_dt_obj(dt);
6038         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6039         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6040         const struct lu_fid *fid = rec1->rec_fid;
6041         const char *name = (const char *)key;
6042         struct osd_thread_info *oti = osd_oti_get(env);
6043         struct inode *child_inode = NULL;
6044         struct osd_idmap_cache *idc;
6045         int rc;
6046
6047         ENTRY;
6048
6049         if (!dt_object_exists(dt))
6050                 RETURN(-ENOENT);
6051
6052         LASSERT(osd_invariant(obj));
6053         LASSERT(!dt_object_remote(dt));
6054         LASSERT(th != NULL);
6055
6056         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6057
6058         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6059
6060         idc = osd_idc_find(env, osd, fid);
6061         if (unlikely(idc == NULL)) {
6062                 idc = osd_idc_find_or_init(env, osd, fid);
6063                 if (IS_ERR(idc)) {
6064                         /*
6065                          * this dt_insert() wasn't declared properly, so
6066                          * FID is missing in OI cache. we better do not
6067                          * lookup FID in FLDB/OI and don't risk to deadlock,
6068                          * but in some special cases (lfsck testing, etc)
6069                          * it's much simpler than fixing a caller.
6070                          *
6071                          * normally this error should be placed after the first
6072                          * find, but migrate may attach source stripes to
6073                          * target, which doesn't create stripes.
6074                          */
6075                         CERROR("%s: "DFID" wasn't declared for insert\n",
6076                                osd_name(osd), PFID(fid));
6077                         dump_stack();
6078                         RETURN(PTR_ERR(idc));
6079                 }
6080         }
6081
6082         if (idc->oic_remote) {
6083                 /* Insert remote entry */
6084                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6085                         child_inode =
6086                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6087                 } else {
6088                         child_inode = osd_create_local_agent_inode(env, osd,
6089                                         obj, fid, rec1->rec_type & S_IFMT, th);
6090                         if (IS_ERR(child_inode))
6091                                 RETURN(PTR_ERR(child_inode));
6092                 }
6093         } else {
6094                 /* Insert local entry */
6095                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6096                         /* for a reason OI cache wasn't filled properly */
6097                         CERROR("%s: OIC for "DFID" isn't filled\n",
6098                                osd_name(osd), PFID(fid));
6099                         RETURN(-EINVAL);
6100                 }
6101                 child_inode = oti->oti_inode;
6102                 if (unlikely(child_inode == NULL)) {
6103                         struct ldiskfs_inode_info *lii;
6104
6105                         OBD_ALLOC_PTR(lii);
6106                         if (lii == NULL)
6107                                 RETURN(-ENOMEM);
6108                         child_inode = oti->oti_inode = &lii->vfs_inode;
6109                 }
6110                 child_inode->i_sb = osd_sb(osd);
6111                 child_inode->i_ino = idc->oic_lid.oii_ino;
6112                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6113         }
6114
6115         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6116
6117         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6118                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6119
6120         if (child_inode && child_inode != oti->oti_inode)
6121                 iput(child_inode);
6122         LASSERT(osd_invariant(obj));
6123         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6124         RETURN(rc);
6125 }
6126
6127 /**
6128  *  Initialize osd Iterator for given osd index object.
6129  *
6130  *  \param  dt      osd index object
6131  */
6132
6133 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6134                                      struct dt_object *dt,
6135                                      __u32 unused)
6136 {
6137         struct osd_it_iam *it;
6138         struct osd_object *obj = osd_dt_obj(dt);
6139         struct lu_object *lo = &dt->do_lu;
6140         struct iam_path_descr *ipd;
6141         struct iam_container *bag = &obj->oo_dir->od_container;
6142
6143         if (!dt_object_exists(dt))
6144                 return ERR_PTR(-ENOENT);
6145
6146         OBD_ALLOC_PTR(it);
6147         if (it == NULL)
6148                 return ERR_PTR(-ENOMEM);
6149
6150         ipd = osd_it_ipd_get(env, bag);
6151         if (likely(ipd != NULL)) {
6152                 it->oi_obj = obj;
6153                 it->oi_ipd = ipd;
6154                 lu_object_get(lo);
6155                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6156                 return (struct dt_it *)it;
6157         } else {
6158                 OBD_FREE_PTR(it);
6159                 return ERR_PTR(-ENOMEM);
6160         }
6161 }
6162
6163 /**
6164  * free given Iterator.
6165  */
6166 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6167 {
6168         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6169         struct osd_object *obj = it->oi_obj;
6170
6171         iam_it_fini(&it->oi_it);
6172         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6173         osd_object_put(env, obj);
6174         OBD_FREE_PTR(it);
6175 }
6176
6177 /**
6178  *  Move Iterator to record specified by \a key
6179  *
6180  *  \param  di      osd iterator
6181  *  \param  key     key for index
6182  *
6183  *  \retval +ve  di points to record with least key not larger than key
6184  *  \retval  0   di points to exact matched key
6185  *  \retval -ve  failure
6186  */
6187
6188 static int osd_it_iam_get(const struct lu_env *env,
6189                           struct dt_it *di, const struct dt_key *key)
6190 {
6191         struct osd_thread_info *oti = osd_oti_get(env);
6192         struct osd_it_iam *it = (struct osd_it_iam *)di;
6193
6194         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6195                 /* swab quota uid/gid */
6196                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6197                 key = (struct dt_key *)&oti->oti_quota_id;
6198         }
6199
6200         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6201 }
6202
6203 /**
6204  *  Release Iterator
6205  *
6206  *  \param  di      osd iterator
6207  */
6208 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6209 {
6210         struct osd_it_iam *it = (struct osd_it_iam *)di;
6211
6212         iam_it_put(&it->oi_it);
6213 }
6214
6215 /**
6216  *  Move iterator by one record
6217  *
6218  *  \param  di      osd iterator
6219  *
6220  *  \retval +1   end of container reached
6221  *  \retval  0   success
6222  *  \retval -ve  failure
6223  */
6224
6225 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6226 {
6227         struct osd_it_iam *it = (struct osd_it_iam *)di;
6228
6229         return iam_it_next(&it->oi_it);
6230 }
6231
6232 /**
6233  * Return pointer to the key under iterator.
6234  */
6235
6236 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6237                                      const struct dt_it *di)
6238 {
6239         struct osd_thread_info *oti = osd_oti_get(env);
6240         struct osd_it_iam *it = (struct osd_it_iam *)di;
6241         struct osd_object *obj = it->oi_obj;
6242         struct dt_key *key;
6243
6244         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6245
6246         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6247                 /* swab quota uid/gid */
6248                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6249                 key = (struct dt_key *)&oti->oti_quota_id;
6250         }
6251
6252         return key;
6253 }
6254
6255 /**
6256  * Return size of key under iterator (in bytes)
6257  */
6258
6259 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6260 {
6261         struct osd_it_iam *it = (struct osd_it_iam *)di;
6262
6263         return iam_it_key_size(&it->oi_it);
6264 }
6265
6266 static inline void
6267 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6268 {
6269         /* check if file type is required */
6270         if (ent->lde_attrs & LUDA_TYPE) {
6271                 struct luda_type *lt;
6272                 int align = sizeof(*lt) - 1;
6273
6274                 len = (len + align) & ~align;
6275                 lt = (struct luda_type *)(ent->lde_name + len);
6276                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6277         }
6278
6279         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6280 }
6281
6282 /**
6283  * build lu direct from backend fs dirent.
6284  */
6285
6286 static inline void
6287 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6288                    char *name, __u16 namelen, __u16 type, __u32 attr)
6289 {
6290         ent->lde_attrs = attr | LUDA_FID;
6291         fid_cpu_to_le(&ent->lde_fid, fid);
6292
6293         ent->lde_hash = cpu_to_le64(offset);
6294         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6295
6296         strncpy(ent->lde_name, name, namelen);
6297         ent->lde_name[namelen] = '\0';
6298         ent->lde_namelen = cpu_to_le16(namelen);
6299
6300         /* append lustre attributes */
6301         osd_it_append_attrs(ent, namelen, type);
6302 }
6303
6304 /**
6305  * Return pointer to the record under iterator.
