lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 499                               "maximum entry limit\n",
 500                               osd_name(osd), parent->i_ino, PFID(fid),
 501                               rc == -ENOSPC ? "has reached" : "is approaching");
 502                 else
 503                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 504                               "maximum entry limit\n",
 505                               osd_name(osd), parent->i_ino,
 506                               rc == -ENOSPC ? "has reached" : "is approaching");
 507
 508                 /* ignore such error now */
 509                 if (rc == -ENOBUFS)
 510                         rc = 0;
 511         }
 512
 513         return rc;
 514 }
 515
 516
 517 struct inode *
 518 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 519              struct osd_inode_id *id, struct lu_fid *fid)
 520 {
 521         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 522         struct inode *inode;
 523         int rc;
 524
 525         inode = osd_iget(info, dev, id);
 526         if (IS_ERR(inode))
 527                 return inode;
 528
 529         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 530         if (!rc) {
 531                 *fid = loa->loa_lma.lma_self_fid;
 532         } else if (rc == -ENODATA) {
 533                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 534                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 535                 else
 536                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 537         } else {
 538                 iput(inode);
 539                 inode = ERR_PTR(rc);
 540         }
 541         return inode;
 542 }
 543
 544 static struct inode *osd_iget_check(struct osd_thread_info *info,
 545                                     struct osd_device *dev,
 546                                     const struct lu_fid *fid,
 547                                     struct osd_inode_id *id,
 548                                     bool trusted)
 549 {
 550         struct inode *inode;
 551         int rc = 0;
 552
 553         ENTRY;
 554
 555         /*
 556          * The cached OI mapping is trustable. If we cannot locate the inode
 557          * via the cached OI mapping, then return the failure to the caller
 558          * directly without further OI checking.
 559          */
 560
 561 again:
 562         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_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         if (result)
1132                 GOTO(out, result);
1133
1134 iget:
1135         obj->oo_inode = NULL;
1136         /* for later passes through checks, not true on first pass */
1137         if (!IS_ERR_OR_NULL(inode))
1138                 iput(inode);
1139
1140         inode = osd_iget_check(info, dev, fid, id, trusted);
1141         if (!IS_ERR(inode)) {
1142                 obj->oo_inode = inode;
1143                 result = 0;
1144                 if (remote)
1145                         goto trigger;
1146
1147                 goto check_lma;
1148         }
1149
1150         result = PTR_ERR(inode);
1151         if (result == -ENOENT || result == -ESTALE)
1152                 GOTO(out, result = 0);
1153
1154         if (result != -EREMCHG)
1155                 GOTO(out, result);
1156
1157 trigger:
1158         /*
1159          * We still have chance to get the valid inode: for the
1160          * object which is referenced by remote name entry, the
1161          * object on the local MDT will be linked under the dir
1162          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1163          *
1164          * We do not know whether the object for the given FID
1165          * is referenced by some remote name entry or not, and
1166          * especially for DNE II, a multiple-linked object may
1167          * have many name entries reside on many MDTs.
1168          *
1169          * To simplify the operation, OSD will not distinguish
1170          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1171          * only happened for the RPC from other MDT during the
1172          * OI scrub, or for the client side RPC with FID only,
1173          * such as FID to path, or from old connected client.
1174          */
1175         if (!remote) {
1176                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1177                 if (!rc1) {
1178                         remote = true;
1179                         trusted = true;
1180                         flags |= SS_AUTO_PARTIAL;
1181                         flags &= ~SS_AUTO_FULL;
1182                         goto iget;
1183                 }
1184         }
1185
1186         if (thread_is_running(&scrub->os_thread)) {
1187                 if (scrub->os_partial_scan && !scrub->os_in_join)
1188                         goto join;
1189
1190                 if (IS_ERR_OR_NULL(inode) || result)
1191                         GOTO(out, result = -EINPROGRESS);
1192
1193                 LASSERT(remote);
1194                 LASSERT(obj->oo_inode == inode);
1195
1196                 osd_add_oi_cache(info, dev, id, fid);
1197                 osd_oii_insert(dev, oic, true);
1198                 goto found;
1199         }
1200
1201         if (dev->od_auto_scrub_interval == AS_NEVER) {
1202                 if (!remote)
1203                         GOTO(out, result = -EREMCHG);
1204
1205                 LASSERT(!result);
1206                 LASSERT(obj->oo_inode == inode);
1207
1208                 osd_add_oi_cache(info, dev, id, fid);
1209                 goto found;
1210         }
1211
1212 join:
1213         rc1 = osd_scrub_start(env, dev, flags);
1214         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1215                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1216         if (rc1 && rc1 != -EALREADY)
1217                 GOTO(out, result = -EREMCHG);
1218
1219         if (IS_ERR_OR_NULL(inode) || result)
1220                 GOTO(out, result = -EINPROGRESS);
1221
1222         LASSERT(remote);
1223         LASSERT(obj->oo_inode == inode);
1224
1225         osd_add_oi_cache(info, dev, id, fid);
1226         osd_oii_insert(dev, oic, true);
1227         goto found;
1228
1229 check_lma:
1230         checked = true;
1231         if (unlikely(obj->oo_header))
1232                 goto found;
1233
1234         result = osd_check_lma(env, obj);
1235         if (!result)
1236                 goto found;
1237
1238         LASSERTF(id->oii_ino == inode->i_ino &&
1239                  id->oii_gen == inode->i_generation,
1240                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1241                  PFID(fid), id->oii_ino, id->oii_gen,
1242                  inode->i_ino, inode->i_generation);
1243
1244         saved_ino = inode->i_ino;
1245         saved_gen = inode->i_generation;
1246
1247         if (unlikely(result == -ENODATA)) {
1248                 /*
1249                  * If the OI scrub updated the OI mapping by race, it
1250                  * must be valid. Trust the inode that has no LMA EA.
1251                  */
1252                 if (updated)
1253                         goto found;
1254
1255                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1256                 if (!result) {
1257                         /*
1258                          * The OI mapping is still there, the inode is still
1259                          * valid. It is just becaues the inode has no LMA EA.
1260                          */
1261                         if (saved_ino == id->oii_ino &&
1262                             saved_gen == id->oii_gen)
1263                                 goto found;
1264
1265                         /*
1266                          * It is the OI scrub updated the OI mapping by race.
1267                          * The new OI mapping must be valid.
1268                          */
1269                         trusted = true;
1270                         updated = true;
1271                         goto iget;
1272                 }
1273
1274                 /*
1275                  * "result == -ENOENT" means that the OI mappinghas been
1276                  * removed by race, so the inode belongs to other object.
1277                  *
1278                  * Others error can be returned  directly.
1279                  */
1280                 if (result == -ENOENT) {
1281                         LASSERT(trusted);
1282
1283                         obj->oo_inode = NULL;
1284                         result = 0;
1285                 }
1286         }
1287
1288         if (result != -EREMCHG)
1289                 GOTO(out, result);
1290
1291         LASSERT(!updated);
1292
1293         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1294         /*
1295          * "result == -ENOENT" means the cached OI mapping has been removed
1296          * from the OI file by race, above inode belongs to other object.
1297          */
1298         if (result == -ENOENT) {
1299                 LASSERT(trusted);
1300
1301                 obj->oo_inode = NULL;
1302                 GOTO(out, result = 0);
1303         }
1304
1305         if (result)
1306                 GOTO(out, result);
1307
1308         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1309                 result = -EREMCHG;
1310                 goto trigger;
1311         }
1312
1313         /*
1314          * It is the OI scrub updated the OI mapping by race.
1315          * The new OI mapping must be valid.
1316          */
1317         trusted = true;
1318         updated = true;
1319         goto iget;
1320
1321 found:
1322         if (!checked) {
1323                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1324                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1325
1326                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1327                 if (!result) {
1328                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1329                             dev->od_is_ost)
1330                                 obj->oo_pfid_in_lma = 1;
1331                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1332                             !dev->od_is_ost)
1333                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1334                 } else if (result != -ENODATA) {
1335                         GOTO(out, result);
1336                 }
1337         }
1338
1339         obj->oo_compat_dot_created = 1;
1340         obj->oo_compat_dotdot_created = 1;
1341
1342         if (S_ISDIR(inode->i_mode) &&
1343             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1344                 osd_check_lmv(info, dev, inode, oic);
1345
1346         result = osd_attach_jinode(inode);
1347         if (result)
1348                 GOTO(out, result);
1349
1350         if (!ldiskfs_pdo)
1351                 GOTO(out, result = 0);
1352
1353         LASSERT(!obj->oo_hl_head);
1354         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1355
1356         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1357
1358 out:
1359         if (result || !obj->oo_inode) {
1360                 if (!IS_ERR_OR_NULL(inode))
1361                         iput(inode);
1362
1363                 obj->oo_inode = NULL;
1364                 if (trusted)
1365                         fid_zero(&oic->oic_fid);
1366         }
1367
1368         LINVRNT(osd_invariant(obj));
1369         return result;
1370 }
1371
1372 /*
1373  * Concurrency: shouldn't matter.
1374  */
1375 static void osd_object_init0(struct osd_object *obj)
1376 {
1377         LASSERT(obj->oo_inode != NULL);
1378         obj->oo_dt.do_body_ops = &osd_body_ops;
1379         obj->oo_dt.do_lu.lo_header->loh_attr |=
1380                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1381 }
1382
1383 /*
1384  * Concurrency: no concurrent access is possible that early in object
1385  * life-cycle.
1386  */
1387 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1388                            const struct lu_object_conf *conf)
1389 {
1390         struct osd_object *obj = osd_obj(l);
1391         int result;
1392
1393         LINVRNT(osd_invariant(obj));
1394
1395         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1396                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1397                 l->lo_header->loh_attr |= LOHA_EXISTS;
1398                 return 0;
1399         }
1400
1401         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1402         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1403         if (result == 0 && obj->oo_inode != NULL) {
1404                 struct osd_thread_info *oti = osd_oti_get(env);
1405                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1406
1407                 osd_object_init0(obj);
1408                 if (unlikely(obj->oo_header))
1409                         return 0;
1410
1411                 result = osd_get_lma(oti, obj->oo_inode,
1412                                      &oti->oti_obj_dentry, loa);
1413                 if (!result) {
1414                         /*
1415                          * Convert LMAI flags to lustre LMA flags
1416                          * and cache it to oo_lma_flags
1417                          */
1418                         obj->oo_lma_flags =
1419                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1420                 } else if (result == -ENODATA) {
1421                         result = 0;
1422                 }
1423         }
1424
1425         LINVRNT(osd_invariant(obj));
1426         return result;
1427 }
1428
1429 /*
1430  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1431  * The left part is for value, binary mode.
1432  */
1433 struct osd_xattr_entry {
1434         struct list_head        oxe_list;
1435         size_t                  oxe_len;
1436         size_t                  oxe_namelen;
1437         bool                    oxe_exist;
1438         struct rcu_head         oxe_rcu;
1439         char                    oxe_buf[0];
1440 };
1441
1442 static int osd_oxc_get(struct osd_object *obj, const char *name,
1443                        struct lu_buf *buf)
1444 {
1445         struct osd_xattr_entry *tmp;
1446         struct osd_xattr_entry *oxe = NULL;
1447         size_t namelen = strlen(name);
1448         int rc;
1449
1450         ENTRY;
1451
1452         rcu_read_lock();
1453         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1454                 if (namelen == tmp->oxe_namelen &&
1455                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1456                         oxe = tmp;
1457                         break;
1458                 }
1459         }
1460
1461         if (oxe == NULL)
1462                 GOTO(out, rc = -ENOENT);
1463
1464         if (!oxe->oxe_exist)
1465                 GOTO(out, rc = -ENODATA);
1466
1467         /* vallen */
1468         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1469         LASSERT(rc > 0);
1470
1471         if (buf->lb_buf == NULL)
1472                 GOTO(out, rc);
1473
1474         if (buf->lb_len < rc)
1475                 GOTO(out, rc = -ERANGE);
1476
1477         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1478         EXIT;
1479 out:
1480         rcu_read_unlock();
1481
1482         return rc;
1483 }
1484
1485 static void osd_oxc_free(struct rcu_head *head)
1486 {
1487         struct osd_xattr_entry *oxe;
1488
1489         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1490         OBD_FREE(oxe, oxe->oxe_len);
1491 }
1492
1493 static void osd_oxc_add(struct osd_object *obj, const char *name,
1494                         const char *buf, int buflen)
1495 {
1496         struct osd_xattr_entry *oxe;
1497         struct osd_xattr_entry *old = NULL;
1498         struct osd_xattr_entry *tmp;
1499         size_t namelen = strlen(name);
1500         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1501
1502         OBD_ALLOC(oxe, len);
1503         if (oxe == NULL)
1504                 return;
1505
1506         INIT_LIST_HEAD(&oxe->oxe_list);
1507         oxe->oxe_len = len;
1508         oxe->oxe_namelen = namelen;
1509         memcpy(oxe->oxe_buf, name, namelen);
1510         if (buflen > 0) {
1511                 LASSERT(buf != NULL);
1512                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1513                 oxe->oxe_exist = true;
1514         } else {
1515                 oxe->oxe_exist = false;
1516         }
1517
1518         /* this should be rarely called, just remove old and add new */
1519         spin_lock(&obj->oo_guard);
1520         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1521                 if (namelen == tmp->oxe_namelen &&
1522                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1523                         old = tmp;
1524                         break;
1525                 }
1526         }
1527         if (old != NULL) {
1528                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1529                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1530         } else {
1531                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1532         }
1533         spin_unlock(&obj->oo_guard);
1534 }
1535
1536 static void osd_oxc_del(struct osd_object *obj, const char *name)
1537 {
1538         struct osd_xattr_entry *oxe;
1539         size_t namelen = strlen(name);
1540
1541         spin_lock(&obj->oo_guard);
1542         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1543                 if (namelen == oxe->oxe_namelen &&
1544                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1545                         list_del_rcu(&oxe->oxe_list);
1546                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1547                         break;
1548                 }
1549         }
1550         spin_unlock(&obj->oo_guard);
1551 }
1552
1553 static void osd_oxc_fini(struct osd_object *obj)
1554 {
1555         struct osd_xattr_entry *oxe, *next;
1556
1557         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1558                 list_del(&oxe->oxe_list);
1559                 OBD_FREE(oxe, oxe->oxe_len);
1560         }
1561 }
1562
1563 /*
1564  * Concurrency: no concurrent access is possible that late in object
1565  * life-cycle.
1566  */
1567 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1568 {
1569         struct osd_object *obj = osd_obj(l);
1570         struct lu_object_header *h = obj->oo_header;
1571
1572         LINVRNT(osd_invariant(obj));
1573
1574         osd_oxc_fini(obj);
1575         dt_object_fini(&obj->oo_dt);
1576         if (obj->oo_hl_head != NULL)
1577                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1578         OBD_FREE_PTR(obj);
1579         if (unlikely(h)) {
1580                 lu_object_header_fini(h);
1581                 OBD_FREE_PTR(h);
1582         }
1583 }
1584
1585 /*
1586  * Concurrency: no concurrent access is possible that late in object
1587  * life-cycle.
1588  */
1589 static void osd_index_fini(struct osd_object *o)
1590 {
1591         struct iam_container *bag;
1592
1593         if (o->oo_dir != NULL) {
1594                 bag = &o->oo_dir->od_container;
1595                 if (o->oo_inode != NULL) {
1596                         if (bag->ic_object == o->oo_inode)
1597                                 iam_container_fini(bag);
1598                 }
1599                 OBD_FREE_PTR(o->oo_dir);
1600                 o->oo_dir = NULL;
1601         }
1602 }
1603
1604 /*
1605  * Concurrency: no concurrent access is possible that late in object
1606  * life-cycle (for all existing callers, that is. New callers have to provide
1607  * their own locking.)
1608  */
1609 static int osd_inode_unlinked(const struct inode *inode)
1610 {
1611         return inode->i_nlink == 0;
1612 }
1613
1614 enum {
1615         OSD_TXN_OI_DELETE_CREDITS    = 20,
1616         OSD_TXN_INODE_DELETE_CREDITS = 20
1617 };
1618
1619 /*
1620  * Journal
1621  */
1622
1623 #if OSD_THANDLE_STATS
1624 /**
1625  * Set time when the handle is allocated
1626  */
1627 static void osd_th_alloced(struct osd_thandle *oth)
1628 {
1629         oth->oth_alloced = ktime_get();
1630 }
1631
1632 /**
1633  * Set time when the handle started
1634  */
1635 static void osd_th_started(struct osd_thandle *oth)
1636 {
1637         oth->oth_started = ktime_get();
1638 }
1639
1640 /**
1641  * Check whether the we deal with this handle for too long.
1642  */
1643 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1644                                 ktime_t alloced, ktime_t started,
1645                                 ktime_t closed)
1646 {
1647         ktime_t now = ktime_get();
1648
1649         LASSERT(dev != NULL);
1650
1651         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1652                             ktime_us_delta(started, alloced));
1653         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1654                             ktime_us_delta(closed, started));
1655         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1656                             ktime_us_delta(now, closed));
1657
1658         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1659                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1660                                 oth, now, alloced, started, closed);
1661                 libcfs_debug_dumpstack(NULL);
1662         }
1663 }
1664
1665 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1666 {                                                                       \
1667         ktime_t __closed = ktime_get();                                 \
1668         ktime_t __alloced = oth->oth_alloced;                           \
1669         ktime_t __started = oth->oth_started;                           \
1670                                                                         \
1671         expr;                                                           \
1672         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1673 }
1674
1675 #else /* OSD_THANDLE_STATS */
1676
1677 #define osd_th_alloced(h)                  do {} while(0)
1678 #define osd_th_started(h)                  do {} while(0)
1679 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1680
1681 #endif /* OSD_THANDLE_STATS */
1682
1683 /*
1684  * Concurrency: doesn't access mutable data.
1685  */
1686 static int osd_param_is_not_sane(const struct osd_device *dev,
1687                                  const struct thandle *th)
1688 {
1689         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1690
1691         return oh->ot_credits > osd_transaction_size(dev);
1692 }
1693
1694 /*
1695  * Concurrency: shouldn't matter.
1696  */
1697 static void osd_trans_commit_cb(struct super_block *sb,
1698                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1699 {
1700         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1701         struct thandle *th = &oh->ot_super;
1702         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1703         struct dt_txn_commit_cb *dcb, *tmp;
1704
1705         LASSERT(oh->ot_handle == NULL);
1706
1707         if (error)
1708                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1709
1710         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1711         /* call per-transaction callbacks if any */
1712         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1713                                  dcb_linkage) {
1714                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1715                          "commit callback entry: magic=%x name='%s'\n",
1716                          dcb->dcb_magic, dcb->dcb_name);
1717                 list_del_init(&dcb->dcb_linkage);
1718                 dcb->dcb_func(NULL, th, dcb, error);
1719         }
1720
1721         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1722         lu_device_put(lud);
1723         th->th_dev = NULL;
1724
1725         OBD_FREE_PTR(oh);
1726 }
1727
1728 #ifndef HAVE_SB_START_WRITE
1729 # define sb_start_write(sb) do {} while (0)
1730 # define sb_end_write(sb) do {} while (0)
1731 #endif
1732
1733 static struct thandle *osd_trans_create(const struct lu_env *env,
1734                                         struct dt_device *d)
1735 {
1736         struct osd_thread_info *oti = osd_oti_get(env);
1737         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1738         struct osd_thandle *oh;
1739         struct thandle *th;
1740
1741         ENTRY;
1742
1743         if (d->dd_rdonly) {
1744                 CERROR("%s: someone try to start transaction under "
1745                        "readonly mode, should be disabled.\n",
1746                        osd_name(osd_dt_dev(d)));
1747                 dump_stack();
1748                 RETURN(ERR_PTR(-EROFS));
1749         }
1750
1751         /* on pending IO in this thread should left from prev. request */
1752         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1753
1754         sb_start_write(osd_sb(osd_dt_dev(d)));
1755
1756         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1757         if (!oh) {
1758                 sb_end_write(osd_sb(osd_dt_dev(d)));
1759                 RETURN(ERR_PTR(-ENOMEM));
1760         }
1761
1762         oh->ot_quota_trans = &oti->oti_quota_trans;
1763         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1764         th = &oh->ot_super;
1765         th->th_dev = d;
1766         th->th_result = 0;
1767         oh->ot_credits = 0;
1768         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1769         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1770         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1771         osd_th_alloced(oh);
1772
1773         memset(oti->oti_declare_ops, 0,
1774                sizeof(oti->oti_declare_ops));
1775         memset(oti->oti_declare_ops_cred, 0,
1776                sizeof(oti->oti_declare_ops_cred));
1777         memset(oti->oti_declare_ops_used, 0,
1778                sizeof(oti->oti_declare_ops_used));
1779
1780         oti->oti_ins_cache_depth++;
1781
1782         RETURN(th);
1783 }
1784
1785 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1786 {
1787         struct osd_thread_info *oti = osd_oti_get(env);
1788         struct osd_thandle *oh;
1789
1790         oh = container_of0(th, struct osd_thandle, ot_super);
1791         LASSERT(oh != NULL);
1792
1793         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1794               oti->oti_declare_ops[OSD_OT_CREATE],
1795               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1796               oti->oti_declare_ops_used[OSD_OT_CREATE],
1797               oti->oti_declare_ops[OSD_OT_DESTROY],
1798               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1799               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1800         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1801               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1802               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1803               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1805               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1806               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1807         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1808               oti->oti_declare_ops[OSD_OT_WRITE],
1809               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1810               oti->oti_declare_ops_used[OSD_OT_WRITE],
1811               oti->oti_declare_ops[OSD_OT_PUNCH],
1812               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1813               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1814               oti->oti_declare_ops[OSD_OT_QUOTA],
1815               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1816               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1817         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_INSERT],
1819               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1820               oti->oti_declare_ops_used[OSD_OT_INSERT],
1821               oti->oti_declare_ops[OSD_OT_DELETE],
1822               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1823               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1824         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1825               oti->oti_declare_ops[OSD_OT_REF_ADD],
1826               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1827               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops[OSD_OT_REF_DEL],
1829               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1830               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1831 }
1832
1833 /*
1834  * Concurrency: shouldn't matter.
1835  */
1836 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1837                            struct thandle *th)
1838 {
1839         struct osd_thread_info *oti = osd_oti_get(env);
1840         struct osd_device *dev = osd_dt_dev(d);
1841         handle_t *jh;
1842         struct osd_thandle *oh;
1843         int rc;
1844
1845         ENTRY;
1846
1847         LASSERT(current->journal_info == NULL);
1848
1849         oh = container_of0(th, struct osd_thandle, ot_super);
1850         LASSERT(oh != NULL);
1851         LASSERT(oh->ot_handle == NULL);
1852
1853         rc = dt_txn_hook_start(env, d, th);
1854         if (rc != 0)
1855                 GOTO(out, rc);
1856
1857         if (unlikely(osd_param_is_not_sane(dev, th))) {
1858                 static unsigned long last_printed;
1859                 static int last_credits;
1860
1861                 /*
1862                  * don't make noise on a tiny testing systems
1863                  * actual credits misuse will be caught anyway
1864                  */
1865                 if (last_credits != oh->ot_credits &&
1866                     time_after(jiffies, last_printed +
1867                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1868                     osd_transaction_size(dev) > 512) {
1869                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1870                               oh->ot_credits, osd_transaction_size(dev));
1871                         osd_trans_dump_creds(env, th);
1872                         libcfs_debug_dumpstack(NULL);
1873                         last_credits = oh->ot_credits;
1874                         last_printed = jiffies;
1875                 }
1876                 /*
1877                  * XXX Limit the credits to 'max_transaction_buffers', and
1878                  *     let the underlying filesystem to catch the error if
1879                  *     we really need so many credits.
1880                  *
1881                  *     This should be removed when we can calculate the
1882                  *     credits precisely.
1883                  */
1884                 oh->ot_credits = osd_transaction_size(dev);
1885         } else if (ldiskfs_track_declares_assert != 0) {
1886                 /*
1887                  * reserve few credits to prevent an assertion in JBD
1888                  * our debugging mechanism will be able to detected
1889                  * overuse. this can help to debug single-update
1890                  * transactions
1891                  */
1892                 oh->ot_credits += 10;
1893                 if (unlikely(osd_param_is_not_sane(dev, th)))
1894                         oh->ot_credits = osd_transaction_size(dev);
1895         }
1896
1897         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1898                 GOTO(out, rc = -EIO);
1899
1900         /*
1901          * XXX temporary stuff. Some abstraction layer should
1902          * be used.
1903          */
1904         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1905         osd_th_started(oh);
1906         if (!IS_ERR(jh)) {
1907                 oh->ot_handle = jh;
1908                 LASSERT(oti->oti_txns == 0);
1909
1910                 lu_device_get(&d->dd_lu_dev);
1911                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1912                               "osd-tx", th);
1913                 oti->oti_txns++;
1914                 rc = 0;
1915         } else {
1916                 rc = PTR_ERR(jh);
1917         }
1918 out:
1919         RETURN(rc);
1920 }
1921
1922 static int osd_seq_exists(const struct lu_env *env,
1923                           struct osd_device *osd, u64 seq)
1924 {
1925         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1926         struct seq_server_site *ss = osd_seq_site(osd);
1927         int rc;
1928
1929         ENTRY;
1930
1931         LASSERT(ss != NULL);
1932         LASSERT(ss->ss_server_fld != NULL);
1933
1934         rc = osd_fld_lookup(env, osd, seq, range);
1935         if (rc != 0) {
1936                 if (rc != -ENOENT)
1937                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1938                                osd_name(osd), seq, rc);
1939                 RETURN(0);
1940         }
1941
1942         RETURN(ss->ss_node_id == range->lsr_index);
1943 }
1944
1945 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1946 {
1947         struct dt_txn_commit_cb *dcb;
1948         struct dt_txn_commit_cb *tmp;
1949
1950         /* call per-transaction stop callbacks if any */
1951         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1952                                  dcb_linkage) {
1953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1954                          "commit callback entry: magic=%x name='%s'\n",
1955                          dcb->dcb_magic, dcb->dcb_name);
1956                 list_del_init(&dcb->dcb_linkage);
1957                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1958         }
1959 }
1960
1961 /*
1962  * Concurrency: shouldn't matter.
