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