6306  */
6307 static int osd_it_iam_rec(const struct lu_env *env,
6308                           const struct dt_it *di,
6309                           struct dt_rec *dtrec, __u32 attr)
6310 {
6311         struct osd_it_iam *it = (struct osd_it_iam *)di;
6312         struct osd_thread_info *info = osd_oti_get(env);
6313
6314         ENTRY;
6315
6316         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6317                 const struct osd_fid_pack *rec;
6318                 struct lu_fid *fid = &info->oti_fid;
6319                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6320                 char *name;
6321                 int namelen;
6322                 __u64 hash;
6323                 int rc;
6324
6325                 name = (char *)iam_it_key_get(&it->oi_it);
6326                 if (IS_ERR(name))
6327                         RETURN(PTR_ERR(name));
6328
6329                 namelen = iam_it_key_size(&it->oi_it);
6330
6331                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6332                 if (IS_ERR(rec))
6333                         RETURN(PTR_ERR(rec));
6334
6335                 rc = osd_fid_unpack(fid, rec);
6336                 if (rc)
6337                         RETURN(rc);
6338
6339                 hash = iam_it_store(&it->oi_it);
6340
6341                 /* IAM does not store object type in IAM index (dir) */
6342                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6343                                    0, LUDA_FID);
6344         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6345                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6346                            (struct iam_rec *)dtrec);
6347                 osd_quota_unpack(it->oi_obj, dtrec);
6348         } else {
6349                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6350                            (struct iam_rec *)dtrec);
6351         }
6352
6353         RETURN(0);
6354 }
6355
6356 /**
6357  * Returns cookie for current Iterator position.
6358  */
6359 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6360 {
6361         struct osd_it_iam *it = (struct osd_it_iam *)di;
6362
6363         return iam_it_store(&it->oi_it);
6364 }
6365
6366 /**
6367  * Restore iterator from cookie.
6368  *
6369  * \param  di      osd iterator
6370  * \param  hash    Iterator location cookie
6371  *
6372  * \retval +ve  di points to record with least key not larger than key.
6373  * \retval  0   di points to exact matched key
6374  * \retval -ve  failure
6375  */
6376
6377 static int osd_it_iam_load(const struct lu_env *env,
6378                            const struct dt_it *di, __u64 hash)
6379 {
6380         struct osd_it_iam *it = (struct osd_it_iam *)di;
6381
6382         return iam_it_load(&it->oi_it, hash);
6383 }
6384
6385 static const struct dt_index_operations osd_index_iam_ops = {
6386         .dio_lookup         = osd_index_iam_lookup,
6387         .dio_declare_insert = osd_index_declare_iam_insert,
6388         .dio_insert         = osd_index_iam_insert,
6389         .dio_declare_delete = osd_index_declare_iam_delete,
6390         .dio_delete         = osd_index_iam_delete,
6391         .dio_it     = {
6392                 .init     = osd_it_iam_init,
6393                 .fini     = osd_it_iam_fini,
6394                 .get      = osd_it_iam_get,
6395                 .put      = osd_it_iam_put,
6396                 .next     = osd_it_iam_next,
6397                 .key      = osd_it_iam_key,
6398                 .key_size = osd_it_iam_key_size,
6399                 .rec      = osd_it_iam_rec,
6400                 .store    = osd_it_iam_store,
6401                 .load     = osd_it_iam_load
6402         }
6403 };
6404
6405
6406 /**
6407  * Creates or initializes iterator context.
6408  *
6409  * \retval struct osd_it_ea, iterator structure on success
6410  *
6411  */
6412 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6413                                     struct dt_object *dt,
6414                                     __u32 attr)
6415 {
6416         struct osd_object *obj = osd_dt_obj(dt);
6417         struct osd_thread_info *info = osd_oti_get(env);
6418         struct osd_it_ea *oie;
6419         struct file *file;
6420         struct lu_object *lo = &dt->do_lu;
6421         struct dentry *obj_dentry;
6422
6423         ENTRY;
6424
6425         if (!dt_object_exists(dt) || obj->oo_destroyed)
6426                 RETURN(ERR_PTR(-ENOENT));
6427
6428         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6429         if (oie == NULL)
6430                 RETURN(ERR_PTR(-ENOMEM));
6431         obj_dentry = &oie->oie_dentry;
6432
6433         obj_dentry->d_inode = obj->oo_inode;
6434         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6435         obj_dentry->d_name.hash = 0;
6436
6437         oie->oie_rd_dirent       = 0;
6438         oie->oie_it_dirent       = 0;
6439         oie->oie_dirent          = NULL;
6440         if (unlikely(!info->oti_it_ea_buf_used)) {
6441                 oie->oie_buf = info->oti_it_ea_buf;
6442                 info->oti_it_ea_buf_used = 1;
6443         } else {
6444                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6445                 if (oie->oie_buf == NULL)
6446                         RETURN(ERR_PTR(-ENOMEM));
6447         }
6448         oie->oie_obj = obj;
6449
6450         file = &oie->oie_file;
6451
6452         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6453         if (attr & LUDA_64BITHASH)
6454                 file->f_mode    = FMODE_64BITHASH;
6455         else
6456                 file->f_mode    = FMODE_32BITHASH;
6457         file->f_path.dentry     = obj_dentry;
6458         file->f_mapping         = obj->oo_inode->i_mapping;
6459         file->f_op              = obj->oo_inode->i_fop;
6460         set_file_inode(file, obj->oo_inode);
6461
6462         lu_object_get(lo);
6463         RETURN((struct dt_it *)oie);
6464 }
6465
6466 /**
6467  * Destroy or finishes iterator context.
6468  *
6469  * \param di iterator structure to be destroyed
6470  */
6471 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6472 {
6473         struct osd_thread_info *info = osd_oti_get(env);
6474         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6475         struct osd_object *obj = oie->oie_obj;
6476         struct inode *inode = obj->oo_inode;
6477
6478         ENTRY;
6479         oie->oie_file.f_op->release(inode, &oie->oie_file);
6480         osd_object_put(env, obj);
6481         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6482                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6483         else
6484                 info->oti_it_ea_buf_used = 0;
6485         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6486         EXIT;
6487 }
6488
6489 /**
6490  * It position the iterator at given key, so that next lookup continues from
6491  * that key Or it is similar to dio_it->load() but based on a key,
6492  * rather than file position.
6493  *
6494  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6495  * to the beginning.
6496  *
6497  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6498  */
6499 static int osd_it_ea_get(const struct lu_env *env,
6500                          struct dt_it *di, const struct dt_key *key)
6501 {
6502         struct osd_it_ea *it = (struct osd_it_ea *)di;
6503
6504         ENTRY;
6505         LASSERT(((const char *)key)[0] == '\0');
6506         it->oie_file.f_pos = 0;
6507         it->oie_rd_dirent = 0;
6508         it->oie_it_dirent = 0;
6509         it->oie_dirent = NULL;
6510
6511         RETURN(+1);
6512 }
6513
6514 /**
6515  * Does nothing
6516  */
6517 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6518 {
6519 }
6520
6521 struct osd_filldir_cbs {
6522 #ifdef HAVE_DIR_CONTEXT
6523         struct dir_context ctx;
6524 #endif
6525         struct osd_it_ea  *it;
6526 };
6527 /**
6528  * It is called internally by ->readdir(). It fills the
6529  * iterator's in-memory data structure with required
6530  * information i.e. name, namelen, rec_size etc.
6531  *
6532  * \param buf in which information to be filled in.
6533  * \param name name of the file in given dir
6534  *
6535  * \retval 0 on success
6536  * \retval 1 on buffer full
6537  */
6538 #ifdef HAVE_FILLDIR_USE_CTX
6539 static int osd_ldiskfs_filldir(struct dir_context *buf,
6540 #else
6541 static int osd_ldiskfs_filldir(void *buf,
6542 #endif
6543                                const char *name, int namelen,
6544                                loff_t offset, __u64 ino, unsigned int d_type)
6545 {
6546         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6547         struct osd_object *obj = it->oie_obj;
6548         struct osd_it_ea_dirent *ent = it->oie_dirent;
6549         struct lu_fid *fid = &ent->oied_fid;
6550         struct osd_fid_pack *rec;
6551
6552         ENTRY;
6553
6554 /* this should never happen */
6555         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6556                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6557                 RETURN(-EIO);
6558         }
6559
6560         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6561             OSD_IT_EA_BUFSIZE)
6562                 RETURN(1);
6563
6564         /* "." is just the object itself. */
6565         if (namelen == 1 && name[0] == '.') {
6566                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6567         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6568                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6569                 if (osd_fid_unpack(fid, rec) != 0)
6570                         fid_zero(fid);
6571         } else {
6572                 fid_zero(fid);
6573         }
6574         d_type &= ~LDISKFS_DIRENT_LUFID;
6575
6576         /* NOT export local root. */
6577         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6578                 ino = obj->oo_inode->i_ino;
6579                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6580         }
6581
6582         ent->oied_ino     = ino;
6583         ent->oied_off     = offset;
6584         ent->oied_namelen = namelen;
6585         ent->oied_type    = d_type;
6586
6587         memcpy(ent->oied_name, name, namelen);
6588
6589         it->oie_rd_dirent++;
6590         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6591         RETURN(0);
6592 }
6593
6594 /**
6595  * Calls ->readdir() to load a directory entry at a time
6596  * and stored it in iterator's in-memory data structure.