1963  */
1964 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1965                           struct thandle *th)
1966 {
1967         struct osd_thread_info *oti = osd_oti_get(env);
1968         struct osd_thandle *oh;
1969         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1970         struct osd_device *osd = osd_dt_dev(th->th_dev);
1971         struct qsd_instance *qsd = osd_def_qsd(osd);
1972         struct lquota_trans *qtrans;
1973         struct list_head truncates = LIST_HEAD_INIT(truncates);
1974         int rc = 0, remove_agents = 0;
1975
1976         ENTRY;
1977
1978         oh = container_of0(th, struct osd_thandle, ot_super);
1979
1980         remove_agents = oh->ot_remove_agents;
1981
1982         qtrans = oh->ot_quota_trans;
1983         oh->ot_quota_trans = NULL;
1984
1985         /* move locks to local list, stop tx, execute truncates */
1986         list_splice(&oh->ot_trunc_locks, &truncates);
1987
1988         if (oh->ot_handle != NULL) {
1989                 int rc2;
1990
1991                 handle_t *hdl = oh->ot_handle;
1992
1993                 /*
1994                  * add commit callback
1995                  * notice we don't do this in osd_trans_start()
1996                  * as underlying transaction can change during truncate
1997                  */
1998                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1999                                              &oh->ot_jcb);
2000
2001                 LASSERT(oti->oti_txns == 1);
2002                 oti->oti_txns--;
2003
2004                 rc = dt_txn_hook_stop(env, th);
2005                 if (rc != 0)
2006                         CERROR("%s: failed in transaction hook: rc = %d\n",
2007                                osd_name(osd), rc);
2008
2009                 osd_trans_stop_cb(oh, rc);
2010                 /* hook functions might modify th_sync */
2011                 hdl->h_sync = th->th_sync;
2012
2013                 oh->ot_handle = NULL;
2014                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2015                 if (rc2 != 0)
2016                         CERROR("%s: failed to stop transaction: rc = %d\n",
2017                                osd_name(osd), rc2);
2018                 if (!rc)
2019                         rc = rc2;
2020
2021                 osd_process_truncates(&truncates);
2022         } else {
2023                 osd_trans_stop_cb(oh, th->th_result);
2024                 OBD_FREE_PTR(oh);
2025         }
2026
2027         osd_trunc_unlock_all(&truncates);
2028
2029         /* inform the quota slave device that the transaction is stopping */
2030         qsd_op_end(env, qsd, qtrans);
2031
2032         /*
2033          * as we want IO to journal and data IO be concurrent, we don't block
2034          * awaiting data IO completion in osd_do_bio(), instead we wait here
2035          * once transaction is submitted to the journal. all reqular requests
2036          * don't do direct IO (except read/write), thus this wait_event becomes
2037          * no-op for them.
2038          *
2039          * IMPORTANT: we have to wait till any IO submited by the thread is
2040          * completed otherwise iobuf may be corrupted by different request
2041          */
2042         wait_event(iobuf->dr_wait,
2043                        atomic_read(&iobuf->dr_numreqs) == 0);
2044         osd_fini_iobuf(osd, iobuf);
2045         if (!rc)
2046                 rc = iobuf->dr_error;
2047
2048         if (unlikely(remove_agents != 0))
2049                 osd_process_scheduled_agent_removals(env, osd);
2050
2051         oti->oti_ins_cache_depth--;
2052         /* reset OI cache for safety */
2053         if (oti->oti_ins_cache_depth == 0)
2054                 oti->oti_ins_cache_used = 0;
2055
2056         sb_end_write(osd_sb(osd));
2057
2058         RETURN(rc);
2059 }
2060
2061 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2062 {
2063         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2064                                                ot_super);
2065
2066         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2067         LASSERT(&dcb->dcb_func != NULL);
2068         if (dcb->dcb_flags & DCB_TRANS_STOP)
2069                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2070         else
2071                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2072
2073         return 0;
2074 }
2075
2076 /*
2077  * Called just before object is freed. Releases all resources except for
2078  * object itself (that is released by osd_object_free()).
2079  *
2080  * Concurrency: no concurrent access is possible that late in object
2081  * life-cycle.
2082  */
2083 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2084 {
2085         struct osd_object *obj = osd_obj(l);
2086         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2087         struct inode *inode = obj->oo_inode;
2088         __u64 projid;
2089         qid_t uid;
2090         qid_t gid;
2091
2092         LINVRNT(osd_invariant(obj));
2093
2094         /*
2095          * If object is unlinked remove fid->ino mapping from object index.
2096          */
2097
2098         osd_index_fini(obj);
2099
2100         if (!inode)
2101                 return;
2102
2103         uid = i_uid_read(inode);
2104         gid = i_gid_read(inode);
2105         projid = i_projid_read(inode);
2106
2107         obj->oo_inode = NULL;
2108         iput(inode);
2109
2110         /* do not rebalance quota if the caller needs to release memory
2111          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2112          * transaction and risk to deadlock - LU-12178 */
2113         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2114                 return;
2115
2116         if (!obj->oo_header && qsd) {
2117                 struct osd_thread_info *info = osd_oti_get(env);
2118                 struct lquota_id_info *qi = &info->oti_qi;
2119
2120                 /* Release granted quota to master if necessary */
2121                 qi->lqi_id.qid_uid = uid;
2122                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2123
2124                 qi->lqi_id.qid_uid = gid;
2125                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2126
2127                 qi->lqi_id.qid_uid = projid;
2128                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2129         }
2130 }
2131
2132 /*
2133  * Concurrency: ->loo_object_release() is called under site spin-lock.
2134  */
2135 static void osd_object_release(const struct lu_env *env,
2136                                struct lu_object *l)
2137 {
2138         struct osd_object *o = osd_obj(l);
2139
2140         /*
2141          * nobody should be releasing a non-destroyed object with nlink=0
2142          * the API allows this, but ldiskfs doesn't like and then report
2143          * this inode as deleted
2144          */
2145         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2146                   o->oo_inode->i_nlink == 0));
2147 }
2148
2149 /*
2150  * Concurrency: shouldn't matter.
2151  */
2152 static int osd_object_print(const struct lu_env *env, void *cookie,
2153                             lu_printer_t p, const struct lu_object *l)
2154 {
2155         struct osd_object *o = osd_obj(l);
2156         struct iam_descr *d;
2157
2158         if (o->oo_dir != NULL)
2159                 d = o->oo_dir->od_container.ic_descr;
2160         else
2161                 d = NULL;
2162         return (*p)(env, cookie,
2163                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2164                     o, o->oo_inode,
2165                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2166                     o->oo_inode ? o->oo_inode->i_generation : 0,
2167                     d ? d->id_ops->id_name : "plain");
2168 }
2169
2170 /*
2171  * Concurrency: shouldn't matter.
2172  */
2173 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2174                 struct obd_statfs *sfs)
2175 {
2176         struct osd_device *osd = osd_dt_dev(d);
2177         struct super_block *sb = osd_sb(osd);
2178         struct kstatfs *ksfs;
2179         __u64 reserved;
2180         int result = 0;
2181
2182         if (unlikely(osd->od_mnt == NULL))
2183                 return -EINPROGRESS;
2184
2185         /* osd_lproc.c call this without env, allocate ksfs for that case */
2186         if (unlikely(env == NULL)) {
2187                 OBD_ALLOC_PTR(ksfs);
2188                 if (ksfs == NULL)
2189                         return -ENOMEM;
2190         } else {
2191                 ksfs = &osd_oti_get(env)->oti_ksfs;
2192         }
2193
2194         result = sb->s_op->statfs(sb->s_root, ksfs);
2195         if (result)
2196                 goto out;
2197
2198         statfs_pack(sfs, ksfs);
2199         if (unlikely(sb->s_flags & MS_RDONLY))
2200                 sfs->os_state |= OS_STATE_READONLY;
2201
2202         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2203
2204         if (ldiskfs_has_feature_extents(sb))
2205                 sfs->os_maxbytes = sb->s_maxbytes;
2206         else
2207                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2208
2209         /*
2210          * Reserve some space so to avoid fragmenting the filesystem too much.
2211          * Fragmentation not only impacts performance, but can also increase
2212          * metadata overhead significantly, causing grant calculation to be
2213          * wrong.
2214          *
2215          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2216          */
2217         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2218         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2219         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2220                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2221
2222         sfs->os_blocks -= reserved;
2223         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2224         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2225
2226 out:
2227         if (unlikely(env == NULL))
2228                 OBD_FREE_PTR(ksfs);
2229         return result;
2230 }
2231
2232 /**
2233  * Estimate space needed for file creations. We assume the largest filename
2234  * which is 2^64 - 1, hence a filename of 20 chars.
2235  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2236  */
2237 #ifdef __LDISKFS_DIR_REC_LEN
2238 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2239 #else
2240 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2241 #endif
2242
2243 /*
2244  * Concurrency: doesn't access mutable data.
2245  */
2246 static void osd_conf_get(const struct lu_env *env,
2247                          const struct dt_device *dev,
2248                          struct dt_device_param *param)
2249 {
2250         struct osd_device *d = osd_dt_dev(dev);
2251         struct super_block *sb = osd_sb(d);
2252         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2253         const char *name;
2254         int ea_overhead;
2255
2256         /*
2257          * XXX should be taken from not-yet-existing fs abstraction layer.
2258          */
2259         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2260         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2261         param->ddp_symlink_max  = sb->s_blocksize;
2262         param->ddp_mount_type   = LDD_MT_LDISKFS;
2263         if (ldiskfs_has_feature_extents(sb))
2264                 param->ddp_maxbytes = sb->s_maxbytes;
2265         else
2266                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2267         /*
2268          * inode are statically allocated, so per-inode space consumption
2269          * is the space consumed by the directory entry
2270          */
2271         param->ddp_inodespace     = PER_OBJ_USAGE;
2272         /*
2273          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2274          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2275          * maximum length to not under-count the actual number of extents
2276          * needed for writing a file if there are sub-optimal block allocations.
2277          */
2278         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2279         /* worst-case extent insertion metadata overhead */
2280         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2281         param->ddp_mntopts = 0;
2282         if (test_opt(sb, XATTR_USER))
2283                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2284         if (test_opt(sb, POSIX_ACL))
2285                 param->ddp_mntopts |= MNTOPT_ACL;
2286
2287         /*
2288          * LOD might calculate the max stripe count based on max_ea_size,
2289          * so we need take account in the overhead as well,
2290          * xattr_header + magic + xattr_entry_head
2291          */
2292         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2293                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2294
2295 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2296         if (ldiskfs_has_feature_ea_inode(sb))
2297                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2298                                                                 ea_overhead;
2299         else
2300 #endif
2301                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2302
2303         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE - ea_overhead)
2304                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE - ea_overhead;
2305
2306         /*
2307          * Preferred RPC size for efficient disk IO.  4MB shows good
2308          * all-around performance for ldiskfs, but use bigalloc chunk size
2309          * by default if larger.
2310          */
2311 #if defined(LDISKFS_CLUSTER_SIZE)
2312         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2313                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2314         else
2315 #endif
2316                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2317
2318         param->ddp_t10_cksum_type = 0;
2319         if (bi) {
2320                 unsigned short interval = blk_integrity_interval(bi);
2321                 name = blk_integrity_name(bi);
2322                 /*
2323                  * Expected values:
2324                  * T10-DIF-TYPE1-CRC
2325                  * T10-DIF-TYPE3-CRC
2326                  * T10-DIF-TYPE1-IP
2327                  * T10-DIF-TYPE3-IP
2328                  */
2329                 if (strncmp(name, "T10-DIF-TYPE",
2330                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2331                         /* also skip "1/3-" at end */
2332                         const int type_off = sizeof("T10-DIF-TYPE.");
2333                         char type_number = name[type_off - 2];
2334
2335                         if (interval != 512 && interval != 4096) {
2336                                 CERROR("%s: unsupported T10PI sector size %u\n",
2337                                        d->od_svname, interval);
2338                         } else if (type_number != '1' && type_number != '3') {
2339                                 CERROR("%s: unsupported T10PI type %s\n",
2340                                        d->od_svname, name);
2341                         } else if (strcmp(name + type_off, "CRC") == 0) {
2342                                 d->od_t10_type = type_number == '1' ?
2343                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2344                                 param->ddp_t10_cksum_type = interval == 512 ?
2345                                         OBD_CKSUM_T10CRC512 :
2346                                         OBD_CKSUM_T10CRC4K;
2347                         } else if (strcmp(name + type_off, "IP") == 0) {
2348                                 d->od_t10_type = type_number == '1' ?
2349                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2350                                 param->ddp_t10_cksum_type = interval == 512 ?
2351                                         OBD_CKSUM_T10IP512 :
2352                                         OBD_CKSUM_T10IP4K;
2353                         } else {
2354                                 CERROR("%s: unsupported checksum type of "
2355                                        "T10PI type '%s'",
2356                                        d->od_svname, name);
2357                         }
2358
2359                 } else {
2360                         CERROR("%s: unsupported T10PI type '%s'",
2361                                d->od_svname, name);
2362                 }
2363         }
2364 }
2365
2366 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2367 {
2368         return osd_sb(osd_dt_dev(d));
2369 }
2370
2371 /*
2372  * Concurrency: shouldn't matter.
2373  */
2374 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2375 {
2376         int rc;
2377         struct super_block *s = osd_sb(osd_dt_dev(d));
2378         ENTRY;
2379
2380         down_read(&s->s_umount);
2381         rc = s->s_op->sync_fs(s, 1);
2382         up_read(&s->s_umount);
2383
2384         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2385                rc);
2386
2387         return rc;
2388 }
2389
2390 /**
2391  * Start commit for OSD device.
2392  *
2393  * An implementation of dt_commit_async method for OSD device.
2394  * Asychronously starts underlayng fs sync and thereby a transaction
2395  * commit.
2396  *
2397  * \param env environment
2398  * \param d dt device
2399  *
2400  * \see dt_device_operations
2401  */
2402 static int osd_commit_async(const struct lu_env *env,
2403                             struct dt_device *d)
2404 {
2405         struct super_block *s = osd_sb(osd_dt_dev(d));
2406         int rc;
2407
2408         ENTRY;
2409
2410         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2411         down_read(&s->s_umount);
2412         rc = s->s_op->sync_fs(s, 0);
2413         up_read(&s->s_umount);
2414
2415         RETURN(rc);
2416 }
2417
2418 /* Our own copy of the set readonly functions if present, or NU if not. */
2419 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2420 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2421 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2422
2423 /*
2424  * Concurrency: shouldn't matter.
2425  */
2426 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2427 {
2428         struct super_block *sb = osd_sb(osd_dt_dev(d));
2429         struct block_device *dev = sb->s_bdev;
2430         int rc = -EOPNOTSUPP;
2431
2432         ENTRY;
2433
2434         if (priv_dev_set_rdonly) {
2435                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2436
2437                 rc = 0;
2438                 CERROR("*** setting %s read-only ***\n",
2439                        osd_dt_dev(d)->od_svname);
2440
2441                 if (sb->s_op->freeze_fs) {
2442                         rc = sb->s_op->freeze_fs(sb);
2443                         if (rc)
2444                                 goto out;
2445                 }
2446
2447                 if (jdev && (jdev != dev)) {
2448                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2449                                (long)jdev);
2450                         priv_dev_set_rdonly(jdev);
2451                 }
2452                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2453                 priv_dev_set_rdonly(dev);
2454
2455                 if (sb->s_op->unfreeze_fs)
2456                         sb->s_op->unfreeze_fs(sb);
2457         }
2458
2459 out:
2460         if (rc)
2461                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2462                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2463
2464         RETURN(rc);
2465 }
2466
2467 /**
2468  * Note: we do not count into QUOTA here.
2469  * If we mount with --data_journal we may need more.
2470  */
2471 const int osd_dto_credits_noquota[DTO_NR] = {
2472         /**
2473          * Insert.
2474          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2475          * SINGLEDATA_TRANS_BLOCKS(8)
2476          * XXX Note: maybe iam need more, since iam have more level than
2477          *           EXT3 htree.
2478          */
2479         [DTO_INDEX_INSERT]  = 16,
2480         /**
2481          * Delete
2482          * just modify a single entry, probably merge few within a block
2483          */
2484         [DTO_INDEX_DELETE]  = 1,
2485         /**
2486          * Used for OI scrub
2487          */
2488         [DTO_INDEX_UPDATE]  = 16,
2489         /**
2490          * 4(inode, inode bits, groups, GDT)
2491          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2492          */
2493         [DTO_OBJECT_CREATE] = 4,
2494         /**
2495          * 4(inode, inode bits, groups, GDT)
2496          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2497          */
2498         [DTO_OBJECT_DELETE] = 4,
2499         /**
2500          * Attr set credits (inode)
2501          */
2502         [DTO_ATTR_SET_BASE] = 1,
2503         /**
2504          * Xattr set. The same as xattr of EXT3.
2505          * DATA_TRANS_BLOCKS(14)
2506          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2507          * are also counted in. Do not know why?
2508          */
2509         [DTO_XATTR_SET]     = 14,
2510         /**
2511          * credits for inode change during write.
2512          */
2513         [DTO_WRITE_BASE]    = 3,
2514         /**
2515          * credits for single block write.
2516          */
2517         [DTO_WRITE_BLOCK]   = 14,
2518         /**
2519          * Attr set credits for chown.
2520          * This is extra credits for setattr, and it is null without quota
2521          */
2522         [DTO_ATTR_SET_CHOWN] = 0
2523 };
2524
2525 static const struct dt_device_operations osd_dt_ops = {
2526         .dt_root_get       = osd_root_get,
2527         .dt_statfs         = osd_statfs,
2528         .dt_trans_create   = osd_trans_create,
2529         .dt_trans_start    = osd_trans_start,
2530         .dt_trans_stop     = osd_trans_stop,
2531         .dt_trans_cb_add   = osd_trans_cb_add,
2532         .dt_conf_get       = osd_conf_get,
2533         .dt_mnt_sb_get     = osd_mnt_sb_get,
2534         .dt_sync           = osd_sync,
2535         .dt_ro             = osd_ro,
2536         .dt_commit_async   = osd_commit_async,
2537 };
2538
2539 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2540                           unsigned int role)
2541 {
2542         struct osd_object *obj = osd_dt_obj(dt);
2543         struct osd_thread_info *oti = osd_oti_get(env);
2544
2545         LINVRNT(osd_invariant(obj));
2546
2547         LASSERT(obj->oo_owner != env);
2548         down_read_nested(&obj->oo_sem, role);
2549
2550         LASSERT(obj->oo_owner == NULL);
2551         oti->oti_r_locks++;
2552 }
2553
2554 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2555                            unsigned int role)
2556 {
2557         struct osd_object *obj = osd_dt_obj(dt);
2558         struct osd_thread_info *oti = osd_oti_get(env);
2559
2560         LINVRNT(osd_invariant(obj));
2561
2562         LASSERT(obj->oo_owner != env);
2563         down_write_nested(&obj->oo_sem, role);
2564
2565         LASSERT(obj->oo_owner == NULL);
2566         obj->oo_owner = env;
2567         oti->oti_w_locks++;
2568 }
2569
2570 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2571 {
2572         struct osd_object *obj = osd_dt_obj(dt);
2573         struct osd_thread_info *oti = osd_oti_get(env);
2574
2575         LINVRNT(osd_invariant(obj));
2576
2577         LASSERT(oti->oti_r_locks > 0);
2578         oti->oti_r_locks--;
2579         up_read(&obj->oo_sem);
2580 }
2581
2582 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2583 {
2584         struct osd_object *obj = osd_dt_obj(dt);
2585         struct osd_thread_info *oti = osd_oti_get(env);
2586
2587         LINVRNT(osd_invariant(obj));
2588
2589         LASSERT(obj->oo_owner == env);
2590         LASSERT(oti->oti_w_locks > 0);
2591         oti->oti_w_locks--;
2592         obj->oo_owner = NULL;
2593         up_write(&obj->oo_sem);
2594 }
2595
2596 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2597 {
2598         struct osd_object *obj = osd_dt_obj(dt);
2599
2600         LINVRNT(osd_invariant(obj));
2601
2602         return obj->oo_owner == env;
2603 }
2604
2605 #ifdef HAVE_INODE_TIMESPEC64
2606 static struct timespec64 osd_inode_time(const struct lu_env *env,
2607                                        struct inode *inode, __u64 seconds)
2608 #else
2609 static struct timespec osd_inode_time(const struct lu_env *env,
2610                                        struct inode *inode, __u64 seconds)
2611 #endif
2612 {
2613         struct osd_thread_info *oti = osd_oti_get(env);
2614         struct timespec *t = &oti->oti_time;
2615
2616         t->tv_sec = seconds;
2617         t->tv_nsec = 0;
2618         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2619 #ifdef HAVE_INODE_TIMESPEC64
2620         return timespec_to_timespec64(*t);
2621 #else
2622         return *t;
2623 #endif
2624 }
2625
2626 static void osd_inode_getattr(const struct lu_env *env,
2627                               struct inode *inode, struct lu_attr *attr)
2628 {
2629         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2630                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2631                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2632                            LA_BLKSIZE | LA_TYPE;
2633
2634         attr->la_atime = inode->i_atime.tv_sec;
2635         attr->la_mtime = inode->i_mtime.tv_sec;
2636         attr->la_ctime = inode->i_ctime.tv_sec;
2637         attr->la_mode    = inode->i_mode;
2638         attr->la_size    = i_size_read(inode);
2639         attr->la_blocks  = inode->i_blocks;
2640         attr->la_uid     = i_uid_read(inode);
2641         attr->la_gid     = i_gid_read(inode);
2642         attr->la_projid  = i_projid_read(inode);
2643         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2644         attr->la_nlink   = inode->i_nlink;
2645         attr->la_rdev    = inode->i_rdev;
2646         attr->la_blksize = 1 << inode->i_blkbits;
2647         attr->la_blkbits = inode->i_blkbits;
2648         /*
2649          * Ext4 did not transfer inherit flags from raw inode
2650          * to inode flags, and ext4 internally test raw inode
2651          * @i_flags directly. Instead of patching ext4, we do it here.
2652          */
2653         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2654                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2655 }
2656
2657 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2658                         struct lu_attr *attr)
2659 {
2660         struct osd_object *obj = osd_dt_obj(dt);
2661
2662         if (unlikely(!dt_object_exists(dt)))
2663                 return -ENOENT;
2664         if (unlikely(obj->oo_destroyed))
2665                 return -ENOENT;
2666
2667         LASSERT(!dt_object_remote(dt));
2668         LINVRNT(osd_invariant(obj));
2669
2670         spin_lock(&obj->oo_guard);
2671         osd_inode_getattr(env, obj->oo_inode, attr);
2672         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2673                 attr->la_valid |= LA_FLAGS;
2674                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2675         }
2676         spin_unlock(&obj->oo_guard);
2677
2678         return 0;
2679 }
2680
2681 static int osd_declare_attr_qid(const struct lu_env *env,
2682                                 struct osd_object *obj,
2683                                 struct osd_thandle *oh, long long bspace,
2684                                 qid_t old_id, qid_t new_id, bool enforce,
2685                                 unsigned int type, bool ignore_edquot)
2686 {
2687         int rc;
2688         struct osd_thread_info *info = osd_oti_get(env);
2689         struct lquota_id_info  *qi = &info->oti_qi;
2690
2691         qi->lqi_type = type;
2692         /* inode accounting */
2693         qi->lqi_is_blk = false;
2694
2695         /* one more inode for the new id ... */
2696         qi->lqi_id.qid_uid = new_id;
2697         qi->lqi_space      = 1;
2698         /* Reserve credits for the new id */
2699         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2700         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2701                 rc = 0;
2702         if (rc)
2703                 RETURN(rc);
2704
2705         /* and one less inode for the current id */
2706         qi->lqi_id.qid_uid = old_id;
2707         qi->lqi_space = -1;
2708         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2709         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2710                 rc = 0;
2711         if (rc)
2712                 RETURN(rc);
2713
2714         /* block accounting */
2715         qi->lqi_is_blk = true;
2716
2717         /* more blocks for the new id ... */
2718         qi->lqi_id.qid_uid = new_id;
2719         qi->lqi_space      = bspace;
2720         /*
2721          * Credits for the new uid has been reserved, re-use "obj"
2722          * to save credit reservation.
2723          */
2724         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2725         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2726                 rc = 0;
2727         if (rc)
2728                 RETURN(rc);
2729
2730         /* and finally less blocks for the current uid */
2731         qi->lqi_id.qid_uid = old_id;
2732         qi->lqi_space      = -bspace;
2733         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2734         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2735                 rc = 0;
2736
2737         RETURN(rc);
2738 }
2739
2740 static int osd_declare_attr_set(const struct lu_env *env,
2741                                 struct dt_object *dt,
2742                                 const struct lu_attr *attr,
2743                                 struct thandle *handle)
2744 {
2745         struct osd_thandle *oh;
2746         struct osd_object *obj;
2747         qid_t uid;
2748         qid_t gid;
2749         long long bspace;
2750         int rc = 0;
2751         bool enforce;
2752
2753         ENTRY;
2754
2755         LASSERT(dt != NULL);
2756         LASSERT(handle != NULL);
2757
2758         obj = osd_dt_obj(dt);
2759         LASSERT(osd_invariant(obj));
2760
2761         oh = container_of0(handle, struct osd_thandle, ot_super);
2762         LASSERT(oh->ot_handle == NULL);
2763
2764         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2765                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2766
2767         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2768                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2769
2770         if (attr == NULL || obj->oo_inode == NULL)
2771                 RETURN(rc);
2772
2773         bspace   = obj->oo_inode->i_blocks << 9;
2774         bspace   = toqb(bspace);
2775
2776         /*
2777          * Changing ownership is always preformed by super user, it should not
2778          * fail with EDQUOT unless required explicitly.
2779          *
2780          * We still need to call the osd_declare_qid() to calculate the journal
2781          * credits for updating quota accounting files and to trigger quota
2782          * space adjustment once the operation is completed.
2783          */
2784         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2785                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2786
2787                 if (!ignore_edquot)
2788                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2789                                "(the quota space is %lld)\n",
2790                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2791                                attr->la_gid, bspace);
2792
2793                 /* USERQUOTA */
2794                 uid = i_uid_read(obj->oo_inode);
2795                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2796                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2797                                           attr->la_uid, enforce, USRQUOTA,
2798                                           true);
2799                 if (rc)
2800                         RETURN(rc);
2801
2802                 gid = i_gid_read(obj->oo_inode);
2803                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2804                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2805                                           i_gid_read(obj->oo_inode),
2806                                           attr->la_gid, enforce, GRPQUOTA,
2807                                           ignore_edquot);
2808                 if (rc)
2809                         RETURN(rc);
2810
2811         }
2812 #ifdef HAVE_PROJECT_QUOTA
2813         if (attr->la_valid & LA_PROJID) {
2814                 __u32 projid = i_projid_read(obj->oo_inode);
2815
2816                 enforce = (attr->la_valid & LA_PROJID) &&
2817                                         (attr->la_projid != projid);
2818                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2819                                           (qid_t)projid, (qid_t)attr->la_projid,
2820                                           enforce, PRJQUOTA, true);
2821                 if (rc)
2822                         RETURN(rc);
2823         }
2824 #endif
2825         RETURN(rc);
2826 }
2827
2828 static int osd_inode_setattr(const struct lu_env *env,
2829                              struct inode *inode, const struct lu_attr *attr)
2830 {
2831         __u64 bits = attr->la_valid;
2832
2833         /* Only allow set size for regular file */
2834         if (!S_ISREG(inode->i_mode))
2835                 bits &= ~(LA_SIZE | LA_BLOCKS);
2836
2837         if (bits == 0)
2838                 return 0;
2839
2840         if (bits & LA_ATIME)
2841                 inode->i_atime  = osd_inode_time(env, inode, attr->la_atime);
2842         if (bits & LA_CTIME)
2843                 inode->i_ctime  = osd_inode_time(env, inode, attr->la_ctime);
2844         if (bits & LA_MTIME)
2845                 inode->i_mtime  = osd_inode_time(env, inode, attr->la_mtime);
2846         if (bits & LA_SIZE) {
2847                 spin_lock(&inode->i_lock);
2848                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2849                 i_size_write(inode, attr->la_size);
2850                 spin_unlock(&inode->i_lock);
2851         }
2852
2853         /*
2854          * OSD should not change "i_blocks" which is used by quota.
2855          * "i_blocks" should be changed by ldiskfs only.
