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