6597  *
6598  * \param di iterator's in memory structure
6599  *
6600  * \retval   0 on success
6601  * \retval -ve on error
6602  * \retval +1 reach the end of entry
6603  */
6604 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6605                                const struct dt_it *di)
6606 {
6607         struct osd_it_ea *it = (struct osd_it_ea *)di;
6608         struct osd_object *obj = it->oie_obj;
6609         struct inode *inode = obj->oo_inode;
6610         struct htree_lock *hlock = NULL;
6611         struct file *filp = &it->oie_file;
6612         int rc = 0;
6613         struct osd_filldir_cbs buf = {
6614 #ifdef HAVE_DIR_CONTEXT
6615                 .ctx.actor = osd_ldiskfs_filldir,
6616 #endif
6617                 .it = it
6618         };
6619
6620         ENTRY;
6621         it->oie_dirent = it->oie_buf;
6622         it->oie_rd_dirent = 0;
6623
6624         if (obj->oo_hl_head != NULL) {
6625                 hlock = osd_oti_get(env)->oti_hlock;
6626                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6627                                    inode, LDISKFS_HLOCK_READDIR);
6628         } else {
6629                 down_read(&obj->oo_ext_idx_sem);
6630         }
6631
6632 #ifdef HAVE_DIR_CONTEXT
6633         buf.ctx.pos = filp->f_pos;
6634 #ifdef HAVE_ITERATE_SHARED
6635         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6636 #else
6637         rc = inode->i_fop->iterate(filp, &buf.ctx);
6638 #endif
6639         filp->f_pos = buf.ctx.pos;
6640 #else
6641         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6642 #endif
6643
6644         if (hlock != NULL)
6645                 ldiskfs_htree_unlock(hlock);
6646         else
6647                 up_read(&obj->oo_ext_idx_sem);
6648
6649         if (it->oie_rd_dirent == 0) {
6650                 /*
6651                  * If it does not get any dirent, it means it has been reached
6652                  * to the end of the dir
6653                  */
6654                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6655                 if (rc == 0)
6656                         rc = 1;
6657         } else {
6658                 it->oie_dirent = it->oie_buf;
6659                 it->oie_it_dirent = 1;
6660         }
6661
6662         RETURN(rc);
6663 }
6664
6665 /**
6666  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6667  * to load a directory entry at a time and stored it in
6668  * iterator's in-memory data structure.
6669  *
6670  * \param di iterator's in memory structure
6671  *
6672  * \retval +ve iterator reached to end
6673  * \retval   0 iterator not reached to end
6674  * \retval -ve on error
6675  */
6676 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6677 {
6678         struct osd_it_ea *it = (struct osd_it_ea *)di;
6679         int rc;
6680
6681         ENTRY;
6682
6683         if (it->oie_it_dirent < it->oie_rd_dirent) {
6684                 it->oie_dirent =
6685                         (void *)it->oie_dirent +
6686                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6687                                        it->oie_dirent->oied_namelen);
6688                 it->oie_it_dirent++;
6689                 rc = 0;
6690         } else {
6691                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6692                         rc = 1;
6693                 else
6694                         rc = osd_ldiskfs_it_fill(env, di);
6695         }
6696
6697         RETURN(rc);
6698 }
6699
6700 /**
6701  * Returns the key at current position from iterator's in memory structure.
6702  *
6703  * \param di iterator's in memory structure
6704  *
6705  * \retval key i.e. struct dt_key on success
6706  */
6707 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6708                                     const struct dt_it *di)
6709 {
6710         struct osd_it_ea *it = (struct osd_it_ea *)di;
6711
6712         return (struct dt_key *)it->oie_dirent->oied_name;
6713 }
6714
6715 /**
6716  * Returns key's size at current position from iterator's in memory structure.
6717  *
6718  * \param di iterator's in memory structure
6719  *
6720  * \retval key_size i.e. struct dt_key on success
6721  */
6722 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6723 {
6724         struct osd_it_ea *it = (struct osd_it_ea *)di;
6725
6726         return it->oie_dirent->oied_namelen;
6727 }
6728
6729 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6730 {
6731         if (LDISKFS_DIR_REC_LEN(de) >=
6732             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6733                 return true;
6734
6735         return false;
6736 }
6737
6738 static inline bool
6739 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6740                      unsigned int blocksize, bool dotdot)
6741 {
6742         if (dotdot)
6743                 return osd_dotdot_has_space(de);
6744
6745         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6746             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6747                 return true;
6748
6749         return false;
6750 }
6751
6752 static int
6753 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6754                     handle_t *jh, struct dentry *dentry,
6755                     const struct lu_fid *fid, struct buffer_head *bh,
6756                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6757                     bool dotdot)
6758 {
6759         struct inode *dir = dentry->d_parent->d_inode;
6760         struct inode *inode = dentry->d_inode;
6761         struct osd_fid_pack *rec;
6762         struct ldiskfs_dentry_param *ldp;
6763         int namelen = dentry->d_name.len;
6764         int rc;
6765         struct osd_thread_info *info = osd_oti_get(env);
6766
6767         ENTRY;
6768
6769         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6770                 RETURN(0);
6771
6772         /* There is enough space to hold the FID-in-dirent. */
6773         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6774                 rc = ldiskfs_journal_get_write_access(jh, bh);
6775                 if (rc != 0)
6776                         RETURN(rc);
6777
6778                 de->name[namelen] = 0;
6779                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6780                 rec->fp_len = sizeof(struct lu_fid) + 1;
6781                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6782                 de->file_type |= LDISKFS_DIRENT_LUFID;
6783                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6784
6785                 RETURN(rc);
6786         }
6787
6788         LASSERT(!dotdot);
6789
6790         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6791         if (rc != 0)
6792                 RETURN(rc);
6793
6794         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6795         osd_get_ldiskfs_dirent_param(ldp, fid);
6796         dentry->d_fsdata = (void *)ldp;
6797         ll_vfs_dq_init(dir);
6798         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6799         /*
6800          * It is too bad, we cannot reinsert the name entry back.
6801          * That means we lose it!
6802          */
6803         if (rc != 0)
6804                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6805                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6806                        osd_ino2name(inode),
6807                        dir->i_ino, dir->i_generation, namelen,
6808                        dentry->d_name.name, PFID(fid), rc);
6809
6810         RETURN(rc);
6811 }
6812
6813 static int
6814 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6815                         struct osd_it_ea *it, struct lu_fid *fid,
6816                         struct osd_inode_id *id, __u32 *attr)
6817 {
6818         struct osd_thread_info *info = osd_oti_get(env);
6819         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6820         struct osd_device *dev = osd_obj2dev(obj);
6821         struct super_block *sb = osd_sb(dev);
6822         const char *devname = osd_name(dev);
6823         struct osd_it_ea_dirent *ent = it->oie_dirent;
6824         struct inode *dir = obj->oo_inode;
6825         struct htree_lock *hlock = NULL;
6826         struct buffer_head *bh = NULL;
6827         handle_t *jh = NULL;
6828         struct ldiskfs_dir_entry_2 *de;
6829         struct dentry *dentry;
6830         struct inode *inode;
6831         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6832         int credits;
6833         int rc;
6834         bool dotdot = false;
6835         bool dirty = false;
6836
6837         ENTRY;
6838
6839         if (ent->oied_name[0] == '.') {
6840                 if (ent->oied_namelen == 1)
6841                         RETURN(0);
6842
6843                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6844                         dotdot = true;
6845         }
6846
6847         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6848         inode = osd_iget(info, dev, id);
6849         if (IS_ERR(inode)) {
6850                 rc = PTR_ERR(inode);
6851                 if (rc == -ENOENT || rc == -ESTALE) {
6852                         /*
6853                          * Maybe dangling name entry, or
6854                          * corrupted directory entry.
6855                          */
6856                         *attr |= LUDA_UNKNOWN;
6857                         rc = 0;
6858                 } else {
6859                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6860                                "check_repair, dir = %lu/%u, name = %.*s, "
6861                                "ino = %llu, rc = %d\n",
6862                                devname, dir->i_ino, dir->i_generation,
6863                                ent->oied_namelen, ent->oied_name,
6864                                ent->oied_ino, rc);
6865                 }
6866
6867                 RETURN(rc);
6868         }
6869
6870         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6871                                            ent->oied_namelen);
6872         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6873         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6874                 lma = NULL;
6875         else if (rc != 0)
6876                 GOTO(out, rc);
6877
6878         /*
6879          * We need to ensure that the name entry is still valid.
6880          * Because it may be removed or renamed by other already.
6881          *
6882          * The unlink or rename operation will start journal before PDO lock,
6883          * so to avoid deadlock, here we need to start journal handle before
6884          * related PDO lock also. But because we do not know whether there
6885          * will be something to be repaired before PDO lock, we just start
6886          * journal without conditions.