2856          */
2857         if (bits & LA_MODE)
2858                 inode->i_mode = (inode->i_mode & S_IFMT) |
2859                                 (attr->la_mode & ~S_IFMT);
2860         if (bits & LA_UID)
2861                 i_uid_write(inode, attr->la_uid);
2862         if (bits & LA_GID)
2863                 i_gid_write(inode, attr->la_gid);
2864         if (bits & LA_PROJID)
2865                 i_projid_write(inode, attr->la_projid);
2866         if (bits & LA_NLINK)
2867                 set_nlink(inode, attr->la_nlink);
2868         if (bits & LA_RDEV)
2869                 inode->i_rdev = attr->la_rdev;
2870
2871         if (bits & LA_FLAGS) {
2872                 /* always keep S_NOCMTIME */
2873                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2874                                  S_NOCMTIME;
2875                 /*
2876                  * Ext4 did not transfer inherit flags from
2877                  * @inode->i_flags to raw inode i_flags when writing
2878                  * flags, we do it explictly here.
2879                  */
2880                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2881                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2882                 else
2883                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2884         }
2885         return 0;
2886 }
2887
2888 #ifdef HAVE_PROJECT_QUOTA
2889 static int osd_transfer_project(struct inode *inode, __u32 projid)
2890 {
2891         struct super_block *sb = inode->i_sb;
2892         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2893         int err;
2894         kprojid_t kprojid;
2895         struct ldiskfs_iloc iloc;
2896         struct ldiskfs_inode *raw_inode;
2897         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2898
2899         if (!ldiskfs_has_feature_project(sb)) {
2900                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2901                         == LDISKFS_DEF_PROJID);
2902                 if (projid != LDISKFS_DEF_PROJID)
2903                         return -EOPNOTSUPP;
2904                 else
2905                         return 0;
2906         }
2907
2908         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2909                 return -EOPNOTSUPP;
2910
2911         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2912         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2913                 return 0;
2914
2915         err = ldiskfs_get_inode_loc(inode, &iloc);
2916         if (err)
2917                 return err;
2918
2919         raw_inode = ldiskfs_raw_inode(&iloc);
2920         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2921                 err = -EOVERFLOW;
2922                 brelse(iloc.bh);
2923                 return err;
2924         }
2925         brelse(iloc.bh);
2926
2927         dquot_initialize(inode);
2928         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2929         if (transfer_to[PRJQUOTA]) {
2930                 err = __dquot_transfer(inode, transfer_to);
2931                 dqput(transfer_to[PRJQUOTA]);
2932                 if (err)
2933                         return err;
2934         }
2935
2936         return err;
2937 }
2938 #endif
2939
2940 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2941 {
2942         int rc;
2943
2944         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2945             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2946                 struct iattr iattr;
2947
2948                 ll_vfs_dq_init(inode);
2949                 iattr.ia_valid = 0;
2950                 if (attr->la_valid & LA_UID)
2951                         iattr.ia_valid |= ATTR_UID;
2952                 if (attr->la_valid & LA_GID)
2953                         iattr.ia_valid |= ATTR_GID;
2954                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2955                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2956
2957                 rc = ll_vfs_dq_transfer(inode, &iattr);
2958                 if (rc) {
2959                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2960                                "enforcement enabled on the ldiskfs "
2961                                "filesystem?\n", inode->i_sb->s_id, rc);
2962                         return rc;
2963                 }
2964         }
2965
2966         /* Handle project id transfer here properly */
2967         if (attr->la_valid & LA_PROJID &&
2968             attr->la_projid != i_projid_read(inode)) {
2969 #ifdef HAVE_PROJECT_QUOTA
2970                 rc = osd_transfer_project(inode, attr->la_projid);
2971 #else
2972                 rc = -ENOTSUPP;
2973 #endif
2974                 if (rc) {
2975                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2976                                "enforcement enabled on the ldiskfs "
2977                                "filesystem?\n", inode->i_sb->s_id, rc);
2978                         return rc;
2979                 }
2980         }
2981         return 0;
2982 }
2983
2984 static int osd_attr_set(const struct lu_env *env,
2985                         struct dt_object *dt,
2986                         const struct lu_attr *attr,
2987                         struct thandle *handle)
2988 {
2989         struct osd_object *obj = osd_dt_obj(dt);
2990         struct inode *inode;
2991         int rc;
2992
2993         if (!dt_object_exists(dt))
2994                 return -ENOENT;
2995
2996         LASSERT(handle != NULL);
2997         LASSERT(!dt_object_remote(dt));
2998         LASSERT(osd_invariant(obj));
2999
3000         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3001
3002         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3003             !osd_obj2dev(obj)->od_is_ost) {
3004                 struct osd_thread_info *oti = osd_oti_get(env);
3005                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3006                 struct lu_fid *fid1 = &oti->oti_fid;
3007                 struct osd_inode_id *id = &oti->oti_id;
3008                 struct iam_path_descr *ipd;
3009                 struct iam_container *bag;
3010                 struct osd_thandle *oh;
3011                 int rc;
3012
3013                 fid_cpu_to_be(fid1, fid0);
3014                 memset(id, 1, sizeof(*id));
3015                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3016                                   fid0)->oi_dir.od_container;
3017                 ipd = osd_idx_ipd_get(env, bag);
3018                 if (unlikely(ipd == NULL))
3019                         RETURN(-ENOMEM);
3020
3021                 oh = container_of0(handle, struct osd_thandle, ot_super);
3022                 rc = iam_update(oh->ot_handle, bag,
3023                                 (const struct iam_key *)fid1,
3024                                 (const struct iam_rec *)id, ipd);
3025                 osd_ipd_put(env, bag, ipd);
3026                 return(rc > 0 ? 0 : rc);
3027         }
3028
3029         inode = obj->oo_inode;
3030
3031         rc = osd_quota_transfer(inode, attr);
3032         if (rc)
3033                 return rc;
3034
3035         spin_lock(&obj->oo_guard);
3036         rc = osd_inode_setattr(env, inode, attr);
3037         spin_unlock(&obj->oo_guard);
3038         if (rc != 0)
3039                 GOTO(out, rc);
3040
3041         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3042
3043         if (!(attr->la_valid & LA_FLAGS))
3044                 GOTO(out, rc);
3045
3046         /* Let's check if there are extra flags need to be set into LMA */
3047         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3048                 struct osd_thread_info *info = osd_oti_get(env);
3049                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3050
3051                 LASSERT(!obj->oo_pfid_in_lma);
3052
3053                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3054                                  &info->oti_ost_attrs);
3055                 if (rc)
3056                         GOTO(out, rc);
3057
3058                 lma->lma_incompat |=
3059                         lustre_to_lma_flags(attr->la_flags);
3060                 lustre_lma_swab(lma);
3061                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3062                                      lma, sizeof(*lma), XATTR_REPLACE);
3063                 if (rc != 0) {
3064                         struct osd_device *osd = osd_obj2dev(obj);
3065
3066                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3067                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3068                               lma->lma_incompat, rc);
3069                 } else {
3070                         obj->oo_lma_flags =
3071                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3072                 }
3073                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3074         }
3075 out:
3076         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3077
3078         return rc;
3079 }
3080
3081 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3082                                            struct osd_object *obj,
3083                                            const char *name, const int namelen)
3084 {
3085         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3086 }
3087
3088 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3089                       umode_t mode, struct dt_allocation_hint *hint,
3090                       struct thandle *th, struct lu_attr *attr)
3091 {
3092         int result;
3093         struct osd_device *osd = osd_obj2dev(obj);
3094         struct osd_thandle *oth;
3095         struct dt_object *parent = NULL;
3096         struct inode *inode;
3097         uid_t owner[2] = {0, 0};
3098
3099         if (attr->la_valid & LA_UID)
3100                 owner[0] = attr->la_uid;
3101         if (attr->la_valid & LA_GID)
3102                 owner[1] = attr->la_gid;
3103
3104         LINVRNT(osd_invariant(obj));
3105         LASSERT(obj->oo_inode == NULL);
3106         LASSERT(obj->oo_hl_head == NULL);
3107
3108         if (S_ISDIR(mode) && ldiskfs_pdo) {
3109                 obj->oo_hl_head =
3110                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3111                 if (obj->oo_hl_head == NULL)
3112                         return -ENOMEM;
3113         }
3114
3115         oth = container_of(th, struct osd_thandle, ot_super);
3116         LASSERT(oth->ot_handle->h_transaction != NULL);
3117
3118         if (hint != NULL && hint->dah_parent != NULL &&
3119             !dt_object_remote(hint->dah_parent))
3120                 parent = hint->dah_parent;
3121
3122         inode = ldiskfs_create_inode(oth->ot_handle,
3123                                      parent ? osd_dt_obj(parent)->oo_inode :
3124                                               osd_sb(osd)->s_root->d_inode,
3125                                      mode, owner);
3126         if (!IS_ERR(inode)) {
3127                 /* Do not update file c/mtime in ldiskfs. */
3128                 inode->i_flags |= S_NOCMTIME;
3129
3130                 /*
3131                  * For new created object, it must be consistent,
3132                  * and it is unnecessary to scrub against it.
3133                  */
3134                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3135
3136                 obj->oo_inode = inode;
3137                 result = 0;
3138         } else {
3139                 if (obj->oo_hl_head != NULL) {
3140                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3141                         obj->oo_hl_head = NULL;
3142                 }
3143                 result = PTR_ERR(inode);
3144         }
3145         LINVRNT(osd_invariant(obj));
3146         return result;
3147 }
3148
3149 enum {
3150         OSD_NAME_LEN = 255
3151 };
3152
3153 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3154                      struct lu_attr *attr,
3155                      struct dt_allocation_hint *hint,
3156                      struct dt_object_format *dof,
3157                      struct thandle *th)
3158 {
3159         int result;
3160         struct osd_thandle *oth;
3161         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3162
3163         LASSERT(S_ISDIR(attr->la_mode));
3164
3165         oth = container_of(th, struct osd_thandle, ot_super);
3166         LASSERT(oth->ot_handle->h_transaction != NULL);
3167         result = osd_mkfile(info, obj, mode, hint, th, attr);
3168
3169         return result;
3170 }
3171
3172 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3173                         struct lu_attr *attr,
3174                         struct dt_allocation_hint *hint,
3175                         struct dt_object_format *dof,
3176                         struct thandle *th)
3177 {
3178         int result;
3179         struct osd_thandle *oth;
3180         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3181
3182         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3183
3184         LASSERT(S_ISREG(attr->la_mode));
3185
3186         oth = container_of(th, struct osd_thandle, ot_super);
3187         LASSERT(oth->ot_handle->h_transaction != NULL);
3188
3189         result = osd_mkfile(info, obj, mode, hint, th, attr);
3190         if (result == 0) {
3191                 LASSERT(obj->oo_inode != NULL);
3192                 if (feat->dif_flags & DT_IND_VARKEY)
3193                         result = iam_lvar_create(obj->oo_inode,
3194                                                  feat->dif_keysize_max,
3195                                                  feat->dif_ptrsize,
3196                                                  feat->dif_recsize_max,
3197                                                  oth->ot_handle);
3198                 else
3199                         result = iam_lfix_create(obj->oo_inode,
3200                                                  feat->dif_keysize_max,
3201                                                  feat->dif_ptrsize,
3202                                                  feat->dif_recsize_max,
3203                                                  oth->ot_handle);
3204         }
3205         return result;
3206 }
3207
3208 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3209                      struct lu_attr *attr,
3210                      struct dt_allocation_hint *hint,
3211                      struct dt_object_format *dof,
3212                      struct thandle *th)
3213 {
3214         LASSERT(S_ISREG(attr->la_mode));
3215         return osd_mkfile(info, obj, (attr->la_mode &
3216                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3217                           attr);
3218 }
3219
3220 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3221                      struct lu_attr *attr,
3222                      struct dt_allocation_hint *hint,
3223                      struct dt_object_format *dof,
3224                      struct thandle *th)
3225 {
3226         LASSERT(S_ISLNK(attr->la_mode));
3227         return osd_mkfile(info, obj, (attr->la_mode &
3228                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3229                           attr);
3230 }
3231
3232 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3233                      struct lu_attr *attr,
3234                      struct dt_allocation_hint *hint,
3235                      struct dt_object_format *dof,
3236                      struct thandle *th)
3237 {
3238         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3239         int result;
3240
3241         LINVRNT(osd_invariant(obj));
3242         LASSERT(obj->oo_inode == NULL);
3243         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3244                 S_ISFIFO(mode) || S_ISSOCK(mode));
3245
3246         result = osd_mkfile(info, obj, mode, hint, th, attr);
3247         if (result == 0) {
3248                 LASSERT(obj->oo_inode != NULL);
3249                 /*
3250                  * This inode should be marked dirty for i_rdev.  Currently
3251                  * that is done in the osd_attr_init().
3252                  */
3253                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3254                                    attr->la_rdev);
3255         }
3256         LINVRNT(osd_invariant(obj));
3257         return result;
3258 }
3259
3260 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3261                               struct lu_attr *,
3262                               struct dt_allocation_hint *hint,
3263                               struct dt_object_format *dof,
3264                               struct thandle *);
3265
3266 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3267 {
3268         osd_obj_type_f result;
3269
3270         switch (type) {
3271         case DFT_DIR:
3272                 result = osd_mkdir;
3273                 break;
3274         case DFT_REGULAR:
3275                 result = osd_mkreg;
3276                 break;
3277         case DFT_SYM:
3278                 result = osd_mksym;
3279                 break;
3280         case DFT_NODE:
3281                 result = osd_mknod;
3282                 break;
3283         case DFT_INDEX:
3284                 result = osd_mk_index;
3285                 break;
3286
3287         default:
3288                 LBUG();
3289                 break;
3290         }
3291         return result;
3292 }
3293
3294
3295 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3296                         struct dt_object *parent, struct dt_object *child,
3297                         umode_t child_mode)
3298 {
3299         LASSERT(ah);
3300
3301         ah->dah_parent = parent;
3302         ah->dah_mode = child_mode;
3303
3304         if (parent != NULL && !dt_object_remote(parent)) {
3305                 /* will help to find FID->ino at dt_insert("..") */
3306                 struct osd_object *pobj = osd_dt_obj(parent);
3307
3308                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3309         }
3310 }
3311
3312 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3313                           struct lu_attr *attr, struct dt_object_format *dof)
3314 {
3315         struct inode *inode = obj->oo_inode;
3316         __u64 valid = attr->la_valid;
3317         int result;
3318
3319         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3320
3321         if (dof->dof_type != DFT_NODE)
3322                 attr->la_valid &= ~LA_RDEV;
3323         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3324                 attr->la_valid &= ~LA_ATIME;
3325         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3326                 attr->la_valid &= ~LA_CTIME;
3327         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3328                 attr->la_valid &= ~LA_MTIME;
3329
3330         result = osd_quota_transfer(inode, attr);
3331         if (result)
3332                 return;
3333
3334         if (attr->la_valid != 0) {
3335                 result = osd_inode_setattr(info->oti_env, inode, attr);
3336                 /*
3337                  * The osd_inode_setattr() should always succeed here.  The
3338                  * only error that could be returned is EDQUOT when we are
3339                  * trying to change the UID or GID of the inode. However, this
3340                  * should not happen since quota enforcement is no longer
3341                  * enabled on ldiskfs (lquota takes care of it).
3342                  */
3343                 LASSERTF(result == 0, "%d\n", result);
3344                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3345         }
3346
3347         attr->la_valid = valid;
3348 }
3349
3350 /**
3351  * Helper function for osd_create()
3352  *
3353  * \retval 0, on success
3354  */
3355 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3356                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3357                         struct dt_object_format *dof, struct thandle *th)
3358 {
3359         int result;
3360         __u32 umask;
3361
3362         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3363
3364         /* we drop umask so that permissions we pass are not affected */
3365         umask = current->fs->umask;
3366         current->fs->umask = 0;
3367
3368         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3369                                                   th);
3370         if (likely(obj->oo_inode != NULL)) {
3371                 LASSERT(obj->oo_inode->i_state & I_NEW);
3372
3373                 /*
3374                  * Unlock the inode before attr initialization to avoid
3375                  * unnecessary dqget operations. LU-6378
3376                  */
3377                 unlock_new_inode(obj->oo_inode);
3378         }
3379
3380         if (likely(result == 0)) {
3381                 osd_attr_init(info, obj, attr, dof);
3382                 osd_object_init0(obj);
3383         }
3384
3385         /* restore previous umask value */
3386         current->fs->umask = umask;
3387
3388         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3389
3390         return result;
3391 }
3392
3393 /**
3394  * Helper function for osd_create()
3395  *
3396  * \retval 0, on success
3397  */
3398 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3399                            const struct lu_fid *fid, struct thandle *th)
3400 {
3401         struct osd_thread_info *info = osd_oti_get(env);
3402         struct osd_inode_id    *id   = &info->oti_id;
3403         struct osd_device      *osd  = osd_obj2dev(obj);
3404         struct osd_thandle     *oh;
3405         int rc;
3406
3407         LASSERT(obj->oo_inode != NULL);
3408
3409         oh = container_of0(th, struct osd_thandle, ot_super);
3410         LASSERT(oh->ot_handle);
3411         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3412
3413         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3414         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3415                            OI_CHECK_FLD, NULL);
3416         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3417
3418         return rc;
3419 }
3420
3421 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3422                    u64 seq, struct lu_seq_range *range)
3423 {
3424         struct seq_server_site *ss = osd_seq_site(osd);
3425
3426         if (fid_seq_is_idif(seq)) {
3427                 fld_range_set_ost(range);
3428                 range->lsr_index = idif_ost_idx(seq);
3429                 return 0;
3430         }
3431
3432         if (!fid_seq_in_fldb(seq)) {
3433                 fld_range_set_mdt(range);
3434                 if (ss != NULL)
3435                         /*
3436                          * FIXME: If ss is NULL, it suppose not get lsr_index
3437                          * at all
3438                          */
3439                         range->lsr_index = ss->ss_node_id;
3440                 return 0;
3441         }
3442
3443         LASSERT(ss != NULL);
3444         fld_range_set_any(range);
3445         /* OSD will only do local fld lookup */
3446         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3447 }
3448
3449 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3450                               struct lu_attr *attr,
3451                               struct dt_allocation_hint *hint,
3452                               struct dt_object_format *dof,
3453                               struct thandle *handle)
3454 {
3455         struct osd_thandle *oh;
3456         int rc;
3457
3458         ENTRY;
3459
3460         LASSERT(handle != NULL);
3461
3462         oh = container_of0(handle, struct osd_thandle, ot_super);
3463         LASSERT(oh->ot_handle == NULL);
3464
3465         /*
3466          * EA object consumes more credits than regular object: osd_mk_index
3467          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3468          * leaf_node, could involves the block, block bitmap, groups, GDT
3469          * change for each block, so add 4 * 2 credits in that case.
3470          */
3471         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3472                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3473                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3474         /*
3475          * Reuse idle OI block may cause additional one OI block
3476          * to be changed.
3477          */
3478         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3479                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3480
3481         /* will help to find FID->ino mapping at dt_insert() */
3482         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3483                                    osd_dt_obj(dt));
3484         if (rc != 0)
3485                 RETURN(rc);
3486
3487         if (!attr)
3488                 RETURN(0);
3489
3490         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3491                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3492                                    NULL, OSD_QID_INODE);
3493         if (rc != 0)
3494                 RETURN(rc);
3495
3496         RETURN(rc);
3497 }
3498
3499 /**
3500  * Called to destroy on-disk representation of the object
3501  *
3502  * Concurrency: must be locked
3503  */
3504 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3505                                struct thandle *th)
3506 {
3507         struct osd_object *obj = osd_dt_obj(dt);
3508         struct inode *inode = obj->oo_inode;
3509         struct osd_thandle *oh;
3510         int rc;
3511
3512         ENTRY;
3513
3514         if (inode == NULL)
3515                 RETURN(-ENOENT);
3516
3517         oh = container_of0(th, struct osd_thandle, ot_super);
3518         LASSERT(oh->ot_handle == NULL);
3519
3520         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3521                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3522
3523         /* For removing agent entry */
3524         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3525                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3526
3527         /*
3528          * Recycle idle OI leaf may cause additional three OI blocks
3529          * to be changed.
3530          */
3531         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3532                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3533                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3534         /* one less inode */
3535         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3536                                    i_projid_read(inode), -1, oh, obj, NULL,
3537                                    OSD_QID_INODE);
3538         if (rc)
3539                 RETURN(rc);
3540         /* data to be truncated */
3541         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3542                                    i_projid_read(inode), 0, oh, obj, NULL,
3543                                    OSD_QID_BLK);
3544         if (rc)
3545                 RETURN(rc);
3546
3547         /*
3548          * will help to find FID->ino when this object is being
3549          * added to PENDING
3550          */
3551         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3552
3553         RETURN(rc);
3554 }
3555
3556 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3557                        struct thandle *th)
3558 {
3559         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3560         struct osd_object *obj = osd_dt_obj(dt);
3561         struct inode *inode = obj->oo_inode;
3562         struct osd_device *osd = osd_obj2dev(obj);
3563         struct osd_thandle *oh;
3564         int result;
3565
3566         ENTRY;
3567
3568         oh = container_of0(th, struct osd_thandle, ot_super);
3569         LASSERT(oh->ot_handle);
3570         LASSERT(inode);
3571         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3572
3573         if (unlikely(fid_is_acct(fid)))
3574                 RETURN(-EPERM);
3575
3576         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3577                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3578                 if (result != 0)
3579                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3580                                osd_name(osd), PFID(fid), result);
3581         }
3582
3583         if (S_ISDIR(inode->i_mode)) {
3584                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3585                         inode->i_nlink == 2);
3586
3587                 spin_lock(&obj->oo_guard);
3588                 clear_nlink(inode);
3589                 spin_unlock(&obj->oo_guard);
3590                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3591         }
3592
3593         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3594
3595         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3596
3597         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3598                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3599                                        oh->ot_handle, OI_CHECK_FLD);
3600
3601         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3602         /* XXX: add to ext3 orphan list */
3603         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3604
3605         /* not needed in the cache anymore */
3606         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3607         obj->oo_destroyed = 1;
3608
3609         RETURN(0);
3610 }
3611
3612 /**
3613  * Put the fid into lustre_mdt_attrs, and then place the structure
3614  * inode's ea. This fid should not be altered during the life time
3615  * of the inode.
3616  *
3617  * \retval +ve, on success
3618  * \retval -ve, on error
3619  *
3620  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3621  */
3622 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3623                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3624 {
3625         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3626         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3627         int rc;
3628
3629         ENTRY;
3630
3631         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3632                 RETURN(0);
3633
3634         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3635                 rc = -ENOMEM;
3636
3637         lustre_loa_init(loa, fid, compat, incompat);
3638         lustre_loa_swab(loa, false);
3639
3640         /*
3641          * For the OST device with 256 bytes inode size by default,
3642          * the PFID EA will be stored together with LMA EA to avoid
3643          * performance trouble. Otherwise the PFID EA can be stored
3644          * independently. LU-8998
3645          */
3646         if ((compat & LMAC_FID_ON_OST) &&
3647             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3648                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3649                                      sizeof(*loa), XATTR_CREATE);
3650         else
3651                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3652                                      sizeof(*lma), XATTR_CREATE);
3653         /*
3654          * LMA may already exist, but we need to check that all the
3655          * desired compat/incompat flags have been added.
3656          */
3657         if (unlikely(rc == -EEXIST)) {
3658                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3659                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3660                 if (rc < 0)
3661                         RETURN(rc);
3662
3663                 if (rc < sizeof(*lma))
3664                         RETURN(-EINVAL);
3665
3666                 lustre_loa_swab(loa, true);
3667                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3668                     ((compat == 0 && incompat == 0) ||
3669                      (!(~lma->lma_compat & compat) &&
3670                       !(~lma->lma_incompat & incompat))))
3671                         RETURN(0);
3672
3673                 lma->lma_self_fid = *fid;
3674                 lma->lma_compat |= compat;
3675                 lma->lma_incompat |= incompat;
3676                 if (rc == sizeof(*lma)) {
3677                         lustre_lma_swab(lma);
3678                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3679                                              sizeof(*lma), XATTR_REPLACE);
3680                 } else {
3681                         lustre_loa_swab(loa, false);
3682                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3683                                              sizeof(*loa), XATTR_REPLACE);
3684                 }
3685         }
3686
3687         RETURN(rc);
3688 }
3689
3690 /**
3691  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3692  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3693  * To have compatilibility with 1.8 ldiskfs driver we need to have
3694  * magic number at start of fid data.
3695  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3696  * its inmemory API.
3697  */
3698 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3699                                          const struct lu_fid *fid)
3700 {
3701         if (!fid_is_namespace_visible(fid) ||
3702             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3703                 param->edp_magic = 0;
3704                 return;
3705         }
3706
3707         param->edp_magic = LDISKFS_LUFID_MAGIC;
3708         param->edp_len =  sizeof(struct lu_fid) + 1;
3709         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3710 }
3711
3712 /**
3713  * Try to read the fid from inode ea into dt_rec.
3714  *
3715  * \param fid object fid.
3716  *
3717  * \retval 0 on success
3718  */
3719 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3720                           __u32 ino, struct lu_fid *fid,
3721                           struct osd_inode_id *id)
3722 {
3723         struct osd_thread_info *info  = osd_oti_get(env);
3724         struct inode *inode;
3725
3726         ENTRY;
3727
3728         osd_id_gen(id, ino, OSD_OII_NOGEN);
3729         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3730         if (IS_ERR(inode))
3731                 RETURN(PTR_ERR(inode));
3732
3733         iput(inode);
3734         RETURN(0);
3735 }
3736
3737 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3738                                         struct inode *dir,
3739                                         struct inode *parent_dir,
3740                                         const struct lu_fid *dot_fid,
3741                                         const struct lu_fid *dot_dot_fid,
3742                                         struct osd_thandle *oth)
3743 {
3744         struct ldiskfs_dentry_param *dot_ldp;
3745         struct ldiskfs_dentry_param *dot_dot_ldp;
3746         __u32 saved_nlink = dir->i_nlink;
3747         int rc;
3748
3749         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3750         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3751
3752         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3753         dot_ldp->edp_magic = 0;
3754
3755         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3756                                     dir, dot_ldp, dot_dot_ldp);
3757         /*
3758          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3759          * the subseqent ref_add() will increase the dir->i_nlink
3760          * as 3. That is incorrect for new created directory.
3761          *
3762          * It looks like hack, because we want to make the OSD API
3763          * to be order-independent for new created directory object
3764          * between dt_insert(..) and ref_add() operations.
3765          *
3766          * Here, we only restore the in-RAM dir-inode's nlink attr,
3767          * becuase if the nlink attr is not 2, then there will be
3768          * ref_add() called following the dt_insert(..), such call
3769          * will make both the in-RAM and on-disk dir-inode's nlink
3770          * attr to be set as 2. LU-7447
3771          */
3772         set_nlink(dir, saved_nlink);
3773         return rc;
3774 }
3775
3776 /**
3777  * Create an local agent inode for remote entry
3778  */
3779 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3780                                                   struct osd_device *osd,
3781                                                   struct osd_object *pobj,
3782                                                   const struct lu_fid *fid,
3783                                                   __u32 type,
3784                                                   struct thandle *th)
3785 {
3786         struct osd_thread_info *info = osd_oti_get(env);
3787         struct inode *local;
3788         struct osd_thandle *oh;
3789         int rc;
3790
3791         ENTRY;
3792
3793         LASSERT(th);
3794         oh = container_of(th, struct osd_thandle, ot_super);
3795         LASSERT(oh->ot_handle->h_transaction != NULL);
3796
3797         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3798                                      NULL);
3799         if (IS_ERR(local)) {
3800                 CERROR("%s: create local error %d\n", osd_name(osd),
3801                        (int)PTR_ERR(local));
3802                 RETURN(local);
3803         }
3804
3805         /*
3806          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3807          * correct gid on remote file, not agent here
3808          */
3809         local->i_gid = current_fsgid();
3810         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3811
3812         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3813          * That gives e2fsck something to look at and be happy, and allows
3814          * debugging if we need to determine where this symlink came from.
3815          */
3816         if (S_ISLNK(type)) {
3817                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3818                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3819                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3820
3821                 i_size_write(local, rc);
3822                 LDISKFS_I(local)->i_disksize = rc;
3823         }
3824         unlock_new_inode(local);
3825
3826         /* Agent inode should not have project ID */
3827 #ifdef  HAVE_PROJECT_QUOTA
3828         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3829             i_projid_read(pobj->oo_inode) != 0) {
3830                 rc = osd_transfer_project(local, 0);
3831                 if (rc) {
3832                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3833                                "quota enforcement enabled on the ldiskfs "
3834                                "filesystem?\n", local->i_sb->s_id, rc);
3835                         RETURN(ERR_PTR(rc));
3836                 }
3837         }
3838 #endif
3839         /* Set special LMA flag for local agent inode */
3840         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3841         if (rc != 0) {
3842                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3843                        osd_name(osd), PFID(fid), rc);
3844                 RETURN(ERR_PTR(rc));
3845         }
3846
3847         if (!S_ISDIR(type))
3848                 RETURN(local);
3849
3850         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3851                                          lu_object_fid(&pobj->oo_dt.do_lu),
3852                                          fid, oh);
3853         if (rc != 0) {
3854                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3855                         osd_name(osd), PFID(fid), rc);
3856                 RETURN(ERR_PTR(rc));
3857         }
3858
3859         RETURN(local);
3860 }
3861
3862 /**
3863  * when direntry is deleted, we have to take care of possible agent inode
3864  * referenced by that. unfortunately we can't do this at that point:
3865  * iget() within a running transaction leads to deadlock and we better do
3866  * not call that every delete declaration to save performance. so we put
3867  * a potention agent inode on a list and process that once the transaction
3868  * is over. Notice it's not any worse in terms of real orphans as regular
3869  * object destroy doesn't put inodes on the on-disk orphan list. this should
3870  * be addressed separately
3871  */
3872 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3873                                             struct osd_thandle *oh,
3874                                             __u32 ino)
3875 {
3876         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3877         struct osd_obj_orphan *oor;
3878
3879         OBD_ALLOC_PTR(oor);
3880         if (oor == NULL)
3881                 return -ENOMEM;
3882
3883         oor->oor_ino = ino;
3884         oor->oor_env = (struct lu_env *)env;
3885         spin_lock(&osd->od_osfs_lock);
3886         list_add(&oor->oor_list, &osd->od_orphan_list);
3887         spin_unlock(&osd->od_osfs_lock);
3888
3889         oh->ot_remove_agents = 1;
3890
3891         return 0;
3892
3893 }
3894
3895 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3896                                                 struct osd_device *osd)
3897 {
3898         struct osd_thread_info *info = osd_oti_get(env);
3899         struct osd_obj_orphan *oor, *tmp;
3900         struct osd_inode_id id;
3901         struct list_head list;
3902         struct inode *inode;
3903         struct lu_fid fid;
3904         handle_t *jh;
3905         __u32 ino;
3906
3907         INIT_LIST_HEAD(&list);
3908
3909         spin_lock(&osd->od_osfs_lock);
3910         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3911                 if (oor->oor_env == env) {
3912                         list_del(&oor->oor_list);
3913                         list_add(&oor->oor_list, &list);
3914                 }
3915         }
3916         spin_unlock(&osd->od_osfs_lock);
3917
3918         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3919
3920                 ino = oor->oor_ino;
3921
3922                 list_del(&oor->oor_list);
3923                 OBD_FREE_PTR(oor);
3924
3925                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3926                 inode = osd_iget_fid(info, osd, &id, &fid);
3927                 if (IS_ERR(inode))
3928                         continue;
3929
3930                 if (!osd_remote_fid(env, osd, &fid)) {
3931                         iput(inode);
3932                         continue;
3933                 }
3934
3935                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3936                 clear_nlink(inode);
3937                 mark_inode_dirty(inode);
3938                 ldiskfs_journal_stop(jh);
3939                 iput(inode);
3940         }
3941
3942         return 0;
3943 }
3944
3945 /**
3946  * OSD layer object create function for OST objects (b=11826).
3947  *
3948  * The FID is inserted into inode xattr here.
3949  *
3950  * \retval   0, on success
3951  * \retval -ve, on error
3952  */
3953 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3954                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3955                       struct dt_object_format *dof, struct thandle *th)
3956 {
3957         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3958         struct osd_object *obj = osd_dt_obj(dt);
3959         struct osd_thread_info *info = osd_oti_get(env);
3960         int result, on_ost = 0;
3961
3962         ENTRY;
3963
3964         if (dt_object_exists(dt))
3965                 RETURN(-EEXIST);
3966
3967         LINVRNT(osd_invariant(obj));
3968         LASSERT(!dt_object_remote(dt));
3969         LASSERT(osd_is_write_locked(env, obj));
3970         LASSERT(th != NULL);
3971
3972         if (unlikely(fid_is_acct(fid)))
3973                 /*
3974                  * Quota files can't be created from the kernel any more,
3975                  * 'tune2fs -O quota' will take care of creating them
3976                  */
3977                 RETURN(-EPERM);
3978
3979         result = __osd_create(info, obj, attr, hint, dof, th);
3980         if (result == 0) {
3981                 if (fid_is_idif(fid) &&
3982                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3983                         struct lu_fid *tfid = &info->oti_fid;
3984                         struct ost_id *oi   = &info->oti_ostid;
3985
3986                         fid_to_ostid(fid, oi);
3987                         ostid_to_fid(tfid, oi, 0);
3988                         on_ost = 1;
3989                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3990                                                 LMAC_FID_ON_OST, 0);
3991                 } else {
3992                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3993                                                fid, OI_CHECK_FLD);
3994                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3995                                                 on_ost ? LMAC_FID_ON_OST : 0,
3996                                                 0);
3997                 }
3998                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3999                         obj->oo_dt.do_body_ops = &osd_body_ops;
4000         }
4001
4002         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4003                 result = __osd_oi_insert(env, obj, fid, th);
4004
4005         /*
4006          * a small optimization - dt_insert() isn't usually applied
4007          * to OST objects, so we don't need to cache OI mapping for
4008          * OST objects
4009          */
4010         if (result == 0 && on_ost == 0) {
4011                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4012
4013                 result = osd_idc_find_and_init(env, osd, obj);
4014                 LASSERT(result == 0);
4015         }
4016
4017         LASSERT(ergo(result == 0,
4018                      dt_object_exists(dt) && !dt_object_remote(dt)));
4019         LINVRNT(osd_invariant(obj));
4020         RETURN(result);
4021 }
4022
4023 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4024                                struct thandle *handle)
4025 {
4026         struct osd_thandle *oh;
4027
4028         /* it's possible that object doesn't exist yet */
4029         LASSERT(handle != NULL);
4030
4031         oh = container_of0(handle, struct osd_thandle, ot_super);
4032         LASSERT(oh->ot_handle == NULL);
4033
4034         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4035                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4036
4037         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4038
4039         return 0;
4040 }
4041
4042 /*
4043  * Concurrency: @dt is write locked.
4044  */
4045 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4046                        struct thandle *th)
4047 {
4048         struct osd_object *obj = osd_dt_obj(dt);
4049         struct inode *inode = obj->oo_inode;
4050         struct osd_thandle *oh;
4051         int rc = 0;
4052
4053         if (!dt_object_exists(dt) || obj->oo_destroyed)
4054                 return -ENOENT;
4055
4056         LINVRNT(osd_invariant(obj));
4057         LASSERT(!dt_object_remote(dt));
4058         LASSERT(osd_is_write_locked(env, obj));
4059         LASSERT(th != NULL);
4060
4061         oh = container_of0(th, struct osd_thandle, ot_super);
4062         LASSERT(oh->ot_handle != NULL);
4063
4064         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4065
4066         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4067                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4068         /*
4069          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4070          * (65000) subdirectories by storing "1" in i_nlink if the link count
4071          * would otherwise overflow. Directory tranversal tools understand
4072          * that (st_nlink == 1) indicates that the filesystem dose not track
4073          * hard links count on the directory, and will not abort subdirectory
4074          * scanning early once (st_nlink - 2) subdirs have been found.
4075          *
4076          * This also has to properly handle the case of inodes with nlink == 0
4077          * in case they are being linked into the PENDING directory
4078          */
4079         spin_lock(&obj->oo_guard);
4080         if (unlikely(inode->i_nlink == 0))
4081                 /* inc_nlink from 0 may cause WARN_ON */
4082                 set_nlink(inode, 1);
4083         else {
4084                 ldiskfs_inc_count(oh->ot_handle, inode);
4085                 if (!S_ISDIR(inode->i_mode))
4086                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4087         }
4088         spin_unlock(&obj->oo_guard);
4089
4090         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4091         LINVRNT(osd_invariant(obj));
4092
4093         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4094
4095         return rc;
4096 }
4097
4098 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4099                                struct thandle *handle)
4100 {
4101         struct osd_thandle *oh;
4102
4103         if (!dt_object_exists(dt))
4104                 return -ENOENT;
4105
4106         LASSERT(!dt_object_remote(dt));
4107         LASSERT(handle != NULL);
4108
4109         oh = container_of0(handle, struct osd_thandle, ot_super);
4110         LASSERT(oh->ot_handle == NULL);
4111
4112         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4113                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4114
4115         return 0;
4116 }
4117
4118 /*
4119  * Concurrency: @dt is write locked.
4120  */
4121 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4122                        struct thandle *th)
4123 {
4124         struct osd_object *obj = osd_dt_obj(dt);
4125         struct inode *inode = obj->oo_inode;
4126         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4127         struct osd_thandle *oh;
4128
4129         if (!dt_object_exists(dt))
4130                 return -ENOENT;
4131
4132         LINVRNT(osd_invariant(obj));
4133         LASSERT(!dt_object_remote(dt));
4134         LASSERT(osd_is_write_locked(env, obj));
4135         LASSERT(th != NULL);
4136
4137         oh = container_of0(th, struct osd_thandle, ot_super);
4138         LASSERT(oh->ot_handle != NULL);
4139
4140         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4141
4142         spin_lock(&obj->oo_guard);
4143         /*
4144          * That can be result of upgrade from old Lustre version and
4145          * applied only to local files.  Just skip this ref_del call.
4146          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4147          */
4148         if (inode->i_nlink == 0) {
4149                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4150                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4151                              ", maybe an upgraded file? (LU-3915)\n",
4152                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4153                 spin_unlock(&obj->oo_guard);
4154                 return 0;
4155         }
4156
4157         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4158                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4159
4160         ldiskfs_dec_count(oh->ot_handle, inode);
4161         spin_unlock(&obj->oo_guard);
4162
4163         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4164         LINVRNT(osd_invariant(obj));
4165
4166         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4167
4168         return 0;
4169 }
4170
4171 /*
4172  * Concurrency: @dt is read locked.
4173  */
4174 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4175                          struct lu_buf *buf, const char *name)
4176 {
4177         struct osd_object      *obj    = osd_dt_obj(dt);
4178         struct inode           *inode  = obj->oo_inode;
4179         struct osd_thread_info *info   = osd_oti_get(env);
4180         struct dentry          *dentry = &info->oti_obj_dentry;
4181         bool cache_xattr = false;
4182         int rc;
4183
4184         LASSERT(buf);
4185
4186         /* version get is not real XATTR but uses xattr API */
4187         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4188                 dt_obj_version_t *ver = buf->lb_buf;
4189
4190                 /*
4191                  * for version we are just using xattr API but change inode
4192                  * field instead
4193                  */
4194                 if (buf->lb_len == 0)
4195                         return sizeof(dt_obj_version_t);
4196
4197                 if (buf->lb_len < sizeof(dt_obj_version_t))
4198                         return -ERANGE;
4199
4200                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4201                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4202
4203                 *ver = LDISKFS_I(inode)->i_fs_version;
4204
4205                 return sizeof(dt_obj_version_t);
4206         }
4207
4208         if (!dt_object_exists(dt))
4209                 return -ENOENT;
4210
4211         LASSERT(!dt_object_remote(dt));
4212         LASSERT(inode->i_op != NULL);
4213 #ifdef HAVE_IOP_XATTR
4214         LASSERT(inode->i_op->getxattr != NULL);
4215 #endif
4216
4217         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4218             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4219                 cache_xattr = true;
4220
4221         if (cache_xattr) {
4222                 rc = osd_oxc_get(obj, name, buf);
4223                 if (rc != -ENOENT)
4224                         return rc;
4225         }
4226
4227         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4228                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4229                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4230                 struct filter_fid *ff;
4231                 struct ost_layout *ol;
4232
4233                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4234                 if (rc)
4235                         return rc;
4236
4237                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4238
4239                 rc = sizeof(*ff);
4240                 if (buf->lb_len == 0 || !buf->lb_buf)
4241                         return rc;
4242
4243                 if (buf->lb_len < rc)
4244                         return -ERANGE;
4245
4246                 ff = buf->lb_buf;
4247                 ol = &ff->ff_layout;
4248                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4249                                                   PFID_STRIPE_IDX_BITS);
4250                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4251                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4252                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4253                 if (lma->lma_compat & LMAC_COMP_INFO) {
4254                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4255                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4256                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4257                 } else {
4258                         ol->ol_comp_start = 0;
4259                         ol->ol_comp_end = 0;
4260                         ol->ol_comp_id = 0;
4261                 }
4262         } else {
4263                 rc = __osd_xattr_get(inode, dentry, name,
4264                                      buf->lb_buf, buf->lb_len);
4265         }
4266
4267         if (cache_xattr) {
4268                 if (rc == -ENOENT || rc == -ENODATA)
4269                         osd_oxc_add(obj, name, NULL, 0);
4270                 else if (rc > 0 && buf->lb_buf != NULL)
4271                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4272         }
4273
4274         return rc;
4275 }
4276
4277 static int osd_declare_xattr_set(const struct lu_env *env,
4278                                  struct dt_object *dt,
4279                                  const struct lu_buf *buf, const char *name,
4280                                  int fl, struct thandle *handle)
4281 {
4282         struct osd_thandle *oh;
4283         int credits = 0;
4284         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4285
4286         LASSERT(handle != NULL);
4287
4288         oh = container_of0(handle, struct osd_thandle, ot_super);
4289         LASSERT(oh->ot_handle == NULL);
4290
4291         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4292                 /*
4293                  * For non-upgrading case, the LMA is set first and
4294                  * usually fit inode. But for upgrade case, the LMA
4295                  * may be in another separated EA block.
4296                  */
4297                 if (dt_object_exists(dt)) {
4298                         if (fl == LU_XATTR_REPLACE)
4299                                 credits = 1;
4300                         else
4301                                 goto upgrade;
4302                 }
4303         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4304                 credits = 1;
4305         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4306                 /* We may need to delete the old PFID EA. */
4307                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4308                 if (fl == LU_XATTR_REPLACE)
4309                         credits += 1;
4310                 else
4311                         goto upgrade;
4312         } else {
4313                 /*
4314                  * If some name entry resides on remote MDT, then will create
4315                  * agent entry under remote parent. On the other hand, if the
4316                  * remote entry will be removed, then related agent entry may
4317                  * need to be removed from the remote parent. So there may be
4318                  * kinds of cases, let's declare enough credits. The credits
4319                  * for create agent entry is enough for remove case.
4320                  */
4321                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4322                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4323                         if (dt_object_exists(dt))
4324                                 credits += 1; /* For updating LMA */
4325                 }
4326
4327 upgrade:
4328                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4329
4330                 if (buf != NULL) {
4331                         ssize_t buflen;
4332
4333                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4334                                 /*
4335                                  * learn xattr size from osd_xattr_get if
4336                                  * attribute has not been read yet
4337                                  */
4338                                 buflen = __osd_xattr_get(
4339                                     osd_dt_obj(dt)->oo_inode,
4340                                     &osd_oti_get(env)->oti_obj_dentry,
4341                                     name, NULL, 0);
4342                                 if (buflen < 0)
4343                                         buflen = 0;
4344                         } else {
4345                                 buflen = buf->lb_len;
4346                         }
4347
4348                         if (buflen > sb->s_blocksize) {
4349                                 credits += osd_calc_bkmap_credits(
4350                                     sb, NULL, 0, -1,
4351                                     (buflen + sb->s_blocksize - 1) >>
4352                                     sb->s_blocksize_bits);
4353                         }
4354                 }
4355                 /*
4356                  * xattr set may involve inode quota change, reserve credits for
4357                  * dquot_initialize()
4358                  */
4359                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4360         }
4361
4362         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4363
4364         return 0;
4365 }
4366
4367 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4368                               const struct lu_buf *buf, int fl,
4369                               struct thandle *handle)
4370 {
4371         struct osd_thread_info *info = osd_oti_get(env);
4372         struct dentry *dentry = &info->oti_obj_dentry;
4373         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4374         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4375         struct inode *inode = obj->oo_inode;
4376         struct filter_fid *ff = buf->lb_buf;
4377         struct ost_layout *ol = &ff->ff_layout;
4378         int flags = XATTR_REPLACE;
4379         int rc;
4380
4381         ENTRY;
4382
4383         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4384                 RETURN(-EINVAL);
4385
4386         rc = osd_get_lma(info, inode, dentry, loa);
4387         if (rc == -ENODATA) {
4388                 /* Usually for upgarding from old device */
4389                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4390                                 LMAC_FID_ON_OST, 0);
4391                 flags = XATTR_CREATE;
4392         } else if (rc) {
4393                 RETURN(rc);
4394         }
4395
4396         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4397                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4398                         RETURN(-EEXIST);
4399
4400                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4401                         /* Separate PFID EA from LMA */
4402                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4403                         lustre_lma_swab(lma);
4404                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4405                                              sizeof(*lma), XATTR_REPLACE);
4406                         if (!rc) {
4407                                 obj->oo_pfid_in_lma = 0;
4408                                 rc = LU_XATTR_CREATE;
4409                         }
4410
4411                         RETURN(rc);
4412                 }
4413         } else {
4414                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4415                         RETURN(fl);
4416
4417                 /*
4418                  * Old client does not send stripe information,
4419                  * then store the PFID EA on disk separatedly.
4420                  */
4421                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4422                              ol->ol_stripe_size == 0))
4423                         RETURN(fl);
4424
4425                 /* Remove old PFID EA entry firstly. */
4426                 ll_vfs_dq_init(inode);
4427                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4428                 if (rc == -ENODATA) {
4429                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4430                                 RETURN(rc);
4431                 } else if (rc) {
4432                         RETURN(rc);
4433                 }
4434         }
4435
4436         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4437         if (likely(ol->ol_stripe_size != 0)) {
4438                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4439                                              PFID_STRIPE_IDX_BITS;
4440                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4441                 lma->lma_compat |= LMAC_STRIPE_INFO;
4442                 if (ol->ol_comp_id != 0) {
4443                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4444                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4445                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4446                         lma->lma_compat |= LMAC_COMP_INFO;
4447                 }
4448         }
4449
4450         lustre_loa_swab(loa, false);
4451
4452         /* Store the PFID EA inside LMA. */
4453         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4454                              flags);
4455         if (!rc)
4456                 obj->oo_pfid_in_lma = 1;
4457
4458         RETURN(rc);
4459 }
4460
4461 /*
4462  * In DNE environment, the object (in spite of regular file or directory)
4463  * and its name entry may reside on different MDTs. Under such case, we will
4464  * create an agent entry on the MDT where the object resides. The agent entry
4465  * references the object locally, that makes the object to be visible to the
4466  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4467  * such as 'tar' can handle the object properly.
4468  *
4469  * We handle the agent entry during set linkEA that is the common interface
4470  * for both regular file and directroy, can handle kinds of cases, such as
4471  * create/link/unlink/rename, and so on.
4472  *
4473  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4474  *
4475  * XXX: There are two known issues:
4476  * 1. For one object, we will create at most one agent entry even if there
4477  *    may be more than one cross-MDTs hard links on the object. So the local
4478  *    e2fsck may claim that the object's nlink is larger than the name entries
4479  *    that reference such inode. And in further, the e2fsck will fix the nlink
4480  *    attribute to match the local references. Then it will cause the object's
4481  *    nlink attribute to be inconsistent with the global references. it is bad
4482  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4483  *    other hand, the global namespace LFSCK can repair the object's attribute
4484  *    according to the linkEA.
4485  * 2. There may be too many hard links on the object as to its linkEA overflow,
4486  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4487  *    case happened, then at this point, we do not know whether there are some
4488  *    cross-MDTs reference. But there are local references, it guarantees that
4489  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4490  */
4491 static int osd_xattr_handle_linkea(const struct lu_env *env,
4492                                    struct osd_device *osd,
4493                                    struct osd_object *obj,
4494                                    const struct lu_buf *buf,
4495                                    struct thandle *handle)
4496 {
4497         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4498         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4499         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4500         struct lu_name tmpname;
4501         struct osd_thandle *oh;
4502         int rc;
4503         bool remote = false;
4504
4505         ENTRY;
4506
4507         oh = container_of0(handle, struct osd_thandle, ot_super);
4508         LASSERT(oh->ot_handle != NULL);
4509
4510         rc = linkea_init_with_rec(&ldata);
4511         if (!rc) {
4512                 linkea_first_entry(&ldata);
4513                 while (ldata.ld_lee != NULL && !remote) {
4514                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4515                                             &tmpname, tfid);
4516                         if (osd_remote_fid(env, osd, tfid) > 0)
4517                                 remote = true;
4518                         else
4519                                 linkea_next_entry(&ldata);
4520                 }
4521         } else if (rc == -ENODATA) {
4522                 rc = 0;
4523         } else {
4524                 RETURN(rc);
4525         }
4526
4527         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4528                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4529                 if (rc)
4530                         CERROR("%s: failed to remove agent entry for "DFID
4531                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4532         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4533                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4534                 if (rc)
4535                         CERROR("%s: failed to create agent entry for "DFID
4536                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4537         }
4538
4539         RETURN(rc);
4540 }
4541
4542 /*
4543  * Concurrency: @dt is write locked.
4544  */
4545 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4546                          const struct lu_buf *buf, const char *name, int fl,
4547                          struct thandle *handle)
4548 {
4549         struct osd_object *obj = osd_dt_obj(dt);
4550         struct osd_device *osd = osd_obj2dev(obj);
4551         struct inode *inode = obj->oo_inode;
4552         struct osd_thread_info *info = osd_oti_get(env);
4553         int fs_flags = 0;
4554         int len;
4555         int rc;
4556
4557         ENTRY;
4558
4559         LASSERT(handle);
4560         LASSERT(buf);
4561
4562         /* version set is not real XATTR */
4563         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4564                 dt_obj_version_t *version = buf->lb_buf;
4565
4566                 /*
4567                  * for version we are just using xattr API but change inode
4568                  * field instead
4569                  */
4570                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4571
4572                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4573                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4574
4575                 LDISKFS_I(inode)->i_fs_version = *version;
4576                 /*
4577                  * Version is set after all inode operations are finished,
4578                  * so we should mark it dirty here
4579                  */
4580                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4581
4582                 RETURN(0);
4583         }
4584
4585         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4586                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4587
4588         len = buf->lb_len;
4589         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4590
4591         /*
4592          * For the OST device with 256 bytes inode size by default,
4593          * the PFID EA will be stored together with LMA EA to avoid
4594          * performance trouble. Otherwise the PFID EA can be stored
4595          * independently. LU-8998
4596          */
4597         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4598             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4599                 LASSERT(buf->lb_buf);
4600
4601                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4602                 if (fl <= 0)
4603                         RETURN(fl);
4604         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4605                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4606                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4607
4608                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4609                 if (rc)
4610                         RETURN(rc);
4611
4612                 lma->lma_incompat |= LMAI_STRIPED;
4613                 lustre_lma_swab(lma);
4614                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4615                                      sizeof(*lma), XATTR_REPLACE);
4616                 if (rc != 0)
4617                         RETURN(rc);
4618         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4619                 LASSERT(!osd->od_is_ost);
4620
4621                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4622                 if (rc)
4623                         RETURN(rc);
4624         }
4625
4626         if (fl & LU_XATTR_REPLACE)
4627                 fs_flags |= XATTR_REPLACE;
4628
4629         if (fl & LU_XATTR_CREATE)
4630                 fs_flags |= XATTR_CREATE;
4631
4632         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4633         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4634
4635         if (rc == 0 &&
4636             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4637              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4638                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4639
4640         return rc;
4641 }
4642
4643 /*
4644  * Concurrency: @dt is read locked.
4645  */
4646 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4647                           const struct lu_buf *buf)
4648 {
4649         struct osd_object *obj = osd_dt_obj(dt);
4650         struct inode *inode = obj->oo_inode;
4651         struct osd_thread_info *info = osd_oti_get(env);
4652         struct dentry *dentry = &info->oti_obj_dentry;
4653
4654         if (!dt_object_exists(dt))
4655                 return -ENOENT;
4656
4657         LASSERT(!dt_object_remote(dt));
4658         LASSERT(inode->i_op != NULL);
4659         LASSERT(inode->i_op->listxattr != NULL);
4660
4661         dentry->d_inode = inode;
4662         dentry->d_sb = inode->i_sb;
4663         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4664 }
4665
4666 static int osd_declare_xattr_del(const struct lu_env *env,
4667                                  struct dt_object *dt, const char *name,
4668                                  struct thandle *handle)
4669 {
4670         struct osd_thandle *oh;
4671         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4672
4673         LASSERT(!dt_object_remote(dt));
4674         LASSERT(handle != NULL);
4675
4676         oh = container_of0(handle, struct osd_thandle, ot_super);
4677         LASSERT(oh->ot_handle == NULL);
4678
4679         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4680                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4681         /*
4682          * xattr del may involve inode quota change, reserve credits for
4683          * dquot_initialize()
4684          */
4685         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4686
4687         return 0;
4688 }
4689
4690 /*
4691  * Concurrency: @dt is write locked.