6887          *
6888          * We may need to remove the name entry firstly, then insert back.
6889          * One credit is for user quota file update.
6890          * One credit is for group quota file update.
6891          * Two credits are for dirty inode.
6892          */
6893         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6894                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6895
6896         if (dev->od_dirent_journal != 0) {
6897
6898 again:
6899                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6900                 if (IS_ERR(jh)) {
6901                         rc = PTR_ERR(jh);
6902                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6903                                "check_repair, dir = %lu/%u, credits = %d, "
6904                                "name = %.*s, ino = %llu: rc = %d\n",
6905                                devname, dir->i_ino, dir->i_generation, credits,
6906                                ent->oied_namelen, ent->oied_name,
6907                                ent->oied_ino, rc);
6908
6909                         GOTO(out_inode, rc);
6910                 }
6911
6912                 if (obj->oo_hl_head != NULL) {
6913                         hlock = osd_oti_get(env)->oti_hlock;
6914                         /*
6915                          * "0" means exclusive lock for the whole directory.
6916                          * We need to prevent others access such name entry
6917                          * during the delete + insert. Neither HLOCK_ADD nor
6918                          * HLOCK_DEL cannot guarantee the atomicity.
6919                          */
6920                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6921                 } else {
6922                         down_write(&obj->oo_ext_idx_sem);
6923                 }
6924         } else {
6925                 if (obj->oo_hl_head != NULL) {
6926                         hlock = osd_oti_get(env)->oti_hlock;
6927                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6928                                            LDISKFS_HLOCK_LOOKUP);
6929                 } else {
6930                         down_read(&obj->oo_ext_idx_sem);
6931                 }
6932         }
6933
6934         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6935         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6936                 *attr |= LUDA_IGNORE;
6937
6938                 GOTO(out, rc = 0);
6939         }
6940
6941         /*
6942          * For dotdot entry, if there is not enough space to hold the
6943          * FID-in-dirent, just keep them there. It only happens when the
6944          * device upgraded from 1.8 or restored from MDT file-level backup.
6945          * For the whole directory, only dotdot entry have no FID-in-dirent
6946          * and needs to get FID from LMA when readdir, it will not affect the
6947          * performance much.
6948          */
6949         if (dotdot && !osd_dotdot_has_space(de)) {
6950                 *attr |= LUDA_UNKNOWN;
6951
6952                 GOTO(out, rc = 0);
6953         }
6954
6955         if (lma != NULL) {
6956                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6957                         GOTO(out, rc = 0);
6958
6959                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6960                         struct lu_fid *tfid = &lma->lma_self_fid;
6961
6962                         if (likely(dotdot &&
6963                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6964                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6965                                 /*
6966                                  * It must be REMOTE_PARENT_DIR and as the
6967                                  * 'dotdot' entry of remote directory
6968                                  */
6969                                 *attr |= LUDA_IGNORE;
6970                         } else {
6971                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6972                                        "parent directory, but got %.*s under "
6973                                        "dir = %lu/%u with the FID "DFID"\n",
6974                                        devname, ent->oied_namelen,
6975                                        ent->oied_name, dir->i_ino,
6976                                        dir->i_generation, PFID(tfid));
6977
6978                                 *attr |= LUDA_UNKNOWN;
6979                         }
6980
6981                         GOTO(out, rc = 0);
6982                 }
6983         }
6984
6985         if (!fid_is_zero(fid)) {
6986                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6987                                               ent->oied_namelen);
6988                 if (rc == -ENOENT ||
6989                     (rc == -ENODATA &&
6990                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6991                         /*
6992                          * linkEA does not recognize the dirent entry,
6993                          * it may because the dirent entry corruption
6994                          * and points to other's inode.
6995                          */
6996                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6997                                "recognize the dirent, dir = %lu/%u, "
6998                                " name = %.*s, ino = %llu, "
6999                                DFID": rc = %d\n", devname, dir->i_ino,
7000                                dir->i_generation, ent->oied_namelen,
7001                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7002                         *attr |= LUDA_UNKNOWN;
7003
7004                         GOTO(out, rc = 0);
7005                 }
7006
7007                 if (rc && rc != -ENODATA) {
7008                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7009                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7010                                DFID": rc = %d\n", devname, dir->i_ino,
7011                                dir->i_generation, ent->oied_namelen,
7012                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7013                         *attr |= LUDA_UNKNOWN;
7014
7015                         GOTO(out, rc = 0);
7016                 }
7017         }
7018
7019         if (lma != NULL) {
7020                 /*
7021                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7022                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7023                  */
7024                 if (*attr & LUDA_VERIFY_DRYRUN) {
7025                         *fid = lma->lma_self_fid;
7026                         *attr |= LUDA_REPAIR;
7027
7028                         GOTO(out, rc = 0);
7029                 }
7030
7031                 if (jh == NULL) {
7032                         brelse(bh);
7033                         dev->od_dirent_journal = 1;
7034                         if (hlock != NULL) {
7035                                 ldiskfs_htree_unlock(hlock);
7036                                 hlock = NULL;
7037                         } else {
7038                                 up_read(&obj->oo_ext_idx_sem);
7039                         }
7040
7041                         goto again;
7042                 }
7043
7044                 *fid = lma->lma_self_fid;
7045                 dirty = true;
7046                 /* Update or append the FID-in-dirent. */
7047                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7048                                          bh, de, hlock, dotdot);
7049                 if (rc == 0)
7050                         *attr |= LUDA_REPAIR;
7051                 else
7052                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7053                                "the dirent, dir = %lu/%u, name = %.*s, "
7054                                "ino = %llu, "DFID": rc = %d\n",
7055                                devname, dir->i_ino, dir->i_generation,
7056                                ent->oied_namelen, ent->oied_name,
7057                                ent->oied_ino, PFID(fid), rc);
7058         } else {
7059                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7060                 if (*attr & LUDA_VERIFY_DRYRUN) {
7061                         if (fid_is_sane(fid)) {
7062                                 *attr |= LUDA_REPAIR;
7063                         } else if (dev->od_index == 0) {
7064                                 lu_igif_build(fid, inode->i_ino,
7065                                               inode->i_generation);
7066                                 *attr |= LUDA_UPGRADE;
7067                         }
7068
7069                         GOTO(out, rc = 0);
7070                 }
7071
7072                 if (jh == NULL) {
7073                         brelse(bh);
7074                         dev->od_dirent_journal = 1;
7075                         if (hlock != NULL) {
7076                                 ldiskfs_htree_unlock(hlock);
7077                                 hlock = NULL;
7078                         } else {
7079                                 up_read(&obj->oo_ext_idx_sem);
7080                         }
7081
7082                         goto again;
7083                 }
7084
7085                 dirty = true;
7086                 if (unlikely(fid_is_sane(fid))) {
7087                         /*
7088                          * FID-in-dirent exists, but FID-in-LMA is lost.
7089                          * Trust the FID-in-dirent, and add FID-in-LMA.
7090                          */
7091                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7092                         if (rc == 0)
7093                                 *attr |= LUDA_REPAIR;
7094                         else
7095                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7096                                        "update dirent, dir = %lu/%u, "
7097                                        "name = %.*s, ino = %llu, "
7098                                        DFID": rc = %d\n",
7099                                        devname, dir->i_ino, dir->i_generation,
7100                                        ent->oied_namelen, ent->oied_name,
7101                                        ent->oied_ino, PFID(fid), rc);
7102                 } else if (dev->od_index == 0) {
7103                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7104                         /*
7105                          * It is probably IGIF object. Only aappend the
7106                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7107                          */
7108                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7109                                                  bh, de, hlock, dotdot);
7110                         if (rc == 0)
7111                                 *attr |= LUDA_UPGRADE;
7112                         else
7113                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7114                                        "after the dirent, dir = %lu/%u, "
7115                                        "name = %.*s, ino = %llu, "
7116                                        DFID": rc = %d\n",
7117                                        devname, dir->i_ino, dir->i_generation,
7118                                        ent->oied_namelen, ent->oied_name,
7119                                        ent->oied_ino, PFID(fid), rc);
7120                 }
7121         }
7122
7123         GOTO(out, rc);
7124
7125 out:
7126         if (!IS_ERR(bh))
7127                 brelse(bh);
7128         if (hlock != NULL) {
7129                 ldiskfs_htree_unlock(hlock);
7130         } else {
7131                 if (dev->od_dirent_journal != 0)
7132                         up_write(&obj->oo_ext_idx_sem);
7133                 else
7134                         up_read(&obj->oo_ext_idx_sem);
7135         }
7136
7137         if (jh != NULL)
7138                 ldiskfs_journal_stop(jh);
7139
7140 out_inode:
7141         iput(inode);
7142         if (rc >= 0 && !dirty)
7143                 dev->od_dirent_journal = 0;
7144
7145         return rc;
7146 }
7147
7148 /**
7149  * Returns the value at current position from iterator's in memory structure.