4692  */
4693 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4694                          const char *name, struct thandle *handle)
4695 {
4696         struct osd_object *obj = osd_dt_obj(dt);
4697         struct inode *inode = obj->oo_inode;
4698         struct osd_thread_info *info = osd_oti_get(env);
4699         struct dentry *dentry = &info->oti_obj_dentry;
4700         int rc;
4701
4702         if (!dt_object_exists(dt))
4703                 return -ENOENT;
4704
4705         LASSERT(!dt_object_remote(dt));
4706         LASSERT(inode->i_op != NULL);
4707         LASSERT(handle != NULL);
4708 #ifdef HAVE_IOP_XATTR
4709         LASSERT(inode->i_op->removexattr != NULL);
4710 #endif
4711
4712         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4713
4714         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4715                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4716
4717                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4718                                  &info->oti_ost_attrs);
4719                 if (!rc) {
4720                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4721
4722                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4723                         lustre_lma_swab(lma);
4724                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4725                                              sizeof(*lma), XATTR_REPLACE);
4726                         if (!rc)
4727                                 obj->oo_pfid_in_lma = 0;
4728                 }
4729         } else {
4730                 ll_vfs_dq_init(inode);
4731                 dentry->d_inode = inode;
4732                 dentry->d_sb = inode->i_sb;
4733                 rc = osd_removexattr(dentry, inode, name);
4734         }
4735
4736         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4737
4738         if (rc == 0 &&
4739             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4740              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4741                 osd_oxc_del(obj, name);
4742
4743         return rc;
4744 }
4745
4746 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4747                            __u64 start, __u64 end)
4748 {
4749         struct osd_object *obj = osd_dt_obj(dt);
4750         struct inode *inode = obj->oo_inode;
4751         struct osd_thread_info *info = osd_oti_get(env);
4752         struct dentry *dentry = &info->oti_obj_dentry;
4753         struct file *file = &info->oti_file;
4754         int rc;
4755
4756         ENTRY;
4757
4758         dentry->d_inode = inode;
4759         dentry->d_sb = inode->i_sb;
4760         file->f_path.dentry = dentry;
4761         file->f_mapping = inode->i_mapping;
4762         file->f_op = inode->i_fop;
4763         set_file_inode(file, inode);
4764
4765         rc = ll_vfs_fsync_range(file, start, end, 0);
4766
4767         RETURN(rc);
4768 }
4769
4770 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4771 {
4772         return 0;
4773 }
4774
4775 /*
4776  * Index operations.
4777  */
4778
4779 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4780                                const struct dt_index_features *feat)
4781 {
4782         struct iam_descr *descr;
4783
4784         if (osd_object_is_root(o))
4785                 return feat == &dt_directory_features;
4786
4787         LASSERT(o->oo_dir != NULL);
4788
4789         descr = o->oo_dir->od_container.ic_descr;
4790         if (feat == &dt_directory_features) {
4791                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4792                         return 1;
4793                 else
4794                         return 0;
4795         } else {
4796                 return feat->dif_keysize_min <= descr->id_key_size &&
4797                        descr->id_key_size <= feat->dif_keysize_max &&
4798                        feat->dif_recsize_min <= descr->id_rec_size &&
4799                        descr->id_rec_size <= feat->dif_recsize_max &&
4800                        !(feat->dif_flags & (DT_IND_VARKEY |
4801                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4802                        ergo(feat->dif_flags & DT_IND_UPDATE,
4803                             1 /* XXX check that object (and fs) is writable */);
4804         }
4805 }
4806
4807 static int osd_iam_container_init(const struct lu_env *env,
4808                                   struct osd_object *obj,
4809                                   struct osd_directory *dir)
4810 {
4811         struct iam_container *bag = &dir->od_container;
4812         int result;
4813
4814         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4815         if (result != 0)
4816                 return result;
4817
4818         result = iam_container_setup(bag);
4819         if (result == 0)
4820                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4821         else
4822                 iam_container_fini(bag);
4823
4824         return result;
4825 }
4826
4827
4828 /*
4829  * Concurrency: no external locking is necessary.
4830  */
4831 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4832                          const struct dt_index_features *feat)
4833 {
4834         int result;
4835         int skip_iam = 0;
4836         struct osd_object *obj = osd_dt_obj(dt);
4837
4838         LINVRNT(osd_invariant(obj));
4839
4840         if (osd_object_is_root(obj)) {
4841                 dt->do_index_ops = &osd_index_ea_ops;
4842                 result = 0;
4843         } else if (feat == &dt_directory_features) {
4844                 dt->do_index_ops = &osd_index_ea_ops;
4845                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4846                         result = 0;
4847                 else
4848                         result = -ENOTDIR;
4849                 skip_iam = 1;
4850         } else if (unlikely(feat == &dt_otable_features)) {
4851                 dt->do_index_ops = &osd_otable_ops;
4852                 return 0;
4853         } else if (unlikely(feat == &dt_acct_features)) {
4854                 dt->do_index_ops = &osd_acct_index_ops;
4855                 result = 0;
4856                 skip_iam = 1;
4857         } else if (!osd_has_index(obj)) {
4858                 struct osd_directory *dir;
4859                 struct osd_device *osd = osd_obj2dev(obj);
4860                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4861
4862                 OBD_ALLOC_PTR(dir);
4863                 if (dir) {
4864
4865                         spin_lock(&obj->oo_guard);
4866                         if (obj->oo_dir == NULL)
4867                                 obj->oo_dir = dir;
4868                         else
4869                                 /*
4870                                  * Concurrent thread allocated container data.
4871                                  */
4872                                 OBD_FREE_PTR(dir);
4873                         spin_unlock(&obj->oo_guard);
4874                         /*
4875                          * Now, that we have container data, serialize its
4876                          * initialization.
4877                          */
4878                         down_write(&obj->oo_ext_idx_sem);
4879                         /*
4880                          * recheck under lock.
4881                          */
4882
4883                         if (osd_has_index(obj)) {
4884                                 result = 0;
4885                                 goto unlock;
4886                         }
4887
4888                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4889                         if (result || feat == &dt_lfsck_namespace_features ||
4890                             feat == &dt_lfsck_layout_orphan_features ||
4891                             feat == &dt_lfsck_layout_dangling_features)
4892                                 goto unlock;
4893
4894                         result = osd_index_register(osd, fid,
4895                                                     feat->dif_keysize_max,
4896                                                     feat->dif_recsize_max);
4897                         if (result < 0)
4898                                 CWARN("%s: failed to register index "
4899                                       DFID": rc = %d\n",
4900                                       osd_name(osd), PFID(fid), result);
4901                         else if (result > 0)
4902                                 result = 0;
4903                         else
4904                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4905                                        " (%d/%d) registered\n",
4906                                        osd_name(osd), PFID(fid),
4907                                        (int)feat->dif_keysize_max,
4908                                        (int)feat->dif_recsize_max);
4909
4910 unlock:
4911                         up_write(&obj->oo_ext_idx_sem);
4912                 } else {
4913                         result = -ENOMEM;
4914                 }
4915         } else {
4916                 result = 0;
4917         }
4918
4919         if (result == 0 && skip_iam == 0) {
4920                 if (!osd_iam_index_probe(env, obj, feat))
4921                         result = -ENOTDIR;
4922         }
4923         LINVRNT(osd_invariant(obj));
4924
4925         return result;
4926 }
4927
4928 static int osd_otable_it_attr_get(const struct lu_env *env,
4929                                  struct dt_object *dt,
4930                                  struct lu_attr *attr)
4931 {
4932         attr->la_valid = 0;
4933         return 0;
4934 }
4935
4936 static const struct dt_object_operations osd_obj_ops = {
4937         .do_read_lock           = osd_read_lock,
4938         .do_write_lock          = osd_write_lock,
4939         .do_read_unlock         = osd_read_unlock,
4940         .do_write_unlock        = osd_write_unlock,
4941         .do_write_locked        = osd_write_locked,
4942         .do_attr_get            = osd_attr_get,
4943         .do_declare_attr_set    = osd_declare_attr_set,
4944         .do_attr_set            = osd_attr_set,
4945         .do_ah_init             = osd_ah_init,
4946         .do_declare_create      = osd_declare_create,
4947         .do_create              = osd_create,
4948         .do_declare_destroy     = osd_declare_destroy,
4949         .do_destroy             = osd_destroy,
4950         .do_index_try           = osd_index_try,
4951         .do_declare_ref_add     = osd_declare_ref_add,
4952         .do_ref_add             = osd_ref_add,
4953         .do_declare_ref_del     = osd_declare_ref_del,
4954         .do_ref_del             = osd_ref_del,
4955         .do_xattr_get           = osd_xattr_get,
4956         .do_declare_xattr_set   = osd_declare_xattr_set,
4957         .do_xattr_set           = osd_xattr_set,
4958         .do_declare_xattr_del   = osd_declare_xattr_del,
4959         .do_xattr_del           = osd_xattr_del,
4960         .do_xattr_list          = osd_xattr_list,
4961         .do_object_sync         = osd_object_sync,
4962         .do_invalidate          = osd_invalidate,
4963 };
4964
4965 static const struct dt_object_operations osd_obj_otable_it_ops = {
4966         .do_attr_get    = osd_otable_it_attr_get,
4967         .do_index_try   = osd_index_try,
4968 };
4969
4970 static int osd_index_declare_iam_delete(const struct lu_env *env,
4971                                         struct dt_object *dt,
4972                                         const struct dt_key *key,
4973                                         struct thandle *handle)
4974 {
4975         struct osd_thandle *oh;
4976
4977         oh = container_of0(handle, struct osd_thandle, ot_super);
4978         LASSERT(oh->ot_handle == NULL);
4979
4980         /* Recycle  may cause additional three blocks to be changed. */
4981         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4982                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4983
4984         return 0;
4985 }
4986
4987 /**
4988  *      delete a (key, value) pair from index \a dt specified by \a key
4989  *
4990  *      \param  dt      osd index object
4991  *      \param  key     key for index
4992  *      \param  rec     record reference
4993  *      \param  handle  transaction handler
4994  *
4995  *      \retval  0  success
4996  *      \retval -ve   failure
4997  */
4998 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4999                                 const struct dt_key *key,
5000                                 struct thandle *handle)
5001 {
5002         struct osd_thread_info *oti = osd_oti_get(env);
5003         struct osd_object *obj = osd_dt_obj(dt);
5004         struct osd_thandle *oh;
5005         struct iam_path_descr *ipd;
5006         struct iam_container *bag = &obj->oo_dir->od_container;
5007         int rc;
5008
5009         ENTRY;
5010
5011         if (!dt_object_exists(dt))
5012                 RETURN(-ENOENT);
5013
5014         LINVRNT(osd_invariant(obj));
5015         LASSERT(!dt_object_remote(dt));
5016         LASSERT(bag->ic_object == obj->oo_inode);
5017         LASSERT(handle != NULL);
5018
5019         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5020
5021         ipd = osd_idx_ipd_get(env, bag);
5022         if (unlikely(ipd == NULL))
5023                 RETURN(-ENOMEM);
5024
5025         oh = container_of0(handle, struct osd_thandle, ot_super);
5026         LASSERT(oh->ot_handle != NULL);
5027         LASSERT(oh->ot_handle->h_transaction != NULL);
5028
5029         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5030                 /* swab quota uid/gid provided by caller */
5031                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5032                 key = (const struct dt_key *)&oti->oti_quota_id;
5033         }
5034
5035         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5036         osd_ipd_put(env, bag, ipd);
5037         LINVRNT(osd_invariant(obj));
5038         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5039         RETURN(rc);
5040 }
5041
5042 static int osd_index_declare_ea_delete(const struct lu_env *env,
5043                                        struct dt_object *dt,
5044                                        const struct dt_key *key,
5045                                        struct thandle *handle)
5046 {
5047         struct osd_thandle *oh;
5048         struct inode *inode;
5049         int rc, credits;
5050
5051         ENTRY;
5052
5053         LASSERT(!dt_object_remote(dt));
5054         LASSERT(handle != NULL);
5055
5056         oh = container_of0(handle, struct osd_thandle, ot_super);
5057         LASSERT(oh->ot_handle == NULL);
5058
5059         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5060         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5061
5062         inode = osd_dt_obj(dt)->oo_inode;
5063         if (inode == NULL)
5064                 RETURN(-ENOENT);
5065
5066         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5067                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5068                                    NULL, OSD_QID_BLK);
5069         RETURN(rc);
5070 }
5071
5072 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5073                                           struct dt_rec *fid)
5074 {
5075         struct osd_fid_pack *rec;
5076         int rc = -ENODATA;
5077
5078         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5079                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5080                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5081                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5082                         rc = -EINVAL;
5083         }
5084         return rc;
5085 }
5086
5087 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5088                           const struct lu_fid *fid)
5089 {
5090         struct seq_server_site *ss = osd_seq_site(osd);
5091
5092         ENTRY;
5093
5094         /* FID seqs not in FLDB, must be local seq */
5095         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5096                 RETURN(0);
5097
5098         /*
5099          * If FLD is not being initialized yet, it only happens during the
5100          * initialization, likely during mgs initialization, and we assume
5101          * this is local FID.
5102          */
5103         if (ss == NULL || ss->ss_server_fld == NULL)
5104                 RETURN(0);
5105
5106         /* Only check the local FLDB here */
5107         if (osd_seq_exists(env, osd, fid_seq(fid)))
5108                 RETURN(0);
5109
5110         RETURN(1);
5111 }
5112
5113 static void osd_take_care_of_agent(const struct lu_env *env,
5114                                    struct osd_device *osd,
5115                                    struct osd_thandle *oh,
5116                                    struct ldiskfs_dir_entry_2 *de)
5117 {
5118         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5119         struct osd_idmap_cache *idc;
5120         int rc, schedule = 0;
5121
5122         LASSERT(de != NULL);
5123
5124         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5125         if (likely(rc == 0)) {
5126                 idc = osd_idc_find_or_init(env, osd, fid);
5127                 if (IS_ERR(idc) || idc->oic_remote)
5128                         schedule = 1;
5129         } else if (rc == -ENODATA) {
5130                 /*
5131                  * can't get FID, postpone to the end of the
5132                  * transaction when iget() is safe
5133                  */
5134                 schedule = 1;
5135         } else {
5136                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5137         }
5138         if (schedule)
5139                 osd_schedule_agent_inode_removal(env, oh,
5140                                                  le32_to_cpu(de->inode));
5141 }
5142
5143 /**
5144  * Index delete function for interoperability mode (b11826).
5145  * It will remove the directory entry added by osd_index_ea_insert().
5146  * This entry is needed to maintain name->fid mapping.
5147  *
5148  * \param key,  key i.e. file entry to be deleted
5149  *
5150  * \retval   0, on success
5151  * \retval -ve, on error
5152  */
5153 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5154                                const struct dt_key *key, struct thandle *handle)
5155 {
5156         struct osd_object *obj = osd_dt_obj(dt);
5157         struct inode *dir = obj->oo_inode;
5158         struct dentry *dentry;
5159         struct osd_thandle *oh;
5160         struct ldiskfs_dir_entry_2 *de = NULL;
5161         struct buffer_head *bh;
5162         struct htree_lock *hlock = NULL;
5163         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5164         int rc;
5165
5166         ENTRY;
5167
5168         if (!dt_object_exists(dt))
5169                 RETURN(-ENOENT);
5170
5171         LINVRNT(osd_invariant(obj));
5172         LASSERT(!dt_object_remote(dt));
5173         LASSERT(handle != NULL);
5174
5175         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5176
5177         oh = container_of(handle, struct osd_thandle, ot_super);
5178         LASSERT(oh->ot_handle != NULL);
5179         LASSERT(oh->ot_handle->h_transaction != NULL);
5180
5181         ll_vfs_dq_init(dir);
5182         dentry = osd_child_dentry_get(env, obj,
5183                                       (char *)key, strlen((char *)key));
5184
5185         if (obj->oo_hl_head != NULL) {
5186                 hlock = osd_oti_get(env)->oti_hlock;
5187                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5188                                    dir, LDISKFS_HLOCK_DEL);
5189         } else {
5190                 down_write(&obj->oo_ext_idx_sem);
5191         }
5192
5193         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5194         if (!IS_ERR(bh)) {
5195                 /*
5196                  * If this is not the ".." entry, it might be a remote DNE
5197                  * entry and  we need to check if the FID is for a remote
5198                  * MDT.  If the FID is  not in the directory entry (e.g.
5199                  * upgraded 1.8 filesystem without dirdata enabled) then
5200                  * we need to get the FID from the LMA. For a remote directory
5201                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5202                  * case.
5203                  *
5204                  * Delete the entry before the agent inode in order to
5205                  * simplify error handling.  At worst an error after deleting
5206                  * the entry first might leak the agent inode afterward. The
5207                  * reverse would need filesystem abort in case of error deleting
5208                  * the entry after the agent had been removed, or leave a
5209                  * dangling entry pointing at a random inode.
5210                  */
5211                 if (strcmp((char *)key, dotdot) != 0)
5212                         osd_take_care_of_agent(env, osd, oh, de);
5213                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5214                 brelse(bh);
5215         } else {
5216                 rc = PTR_ERR(bh);
5217         }
5218         if (hlock != NULL)
5219                 ldiskfs_htree_unlock(hlock);
5220         else
5221                 up_write(&obj->oo_ext_idx_sem);
5222
5223         GOTO(out, rc);
5224 out:
5225         LASSERT(osd_invariant(obj));
5226         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5227         RETURN(rc);
5228 }
5229
5230 /**
5231  *      Lookup index for \a key and copy record to \a rec.
5232  *
5233  *      \param  dt      osd index object
5234  *      \param  key     key for index
5235  *      \param  rec     record reference
5236  *
5237  *      \retval  +ve  success : exact mach
5238  *      \retval  0    return record with key not greater than \a key
5239  *      \retval -ve   failure
5240  */
5241 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5242                                 struct dt_rec *rec, const struct dt_key *key)
5243 {
5244         struct osd_object *obj = osd_dt_obj(dt);
5245         struct iam_path_descr *ipd;
5246         struct iam_container *bag = &obj->oo_dir->od_container;
5247         struct osd_thread_info *oti = osd_oti_get(env);
5248         struct iam_iterator *it = &oti->oti_idx_it;
5249         struct iam_rec *iam_rec;
5250         int rc;
5251
5252         ENTRY;
5253
5254         if (!dt_object_exists(dt))
5255                 RETURN(-ENOENT);
5256
5257         LASSERT(osd_invariant(obj));
5258         LASSERT(!dt_object_remote(dt));
5259         LASSERT(bag->ic_object == obj->oo_inode);
5260
5261         ipd = osd_idx_ipd_get(env, bag);
5262         if (IS_ERR(ipd))
5263                 RETURN(-ENOMEM);
5264
5265         /* got ipd now we can start iterator. */
5266         iam_it_init(it, bag, 0, ipd);
5267
5268         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5269                 /* swab quota uid/gid provided by caller */
5270                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5271                 key = (const struct dt_key *)&oti->oti_quota_id;
5272         }
5273
5274         rc = iam_it_get(it, (struct iam_key *)key);
5275         if (rc >= 0) {
5276                 if (S_ISDIR(obj->oo_inode->i_mode))
5277                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5278                 else
5279                         iam_rec = (struct iam_rec *)rec;
5280
5281                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5282
5283                 if (S_ISDIR(obj->oo_inode->i_mode))
5284                         osd_fid_unpack((struct lu_fid *)rec,
5285                                        (struct osd_fid_pack *)iam_rec);
5286                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5287                         osd_quota_unpack(obj, rec);
5288         }
5289
5290         iam_it_put(it);
5291         iam_it_fini(it);
5292         osd_ipd_put(env, bag, ipd);
5293
5294         LINVRNT(osd_invariant(obj));
5295
5296         RETURN(rc);
5297 }
5298
5299 static int osd_index_declare_iam_insert(const struct lu_env *env,
5300                                         struct dt_object *dt,
5301                                         const struct dt_rec *rec,
5302                                         const struct dt_key *key,
5303                                         struct thandle *handle)
5304 {
5305         struct osd_thandle *oh;
5306
5307         LASSERT(handle != NULL);
5308
5309         oh = container_of0(handle, struct osd_thandle, ot_super);
5310         LASSERT(oh->ot_handle == NULL);
5311
5312         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5313                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5314
5315         return 0;
5316 }
5317
5318 /**
5319  *      Inserts (key, value) pair in \a dt index object.
5320  *
5321  *      \param  dt      osd index object
5322  *      \param  key     key for index
5323  *      \param  rec     record reference
5324  *      \param  th      transaction handler
5325  *
5326  *      \retval  0  success
5327  *      \retval -ve failure
5328  */
5329 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5330                                 const struct dt_rec *rec,
5331                                 const struct dt_key *key, struct thandle *th)
5332 {
5333         struct osd_object *obj = osd_dt_obj(dt);
5334         struct iam_path_descr *ipd;
5335         struct osd_thandle *oh;
5336         struct iam_container *bag;
5337         struct osd_thread_info *oti = osd_oti_get(env);
5338         struct iam_rec *iam_rec;
5339         int rc;
5340
5341         ENTRY;
5342
5343         if (!dt_object_exists(dt))
5344                 RETURN(-ENOENT);
5345
5346         LINVRNT(osd_invariant(obj));
5347         LASSERT(!dt_object_remote(dt));
5348
5349         bag = &obj->oo_dir->od_container;
5350         LASSERT(bag->ic_object == obj->oo_inode);
5351         LASSERT(th != NULL);
5352
5353         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5354
5355         ipd = osd_idx_ipd_get(env, bag);
5356         if (unlikely(ipd == NULL))
5357                 RETURN(-ENOMEM);
5358
5359         oh = container_of0(th, struct osd_thandle, ot_super);
5360         LASSERT(oh->ot_handle != NULL);
5361         LASSERT(oh->ot_handle->h_transaction != NULL);
5362         if (S_ISDIR(obj->oo_inode->i_mode)) {
5363                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5364                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5365                              &oti->oti_fid);
5366         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5367                 /* pack quota uid/gid */
5368                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5369                 key = (const struct dt_key *)&oti->oti_quota_id;
5370                 /* pack quota record */
5371                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5372                 iam_rec = (struct iam_rec *)rec;
5373         } else {
5374                 iam_rec = (struct iam_rec *)rec;
5375         }
5376
5377         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5378                         iam_rec, ipd);
5379         osd_ipd_put(env, bag, ipd);
5380         LINVRNT(osd_invariant(obj));
5381         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5382         RETURN(rc);
5383 }
5384
5385 /**
5386  * Calls ldiskfs_add_entry() to add directory entry
5387  * into the directory. This is required for
5388  * interoperability mode (b11826)
5389  *
5390  * \retval   0, on success
5391  * \retval -ve, on error
5392  */
5393 static int __osd_ea_add_rec(struct osd_thread_info *info,
5394                             struct osd_object *pobj, struct inode  *cinode,
5395                             const char *name, const struct lu_fid *fid,
5396                             struct htree_lock *hlock, struct thandle *th)
5397 {
5398         struct ldiskfs_dentry_param *ldp;
5399         struct dentry *child;
5400         struct osd_thandle *oth;
5401         int rc;
5402
5403         oth = container_of(th, struct osd_thandle, ot_super);
5404         LASSERT(oth->ot_handle != NULL);
5405         LASSERT(oth->ot_handle->h_transaction != NULL);
5406         LASSERT(pobj->oo_inode);
5407
5408         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5409         if (unlikely(pobj->oo_inode ==
5410                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5411                 ldp->edp_magic = 0;
5412         else
5413                 osd_get_ldiskfs_dirent_param(ldp, fid);
5414         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5415         child->d_fsdata = (void *)ldp;
5416         ll_vfs_dq_init(pobj->oo_inode);
5417         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5418                                    child, cinode, hlock);
5419         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5420                 struct ldiskfs_dir_entry_2      *de;
5421                 struct buffer_head              *bh;
5422                 int                              rc1;
5423
5424                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5425                                             NULL, hlock);
5426                 if (!IS_ERR(bh)) {
5427                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5428                                                                bh);
5429                         if (rc1 == 0) {
5430                                 if (S_ISDIR(cinode->i_mode))
5431                                         de->file_type = LDISKFS_DIRENT_LUFID |
5432                                                         LDISKFS_FT_REG_FILE;
5433                                 else
5434                                         de->file_type = LDISKFS_DIRENT_LUFID |
5435                                                         LDISKFS_FT_DIR;
5436                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5437                                                               NULL, bh);
5438                         }
5439                         brelse(bh);
5440                 }
5441         }
5442
5443         RETURN(rc);
5444 }
5445
5446 /**
5447  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5448  * into the directory.Also sets flags into osd object to
5449  * indicate dot and dotdot are created. This is required for
5450  * interoperability mode (b11826)
5451  *
5452  * \param dir   directory for dot and dotdot fixup.
5453  * \param obj   child object for linking
5454  *
5455  * \retval   0, on success
5456  * \retval -ve, on error
5457  */
5458 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5459                               struct osd_object *dir,
5460                               struct inode *parent_dir, const char *name,
5461                               const struct lu_fid *dot_fid,
5462                               const struct lu_fid *dot_dot_fid,
5463                               struct thandle *th)
5464 {
5465         struct inode *inode = dir->oo_inode;
5466         struct osd_thandle *oth;
5467         int result = 0;
5468
5469         oth = container_of(th, struct osd_thandle, ot_super);
5470         LASSERT(oth->ot_handle->h_transaction != NULL);
5471         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5472
5473         if (strcmp(name, dot) == 0) {
5474                 if (dir->oo_compat_dot_created) {
5475                         result = -EEXIST;
5476                 } else {
5477                         LASSERT(inode->i_ino == parent_dir->i_ino);
5478                         dir->oo_compat_dot_created = 1;
5479                         result = 0;
5480                 }
5481         } else if (strcmp(name, dotdot) == 0) {
5482                 if (!dir->oo_compat_dot_created)
5483                         return -EINVAL;
5484                 /* in case of rename, dotdot is already created */
5485                 if (dir->oo_compat_dotdot_created) {
5486                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5487                                                 dot_dot_fid, NULL, th);
5488                 }
5489
5490                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5491                         struct lu_fid tfid = *dot_dot_fid;
5492
5493                         tfid.f_oid--;
5494                         result = osd_add_dot_dotdot_internal(info,
5495                                         dir->oo_inode, parent_dir, dot_fid,
5496                                         &tfid, oth);
5497                 } else {
5498                         result = osd_add_dot_dotdot_internal(info,
5499                                         dir->oo_inode, parent_dir, dot_fid,
5500                                         dot_dot_fid, oth);
5501                 }
5502
5503                 if (result == 0)
5504                         dir->oo_compat_dotdot_created = 1;
5505         }
5506
5507         return result;
5508 }
5509
5510
5511 /**
5512  * It will call the appropriate osd_add* function and return the
5513  * value, return by respective functions.
5514  */
5515 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5516                           struct inode *cinode, const char *name,
5517                           const struct lu_fid *fid, struct thandle *th)
5518 {
5519         struct osd_thread_info *info = osd_oti_get(env);
5520         struct htree_lock *hlock;
5521         int rc;
5522
5523         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5524
5525         if (name[0] == '.' && (name[1] == '\0' ||
5526                                (name[1] == '.' && name[2] == '\0'))) {
5527                 if (hlock != NULL) {
5528                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5529                                            pobj->oo_inode, 0);
5530                 } else {
5531                         down_write(&pobj->oo_ext_idx_sem);
5532                 }
5533
5534                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5535                                         lu_object_fid(&pobj->oo_dt.do_lu),
5536                                         fid, th);
5537         } else {
5538                 if (hlock != NULL) {
5539                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5540                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5541                 } else {
5542                         down_write(&pobj->oo_ext_idx_sem);
5543                 }
5544
5545                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5546                         struct lu_fid *tfid = &info->oti_fid;
5547
5548                         *tfid = *fid;
5549                         tfid->f_ver = ~0;
5550                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5551                                               tfid, hlock, th);
5552                 } else {
5553                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5554                                               hlock, th);
5555                 }
5556         }
5557         if (hlock != NULL)
5558                 ldiskfs_htree_unlock(hlock);
5559         else
5560                 up_write(&pobj->oo_ext_idx_sem);
5561
5562         return rc;
5563 }
5564
5565 static int
5566 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5567                       struct osd_idmap_cache *oic)
5568 {
5569         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5570         struct lu_fid *fid = &oic->oic_fid;
5571         struct osd_inode_id *id = &oic->oic_lid;
5572         struct inode *inode = NULL;
5573         int once = 0;
5574         bool insert;
5575         int rc;
5576
5577         ENTRY;
5578
5579         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5580                 RETURN(0);
5581
5582         if (thread_is_running(&scrub->os_thread) &&
5583             scrub->os_pos_current > id->oii_ino)
5584                 RETURN(0);
5585
5586         if (dev->od_auto_scrub_interval == AS_NEVER ||
5587             ktime_get_real_seconds() <
5588             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5589                 RETURN(0);
5590
5591 again:
5592         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5593         if (rc == -ENOENT) {
5594                 __u32 gen = id->oii_gen;
5595
5596                 insert = true;
5597                 if (inode != NULL)
5598                         goto trigger;
5599
5600                 inode = osd_iget(oti, dev, id);
5601                 /* The inode has been removed (by race maybe). */
5602                 if (IS_ERR(inode)) {
5603                         rc = PTR_ERR(inode);
5604
5605                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5606                 }
5607
5608                 /* The OI mapping is lost. */
5609                 if (gen != OSD_OII_NOGEN)
5610                         goto trigger;
5611
5612                 /*
5613                  * The inode may has been reused by others, we do not know,
5614                  * leave it to be handled by subsequent osd_fid_lookup().
5615                  */
5616                 GOTO(out, rc = 0);
5617         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5618                 GOTO(out, rc);
5619         }
5620
5621         insert = false;
5622
5623 trigger:
5624         if (thread_is_running(&scrub->os_thread)) {
5625                 if (inode == NULL) {
5626                         inode = osd_iget(oti, dev, id);
5627                         /* The inode has been removed (by race maybe). */
5628                         if (IS_ERR(inode)) {
5629                                 rc = PTR_ERR(inode);
5630
5631                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5632                         }
5633                 }
5634
5635                 rc = osd_oii_insert(dev, oic, insert);
5636                 /*
5637                  * There is race condition between osd_oi_lookup and OI scrub.
5638                  * The OI scrub finished just after osd_oi_lookup() failure.
5639                  * Under such case, it is unnecessary to trigger OI scrub again,
5640                  * but try to call osd_oi_lookup() again.
5641                  */
5642                 if (unlikely(rc == -EAGAIN))
5643                         goto again;
5644
5645                 if (!S_ISDIR(inode->i_mode))
5646                         rc = 0;
5647                 else
5648                         rc = osd_check_lmv(oti, dev, inode, oic);
5649
5650                 GOTO(out, rc);
5651         }
5652
5653         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5654                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5655                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5656                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5657                        "%s: trigger partial OI scrub for RPC inconsistency "
5658                        "checking FID "DFID": rc = %d\n",
5659                        osd_dev2name(dev), PFID(fid), rc);
5660                 if (rc == 0 || rc == -EALREADY)
5661                         goto again;
5662         }
5663
5664         GOTO(out, rc);
5665
5666 out:
5667         if (inode)
5668                 iput(inode);
5669
5670         RETURN(rc);
5671 }
5672
5673 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5674                                struct osd_device *dev,
5675                                struct osd_idmap_cache *oic,
5676                                struct lu_fid *fid, __u32 ino)
5677 {
5678         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5679         struct inode *inode;
5680         int rc;
5681
5682         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5683         inode = osd_iget(oti, dev, &oic->oic_lid);
5684         if (IS_ERR(inode)) {
5685                 fid_zero(&oic->oic_fid);
5686                 return PTR_ERR(inode);
5687         }
5688
5689         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5690         iput(inode);
5691         if (rc != 0)
5692                 fid_zero(&oic->oic_fid);
5693         else
5694                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5695         return rc;
5696 }
5697
5698 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5699                       struct osd_inode_id *id, const struct lu_fid *fid)
5700 {
5701         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5702                id->oii_ino, id->oii_gen, info);
5703         info->oti_cache.oic_lid = *id;
5704         info->oti_cache.oic_fid = *fid;
5705         info->oti_cache.oic_dev = osd;
5706 }
5707
5708 /**
5709  * Get parent FID from the linkEA.
5710  *
5711  * For a directory which parent resides on remote MDT, to satisfy the
5712  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5713  * the other hand, to make the lookup(..) on the directory can return
5714  * the real parent FID, we append the real parent FID after its ".."
5715  * name entry in the /REMOTE_PARENT_DIR.
5716  *
5717  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5718  * backup. So after the restore, we cannot get the right parent FID from
5719  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5720  * we have stored the real parent FID in the directory object's linkEA,
5721  * we can parse the linkEA for the real parent FID.
5722  *
5723  * \param[in] env       pointer to the thread context
5724  * \param[in] obj       pointer to the object to be handled
5725  * \param[out]fid       pointer to the buffer to hold the parent FID
5726  *
5727  * \retval              0 for getting the real parent FID successfully
5728  * \retval              negative error number on failure
5729  */
5730 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5731                                     struct osd_object *obj,
5732                                     struct lu_fid *fid)
5733 {
5734         struct osd_thread_info *oti = osd_oti_get(env);
5735         struct lu_buf *buf = &oti->oti_big_buf;
5736         struct dentry *dentry = &oti->oti_obj_dentry;
5737         struct inode *inode = obj->oo_inode;
5738         struct linkea_data ldata = { NULL };
5739         int rc;
5740
5741         ENTRY;
5742
5743         fid_zero(fid);
5744         if (!S_ISDIR(inode->i_mode))
5745                 RETURN(-EIO);
5746
5747 again:
5748         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5749                              buf->lb_buf, buf->lb_len);
5750         if (rc == -ERANGE) {
5751                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5752                                      NULL, 0);
5753                 if (rc > 0) {
5754                         lu_buf_realloc(buf, rc);
5755                         if (buf->lb_buf == NULL)
5756                                 RETURN(-ENOMEM);
5757
5758                         goto again;
5759                 }
5760         }
5761
5762         if (unlikely(rc == 0))
5763                 RETURN(-ENODATA);
5764
5765         if (rc < 0)
5766                 RETURN(rc);
5767
5768         if (unlikely(buf->lb_buf == NULL)) {
5769                 lu_buf_realloc(buf, rc);
5770                 if (buf->lb_buf == NULL)
5771                         RETURN(-ENOMEM);
5772
5773                 goto again;
5774         }
5775
5776         ldata.ld_buf = buf;
5777         rc = linkea_init_with_rec(&ldata);
5778         if (!rc) {
5779                 linkea_first_entry(&ldata);
5780                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5781         }
5782
5783         RETURN(rc);
5784 }
5785
5786 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5787                                     struct inode *inode,
5788                                     const struct lu_fid *pfid,
5789                                     const char *name, const int namelen)
5790 {
5791         struct osd_thread_info *oti = osd_oti_get(env);
5792         struct lu_buf *buf = &oti->oti_big_buf;
5793         struct dentry *dentry = &oti->oti_obj_dentry;
5794         struct linkea_data ldata = { NULL };
5795         struct lu_name cname = { .ln_name = name,
5796                                  .ln_namelen = namelen };
5797         int rc;
5798
5799         ENTRY;
5800
5801 again:
5802         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5803                              buf->lb_buf, buf->lb_len);
5804         if (rc == -ERANGE)
5805                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5806
5807         if (rc < 0)
5808                 RETURN(rc);
5809
5810         if (unlikely(rc == 0))
5811                 RETURN(-ENODATA);
5812
5813         if (buf->lb_len < rc) {
5814                 lu_buf_realloc(buf, rc);
5815                 if (buf->lb_buf == NULL)
5816                         RETURN(-ENOMEM);
5817
5818                 goto again;
5819         }
5820
5821         ldata.ld_buf = buf;
5822         rc = linkea_init_with_rec(&ldata);
5823         if (!rc)
5824                 rc = linkea_links_find(&ldata, &cname, pfid);
5825
5826         RETURN(rc);
5827 }
5828
5829 /**
5830  * Calls ->lookup() to find dentry. From dentry get inode and
5831  * read inode's ea to get fid. This is required for  interoperability
5832  * mode (b11826)
5833  *
5834  * \retval   0, on success
5835  * \retval -ve, on error
5836  */
5837 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5838                              struct dt_rec *rec, const struct dt_key *key)
5839 {
5840         struct inode *dir = obj->oo_inode;
5841         struct dentry *dentry;
5842         struct ldiskfs_dir_entry_2 *de;
5843         struct buffer_head *bh;
5844         struct lu_fid *fid = (struct lu_fid *)rec;
5845         struct htree_lock *hlock = NULL;
5846         int ino;
5847         int rc;
5848
5849         ENTRY;
5850
5851         LASSERT(dir->i_op != NULL);
5852         LASSERT(dir->i_op->lookup != NULL);
5853
5854         dentry = osd_child_dentry_get(env, obj,
5855                                       (char *)key, strlen((char *)key));
5856
5857         if (obj->oo_hl_head != NULL) {
5858                 hlock = osd_oti_get(env)->oti_hlock;
5859                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5860                                    dir, LDISKFS_HLOCK_LOOKUP);
5861         } else {
5862                 down_read(&obj->oo_ext_idx_sem);
5863         }
5864
5865         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5866         if (!IS_ERR(bh)) {
5867                 struct osd_thread_info *oti = osd_oti_get(env);
5868                 struct osd_inode_id *id = &oti->oti_id;
5869                 struct osd_idmap_cache *oic = &oti->oti_cache;
5870                 struct osd_device *dev = osd_obj2dev(obj);
5871
5872                 ino = le32_to_cpu(de->inode);
5873                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5874                         brelse(bh);
5875                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5876                         GOTO(out, rc);
5877                 }
5878
5879                 rc = osd_get_fid_from_dentry(de, rec);
5880
5881                 /* done with de, release bh */
5882                 brelse(bh);
5883                 if (rc != 0) {
5884                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5885                                 const char *name = (const char *)key;
5886
5887                                 /*
5888                                  * If the parent is on remote MDT, and there
5889                                  * is no FID-in-dirent, then we have to get
5890                                  * the parent FID from the linkEA.
5891                                  */
5892                                 if (likely(strlen(name) == 2 &&
5893                                            name[0] == '.' && name[1] == '.'))
5894                                         rc = osd_get_pfid_from_linkea(env, obj,
5895                                                                       fid);
5896                         } else {
5897                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5898                         }
5899                 } else {
5900                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5901                 }
5902
5903                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5904                         fid_zero(&oic->oic_fid);
5905
5906                         GOTO(out, rc);
5907                 }
5908
5909                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5910                 rc = osd_consistency_check(oti, dev, oic);
5911                 if (rc == -ENOENT)
5912                         fid_zero(&oic->oic_fid);
5913                 else
5914                         /* Other error should not affect lookup result. */
5915                         rc = 0;
5916         } else {
5917                 rc = PTR_ERR(bh);
5918         }
5919
5920         GOTO(out, rc);
5921
5922 out:
5923         if (hlock != NULL)
5924                 ldiskfs_htree_unlock(hlock);
5925         else
5926                 up_read(&obj->oo_ext_idx_sem);
5927         return rc;
5928 }
5929
5930 static int osd_index_declare_ea_insert(const struct lu_env *env,
5931                                        struct dt_object *dt,
5932                                        const struct dt_rec *rec,
5933                                        const struct dt_key *key,
5934                                        struct thandle *handle)
5935 {
5936         struct osd_thandle *oh;
5937         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5938         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5939         const struct lu_fid *fid = rec1->rec_fid;
5940         int credits, rc = 0;
5941         struct osd_idmap_cache *idc;
5942
5943         ENTRY;
5944
5945         LASSERT(!dt_object_remote(dt));
5946         LASSERT(handle != NULL);
5947         LASSERT(fid != NULL);
5948         LASSERT(rec1->rec_type != 0);
5949
5950         oh = container_of0(handle, struct osd_thandle, ot_super);
5951         LASSERT(oh->ot_handle == NULL);
5952
5953         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5954
5955         /*
5956          * we can't call iget() while a transactions is running
5957          * (this can lead to a deadlock), but we need to know
5958          * inum and object type. so we find this information at
5959          * declaration and cache in per-thread info
5960          */
5961         idc = osd_idc_find_or_init(env, osd, fid);
5962         if (IS_ERR(idc))
5963                 RETURN(PTR_ERR(idc));
5964         if (idc->oic_remote) {
5965                 /*
5966                  * a reference to remote inode is represented by an
5967                  * agent inode which we have to create
5968                  */
5969                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5970                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5971         }
5972
5973         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5974
5975         if (osd_dt_obj(dt)->oo_inode != NULL) {
5976                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5977
5978                 /*
5979                  * We ignore block quota on meta pool (MDTs), so needn't
5980                  * calculate how many blocks will be consumed by this index
5981                  * insert
5982                  */
5983                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5984                                            i_gid_read(inode),
5985                                            i_projid_read(inode), 0,
5986                                            oh, osd_dt_obj(dt), NULL,
5987                                            OSD_QID_BLK);
5988                 if (rc)
5989                         RETURN(rc);
5990
5991 #ifdef HAVE_PROJECT_QUOTA
5992                 /*
5993                  * Reserve credits for local agent inode to transfer
5994                  * to 0, quota enforcement is ignored in this case.
5995                  */
5996                 if (idc->oic_remote &&
5997                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5998                     i_projid_read(inode) != 0)
5999                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6000                                                   0, i_projid_read(inode),
6001                                                   0, false, PRJQUOTA, true);
6002 #endif
6003         }
6004
6005         RETURN(rc);
6006 }
6007
6008 /**
6009  * Index add function for interoperability mode (b11826).
6010  * It will add the directory entry.This entry is needed to
6011  * maintain name->fid mapping.
6012  *
6013  * \param key it is key i.e. file entry to be inserted
6014  * \param rec it is value of given key i.e. fid
6015  *
6016  * \retval   0, on success
6017  * \retval -ve, on error
6018  */
6019 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6020                                const struct dt_rec *rec,
6021                                const struct dt_key *key, struct thandle *th)
6022 {
6023         struct osd_object *obj = osd_dt_obj(dt);
6024         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6025         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6026         const struct lu_fid *fid = rec1->rec_fid;
6027         const char *name = (const char *)key;
6028         struct osd_thread_info *oti = osd_oti_get(env);
6029         struct inode *child_inode = NULL;
6030         struct osd_idmap_cache *idc;
6031         int rc;
6032
6033         ENTRY;
6034
6035         if (!dt_object_exists(dt))
6036                 RETURN(-ENOENT);
6037
6038         LASSERT(osd_invariant(obj));
6039         LASSERT(!dt_object_remote(dt));
6040         LASSERT(th != NULL);
6041
6042         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6043
6044         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6045
6046         idc = osd_idc_find(env, osd, fid);
6047         if (unlikely(idc == NULL)) {
6048                 idc = osd_idc_find_or_init(env, osd, fid);
6049                 if (IS_ERR(idc)) {
6050                         /*
6051                          * this dt_insert() wasn't declared properly, so
6052                          * FID is missing in OI cache. we better do not
6053                          * lookup FID in FLDB/OI and don't risk to deadlock,
6054                          * but in some special cases (lfsck testing, etc)
6055                          * it's much simpler than fixing a caller.
6056                          *
6057                          * normally this error should be placed after the first
6058                          * find, but migrate may attach source stripes to
6059                          * target, which doesn't create stripes.
6060                          */
6061                         CERROR("%s: "DFID" wasn't declared for insert\n",
6062                                osd_name(osd), PFID(fid));
6063                         dump_stack();
6064                         RETURN(PTR_ERR(idc));
6065                 }
6066         }
6067
6068         if (idc->oic_remote) {
6069                 /* Insert remote entry */
6070                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6071                         child_inode =
6072                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6073                 } else {
6074                         child_inode = osd_create_local_agent_inode(env, osd,
6075                                         obj, fid, rec1->rec_type & S_IFMT, th);
6076                         if (IS_ERR(child_inode))
6077                                 RETURN(PTR_ERR(child_inode));
6078                 }
6079         } else {
6080                 /* Insert local entry */
6081                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6082                         /* for a reason OI cache wasn't filled properly */
6083                         CERROR("%s: OIC for "DFID" isn't filled\n",
6084                                osd_name(osd), PFID(fid));
6085                         RETURN(-EINVAL);
6086                 }
6087                 child_inode = oti->oti_inode;
6088                 if (unlikely(child_inode == NULL)) {
6089                         struct ldiskfs_inode_info *lii;
6090
6091                         OBD_ALLOC_PTR(lii);
6092                         if (lii == NULL)
6093                                 RETURN(-ENOMEM);
6094                         child_inode = oti->oti_inode = &lii->vfs_inode;
6095                 }
6096                 child_inode->i_sb = osd_sb(osd);
6097                 child_inode->i_ino = idc->oic_lid.oii_ino;
6098                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6099         }
6100
6101         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6102
6103         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6104                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6105
6106         if (child_inode && child_inode != oti->oti_inode)
6107                 iput(child_inode);
6108         LASSERT(osd_invariant(obj));
6109         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6110         RETURN(rc);
6111 }
6112
6113 /**
6114  *  Initialize osd Iterator for given osd index object.
6115  *
6116  *  \param  dt      osd index object
6117  */
6118
6119 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6120                                      struct dt_object *dt,
6121                                      __u32 unused)
6122 {
6123         struct osd_it_iam *it;
6124         struct osd_object *obj = osd_dt_obj(dt);
6125         struct lu_object *lo = &dt->do_lu;
6126         struct iam_path_descr *ipd;
6127         struct iam_container *bag = &obj->oo_dir->od_container;
6128
6129         if (!dt_object_exists(dt))
6130                 return ERR_PTR(-ENOENT);
6131
6132         OBD_ALLOC_PTR(it);
6133         if (it == NULL)
6134                 return ERR_PTR(-ENOMEM);
6135
6136         ipd = osd_it_ipd_get(env, bag);
6137         if (likely(ipd != NULL)) {
6138                 it->oi_obj = obj;
6139                 it->oi_ipd = ipd;
6140                 lu_object_get(lo);
6141                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6142                 return (struct dt_it *)it;
6143         } else {
6144                 OBD_FREE_PTR(it);
6145                 return ERR_PTR(-ENOMEM);
6146         }
6147 }
6148
6149 /**
6150  * free given Iterator.
6151  */
6152 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6153 {
6154         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6155         struct osd_object *obj = it->oi_obj;
6156
6157         iam_it_fini(&it->oi_it);
6158         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6159         osd_object_put(env, obj);
6160         OBD_FREE_PTR(it);
6161 }
6162
6163 /**
6164  *  Move Iterator to record specified by \a key
6165  *
6166  *  \param  di      osd iterator
6167  *  \param  key     key for index
6168  *
6169  *  \retval +ve  di points to record with least key not larger than key
6170  *  \retval  0   di points to exact matched key
6171  *  \retval -ve  failure
6172  */
6173
6174 static int osd_it_iam_get(const struct lu_env *env,
6175                           struct dt_it *di, const struct dt_key *key)
6176 {
6177         struct osd_thread_info *oti = osd_oti_get(env);
6178         struct osd_it_iam *it = (struct osd_it_iam *)di;
6179
6180         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6181                 /* swab quota uid/gid */
6182                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6183                 key = (struct dt_key *)&oti->oti_quota_id;
6184         }
6185
6186         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6187 }
6188
6189 /**
6190  *  Release Iterator
6191  *
6192  *  \param  di      osd iterator
6193  */
6194 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6195 {
6196         struct osd_it_iam *it = (struct osd_it_iam *)di;
6197
6198         iam_it_put(&it->oi_it);
6199 }
6200
6201 /**
6202  *  Move iterator by one record
6203  *
6204  *  \param  di      osd iterator
6205  *
6206  *  \retval +1   end of container reached
6207  *  \retval  0   success
6208  *  \retval -ve  failure
6209  */
6210
6211 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6212 {
6213         struct osd_it_iam *it = (struct osd_it_iam *)di;
6214
6215         return iam_it_next(&it->oi_it);
6216 }
6217
6218 /**
6219  * Return pointer to the key under iterator.
6220  */
6221
6222 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6223                                      const struct dt_it *di)
6224 {
6225         struct osd_thread_info *oti = osd_oti_get(env);
6226         struct osd_it_iam *it = (struct osd_it_iam *)di;
6227         struct osd_object *obj = it->oi_obj;
6228         struct dt_key *key;
6229
6230         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6231
6232         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6233                 /* swab quota uid/gid */
6234                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6235                 key = (struct dt_key *)&oti->oti_quota_id;
6236         }
6237
6238         return key;
6239 }
6240
6241 /**
6242  * Return size of key under iterator (in bytes)
6243  */
6244
6245 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6246 {
6247         struct osd_it_iam *it = (struct osd_it_iam *)di;
6248
6249         return iam_it_key_size(&it->oi_it);
6250 }
6251
6252 static inline void
6253 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6254 {
6255         /* check if file type is required */
6256         if (ent->lde_attrs & LUDA_TYPE) {
6257                 struct luda_type *lt;
6258                 int align = sizeof(*lt) - 1;
6259
6260                 len = (len + align) & ~align;
6261                 lt = (struct luda_type *)(ent->lde_name + len);
6262                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6263         }
6264
6265         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6266 }
6267
6268 /**
6269  * build lu direct from backend fs dirent.
6270  */
6271
6272 static inline void
6273 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6274                    char *name, __u16 namelen, __u16 type, __u32 attr)
6275 {
6276         ent->lde_attrs = attr | LUDA_FID;
6277         fid_cpu_to_le(&ent->lde_fid, fid);
6278
6279         ent->lde_hash = cpu_to_le64(offset);
6280         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6281
6282         strncpy(ent->lde_name, name, namelen);
6283         ent->lde_name[namelen] = '\0';
6284         ent->lde_namelen = cpu_to_le16(namelen);
6285
6286         /* append lustre attributes */
6287         osd_it_append_attrs(ent, namelen, type);
6288 }
6289
6290 /**
6291  * Return pointer to the record under iterator.
6292  */
6293 static int osd_it_iam_rec(const struct lu_env *env,
6294                           const struct dt_it *di,
6295                           struct dt_rec *dtrec, __u32 attr)
6296 {
6297         struct osd_it_iam *it = (struct osd_it_iam *)di;
6298         struct osd_thread_info *info = osd_oti_get(env);
6299
6300         ENTRY;
6301
6302         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6303                 const struct osd_fid_pack *rec;
6304                 struct lu_fid *fid = &info->oti_fid;
6305                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6306                 char *name;
6307                 int namelen;
6308                 __u64 hash;
6309                 int rc;
6310
6311                 name = (char *)iam_it_key_get(&it->oi_it);
6312                 if (IS_ERR(name))
6313                         RETURN(PTR_ERR(name));
6314
6315                 namelen = iam_it_key_size(&it->oi_it);
6316
6317                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6318                 if (IS_ERR(rec))
6319                         RETURN(PTR_ERR(rec));
6320
6321                 rc = osd_fid_unpack(fid, rec);
6322                 if (rc)
6323                         RETURN(rc);
6324
6325                 hash = iam_it_store(&it->oi_it);
6326
6327                 /* IAM does not store object type in IAM index (dir) */
6328                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6329                                    0, LUDA_FID);
6330         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6331                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6332                            (struct iam_rec *)dtrec);
6333                 osd_quota_unpack(it->oi_obj, dtrec);
6334         } else {
6335                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6336                            (struct iam_rec *)dtrec);
6337         }
6338
6339         RETURN(0);
6340 }
6341
6342 /**
6343  * Returns cookie for current Iterator position.
6344  */
6345 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6346 {
6347         struct osd_it_iam *it = (struct osd_it_iam *)di;
6348
6349         return iam_it_store(&it->oi_it);
6350 }
6351
6352 /**
6353  * Restore iterator from cookie.
6354  *
6355  * \param  di      osd iterator
6356  * \param  hash    Iterator location cookie
6357  *
6358  * \retval +ve  di points to record with least key not larger than key.
6359  * \retval  0   di points to exact matched key
6360  * \retval -ve  failure
6361  */
6362
6363 static int osd_it_iam_load(const struct lu_env *env,
6364                            const struct dt_it *di, __u64 hash)
6365 {
6366         struct osd_it_iam *it = (struct osd_it_iam *)di;
6367
6368         return iam_it_load(&it->oi_it, hash);
6369 }
6370
6371 static const struct dt_index_operations osd_index_iam_ops = {
6372         .dio_lookup         = osd_index_iam_lookup,
6373         .dio_declare_insert = osd_index_declare_iam_insert,
6374         .dio_insert         = osd_index_iam_insert,
6375         .dio_declare_delete = osd_index_declare_iam_delete,
6376         .dio_delete         = osd_index_iam_delete,
6377         .dio_it     = {
6378                 .init     = osd_it_iam_init,
6379                 .fini     = osd_it_iam_fini,
6380                 .get      = osd_it_iam_get,
6381                 .put      = osd_it_iam_put,
6382                 .next     = osd_it_iam_next,
6383                 .key      = osd_it_iam_key,
6384                 .key_size = osd_it_iam_key_size,
6385                 .rec      = osd_it_iam_rec,
6386                 .store    = osd_it_iam_store,
6387                 .load     = osd_it_iam_load
6388         }
6389 };
6390
6391
6392 /**
6393  * Creates or initializes iterator context.
6394  *
6395  * \retval struct osd_it_ea, iterator structure on success
6396  *
6397  */
6398 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6399                                     struct dt_object *dt,
6400                                     __u32 attr)
6401 {
6402         struct osd_object *obj = osd_dt_obj(dt);
6403         struct osd_thread_info *info = osd_oti_get(env);
6404         struct osd_it_ea *oie;
6405         struct file *file;
6406         struct lu_object *lo = &dt->do_lu;
6407         struct dentry *obj_dentry;
6408
6409         ENTRY;
6410
6411         if (!dt_object_exists(dt) || obj->oo_destroyed)
6412                 RETURN(ERR_PTR(-ENOENT));
6413
6414         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6415         if (oie == NULL)
6416                 RETURN(ERR_PTR(-ENOMEM));
6417         obj_dentry = &oie->oie_dentry;
6418
6419         obj_dentry->d_inode = obj->oo_inode;
6420         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6421         obj_dentry->d_name.hash = 0;
6422
6423         oie->oie_rd_dirent       = 0;
6424         oie->oie_it_dirent       = 0;
6425         oie->oie_dirent          = NULL;
6426         if (unlikely(!info->oti_it_ea_buf_used)) {
6427                 oie->oie_buf = info->oti_it_ea_buf;
6428                 info->oti_it_ea_buf_used = 1;
6429         } else {
6430                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6431                 if (oie->oie_buf == NULL)
6432                         RETURN(ERR_PTR(-ENOMEM));
6433         }
6434         oie->oie_obj = obj;
6435
6436         file = &oie->oie_file;
6437
6438         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6439         if (attr & LUDA_64BITHASH)
6440                 file->f_mode    = FMODE_64BITHASH;
6441         else
6442                 file->f_mode    = FMODE_32BITHASH;
6443         file->f_path.dentry     = obj_dentry;
6444         file->f_mapping         = obj->oo_inode->i_mapping;
6445         file->f_op              = obj->oo_inode->i_fop;
6446         set_file_inode(file, obj->oo_inode);
6447
6448         lu_object_get(lo);
6449         RETURN((struct dt_it *)oie);
6450 }
6451
6452 /**
6453  * Destroy or finishes iterator context.