7150  *
7151  * \param di struct osd_it_ea, iterator's in memory structure
7152  * \param attr attr requested for dirent.
7153  * \param lde lustre dirent
7154  *
7155  * \retval   0 no error and \param lde has correct lustre dirent.
7156  * \retval -ve on error
7157  */
7158 static inline int osd_it_ea_rec(const struct lu_env *env,
7159                                 const struct dt_it *di,
7160                                 struct dt_rec *dtrec, __u32 attr)
7161 {
7162         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7163         struct osd_object      *obj   = it->oie_obj;
7164         struct osd_device      *dev   = osd_obj2dev(obj);
7165         struct osd_thread_info *oti   = osd_oti_get(env);
7166         struct osd_inode_id    *id    = &oti->oti_id;
7167         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7168         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7169         __u32 ino = it->oie_dirent->oied_ino;
7170         int rc = 0;
7171
7172         ENTRY;
7173
7174         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7175
7176         if (attr & LUDA_VERIFY) {
7177                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7178                         attr |= LUDA_IGNORE;
7179                         /*
7180                          * If the parent is on remote MDT, and there
7181                          * is no FID-in-dirent, then we have to get
7182                          * the parent FID from the linkEA.
7183                          */
7184                         if (!fid_is_sane(fid) &&
7185                             it->oie_dirent->oied_namelen == 2 &&
7186                             it->oie_dirent->oied_name[0] == '.' &&
7187                             it->oie_dirent->oied_name[1] == '.')
7188                                 osd_get_pfid_from_linkea(env, obj, fid);
7189                 } else {
7190                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7191                                                      &attr);
7192                 }
7193
7194                 if (!fid_is_sane(fid))
7195                         attr |= LUDA_UNKNOWN;
7196         } else {
7197                 attr &= ~LU_DIRENT_ATTRS_MASK;
7198                 if (!fid_is_sane(fid)) {
7199                         bool is_dotdot = false;
7200
7201                         if (it->oie_dirent->oied_namelen == 2 &&
7202                             it->oie_dirent->oied_name[0] == '.' &&
7203                             it->oie_dirent->oied_name[1] == '.')
7204                                 is_dotdot = true;
7205                         /*
7206                          * If the parent is on remote MDT, and there
7207                          * is no FID-in-dirent, then we have to get
7208                          * the parent FID from the linkEA.
7209                          */
7210                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7211                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7212                         } else {
7213                                 if (is_dotdot == false &&
7214                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7215                                         RETURN(-ENOENT);
7216
7217                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7218                         }
7219                 } else {
7220                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7221                 }
7222         }
7223
7224         /* Pack the entry anyway, at least the offset is right. */
7225         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7226                            it->oie_dirent->oied_name,
7227                            it->oie_dirent->oied_namelen,
7228                            it->oie_dirent->oied_type, attr);
7229
7230         if (rc < 0)
7231                 RETURN(rc);
7232
7233         if (osd_remote_fid(env, dev, fid))
7234                 RETURN(0);
7235
7236         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7237                 osd_add_oi_cache(oti, dev, id, fid);
7238
7239         RETURN(rc > 0 ? 0 : rc);
7240 }
7241
7242 /**
7243  * Returns the record size size at current position.
7244  *
7245  * This function will return record(lu_dirent) size in bytes.
7246  *
7247  * \param[in] env       execution environment
7248  * \param[in] di        iterator's in memory structure
7249  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7250  *                      calculate the lu_dirent size.
7251  *
7252  * \retval      record size(in bytes & in memory) of the current lu_dirent
7253  *              entry.
7254  */
7255 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7256                               __u32 attr)
7257 {
7258         struct osd_it_ea *it = (struct osd_it_ea *)di;
7259
7260         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7261 }
7262
7263 /**
7264  * Returns a cookie for current position of the iterator head, so that
7265  * user can use this cookie to load/start the iterator next time.
7266  *
7267  * \param di iterator's in memory structure
7268  *
7269  * \retval cookie for current position, on success
7270  */
7271 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7272 {
7273         struct osd_it_ea *it = (struct osd_it_ea *)di;
7274
7275         return it->oie_dirent->oied_off;
7276 }
7277
7278 /**
7279  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7280  * to load a directory entry at a time and stored it i inn,
7281  * in iterator's in-memory data structure.
7282  *
7283  * \param di struct osd_it_ea, iterator's in memory structure
7284  *
7285  * \retval +ve on success
7286  * \retval -ve on error
7287  */
7288 static int osd_it_ea_load(const struct lu_env *env,
7289                           const struct dt_it *di, __u64 hash)
7290 {
7291         struct osd_it_ea *it = (struct osd_it_ea *)di;
7292         int rc;
7293
7294         ENTRY;
7295         it->oie_file.f_pos = hash;
7296
7297         rc =  osd_ldiskfs_it_fill(env, di);
7298         if (rc > 0)
7299                 rc = -ENODATA;
7300
7301         if (rc == 0)
7302                 rc = 1;
7303
7304         RETURN(rc);
7305 }
7306
7307 /**
7308  * Index lookup function for interoperability mode (b11826).
7309  *
7310  * \param key,  key i.e. file name to be searched
7311  *
7312  * \retval +ve, on success
7313  * \retval -ve, on error
7314  */
7315 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7316                                struct dt_rec *rec, const struct dt_key *key)
7317 {
7318         struct osd_object *obj = osd_dt_obj(dt);
7319         int rc = 0;
7320
7321         ENTRY;
7322
7323         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7324         LINVRNT(osd_invariant(obj));
7325
7326         rc = osd_ea_lookup_rec(env, obj, rec, key);
7327         if (rc == 0)
7328                 rc = 1;
7329         RETURN(rc);
7330 }
7331
7332 /**
7333  * Index and Iterator operations for interoperability
7334  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7335  */
7336 static const struct dt_index_operations osd_index_ea_ops = {
7337         .dio_lookup         = osd_index_ea_lookup,
7338         .dio_declare_insert = osd_index_declare_ea_insert,
7339         .dio_insert         = osd_index_ea_insert,
7340         .dio_declare_delete = osd_index_declare_ea_delete,
7341         .dio_delete         = osd_index_ea_delete,
7342         .dio_it     = {
7343                 .init     = osd_it_ea_init,
7344                 .fini     = osd_it_ea_fini,
7345                 .get      = osd_it_ea_get,
7346                 .put      = osd_it_ea_put,
7347                 .next     = osd_it_ea_next,
7348                 .key      = osd_it_ea_key,
7349                 .key_size = osd_it_ea_key_size,
7350                 .rec      = osd_it_ea_rec,
7351                 .rec_size = osd_it_ea_rec_size,
7352                 .store    = osd_it_ea_store,
7353                 .load     = osd_it_ea_load
7354         }
7355 };
7356
7357 static void *osd_key_init(const struct lu_context *ctx,
7358                           struct lu_context_key *key)
7359 {
7360         struct osd_thread_info *info;
7361
7362         OBD_ALLOC_PTR(info);
7363         if (info == NULL)
7364                 return ERR_PTR(-ENOMEM);
7365
7366         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7367         if (info->oti_it_ea_buf == NULL)
7368                 goto out_free_info;
7369
7370         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7371
7372         info->oti_hlock = ldiskfs_htree_lock_alloc();
7373         if (info->oti_hlock == NULL)
7374                 goto out_free_ea;
7375
7376         return info;
7377
7378 out_free_ea:
7379         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7380 out_free_info:
7381         OBD_FREE_PTR(info);
7382         return ERR_PTR(-ENOMEM);
7383 }
7384
7385 static void osd_key_fini(const struct lu_context *ctx,
7386                          struct lu_context_key *key, void *data)
7387 {
7388         struct osd_thread_info *info = data;
7389         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7390         struct osd_idmap_cache *idc = info->oti_ins_cache;
7391
7392         if (info->oti_dio_pages) {
7393                 int i;
7394                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7395                         if (info->oti_dio_pages[i])
7396                                 __free_page(info->oti_dio_pages[i]);
7397                 }
7398                 OBD_FREE(info->oti_dio_pages,
7399                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7400         }
7401
7402         if (info->oti_inode != NULL)
7403                 OBD_FREE_PTR(lli);
7404         if (info->oti_hlock != NULL)
7405                 ldiskfs_htree_lock_free(info->oti_hlock);
7406         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7407         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7408         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7409         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7410         lu_buf_free(&info->oti_big_buf);
7411         if (idc != NULL) {
7412                 LASSERT(info->oti_ins_cache_size > 0);
7413                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7414                 info->oti_ins_cache = NULL;
7415                 info->oti_ins_cache_size = 0;
7416         }
7417         OBD_FREE_PTR(info);
7418 }
7419
7420 static void osd_key_exit(const struct lu_context *ctx,
7421                          struct lu_context_key *key, void *data)
7422 {
7423         struct osd_thread_info *info = data;
7424
7425         LASSERT(info->oti_r_locks == 0);
7426         LASSERT(info->oti_w_locks == 0);
7427         LASSERT(info->oti_txns    == 0);
7428 }
7429
7430 /* type constructor/destructor: osd_type_init, osd_type_fini */
7431 LU_TYPE_INIT_FINI(osd, &osd_key);
7432
7433 struct lu_context_key osd_key = {
7434         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7435         .lct_init = osd_key_init,
7436         .lct_fini = osd_key_fini,
7437         .lct_exit = osd_key_exit
7438 };
7439
7440
7441 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7442                            const char *name, struct lu_device *next)
7443 {
7444         struct osd_device *osd = osd_dev(d);
7445
7446         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7447             sizeof(osd->od_svname))
7448                 return -E2BIG;
7449         return osd_procfs_init(osd, name);
7450 }
7451
7452 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7453 {
7454         struct seq_server_site *ss = osd_seq_site(osd);
7455         int rc;
7456
7457         ENTRY;
7458
7459         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7460                 RETURN(0);
7461
7462         if (unlikely(ss == NULL))
7463                 RETURN(-ENODEV);
7464
7465         OBD_ALLOC_PTR(osd->od_cl_seq);
7466         if (osd->od_cl_seq == NULL)
7467                 RETURN(-ENOMEM);
7468
7469         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7470                              osd->od_svname, ss->ss_server_seq);
7471         if (rc != 0) {
7472                 OBD_FREE_PTR(osd->od_cl_seq);
7473                 osd->od_cl_seq = NULL;
7474                 RETURN(rc);
7475         }
7476
7477         if (ss->ss_node_id == 0) {
7478                 /*
7479                  * If the OSD on the sequence controller(MDT0), then allocate
7480                  * sequence here, otherwise allocate sequence after connected
7481                  * to MDT0 (see mdt_register_lwp_callback()).