6454  *
6455  * \param di iterator structure to be destroyed
6456  */
6457 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6458 {
6459         struct osd_thread_info *info = osd_oti_get(env);
6460         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6461         struct osd_object *obj = oie->oie_obj;
6462         struct inode *inode = obj->oo_inode;
6463
6464         ENTRY;
6465         oie->oie_file.f_op->release(inode, &oie->oie_file);
6466         osd_object_put(env, obj);
6467         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6468                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6469         else
6470                 info->oti_it_ea_buf_used = 0;
6471         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6472         EXIT;
6473 }
6474
6475 /**
6476  * It position the iterator at given key, so that next lookup continues from
6477  * that key Or it is similar to dio_it->load() but based on a key,
6478  * rather than file position.
6479  *
6480  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6481  * to the beginning.
6482  *
6483  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6484  */
6485 static int osd_it_ea_get(const struct lu_env *env,
6486                          struct dt_it *di, const struct dt_key *key)
6487 {
6488         struct osd_it_ea *it = (struct osd_it_ea *)di;
6489
6490         ENTRY;
6491         LASSERT(((const char *)key)[0] == '\0');
6492         it->oie_file.f_pos = 0;
6493         it->oie_rd_dirent = 0;
6494         it->oie_it_dirent = 0;
6495         it->oie_dirent = NULL;
6496
6497         RETURN(+1);
6498 }
6499
6500 /**
6501  * Does nothing
6502  */
6503 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6504 {
6505 }
6506
6507 struct osd_filldir_cbs {
6508 #ifdef HAVE_DIR_CONTEXT
6509         struct dir_context ctx;
6510 #endif
6511         struct osd_it_ea  *it;
6512 };
6513 /**
6514  * It is called internally by ->readdir(). It fills the
6515  * iterator's in-memory data structure with required
6516  * information i.e. name, namelen, rec_size etc.
6517  *
6518  * \param buf in which information to be filled in.
6519  * \param name name of the file in given dir
6520  *
6521  * \retval 0 on success
6522  * \retval 1 on buffer full
6523  */
6524 #ifdef HAVE_FILLDIR_USE_CTX
6525 static int osd_ldiskfs_filldir(struct dir_context *buf,
6526 #else
6527 static int osd_ldiskfs_filldir(void *buf,
6528 #endif
6529                                const char *name, int namelen,
6530                                loff_t offset, __u64 ino, unsigned int d_type)
6531 {
6532         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6533         struct osd_object *obj = it->oie_obj;
6534         struct osd_it_ea_dirent *ent = it->oie_dirent;
6535         struct lu_fid *fid = &ent->oied_fid;
6536         struct osd_fid_pack *rec;
6537
6538         ENTRY;
6539
6540 /* this should never happen */
6541         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6542                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6543                 RETURN(-EIO);
6544         }
6545
6546         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6547             OSD_IT_EA_BUFSIZE)
6548                 RETURN(1);
6549
6550         /* "." is just the object itself. */
6551         if (namelen == 1 && name[0] == '.') {
6552                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6553         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6554                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6555                 if (osd_fid_unpack(fid, rec) != 0)
6556                         fid_zero(fid);
6557         } else {
6558                 fid_zero(fid);
6559         }
6560         d_type &= ~LDISKFS_DIRENT_LUFID;
6561
6562         /* NOT export local root. */
6563         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6564                 ino = obj->oo_inode->i_ino;
6565                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6566         }
6567
6568         ent->oied_ino     = ino;
6569         ent->oied_off     = offset;
6570         ent->oied_namelen = namelen;
6571         ent->oied_type    = d_type;
6572
6573         memcpy(ent->oied_name, name, namelen);
6574
6575         it->oie_rd_dirent++;
6576         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6577         RETURN(0);
6578 }
6579
6580 /**
6581  * Calls ->readdir() to load a directory entry at a time
6582  * and stored it in iterator's in-memory data structure.
6583  *
6584  * \param di iterator's in memory structure
6585  *
6586  * \retval   0 on success
6587  * \retval -ve on error
6588  * \retval +1 reach the end of entry
6589  */
6590 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6591                                const struct dt_it *di)
6592 {
6593         struct osd_it_ea *it = (struct osd_it_ea *)di;
6594         struct osd_object *obj = it->oie_obj;
6595         struct inode *inode = obj->oo_inode;
6596         struct htree_lock *hlock = NULL;
6597         struct file *filp = &it->oie_file;
6598         int rc = 0;
6599         struct osd_filldir_cbs buf = {
6600 #ifdef HAVE_DIR_CONTEXT
6601                 .ctx.actor = osd_ldiskfs_filldir,
6602 #endif
6603                 .it = it
6604         };
6605
6606         ENTRY;
6607         it->oie_dirent = it->oie_buf;
6608         it->oie_rd_dirent = 0;
6609
6610         if (obj->oo_hl_head != NULL) {
6611                 hlock = osd_oti_get(env)->oti_hlock;
6612                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6613                                    inode, LDISKFS_HLOCK_READDIR);
6614         } else {
6615                 down_read(&obj->oo_ext_idx_sem);
6616         }
6617
6618 #ifdef HAVE_DIR_CONTEXT
6619         buf.ctx.pos = filp->f_pos;
6620 #ifdef HAVE_ITERATE_SHARED
6621         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6622 #else
6623         rc = inode->i_fop->iterate(filp, &buf.ctx);
6624 #endif
6625         filp->f_pos = buf.ctx.pos;
6626 #else
6627         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6628 #endif
6629
6630         if (hlock != NULL)
6631                 ldiskfs_htree_unlock(hlock);
6632         else
6633                 up_read(&obj->oo_ext_idx_sem);
6634
6635         if (it->oie_rd_dirent == 0) {
6636                 /*
6637                  * If it does not get any dirent, it means it has been reached
6638                  * to the end of the dir
6639                  */
6640                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6641                 if (rc == 0)
6642                         rc = 1;
6643         } else {
6644                 it->oie_dirent = it->oie_buf;
6645                 it->oie_it_dirent = 1;
6646         }
6647
6648         RETURN(rc);
6649 }
6650
6651 /**
6652  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6653  * to load a directory entry at a time and stored it in
6654  * iterator's in-memory data structure.
6655  *
6656  * \param di iterator's in memory structure
6657  *
6658  * \retval +ve iterator reached to end
6659  * \retval   0 iterator not reached to end
6660  * \retval -ve on error
6661  */
6662 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6663 {
6664         struct osd_it_ea *it = (struct osd_it_ea *)di;
6665         int rc;
6666
6667         ENTRY;
6668
6669         if (it->oie_it_dirent < it->oie_rd_dirent) {
6670                 it->oie_dirent =
6671                         (void *)it->oie_dirent +
6672                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6673                                        it->oie_dirent->oied_namelen);
6674                 it->oie_it_dirent++;
6675                 rc = 0;
6676         } else {
6677                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6678                         rc = 1;
6679                 else
6680                         rc = osd_ldiskfs_it_fill(env, di);
6681         }
6682
6683         RETURN(rc);
6684 }
6685
6686 /**
6687  * Returns the key at current position from iterator's in memory structure.
6688  *
6689  * \param di iterator's in memory structure
6690  *
6691  * \retval key i.e. struct dt_key on success
6692  */
6693 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6694                                     const struct dt_it *di)
6695 {
6696         struct osd_it_ea *it = (struct osd_it_ea *)di;
6697
6698         return (struct dt_key *)it->oie_dirent->oied_name;
6699 }
6700
6701 /**
6702  * Returns key's size at current position from iterator's in memory structure.
6703  *
6704  * \param di iterator's in memory structure
6705  *
6706  * \retval key_size i.e. struct dt_key on success
6707  */
6708 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6709 {
6710         struct osd_it_ea *it = (struct osd_it_ea *)di;
6711
6712         return it->oie_dirent->oied_namelen;
6713 }
6714
6715 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6716 {
6717         if (LDISKFS_DIR_REC_LEN(de) >=
6718             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6719                 return true;
6720
6721         return false;
6722 }
6723
6724 static inline bool
6725 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6726                      unsigned int blocksize, bool dotdot)
6727 {
6728         if (dotdot)
6729                 return osd_dotdot_has_space(de);
6730
6731         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6732             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6733                 return true;
6734
6735         return false;
6736 }
6737
6738 static int
6739 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6740                     handle_t *jh, struct dentry *dentry,
6741                     const struct lu_fid *fid, struct buffer_head *bh,
6742                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6743                     bool dotdot)
6744 {
6745         struct inode *dir = dentry->d_parent->d_inode;
6746         struct inode *inode = dentry->d_inode;
6747         struct osd_fid_pack *rec;
6748         struct ldiskfs_dentry_param *ldp;
6749         int namelen = dentry->d_name.len;
6750         int rc;
6751         struct osd_thread_info *info = osd_oti_get(env);
6752
6753         ENTRY;
6754
6755         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6756                 RETURN(0);
6757
6758         /* There is enough space to hold the FID-in-dirent. */
6759         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6760                 rc = ldiskfs_journal_get_write_access(jh, bh);
6761                 if (rc != 0)
6762                         RETURN(rc);
6763
6764                 de->name[namelen] = 0;
6765                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6766                 rec->fp_len = sizeof(struct lu_fid) + 1;
6767                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6768                 de->file_type |= LDISKFS_DIRENT_LUFID;
6769                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6770
6771                 RETURN(rc);
6772         }
6773
6774         LASSERT(!dotdot);
6775
6776         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6777         if (rc != 0)
6778                 RETURN(rc);
6779
6780         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6781         osd_get_ldiskfs_dirent_param(ldp, fid);
6782         dentry->d_fsdata = (void *)ldp;
6783         ll_vfs_dq_init(dir);
6784         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6785         /*
6786          * It is too bad, we cannot reinsert the name entry back.
6787          * That means we lose it!
6788          */
6789         if (rc != 0)
6790                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6791                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6792                        osd_ino2name(inode),
6793                        dir->i_ino, dir->i_generation, namelen,
6794                        dentry->d_name.name, PFID(fid), rc);
6795
6796         RETURN(rc);
6797 }
6798
6799 static int
6800 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6801                         struct osd_it_ea *it, struct lu_fid *fid,
6802                         struct osd_inode_id *id, __u32 *attr)
6803 {
6804         struct osd_thread_info *info = osd_oti_get(env);
6805         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6806         struct osd_device *dev = osd_obj2dev(obj);
6807         struct super_block *sb = osd_sb(dev);
6808         const char *devname = osd_name(dev);
6809         struct osd_it_ea_dirent *ent = it->oie_dirent;
6810         struct inode *dir = obj->oo_inode;
6811         struct htree_lock *hlock = NULL;
6812         struct buffer_head *bh = NULL;
6813         handle_t *jh = NULL;
6814         struct ldiskfs_dir_entry_2 *de;
6815         struct dentry *dentry;
6816         struct inode *inode;
6817         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6818         int credits;
6819         int rc;
6820         bool dotdot = false;
6821         bool dirty = false;
6822
6823         ENTRY;
6824
6825         if (ent->oied_name[0] == '.') {
6826                 if (ent->oied_namelen == 1)
6827                         RETURN(0);
6828
6829                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6830                         dotdot = true;
6831         }
6832
6833         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6834         inode = osd_iget(info, dev, id);
6835         if (IS_ERR(inode)) {
6836                 rc = PTR_ERR(inode);
6837                 if (rc == -ENOENT || rc == -ESTALE) {
6838                         /*
6839                          * Maybe dangling name entry, or
6840                          * corrupted directory entry.
6841                          */
6842                         *attr |= LUDA_UNKNOWN;
6843                         rc = 0;
6844                 } else {
6845                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6846                                "check_repair, dir = %lu/%u, name = %.*s, "
6847                                "ino = %llu, rc = %d\n",
6848                                devname, dir->i_ino, dir->i_generation,
6849                                ent->oied_namelen, ent->oied_name,
6850                                ent->oied_ino, rc);
6851                 }
6852
6853                 RETURN(rc);
6854         }
6855
6856         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6857                                            ent->oied_namelen);
6858         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6859         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6860                 lma = NULL;
6861         else if (rc != 0)
6862                 GOTO(out, rc);
6863
6864         /*
6865          * We need to ensure that the name entry is still valid.
6866          * Because it may be removed or renamed by other already.
6867          *
6868          * The unlink or rename operation will start journal before PDO lock,
6869          * so to avoid deadlock, here we need to start journal handle before
6870          * related PDO lock also. But because we do not know whether there
6871          * will be something to be repaired before PDO lock, we just start
6872          * journal without conditions.
6873          *
6874          * We may need to remove the name entry firstly, then insert back.
6875          * One credit is for user quota file update.
6876          * One credit is for group quota file update.
6877          * Two credits are for dirty inode.
6878          */
6879         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6880                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6881
6882         if (dev->od_dirent_journal != 0) {
6883
6884 again:
6885                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6886                 if (IS_ERR(jh)) {
6887                         rc = PTR_ERR(jh);
6888                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6889                                "check_repair, dir = %lu/%u, credits = %d, "
6890                                "name = %.*s, ino = %llu: rc = %d\n",
6891                                devname, dir->i_ino, dir->i_generation, credits,
6892                                ent->oied_namelen, ent->oied_name,
6893                                ent->oied_ino, rc);
6894
6895                         GOTO(out_inode, rc);
6896                 }
6897
6898                 if (obj->oo_hl_head != NULL) {
6899                         hlock = osd_oti_get(env)->oti_hlock;
6900                         /*
6901                          * "0" means exclusive lock for the whole directory.
6902                          * We need to prevent others access such name entry
6903                          * during the delete + insert. Neither HLOCK_ADD nor
6904                          * HLOCK_DEL cannot guarantee the atomicity.
6905                          */
6906                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6907                 } else {
6908                         down_write(&obj->oo_ext_idx_sem);
6909                 }
6910         } else {
6911                 if (obj->oo_hl_head != NULL) {
6912                         hlock = osd_oti_get(env)->oti_hlock;
6913                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6914                                            LDISKFS_HLOCK_LOOKUP);
6915                 } else {
6916                         down_read(&obj->oo_ext_idx_sem);
6917                 }
6918         }
6919
6920         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6921         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6922                 *attr |= LUDA_IGNORE;
6923
6924                 GOTO(out, rc = 0);
6925         }
6926
6927         /*
6928          * For dotdot entry, if there is not enough space to hold the
6929          * FID-in-dirent, just keep them there. It only happens when the
6930          * device upgraded from 1.8 or restored from MDT file-level backup.
6931          * For the whole directory, only dotdot entry have no FID-in-dirent
6932          * and needs to get FID from LMA when readdir, it will not affect the
6933          * performance much.
6934          */
6935         if (dotdot && !osd_dotdot_has_space(de)) {
6936                 *attr |= LUDA_UNKNOWN;
6937
6938                 GOTO(out, rc = 0);
6939         }
6940
6941         if (lma != NULL) {
6942                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6943                         GOTO(out, rc = 0);
6944
6945                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6946                         struct lu_fid *tfid = &lma->lma_self_fid;
6947
6948                         if (likely(dotdot &&
6949                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6950                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6951                                 /*
6952                                  * It must be REMOTE_PARENT_DIR and as the
6953                                  * 'dotdot' entry of remote directory
6954                                  */
6955                                 *attr |= LUDA_IGNORE;
6956                         } else {
6957                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6958                                        "parent directory, but got %.*s under "
6959                                        "dir = %lu/%u with the FID "DFID"\n",
6960                                        devname, ent->oied_namelen,
6961                                        ent->oied_name, dir->i_ino,
6962                                        dir->i_generation, PFID(tfid));
6963
6964                                 *attr |= LUDA_UNKNOWN;
6965                         }
6966
6967                         GOTO(out, rc = 0);
6968                 }
6969         }
6970
6971         if (!fid_is_zero(fid)) {
6972                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6973                                               ent->oied_namelen);
6974                 if (rc == -ENOENT ||
6975                     (rc == -ENODATA &&
6976                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6977                         /*
6978                          * linkEA does not recognize the dirent entry,
6979                          * it may because the dirent entry corruption
6980                          * and points to other's inode.
6981                          */
6982                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6983                                "recognize the dirent, dir = %lu/%u, "
6984                                " name = %.*s, ino = %llu, "
6985                                DFID": rc = %d\n", devname, dir->i_ino,
6986                                dir->i_generation, ent->oied_namelen,
6987                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6988                         *attr |= LUDA_UNKNOWN;
6989
6990                         GOTO(out, rc = 0);
6991                 }
6992
6993                 if (rc && rc != -ENODATA) {
6994                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6995                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6996                                DFID": rc = %d\n", devname, dir->i_ino,
6997                                dir->i_generation, ent->oied_namelen,
6998                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6999                         *attr |= LUDA_UNKNOWN;
7000
7001                         GOTO(out, rc = 0);
7002                 }
7003         }
7004
7005         if (lma != NULL) {
7006                 /*
7007                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7008                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7009                  */
7010                 if (*attr & LUDA_VERIFY_DRYRUN) {
7011                         *fid = lma->lma_self_fid;
7012                         *attr |= LUDA_REPAIR;
7013
7014                         GOTO(out, rc = 0);
7015                 }
7016
7017                 if (jh == NULL) {
7018                         brelse(bh);
7019                         dev->od_dirent_journal = 1;
7020                         if (hlock != NULL) {
7021                                 ldiskfs_htree_unlock(hlock);
7022                                 hlock = NULL;
7023                         } else {
7024                                 up_read(&obj->oo_ext_idx_sem);
7025                         }
7026
7027                         goto again;
7028                 }
7029
7030                 *fid = lma->lma_self_fid;
7031                 dirty = true;
7032                 /* Update or append the FID-in-dirent. */
7033                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7034                                          bh, de, hlock, dotdot);
7035                 if (rc == 0)
7036                         *attr |= LUDA_REPAIR;
7037                 else
7038                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7039                                "the dirent, dir = %lu/%u, name = %.*s, "
7040                                "ino = %llu, "DFID": rc = %d\n",
7041                                devname, dir->i_ino, dir->i_generation,
7042                                ent->oied_namelen, ent->oied_name,
7043                                ent->oied_ino, PFID(fid), rc);
7044         } else {
7045                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7046                 if (*attr & LUDA_VERIFY_DRYRUN) {
7047                         if (fid_is_sane(fid)) {
7048                                 *attr |= LUDA_REPAIR;
7049                         } else if (dev->od_index == 0) {
7050                                 lu_igif_build(fid, inode->i_ino,
7051                                               inode->i_generation);
7052                                 *attr |= LUDA_UPGRADE;
7053                         }
7054
7055                         GOTO(out, rc = 0);
7056                 }
7057
7058                 if (jh == NULL) {
7059                         brelse(bh);
7060                         dev->od_dirent_journal = 1;
7061                         if (hlock != NULL) {
7062                                 ldiskfs_htree_unlock(hlock);
7063                                 hlock = NULL;
7064                         } else {
7065                                 up_read(&obj->oo_ext_idx_sem);
7066                         }
7067
7068                         goto again;
7069                 }
7070
7071                 dirty = true;
7072                 if (unlikely(fid_is_sane(fid))) {
7073                         /*
7074                          * FID-in-dirent exists, but FID-in-LMA is lost.
7075                          * Trust the FID-in-dirent, and add FID-in-LMA.
7076                          */
7077                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7078                         if (rc == 0)
7079                                 *attr |= LUDA_REPAIR;
7080                         else
7081                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7082                                        "update dirent, dir = %lu/%u, "
7083                                        "name = %.*s, ino = %llu, "
7084                                        DFID": rc = %d\n",
7085                                        devname, dir->i_ino, dir->i_generation,
7086                                        ent->oied_namelen, ent->oied_name,
7087                                        ent->oied_ino, PFID(fid), rc);
7088                 } else if (dev->od_index == 0) {
7089                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7090                         /*
7091                          * It is probably IGIF object. Only aappend the
7092                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7093                          */
7094                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7095                                                  bh, de, hlock, dotdot);
7096                         if (rc == 0)
7097                                 *attr |= LUDA_UPGRADE;
7098                         else
7099                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7100                                        "after the dirent, dir = %lu/%u, "
7101                                        "name = %.*s, ino = %llu, "
7102                                        DFID": rc = %d\n",
7103                                        devname, dir->i_ino, dir->i_generation,
7104                                        ent->oied_namelen, ent->oied_name,
7105                                        ent->oied_ino, PFID(fid), rc);
7106                 }
7107         }
7108
7109         GOTO(out, rc);
7110
7111 out:
7112         if (!IS_ERR(bh))
7113                 brelse(bh);
7114         if (hlock != NULL) {
7115                 ldiskfs_htree_unlock(hlock);
7116         } else {
7117                 if (dev->od_dirent_journal != 0)
7118                         up_write(&obj->oo_ext_idx_sem);
7119                 else
7120                         up_read(&obj->oo_ext_idx_sem);
7121         }
7122
7123         if (jh != NULL)
7124                 ldiskfs_journal_stop(jh);
7125
7126 out_inode:
7127         iput(inode);
7128         if (rc >= 0 && !dirty)
7129                 dev->od_dirent_journal = 0;
7130
7131         return rc;
7132 }
7133
7134 /**
7135  * Returns the value at current position from iterator's in memory structure.
7136  *
7137  * \param di struct osd_it_ea, iterator's in memory structure
7138  * \param attr attr requested for dirent.
7139  * \param lde lustre dirent
7140  *
7141  * \retval   0 no error and \param lde has correct lustre dirent.
7142  * \retval -ve on error
7143  */
7144 static inline int osd_it_ea_rec(const struct lu_env *env,
7145                                 const struct dt_it *di,
7146                                 struct dt_rec *dtrec, __u32 attr)
7147 {
7148         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7149         struct osd_object      *obj   = it->oie_obj;
7150         struct osd_device      *dev   = osd_obj2dev(obj);
7151         struct osd_thread_info *oti   = osd_oti_get(env);
7152         struct osd_inode_id    *id    = &oti->oti_id;
7153         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7154         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7155         __u32 ino = it->oie_dirent->oied_ino;
7156         int rc = 0;
7157
7158         ENTRY;
7159
7160         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7161
7162         if (attr & LUDA_VERIFY) {
7163                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7164                         attr |= LUDA_IGNORE;
7165                         /*
7166                          * If the parent is on remote MDT, and there
7167                          * is no FID-in-dirent, then we have to get
7168                          * the parent FID from the linkEA.
7169                          */
7170                         if (!fid_is_sane(fid) &&
7171                             it->oie_dirent->oied_namelen == 2 &&
7172                             it->oie_dirent->oied_name[0] == '.' &&
7173                             it->oie_dirent->oied_name[1] == '.')
7174                                 osd_get_pfid_from_linkea(env, obj, fid);
7175                 } else {
7176                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7177                                                      &attr);
7178                 }
7179
7180                 if (!fid_is_sane(fid))
7181                         attr |= LUDA_UNKNOWN;
7182         } else {
7183                 attr &= ~LU_DIRENT_ATTRS_MASK;
7184                 if (!fid_is_sane(fid)) {
7185                         bool is_dotdot = false;
7186
7187                         if (it->oie_dirent->oied_namelen == 2 &&
7188                             it->oie_dirent->oied_name[0] == '.' &&
7189                             it->oie_dirent->oied_name[1] == '.')
7190                                 is_dotdot = true;
7191                         /*
7192                          * If the parent is on remote MDT, and there
7193                          * is no FID-in-dirent, then we have to get
7194                          * the parent FID from the linkEA.
7195                          */
7196                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7197                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7198                         } else {
7199                                 if (is_dotdot == false &&
7200                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7201                                         RETURN(-ENOENT);
7202
7203                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7204                         }
7205                 } else {
7206                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7207                 }
7208         }
7209
7210         /* Pack the entry anyway, at least the offset is right. */
7211         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7212                            it->oie_dirent->oied_name,
7213                            it->oie_dirent->oied_namelen,
7214                            it->oie_dirent->oied_type, attr);
7215
7216         if (rc < 0)
7217                 RETURN(rc);
7218
7219         if (osd_remote_fid(env, dev, fid))
7220                 RETURN(0);
7221
7222         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7223                 osd_add_oi_cache(oti, dev, id, fid);
7224
7225         RETURN(rc > 0 ? 0 : rc);
7226 }
7227
7228 /**
7229  * Returns the record size size at current position.
7230  *
7231  * This function will return record(lu_dirent) size in bytes.
7232  *
7233  * \param[in] env       execution environment
7234  * \param[in] di        iterator's in memory structure
7235  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7236  *                      calculate the lu_dirent size.
7237  *
7238  * \retval      record size(in bytes & in memory) of the current lu_dirent
7239  *              entry.
7240  */
7241 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7242                               __u32 attr)
7243 {
7244         struct osd_it_ea *it = (struct osd_it_ea *)di;
7245
7246         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7247 }
7248
7249 /**
7250  * Returns a cookie for current position of the iterator head, so that
7251  * user can use this cookie to load/start the iterator next time.
7252  *
7253  * \param di iterator's in memory structure
7254  *
7255  * \retval cookie for current position, on success
7256  */
7257 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7258 {
7259         struct osd_it_ea *it = (struct osd_it_ea *)di;
7260
7261         return it->oie_dirent->oied_off;
7262 }
7263
7264 /**
7265  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7266  * to load a directory entry at a time and stored it i inn,
7267  * in iterator's in-memory data structure.
7268  *
7269  * \param di struct osd_it_ea, iterator's in memory structure
7270  *
7271  * \retval +ve on success
7272  * \retval -ve on error
7273  */
7274 static int osd_it_ea_load(const struct lu_env *env,
7275                           const struct dt_it *di, __u64 hash)
7276 {
7277         struct osd_it_ea *it = (struct osd_it_ea *)di;
7278         int rc;
7279
7280         ENTRY;
7281         it->oie_file.f_pos = hash;
7282
7283         rc =  osd_ldiskfs_it_fill(env, di);
7284         if (rc > 0)
7285                 rc = -ENODATA;
7286
7287         if (rc == 0)
7288                 rc = 1;
7289
7290         RETURN(rc);
7291 }
7292
7293 /**
7294  * Index lookup function for interoperability mode (b11826).