7482                  */
7483                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7484                                    &osd->od_cl_seq->lcs_space, env);
7485         }
7486
7487         RETURN(rc);
7488 }
7489
7490 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7491 {
7492         if (osd->od_cl_seq == NULL)
7493                 return;
7494
7495         seq_client_fini(osd->od_cl_seq);
7496         OBD_FREE_PTR(osd->od_cl_seq);
7497         osd->od_cl_seq = NULL;
7498 }
7499
7500 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7501 {
7502         ENTRY;
7503
7504         /* shutdown quota slave instance associated with the device */
7505         if (o->od_quota_slave_md != NULL) {
7506                 struct qsd_instance *qsd = o->od_quota_slave_md;
7507
7508                 o->od_quota_slave_md = NULL;
7509                 qsd_fini(env, qsd);
7510         }
7511
7512         if (o->od_quota_slave_dt != NULL) {
7513                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7514
7515                 o->od_quota_slave_dt = NULL;
7516                 qsd_fini(env, qsd);
7517         }
7518
7519         osd_fid_fini(env, o);
7520         osd_scrub_cleanup(env, o);
7521
7522         RETURN(0);
7523 }
7524
7525 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7526 {
7527         ENTRY;
7528
7529         if (o->od_mnt != NULL) {
7530                 shrink_dcache_sb(osd_sb(o));
7531                 osd_sync(env, &o->od_dt_dev);
7532
7533                 mntput(o->od_mnt);
7534                 o->od_mnt = NULL;
7535         }
7536
7537         EXIT;
7538 }
7539
7540 static int osd_mount(const struct lu_env *env,
7541                      struct osd_device *o, struct lustre_cfg *cfg)
7542 {
7543         const char *name = lustre_cfg_string(cfg, 0);
7544         const char *dev = lustre_cfg_string(cfg, 1);
7545         const char *opts;
7546         unsigned long page, s_flags, lmd_flags = 0;
7547         struct page *__page;
7548         struct file_system_type *type;
7549         char *options = NULL;
7550         char *str;
7551         struct osd_thread_info *info = osd_oti_get(env);
7552         struct lu_fid *fid = &info->oti_fid;
7553         struct inode *inode;
7554         int rc = 0, force_over_512tb = 0;
7555
7556         ENTRY;
7557
7558         if (o->od_mnt != NULL)
7559                 RETURN(0);
7560
7561         if (strlen(dev) >= sizeof(o->od_mntdev))
7562                 RETURN(-E2BIG);
7563         strcpy(o->od_mntdev, dev);
7564
7565         str = lustre_cfg_string(cfg, 2);
7566         s_flags = simple_strtoul(str, NULL, 0);
7567         str = strstr(str, ":");
7568         if (str)
7569                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7570         opts = lustre_cfg_string(cfg, 3);
7571 #ifdef __BIG_ENDIAN
7572         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7573                 CERROR("%s: device %s extents feature is not guaranteed to "
7574                        "work on big-endian systems. Use \"bigendian_extents\" "
7575                        "mount option to override.\n", name, dev);
7576                 RETURN(-EINVAL);
7577         }
7578 #endif
7579 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7580         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7581                 CWARN("force_over_128tb option is deprecated. "
7582                       "Filesystems less than 512TB can be created without any "
7583                       "force options. Use force_over_512tb option for "
7584                       "filesystems greater than 512TB.\n");
7585         }
7586 #endif
7587 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7588         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7589                 CWARN("force_over_256tb option is deprecated. "
7590                       "Filesystems less than 512TB can be created without any "
7591                       "force options. Use force_over_512tb option for "
7592                       "filesystems greater than 512TB.\n");
7593         }
7594 #endif
7595
7596         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7597                 force_over_512tb = 1;
7598
7599         __page = alloc_page(GFP_KERNEL);
7600         if (__page == NULL)
7601                 GOTO(out, rc = -ENOMEM);
7602         page = (unsigned long)page_address(__page);
7603         options = (char *)page;
7604         *options = '\0';
7605         if (opts != NULL) {
7606                 /* strip out the options for back compatiblity */
7607                 static char *sout[] = {
7608                         "mballoc",
7609                         "iopen",
7610                         "noiopen",
7611                         "iopen_nopriv",
7612                         "extents",
7613                         "noextents",
7614                         /* strip out option we processed in osd */
7615                         "bigendian_extents",
7616                         "force_over_128tb",
7617                         "force_over_256tb",
7618                         "force_over_512tb",
7619                         NULL
7620                 };
7621                 strncat(options, opts, PAGE_SIZE);
7622                 for (rc = 0, str = options; sout[rc]; ) {
7623                         char *op = strstr(str, sout[rc]);
7624
7625                         if (op == NULL) {
7626                                 rc++;
7627                                 str = options;
7628                                 continue;
7629                         }
7630                         if (op == options || *(op - 1) == ',') {
7631                                 str = op + strlen(sout[rc]);
7632                                 if (*str == ',' || *str == '\0') {
7633                                         *str == ',' ? str++ : str;
7634                                         memmove(op, str, strlen(str) + 1);
7635                                 }
7636                         }
7637                         for (str = op; *str != ',' && *str != '\0'; str++)
7638                                 ;
7639                 }
7640         } else {
7641                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7642         }
7643
7644         /* Glom up mount options */
7645         if (*options != '\0')
7646                 strncat(options, ",", PAGE_SIZE);
7647         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7648
7649         type = get_fs_type("ldiskfs");
7650         if (!type) {
7651                 CERROR("%s: cannot find ldiskfs module\n", name);
7652                 GOTO(out, rc = -ENODEV);
7653         }
7654
7655         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7656         module_put(type->owner);
7657
7658         if (IS_ERR(o->od_mnt)) {
7659                 rc = PTR_ERR(o->od_mnt);
7660                 o->od_mnt = NULL;
7661                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7662                 GOTO(out, rc);
7663         }
7664
7665         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7666                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7667                                  force_over_512tb == 0) {
7668                 CERROR("%s: device %s LDISKFS does not support filesystems "
7669                        "greater than 512TB and can cause data corruption. "
7670                        "Use \"force_over_512tb\" mount option to override.\n",
7671                        name, dev);
7672                 GOTO(out_mnt, rc = -EINVAL);
7673         }
7674
7675         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7676                 if (priv_dev_set_rdonly) {
7677                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7678                         o->od_dt_dev.dd_rdonly = 1;
7679                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7680                                       name);
7681                 } else {
7682                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7683                                       name);
7684
7685                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7686                 }
7687         } else if (priv_dev_check_rdonly &&
7688                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7689                 CERROR("%s: underlying device %s is marked as "
7690                        "read-only. Setup failed\n", name, dev);
7691
7692                 GOTO(out_mnt, rc = -EROFS);
7693         }
7694
7695         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7696                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7697                 GOTO(out_mnt, rc = -EINVAL);
7698         }
7699
7700 #ifdef LDISKFS_MOUNT_DIRDATA
7701         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7702                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7703         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7704                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7705                       "enabling the dirdata feature. If you do not want to "
7706                       "downgrade to Lustre-1.x again, you can enable it via "
7707                       "'tune2fs -O dirdata device'\n", name, dev);
7708 #endif
7709         inode = osd_sb(o)->s_root->d_inode;
7710         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7711         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7712         if (rc != 0) {
7713                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7714                 GOTO(out_mnt, rc);
7715         }
7716
7717         if (lmd_flags & LMD_FLG_NOSCRUB)
7718                 o->od_auto_scrub_interval = AS_NEVER;
7719
7720         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7721                 /* do not use pagecache with flash-backed storage */
7722                 o->od_writethrough_cache = 0;
7723                 o->od_read_cache = 0;
7724         }
7725
7726         GOTO(out, rc = 0);
7727
7728 out_mnt:
7729         mntput(o->od_mnt);
7730         o->od_mnt = NULL;
7731
7732 out:
7733         if (__page)
7734                 __free_page(__page);
7735
7736         return rc;
7737 }
7738
7739 static struct lu_device *osd_device_fini(const struct lu_env *env,
7740                                          struct lu_device *d)
7741 {
7742         struct osd_device *o = osd_dev(d);
7743
7744         ENTRY;
7745
7746         osd_index_backup(env, o, false);
7747         osd_shutdown(env, o);
7748         osd_procfs_fini(o);
7749         osd_obj_map_fini(o);
7750         osd_umount(env, o);
7751
7752         RETURN(NULL);
7753 }
7754
7755 static int osd_device_init0(const struct lu_env *env,
7756                             struct osd_device *o,
7757                             struct lustre_cfg *cfg)
7758 {
7759         struct lu_device *l = osd2lu_dev(o);
7760         struct osd_thread_info *info;
7761         int rc;
7762         int cplen = 0;
7763
7764         /* if the module was re-loaded, env can loose its keys */
7765         rc = lu_env_refill((struct lu_env *)env);
7766         if (rc)
7767                 GOTO(out, rc);
7768         info = osd_oti_get(env);
7769         LASSERT(info);
7770
7771         l->ld_ops = &osd_lu_ops;
7772         o->od_dt_dev.dd_ops = &osd_dt_ops;
7773
7774         spin_lock_init(&o->od_osfs_lock);
7775         mutex_init(&o->od_otable_mutex);
7776         INIT_LIST_HEAD(&o->od_orphan_list);
7777         INIT_LIST_HEAD(&o->od_index_backup_list);
7778         INIT_LIST_HEAD(&o->od_index_restore_list);
7779         spin_lock_init(&o->od_lock);
7780         o->od_index_backup_policy = LIBP_NONE;
7781         o->od_t10_type = 0;
7782
7783         o->od_read_cache = 1;
7784         o->od_writethrough_cache = 1;
7785         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7786
7787         o->od_auto_scrub_interval = AS_DEFAULT;
7788
7789         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7790                         sizeof(o->od_svname));
7791         if (cplen >= sizeof(o->od_svname)) {
7792                 rc = -E2BIG;
7793                 GOTO(out, rc);
7794         }
7795
7796         o->od_index_backup_stop = 0;
7797         o->od_index = -1; /* -1 means index is invalid */
7798         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7799         if (rc == LDD_F_SV_TYPE_OST)
7800                 o->od_is_ost = 1;
7801
7802         o->od_full_scrub_ratio = OFSR_DEFAULT;
7803         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7804         rc = osd_mount(env, o, cfg);
7805         if (rc != 0)
7806                 GOTO(out, rc);
7807
7808         /* Can only check block device after mount */
7809         o->od_nonrotational =
7810                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7811
7812         rc = osd_obj_map_init(env, o);
7813         if (rc != 0)
7814                 GOTO(out_mnt, rc);
7815
7816         rc = lu_site_init(&o->od_site, l);
7817         if (rc != 0)
7818                 GOTO(out_compat, rc);
7819         o->od_site.ls_bottom_dev = l;
7820
7821         rc = lu_site_init_finish(&o->od_site);
7822         if (rc != 0)
7823                 GOTO(out_site, rc);
7824
7825         INIT_LIST_HEAD(&o->od_ios_list);
7826         /* setup scrub, including OI files initialization */
7827         o->od_in_init = 1;
7828         rc = osd_scrub_setup(env, o);
7829         o->od_in_init = 0;
7830         if (rc < 0)
7831                 GOTO(out_site, rc);
7832
7833         rc = osd_procfs_init(o, o->od_svname);
7834         if (rc != 0) {
7835                 CERROR("%s: can't initialize procfs: rc = %d\n",
7836                        o->od_svname, rc);
7837                 GOTO(out_scrub, rc);
7838         }
7839
7840         LASSERT(l->ld_site->ls_linkage.next != NULL);
7841         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7842
7843         /* initialize quota slave instance */
7844         /* currently it's no need to prepare qsd_instance_md for OST */
7845         if (!o->od_is_ost) {
7846                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7847                                                 &o->od_dt_dev,
7848                                                 o->od_proc_entry, true);
7849                 if (IS_ERR(o->od_quota_slave_md)) {
7850                         rc = PTR_ERR(o->od_quota_slave_md);
7851                         o->od_quota_slave_md = NULL;
7852                         GOTO(out_procfs, rc);
7853                 }
7854         }
7855
7856         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7857                                         o->od_proc_entry, false);
7858
7859         if (IS_ERR(o->od_quota_slave_dt)) {
7860                 if (o->od_quota_slave_md != NULL) {
7861                         qsd_fini(env, o->od_quota_slave_md);
7862                         o->od_quota_slave_md = NULL;
7863                 }
7864
7865                 rc = PTR_ERR(o->od_quota_slave_dt);
7866                 o->od_quota_slave_dt = NULL;
7867                 GOTO(out_procfs, rc);
7868         }
7869
7870         RETURN(0);
7871
7872 out_procfs:
7873         osd_procfs_fini(o);
7874 out_scrub:
7875         osd_scrub_cleanup(env, o);
7876 out_site:
7877         lu_site_fini(&o->od_site);
7878 out_compat:
7879         osd_obj_map_fini(o);
7880 out_mnt:
7881         osd_umount(env, o);
7882 out:
7883         return rc;
7884 }
7885
7886 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7887                                           struct lu_device_type *t,
7888                                           struct lustre_cfg *cfg)
7889 {
7890         struct osd_device *o;
7891         int rc;
7892
7893         OBD_ALLOC_PTR(o);
7894         if (o == NULL)
7895                 return ERR_PTR(-ENOMEM);
7896
7897         rc = dt_device_init(&o->od_dt_dev, t);
7898         if (rc == 0) {
7899                 /*
7900                  * Because the ctx might be revived in dt_device_init,
7901                  * refill the env here
7902                  */
7903                 lu_env_refill((struct lu_env *)env);
7904                 rc = osd_device_init0(env, o, cfg);
7905                 if (rc)
7906                         dt_device_fini(&o->od_dt_dev);
7907         }
7908
7909         if (unlikely(rc != 0))
7910                 OBD_FREE_PTR(o);
7911
7912         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7913 }
7914
7915 static struct lu_device *osd_device_free(const struct lu_env *env,
7916                                          struct lu_device *d)
7917 {
7918         struct osd_device *o = osd_dev(d);
7919
7920         ENTRY;
7921
7922         /* XXX: make osd top device in order to release reference */
7923         d->ld_site->ls_top_dev = d;
7924         lu_site_purge(env, d->ld_site, -1);
7925         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7926                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7927                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7928         }
7929         lu_site_fini(&o->od_site);
7930         dt_device_fini(&o->od_dt_dev);
7931         OBD_FREE_PTR(o);
7932         RETURN(NULL);
7933 }
7934
7935 static int osd_process_config(const struct lu_env *env,
7936                               struct lu_device *d, struct lustre_cfg *cfg)
7937 {
7938         struct osd_device *o = osd_dev(d);
7939         ssize_t count;
7940         int rc;
7941
7942         ENTRY;
7943
7944         switch (cfg->lcfg_command) {
7945         case LCFG_SETUP:
7946                 rc = osd_mount(env, o, cfg);
7947                 break;
7948         case LCFG_CLEANUP:
7949                 /*
7950                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7951                  * that may happend if hit failure during mount process.