7295  *
7296  * \param key,  key i.e. file name to be searched
7297  *
7298  * \retval +ve, on success
7299  * \retval -ve, on error
7300  */
7301 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7302                                struct dt_rec *rec, const struct dt_key *key)
7303 {
7304         struct osd_object *obj = osd_dt_obj(dt);
7305         int rc = 0;
7306
7307         ENTRY;
7308
7309         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7310         LINVRNT(osd_invariant(obj));
7311
7312         rc = osd_ea_lookup_rec(env, obj, rec, key);
7313         if (rc == 0)
7314                 rc = 1;
7315         RETURN(rc);
7316 }
7317
7318 /**
7319  * Index and Iterator operations for interoperability
7320  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7321  */
7322 static const struct dt_index_operations osd_index_ea_ops = {
7323         .dio_lookup         = osd_index_ea_lookup,
7324         .dio_declare_insert = osd_index_declare_ea_insert,
7325         .dio_insert         = osd_index_ea_insert,
7326         .dio_declare_delete = osd_index_declare_ea_delete,
7327         .dio_delete         = osd_index_ea_delete,
7328         .dio_it     = {
7329                 .init     = osd_it_ea_init,
7330                 .fini     = osd_it_ea_fini,
7331                 .get      = osd_it_ea_get,
7332                 .put      = osd_it_ea_put,
7333                 .next     = osd_it_ea_next,
7334                 .key      = osd_it_ea_key,
7335                 .key_size = osd_it_ea_key_size,
7336                 .rec      = osd_it_ea_rec,
7337                 .rec_size = osd_it_ea_rec_size,
7338                 .store    = osd_it_ea_store,
7339                 .load     = osd_it_ea_load
7340         }
7341 };
7342
7343 static void *osd_key_init(const struct lu_context *ctx,
7344                           struct lu_context_key *key)
7345 {
7346         struct osd_thread_info *info;
7347
7348         OBD_ALLOC_PTR(info);
7349         if (info == NULL)
7350                 return ERR_PTR(-ENOMEM);
7351
7352         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7353         if (info->oti_it_ea_buf == NULL)
7354                 goto out_free_info;
7355
7356         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7357
7358         info->oti_hlock = ldiskfs_htree_lock_alloc();
7359         if (info->oti_hlock == NULL)
7360                 goto out_free_ea;
7361
7362         return info;
7363
7364 out_free_ea:
7365         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7366 out_free_info:
7367         OBD_FREE_PTR(info);
7368         return ERR_PTR(-ENOMEM);
7369 }
7370
7371 static void osd_key_fini(const struct lu_context *ctx,
7372                          struct lu_context_key *key, void *data)
7373 {
7374         struct osd_thread_info *info = data;
7375         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7376         struct osd_idmap_cache *idc = info->oti_ins_cache;
7377
7378         if (info->oti_dio_pages) {
7379                 int i;
7380                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7381                         if (info->oti_dio_pages[i])
7382                                 __free_page(info->oti_dio_pages[i]);
7383                 }
7384                 OBD_FREE(info->oti_dio_pages,
7385                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7386         }
7387
7388         if (info->oti_inode != NULL)
7389                 OBD_FREE_PTR(lli);
7390         if (info->oti_hlock != NULL)
7391                 ldiskfs_htree_lock_free(info->oti_hlock);
7392         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7393         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7394         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7395         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7396         lu_buf_free(&info->oti_big_buf);
7397         if (idc != NULL) {
7398                 LASSERT(info->oti_ins_cache_size > 0);
7399                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7400                 info->oti_ins_cache = NULL;
7401                 info->oti_ins_cache_size = 0;
7402         }
7403         OBD_FREE_PTR(info);
7404 }
7405
7406 static void osd_key_exit(const struct lu_context *ctx,
7407                          struct lu_context_key *key, void *data)
7408 {
7409         struct osd_thread_info *info = data;
7410
7411         LASSERT(info->oti_r_locks == 0);
7412         LASSERT(info->oti_w_locks == 0);
7413         LASSERT(info->oti_txns    == 0);
7414 }
7415
7416 /* type constructor/destructor: osd_type_init, osd_type_fini */
7417 LU_TYPE_INIT_FINI(osd, &osd_key);
7418
7419 struct lu_context_key osd_key = {
7420         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7421         .lct_init = osd_key_init,
7422         .lct_fini = osd_key_fini,
7423         .lct_exit = osd_key_exit
7424 };
7425
7426
7427 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7428                            const char *name, struct lu_device *next)
7429 {
7430         struct osd_device *osd = osd_dev(d);
7431
7432         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7433             sizeof(osd->od_svname))
7434                 return -E2BIG;
7435         return osd_procfs_init(osd, name);
7436 }
7437
7438 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7439 {
7440         struct seq_server_site *ss = osd_seq_site(osd);
7441         int rc;
7442
7443         ENTRY;
7444
7445         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7446                 RETURN(0);
7447
7448         if (unlikely(ss == NULL))
7449                 RETURN(-ENODEV);
7450
7451         OBD_ALLOC_PTR(osd->od_cl_seq);
7452         if (osd->od_cl_seq == NULL)
7453                 RETURN(-ENOMEM);
7454
7455         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7456                              osd->od_svname, ss->ss_server_seq);
7457         if (rc != 0) {
7458                 OBD_FREE_PTR(osd->od_cl_seq);
7459                 osd->od_cl_seq = NULL;
7460                 RETURN(rc);
7461         }
7462
7463         if (ss->ss_node_id == 0) {
7464                 /*
7465                  * If the OSD on the sequence controller(MDT0), then allocate
7466                  * sequence here, otherwise allocate sequence after connected
7467                  * to MDT0 (see mdt_register_lwp_callback()).
7468                  */
7469                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7470                                    &osd->od_cl_seq->lcs_space, env);
7471         }
7472
7473         RETURN(rc);
7474 }
7475
7476 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7477 {
7478         if (osd->od_cl_seq == NULL)
7479                 return;
7480
7481         seq_client_fini(osd->od_cl_seq);
7482         OBD_FREE_PTR(osd->od_cl_seq);
7483         osd->od_cl_seq = NULL;
7484 }
7485
7486 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7487 {
7488         ENTRY;
7489
7490         /* shutdown quota slave instance associated with the device */
7491         if (o->od_quota_slave_md != NULL) {
7492                 struct qsd_instance *qsd = o->od_quota_slave_md;
7493
7494                 o->od_quota_slave_md = NULL;
7495                 qsd_fini(env, qsd);
7496         }
7497
7498         if (o->od_quota_slave_dt != NULL) {
7499                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7500
7501                 o->od_quota_slave_dt = NULL;
7502                 qsd_fini(env, qsd);
7503         }
7504
7505         osd_fid_fini(env, o);
7506         osd_scrub_cleanup(env, o);
7507
7508         RETURN(0);
7509 }
7510
7511 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7512 {
7513         ENTRY;
7514
7515         if (o->od_mnt != NULL) {
7516                 shrink_dcache_sb(osd_sb(o));
7517                 osd_sync(env, &o->od_dt_dev);
7518
7519                 mntput(o->od_mnt);
7520                 o->od_mnt = NULL;
7521         }
7522
7523         EXIT;
7524 }
7525
7526 static int osd_mount(const struct lu_env *env,
7527                      struct osd_device *o, struct lustre_cfg *cfg)
7528 {
7529         const char *name = lustre_cfg_string(cfg, 0);
7530         const char *dev = lustre_cfg_string(cfg, 1);
7531         const char *opts;
7532         unsigned long page, s_flags, lmd_flags = 0;
7533         struct page *__page;
7534         struct file_system_type *type;
7535         char *options = NULL;
7536         char *str;
7537         struct osd_thread_info *info = osd_oti_get(env);
7538         struct lu_fid *fid = &info->oti_fid;
7539         struct inode *inode;
7540         int rc = 0, force_over_512tb = 0;
7541
7542         ENTRY;
7543
7544         if (o->od_mnt != NULL)
7545                 RETURN(0);
7546
7547         if (strlen(dev) >= sizeof(o->od_mntdev))
7548                 RETURN(-E2BIG);
7549         strcpy(o->od_mntdev, dev);
7550
7551         str = lustre_cfg_string(cfg, 2);
7552         s_flags = simple_strtoul(str, NULL, 0);
7553         str = strstr(str, ":");
7554         if (str)
7555                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7556         opts = lustre_cfg_string(cfg, 3);
7557 #ifdef __BIG_ENDIAN
7558         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7559                 CERROR("%s: device %s extents feature is not guaranteed to "
7560                        "work on big-endian systems. Use \"bigendian_extents\" "
7561                        "mount option to override.\n", name, dev);
7562                 RETURN(-EINVAL);
7563         }
7564 #endif
7565 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7566         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7567                 CWARN("force_over_128tb option is deprecated. "
7568                       "Filesystems less than 512TB can be created without any "
7569                       "force options. Use force_over_512tb option for "
7570                       "filesystems greater than 512TB.\n");
7571         }
7572 #endif
7573 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7574         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7575                 CWARN("force_over_256tb option is deprecated. "
7576                       "Filesystems less than 512TB can be created without any "
7577                       "force options. Use force_over_512tb option for "
7578                       "filesystems greater than 512TB.\n");
7579         }
7580 #endif
7581
7582         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7583                 force_over_512tb = 1;
7584
7585         __page = alloc_page(GFP_KERNEL);
7586         if (__page == NULL)
7587                 GOTO(out, rc = -ENOMEM);
7588         page = (unsigned long)page_address(__page);
7589         options = (char *)page;
7590         *options = '\0';
7591         if (opts != NULL) {
7592                 /* strip out the options for back compatiblity */
7593                 static char *sout[] = {
7594                         "mballoc",
7595                         "iopen",
7596                         "noiopen",
7597                         "iopen_nopriv",
7598                         "extents",
7599                         "noextents",
7600                         /* strip out option we processed in osd */
7601                         "bigendian_extents",
7602                         "force_over_128tb",
7603                         "force_over_256tb",
7604                         "force_over_512tb",
7605                         NULL
7606                 };
7607                 strncat(options, opts, PAGE_SIZE);
7608                 for (rc = 0, str = options; sout[rc]; ) {
7609                         char *op = strstr(str, sout[rc]);
7610
7611                         if (op == NULL) {
7612                                 rc++;
7613                                 str = options;
7614                                 continue;
7615                         }
7616                         if (op == options || *(op - 1) == ',') {
7617                                 str = op + strlen(sout[rc]);
7618                                 if (*str == ',' || *str == '\0') {
7619                                         *str == ',' ? str++ : str;
7620                                         memmove(op, str, strlen(str) + 1);
7621                                 }
7622                         }
7623                         for (str = op; *str != ',' && *str != '\0'; str++)
7624                                 ;
7625                 }
7626         } else {
7627                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7628         }
7629
7630         /* Glom up mount options */
7631         if (*options != '\0')
7632                 strncat(options, ",", PAGE_SIZE);
7633         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7634
7635         type = get_fs_type("ldiskfs");
7636         if (!type) {
7637                 CERROR("%s: cannot find ldiskfs module\n", name);
7638                 GOTO(out, rc = -ENODEV);
7639         }
7640
7641         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7642         module_put(type->owner);
7643
7644         if (IS_ERR(o->od_mnt)) {
7645                 rc = PTR_ERR(o->od_mnt);
7646                 o->od_mnt = NULL;
7647                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7648                 GOTO(out, rc);
7649         }
7650
7651         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7652                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7653                                  force_over_512tb == 0) {
7654                 CERROR("%s: device %s LDISKFS does not support filesystems "
7655                        "greater than 512TB and can cause data corruption. "
7656                        "Use \"force_over_512tb\" mount option to override.\n",
7657                        name, dev);
7658                 GOTO(out_mnt, rc = -EINVAL);
7659         }
7660
7661         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7662                 if (priv_dev_set_rdonly) {
7663                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7664                         o->od_dt_dev.dd_rdonly = 1;
7665                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7666                                       name);
7667                 } else {
7668                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7669                                       name);
7670
7671                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7672                 }
7673         } else if (priv_dev_check_rdonly &&
7674                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7675                 CERROR("%s: underlying device %s is marked as "
7676                        "read-only. Setup failed\n", name, dev);
7677
7678                 GOTO(out_mnt, rc = -EROFS);
7679         }
7680
7681         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7682                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7683                 GOTO(out_mnt, rc = -EINVAL);
7684         }
7685
7686 #ifdef LDISKFS_MOUNT_DIRDATA
7687         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7688                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7689         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7690                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7691                       "enabling the dirdata feature. If you do not want to "
7692                       "downgrade to Lustre-1.x again, you can enable it via "
7693                       "'tune2fs -O dirdata device'\n", name, dev);
7694 #endif
7695         inode = osd_sb(o)->s_root->d_inode;
7696         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7697         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7698         if (rc != 0) {
7699                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7700                 GOTO(out_mnt, rc);
7701         }
7702
7703         if (lmd_flags & LMD_FLG_NOSCRUB)
7704                 o->od_auto_scrub_interval = AS_NEVER;
7705
7706         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7707                 /* do not use pagecache with flash-backed storage */
7708                 o->od_writethrough_cache = 0;
7709                 o->od_read_cache = 0;
7710         }
7711
7712         GOTO(out, rc = 0);
7713
7714 out_mnt:
7715         mntput(o->od_mnt);
7716         o->od_mnt = NULL;
7717
7718 out:
7719         if (__page)
7720                 __free_page(__page);
7721
7722         return rc;
7723 }
7724
7725 static struct lu_device *osd_device_fini(const struct lu_env *env,
7726                                          struct lu_device *d)
7727 {
7728         struct osd_device *o = osd_dev(d);
7729
7730         ENTRY;
7731
7732         osd_index_backup(env, o, false);
7733         osd_shutdown(env, o);
7734         osd_procfs_fini(o);
7735         osd_obj_map_fini(o);
7736         osd_umount(env, o);
7737
7738         RETURN(NULL);
7739 }
7740
7741 static int osd_device_init0(const struct lu_env *env,
7742                             struct osd_device *o,
7743                             struct lustre_cfg *cfg)
7744 {
7745         struct lu_device *l = osd2lu_dev(o);
7746         struct osd_thread_info *info;
7747         int rc;
7748         int cplen = 0;
7749
7750         /* if the module was re-loaded, env can loose its keys */
7751         rc = lu_env_refill((struct lu_env *)env);
7752         if (rc)
7753                 GOTO(out, rc);
7754         info = osd_oti_get(env);
7755         LASSERT(info);
7756
7757         l->ld_ops = &osd_lu_ops;
7758         o->od_dt_dev.dd_ops = &osd_dt_ops;
7759
7760         spin_lock_init(&o->od_osfs_lock);
7761         mutex_init(&o->od_otable_mutex);
7762         INIT_LIST_HEAD(&o->od_orphan_list);
7763         INIT_LIST_HEAD(&o->od_index_backup_list);
7764         INIT_LIST_HEAD(&o->od_index_restore_list);
7765         spin_lock_init(&o->od_lock);
7766         o->od_index_backup_policy = LIBP_NONE;
7767         o->od_t10_type = 0;
7768
7769         o->od_read_cache = 1;
7770         o->od_writethrough_cache = 1;
7771         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7772
7773         o->od_auto_scrub_interval = AS_DEFAULT;
7774
7775         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7776                         sizeof(o->od_svname));
7777         if (cplen >= sizeof(o->od_svname)) {
7778                 rc = -E2BIG;
7779                 GOTO(out, rc);
7780         }
7781
7782         o->od_index_backup_stop = 0;
7783         o->od_index = -1; /* -1 means index is invalid */
7784         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7785         if (rc == LDD_F_SV_TYPE_OST)
7786                 o->od_is_ost = 1;
7787
7788         o->od_full_scrub_ratio = OFSR_DEFAULT;
7789         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7790         rc = osd_mount(env, o, cfg);
7791         if (rc != 0)
7792                 GOTO(out, rc);
7793
7794         /* Can only check block device after mount */
7795         o->od_nonrotational =
7796                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7797
7798         rc = osd_obj_map_init(env, o);
7799         if (rc != 0)
7800                 GOTO(out_mnt, rc);
7801
7802         rc = lu_site_init(&o->od_site, l);
7803         if (rc != 0)
7804                 GOTO(out_compat, rc);
7805         o->od_site.ls_bottom_dev = l;
7806
7807         rc = lu_site_init_finish(&o->od_site);
7808         if (rc != 0)
7809                 GOTO(out_site, rc);
7810
7811         INIT_LIST_HEAD(&o->od_ios_list);
7812         /* setup scrub, including OI files initialization */
7813         o->od_in_init = 1;
7814         rc = osd_scrub_setup(env, o);
7815         o->od_in_init = 0;
7816         if (rc < 0)
7817                 GOTO(out_site, rc);
7818
7819         rc = osd_procfs_init(o, o->od_svname);
7820         if (rc != 0) {
7821                 CERROR("%s: can't initialize procfs: rc = %d\n",
7822                        o->od_svname, rc);
7823                 GOTO(out_scrub, rc);
7824         }
7825
7826         LASSERT(l->ld_site->ls_linkage.next != NULL);
7827         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7828
7829         /* initialize quota slave instance */
7830         /* currently it's no need to prepare qsd_instance_md for OST */
7831         if (!o->od_is_ost) {
7832                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7833                                                 &o->od_dt_dev,
7834                                                 o->od_proc_entry, true);
7835                 if (IS_ERR(o->od_quota_slave_md)) {
7836                         rc = PTR_ERR(o->od_quota_slave_md);
7837                         o->od_quota_slave_md = NULL;
7838                         GOTO(out_procfs, rc);
7839                 }
7840         }
7841
7842         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7843                                         o->od_proc_entry, false);
7844
7845         if (IS_ERR(o->od_quota_slave_dt)) {
7846                 if (o->od_quota_slave_md != NULL) {
7847                         qsd_fini(env, o->od_quota_slave_md);
7848                         o->od_quota_slave_md = NULL;
7849                 }
7850
7851                 rc = PTR_ERR(o->od_quota_slave_dt);
7852                 o->od_quota_slave_dt = NULL;
7853                 GOTO(out_procfs, rc);
7854         }
7855
7856         RETURN(0);
7857
7858 out_procfs:
7859         osd_procfs_fini(o);
7860 out_scrub:
7861         osd_scrub_cleanup(env, o);
7862 out_site:
7863         lu_site_fini(&o->od_site);
7864 out_compat:
7865         osd_obj_map_fini(o);
7866 out_mnt:
7867         osd_umount(env, o);
7868 out:
7869         return rc;
7870 }
7871
7872 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7873                                           struct lu_device_type *t,
7874                                           struct lustre_cfg *cfg)
7875 {
7876         struct osd_device *o;
7877         int rc;
7878
7879         OBD_ALLOC_PTR(o);
7880         if (o == NULL)
7881                 return ERR_PTR(-ENOMEM);
7882
7883         rc = dt_device_init(&o->od_dt_dev, t);
7884         if (rc == 0) {
7885                 /*
7886                  * Because the ctx might be revived in dt_device_init,
7887                  * refill the env here
7888                  */
7889                 lu_env_refill((struct lu_env *)env);
7890                 rc = osd_device_init0(env, o, cfg);
7891                 if (rc)
7892                         dt_device_fini(&o->od_dt_dev);
7893         }
7894
7895         if (unlikely(rc != 0))
7896                 OBD_FREE_PTR(o);
7897
7898         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7899 }
7900
7901 static struct lu_device *osd_device_free(const struct lu_env *env,
7902                                          struct lu_device *d)
7903 {
7904         struct osd_device *o = osd_dev(d);
7905
7906         ENTRY;
7907
7908         /* XXX: make osd top device in order to release reference */
7909         d->ld_site->ls_top_dev = d;
7910         lu_site_purge(env, d->ld_site, -1);
7911         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7912                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7913                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7914         }
7915         lu_site_fini(&o->od_site);
7916         dt_device_fini(&o->od_dt_dev);
7917         OBD_FREE_PTR(o);
7918         RETURN(NULL);
7919 }
7920
7921 static int osd_process_config(const struct lu_env *env,
7922                               struct lu_device *d, struct lustre_cfg *cfg)
7923 {
7924         struct osd_device *o = osd_dev(d);
7925         ssize_t count;
7926         int rc;
7927
7928         ENTRY;
7929
7930         switch (cfg->lcfg_command) {
7931         case LCFG_SETUP:
7932                 rc = osd_mount(env, o, cfg);
7933                 break;
7934         case LCFG_CLEANUP:
7935                 /*
7936                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7937                  * that may happend if hit failure during mount process.
7938                  */
7939                 osd_index_backup(env, o, false);
7940                 lu_dev_del_linkage(d->ld_site, d);
7941                 rc = osd_shutdown(env, o);
7942                 break;
7943         case LCFG_PARAM:
7944                 LASSERT(&o->od_dt_dev);
7945                 count  = class_modify_config(cfg, PARAM_OSD,
7946                                              &o->od_dt_dev.dd_kobj);
7947                 if (count < 0)
7948                         count = class_modify_config(cfg, PARAM_OST,
7949                                                     &o->od_dt_dev.dd_kobj);
7950                 rc = count > 0 ? 0 : count;
7951                 break;
7952         case LCFG_PRE_CLEANUP:
7953                 osd_scrub_stop(o);
7954                 osd_index_backup(env, o,
7955                                  o->od_index_backup_policy != LIBP_NONE);
7956                 rc = 0;
7957                 break;
7958         default:
7959                 rc = -ENOSYS;
7960         }
7961
7962         RETURN(rc);
7963 }
7964
7965 static int osd_recovery_complete(const struct lu_env *env,
7966                                  struct lu_device *d)
7967 {
7968         struct osd_device *osd = osd_dev(d);
7969         int rc = 0;
7970
7971         ENTRY;
7972
7973         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7974                 RETURN(0);
7975
7976         /*
7977          * start qsd instance on recovery completion, this notifies the quota
7978          * slave code that we are about to process new requests now
7979          */
7980         rc = qsd_start(env, osd->od_quota_slave_dt);
7981         if (rc == 0 && osd->od_quota_slave_md != NULL)
7982                 rc = qsd_start(env, osd->od_quota_slave_md);
7983
7984         RETURN(rc);
7985 }
7986
7987 /*
7988  * we use exports to track all osd users
7989  */
7990 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7991                            struct obd_device *obd, struct obd_uuid *cluuid,
7992                            struct obd_connect_data *data, void *localdata)
7993 {
7994         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7995         struct lustre_handle conn;
7996         int rc;
7997
7998         ENTRY;
7999
8000         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8001
8002         rc = class_connect(&conn, obd, cluuid);
8003         if (rc)
8004                 RETURN(rc);
8005
8006         *exp = class_conn2export(&conn);
8007
8008         spin_lock(&osd->od_osfs_lock);
8009         osd->od_connects++;
8010         spin_unlock(&osd->od_osfs_lock);
8011
8012         RETURN(0);
8013 }
8014
8015 /*
8016  * once last export (we don't count self-export) disappeared
8017  * osd can be released
8018  */
8019 static int osd_obd_disconnect(struct obd_export *exp)
8020 {
8021         struct obd_device *obd = exp->exp_obd;
8022         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8023         int rc, release = 0;
8024
8025         ENTRY;
8026
8027         /* Only disconnect the underlying layers on the final disconnect. */
8028         spin_lock(&osd->od_osfs_lock);
8029         osd->od_connects--;
8030         if (osd->od_connects == 0)
8031                 release = 1;
8032         spin_unlock(&osd->od_osfs_lock);
8033
8034         rc = class_disconnect(exp); /* bz 9811 */
8035
8036         if (rc == 0 && release)
8037                 class_manual_cleanup(obd);
8038         RETURN(rc);
8039 }
8040
8041 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8042                        struct lu_device *dev)
8043 {
8044         struct osd_device *osd = osd_dev(dev);
8045         struct lr_server_data *lsd =
8046                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8047         int result = 0;
8048
8049         ENTRY;
8050
8051         if (osd->od_quota_slave_md != NULL) {
8052                 /* set up quota slave objects for inode */
8053                 result = qsd_prepare(env, osd->od_quota_slave_md);
8054                 if (result != 0)
8055                         RETURN(result);
8056         }
8057
8058         if (osd->od_quota_slave_dt != NULL) {
8059                 /* set up quota slave objects for block */
8060                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8061                 if (result != 0)
8062                         RETURN(result);
8063         }
8064
8065
8066         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8067 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8068                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8069                         osd->od_index_in_idif = 1;
8070                 } else {
8071                         osd->od_index_in_idif = 0;
8072                         result = osd_register_proc_index_in_idif(osd);
8073                         if (result != 0)
8074                                 RETURN(result);
8075                 }
8076 #else
8077                 osd->od_index_in_idif = 1;
8078 #endif
8079         }
8080
8081         result = osd_fid_init(env, osd);
8082
8083         RETURN(result);
8084 }
8085
8086 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8087                          struct lu_fid *fid, struct md_op_data *op_data)
8088 {
8089         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8090
8091         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8092 }
8093
8094 static const struct lu_object_operations osd_lu_obj_ops = {
8095         .loo_object_init      = osd_object_init,
8096         .loo_object_delete    = osd_object_delete,
8097         .loo_object_release   = osd_object_release,
8098         .loo_object_free      = osd_object_free,
8099         .loo_object_print     = osd_object_print,
8100         .loo_object_invariant = osd_object_invariant
8101 };
8102
8103 const struct lu_device_operations osd_lu_ops = {
8104         .ldo_object_alloc      = osd_object_alloc,
8105         .ldo_process_config    = osd_process_config,
8106         .ldo_recovery_complete = osd_recovery_complete,
8107         .ldo_prepare           = osd_prepare,
8108 };
8109
8110 static const struct lu_device_type_operations osd_device_type_ops = {
8111         .ldto_init = osd_type_init,
8112         .ldto_fini = osd_type_fini,
8113
8114         .ldto_start = osd_type_start,
8115         .ldto_stop  = osd_type_stop,
8116
8117         .ldto_device_alloc = osd_device_alloc,
8118         .ldto_device_free  = osd_device_free,
8119
8120         .ldto_device_init = osd_device_init,
8121         .ldto_device_fini = osd_device_fini
8122 };
8123
8124 static struct lu_device_type osd_device_type = {
8125         .ldt_tags     = LU_DEVICE_DT,
8126         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8127         .ldt_ops      = &osd_device_type_ops,
8128         .ldt_ctx_tags = LCT_LOCAL,
8129 };
8130
8131 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8132 {
8133         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8134         struct super_block *sb = osd_sb(osd);
8135
8136         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8137 }
8138
8139 /*
8140  * lprocfs legacy support.
8141  */
8142 static struct obd_ops osd_obd_device_ops = {
8143         .o_owner = THIS_MODULE,
8144         .o_connect      = osd_obd_connect,
8145         .o_disconnect   = osd_obd_disconnect,
8146         .o_fid_alloc    = osd_fid_alloc,
8147         .o_health_check = osd_health_check,
8148 };
8149
8150 static ssize_t track_declares_assert_show(struct kobject *kobj,
8151                                    struct attribute *attr,
8152                                    char *buf)
8153 {
8154         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8155 }
8156
8157 static ssize_t track_declares_assert_store(struct kobject *kobj,
8158                                            struct attribute *attr,
8159                                            const char *buffer, size_t count)
8160 {
8161         bool track_declares_assert;
8162         int rc;
8163
8164         rc = kstrtobool(buffer, &track_declares_assert);
8165         if (rc)
8166                 return rc;
8167
8168         ldiskfs_track_declares_assert = track_declares_assert;
8169
8170         return count;
8171 }
8172 LUSTRE_RW_ATTR(track_declares_assert);
8173
8174 static int __init osd_init(void)
8175 {
8176         struct kobject *kobj;
8177         int rc;
8178
8179         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8180 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8181         /* please, try to keep osd_thread_info smaller than a page */
8182         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8183 #endif
8184
8185         osd_oi_mod_init();
8186
8187         rc = lu_kmem_init(ldiskfs_caches);
8188         if (rc)
8189                 return rc;
8190
8191 #ifdef CONFIG_KALLSYMS
8192         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8193         priv_dev_check_rdonly =
8194                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8195 #endif
8196
8197         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8198                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8199         if (rc) {
8200                 lu_kmem_fini(ldiskfs_caches);
8201                 return rc;
8202         }
8203
8204         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8205         if (kobj) {
8206                 rc = sysfs_create_file(kobj,
8207                                        &lustre_attr_track_declares_assert.attr);
8208                 kobject_put(kobj);
8209                 if (rc) {
8210                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8211                         rc = 0;
8212                 }
8213         }
8214         return rc;
8215 }
8216
8217 static void __exit osd_exit(void)
8218 {
8219         struct kobject *kobj;
8220
8221         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8222         if (kobj) {
8223                 sysfs_remove_file(kobj,
8224                                   &lustre_attr_track_declares_assert.attr);
8225                 kobject_put(kobj);
8226         }
8227         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8228         lu_kmem_fini(ldiskfs_caches);
8229 }
8230
8231 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8232 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8233 MODULE_VERSION(LUSTRE_VERSION_STRING);
8234 MODULE_LICENSE("GPL");
8235
8236 module_init(osd_init);
8237 module_exit(osd_exit);