7952                  */
7953                 osd_index_backup(env, o, false);
7954                 lu_dev_del_linkage(d->ld_site, d);
7955                 rc = osd_shutdown(env, o);
7956                 break;
7957         case LCFG_PARAM:
7958                 LASSERT(&o->od_dt_dev);
7959                 count  = class_modify_config(cfg, PARAM_OSD,
7960                                              &o->od_dt_dev.dd_kobj);
7961                 if (count < 0)
7962                         count = class_modify_config(cfg, PARAM_OST,
7963                                                     &o->od_dt_dev.dd_kobj);
7964                 rc = count > 0 ? 0 : count;
7965                 break;
7966         case LCFG_PRE_CLEANUP:
7967                 osd_scrub_stop(o);
7968                 osd_index_backup(env, o,
7969                                  o->od_index_backup_policy != LIBP_NONE);
7970                 rc = 0;
7971                 break;
7972         default:
7973                 rc = -ENOSYS;
7974         }
7975
7976         RETURN(rc);
7977 }
7978
7979 static int osd_recovery_complete(const struct lu_env *env,
7980                                  struct lu_device *d)
7981 {
7982         struct osd_device *osd = osd_dev(d);
7983         int rc = 0;
7984
7985         ENTRY;
7986
7987         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7988                 RETURN(0);
7989
7990         /*
7991          * start qsd instance on recovery completion, this notifies the quota
7992          * slave code that we are about to process new requests now
7993          */
7994         rc = qsd_start(env, osd->od_quota_slave_dt);
7995         if (rc == 0 && osd->od_quota_slave_md != NULL)
7996                 rc = qsd_start(env, osd->od_quota_slave_md);
7997
7998         RETURN(rc);
7999 }
8000
8001 /*
8002  * we use exports to track all osd users
8003  */
8004 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8005                            struct obd_device *obd, struct obd_uuid *cluuid,
8006                            struct obd_connect_data *data, void *localdata)
8007 {
8008         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8009         struct lustre_handle conn;
8010         int rc;
8011
8012         ENTRY;
8013
8014         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8015
8016         rc = class_connect(&conn, obd, cluuid);
8017         if (rc)
8018                 RETURN(rc);
8019
8020         *exp = class_conn2export(&conn);
8021
8022         spin_lock(&osd->od_osfs_lock);
8023         osd->od_connects++;
8024         spin_unlock(&osd->od_osfs_lock);
8025
8026         RETURN(0);
8027 }
8028
8029 /*
8030  * once last export (we don't count self-export) disappeared
8031  * osd can be released
8032  */
8033 static int osd_obd_disconnect(struct obd_export *exp)
8034 {
8035         struct obd_device *obd = exp->exp_obd;
8036         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8037         int rc, release = 0;
8038
8039         ENTRY;
8040
8041         /* Only disconnect the underlying layers on the final disconnect. */
8042         spin_lock(&osd->od_osfs_lock);
8043         osd->od_connects--;
8044         if (osd->od_connects == 0)
8045                 release = 1;
8046         spin_unlock(&osd->od_osfs_lock);
8047
8048         rc = class_disconnect(exp); /* bz 9811 */
8049
8050         if (rc == 0 && release)
8051                 class_manual_cleanup(obd);
8052         RETURN(rc);
8053 }
8054
8055 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8056                        struct lu_device *dev)
8057 {
8058         struct osd_device *osd = osd_dev(dev);
8059         struct lr_server_data *lsd =
8060                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8061         int result = 0;
8062
8063         ENTRY;
8064
8065         if (osd->od_quota_slave_md != NULL) {
8066                 /* set up quota slave objects for inode */
8067                 result = qsd_prepare(env, osd->od_quota_slave_md);
8068                 if (result != 0)
8069                         RETURN(result);
8070         }
8071
8072         if (osd->od_quota_slave_dt != NULL) {
8073                 /* set up quota slave objects for block */
8074                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8075                 if (result != 0)
8076                         RETURN(result);
8077         }
8078
8079
8080         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8081 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8082                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8083                         osd->od_index_in_idif = 1;
8084                 } else {
8085                         osd->od_index_in_idif = 0;
8086                         result = osd_register_proc_index_in_idif(osd);
8087                         if (result != 0)
8088                                 RETURN(result);
8089                 }
8090 #else
8091                 osd->od_index_in_idif = 1;
8092 #endif
8093         }
8094
8095         result = osd_fid_init(env, osd);
8096
8097         RETURN(result);
8098 }
8099
8100 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8101                          struct lu_fid *fid, struct md_op_data *op_data)
8102 {
8103         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8104
8105         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8106 }
8107
8108 static const struct lu_object_operations osd_lu_obj_ops = {
8109         .loo_object_init      = osd_object_init,
8110         .loo_object_delete    = osd_object_delete,
8111         .loo_object_release   = osd_object_release,
8112         .loo_object_free      = osd_object_free,
8113         .loo_object_print     = osd_object_print,
8114         .loo_object_invariant = osd_object_invariant
8115 };
8116
8117 const struct lu_device_operations osd_lu_ops = {
8118         .ldo_object_alloc      = osd_object_alloc,
8119         .ldo_process_config    = osd_process_config,
8120         .ldo_recovery_complete = osd_recovery_complete,
8121         .ldo_prepare           = osd_prepare,
8122 };
8123
8124 static const struct lu_device_type_operations osd_device_type_ops = {
8125         .ldto_init = osd_type_init,
8126         .ldto_fini = osd_type_fini,
8127
8128         .ldto_start = osd_type_start,
8129         .ldto_stop  = osd_type_stop,
8130
8131         .ldto_device_alloc = osd_device_alloc,
8132         .ldto_device_free  = osd_device_free,
8133
8134         .ldto_device_init = osd_device_init,
8135         .ldto_device_fini = osd_device_fini
8136 };
8137
8138 static struct lu_device_type osd_device_type = {
8139         .ldt_tags     = LU_DEVICE_DT,
8140         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8141         .ldt_ops      = &osd_device_type_ops,
8142         .ldt_ctx_tags = LCT_LOCAL,
8143 };
8144
8145 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8146 {
8147         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8148         struct super_block *sb = osd_sb(osd);
8149
8150         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8151 }
8152
8153 /*
8154  * lprocfs legacy support.
8155  */
8156 static struct obd_ops osd_obd_device_ops = {
8157         .o_owner = THIS_MODULE,
8158         .o_connect      = osd_obd_connect,
8159         .o_disconnect   = osd_obd_disconnect,
8160         .o_fid_alloc    = osd_fid_alloc,
8161         .o_health_check = osd_health_check,
8162 };
8163
8164 static ssize_t track_declares_assert_show(struct kobject *kobj,
8165                                    struct attribute *attr,
8166                                    char *buf)
8167 {
8168         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8169 }
8170
8171 static ssize_t track_declares_assert_store(struct kobject *kobj,
8172                                            struct attribute *attr,
8173                                            const char *buffer, size_t count)
8174 {
8175         bool track_declares_assert;
8176         int rc;
8177
8178         rc = kstrtobool(buffer, &track_declares_assert);
8179         if (rc)
8180                 return rc;
8181
8182         ldiskfs_track_declares_assert = track_declares_assert;
8183
8184         return count;
8185 }
8186 LUSTRE_RW_ATTR(track_declares_assert);
8187
8188 static int __init osd_init(void)
8189 {
8190         struct kobject *kobj;
8191         int rc;
8192
8193         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8194 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8195         /* please, try to keep osd_thread_info smaller than a page */
8196         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8197 #endif
8198
8199         osd_oi_mod_init();
8200
8201         rc = lu_kmem_init(ldiskfs_caches);
8202         if (rc)
8203                 return rc;
8204
8205 #ifdef CONFIG_KALLSYMS
8206         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8207         priv_dev_check_rdonly =
8208                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8209 #endif
8210
8211         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8212                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8213         if (rc) {
8214                 lu_kmem_fini(ldiskfs_caches);
8215                 return rc;
8216         }
8217
8218         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8219         if (kobj) {
8220                 rc = sysfs_create_file(kobj,
8221                                        &lustre_attr_track_declares_assert.attr);
8222                 kobject_put(kobj);
8223                 if (rc) {
8224                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8225                         rc = 0;
8226                 }
8227         }
8228         return rc;
8229 }
8230
8231 static void __exit osd_exit(void)
8232 {
8233         struct kobject *kobj;
8234
8235         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8236         if (kobj) {
8237                 sysfs_remove_file(kobj,
8238                                   &lustre_attr_track_declares_assert.attr);
8239                 kobject_put(kobj);
8240         }
8241         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8242         lu_kmem_fini(ldiskfs_caches);
8243 }
8244
8245 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8246 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8247 MODULE_VERSION(LUSTRE_VERSION_STRING);
8248 MODULE_LICENSE("GPL");
8249
8250 module_init(osd_init);
8251 module_exit(osd_exit);