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 #include <linux/uidgid.h>
  46
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/types.h>
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/fs.h>
  51 /* XATTR_{REPLACE,CREATE} */
  52 #include <linux/xattr.h>
  53
  54 #include <ldiskfs/ldiskfs.h>
  55 #include <ldiskfs/xattr.h>
  56 #include <ldiskfs/ldiskfs_extents.h>
  57 #undef ENTRY
  58 /*
  59  * struct OBD_{ALLOC,FREE}*()
  60  * OBD_FAIL_CHECK
  61  */
  62 #include <obd_support.h>
  63 /* struct ptlrpc_thread */
  64 #include <lustre_net.h>
  65 #include <lustre_fid.h>
  66 /* process_config */
  67 #include <uapi/linux/lustre/lustre_param.h>
  68
  69 #include "osd_internal.h"
  70 #include "osd_dynlocks.h"
  71
  72 /* llo_* api support */
  73 #include <md_object.h>
  74 #include <lustre_quota.h>
  75
  76 #include <lustre_linkea.h>
  77
  78 /* Maximum EA size is limited by LNET_MTU for remote objects */
  79 #define OSD_MAX_EA_SIZE 1048364
  80
  81 int ldiskfs_pdo = 1;
  82 module_param(ldiskfs_pdo, int, 0644);
  83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  84
  85 int ldiskfs_track_declares_assert;
  86 module_param(ldiskfs_track_declares_assert, int, 0644);
  87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  88
  89 /* Slab to allocate dynlocks */
  90 struct kmem_cache *dynlock_cachep;
  91
  92 /* Slab to allocate osd_it_ea */
  93 struct kmem_cache *osd_itea_cachep;
  94
  95 static struct lu_kmem_descr ldiskfs_caches[] = {
  96         {
  97                 .ckd_cache = &dynlock_cachep,
  98                 .ckd_name  = "dynlock_cache",
  99                 .ckd_size  = sizeof(struct dynlock_handle)
 100         },
 101         {
 102                 .ckd_cache = &osd_itea_cachep,
 103                 .ckd_name  = "osd_itea_cache",
 104                 .ckd_size  = sizeof(struct osd_it_ea)
 105         },
 106         {
 107                 .ckd_cache = NULL
 108         }
 109 };
 110
 111 static const char dot[] = ".";
 112 static const char dotdot[] = "..";
 113
 114 static const struct lu_object_operations      osd_lu_obj_ops;
 115 static const struct dt_object_operations      osd_obj_ops;
 116 static const struct dt_object_operations      osd_obj_otable_it_ops;
 117 static const struct dt_index_operations       osd_index_iam_ops;
 118 static const struct dt_index_operations       osd_index_ea_ops;
 119
 120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 121                           const struct lu_fid *fid);
 122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 123                                                 struct osd_device *osd);
 124
 125 int osd_trans_declare_op2rb[] = {
 126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 137 };
 138
 139 static int osd_has_index(const struct osd_object *obj)
 140 {
 141         return obj->oo_dt.do_index_ops != NULL;
 142 }
 143
 144 static int osd_object_invariant(const struct lu_object *l)
 145 {
 146         return osd_invariant(osd_obj(l));
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 153 {
 154         struct osd_thread_info *oti = osd_oti_get(env);
 155
 156         return oti->oti_w_locks > 0 && o->oo_owner == env;
 157 }
 158
 159 /*
 160  * Concurrency: doesn't access mutable data
 161  */
 162 static int osd_root_get(const struct lu_env *env,
 163                         struct dt_device *dev, struct lu_fid *f)
 164 {
 165         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 166         return 0;
 167 }
 168
 169 /*
 170  * the following set of functions are used to maintain per-thread
 171  * cache of FID->ino mapping. this mechanism is needed to resolve
 172  * FID to inode at dt_insert() which in turn stores ino in the
 173  * directory entries to keep ldiskfs compatible with ext[34].
 174  * due to locking-originated restrictions we can't lookup ino
 175  * using LU cache (deadlock is possible). lookup using OI is quite
 176  * expensive. so instead we maintain this cache and methods like
 177  * dt_create() fill it. so in the majority of cases dt_insert() is
 178  * able to find needed mapping in lockless manner.
 179  */
 180 static struct osd_idmap_cache *
 181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 182              const struct lu_fid *fid)
 183 {
 184         struct osd_thread_info *oti = osd_oti_get(env);
 185         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 186         int i;
 187
 188         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 189                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 190                         continue;
 191                 if (idc[i].oic_dev != osd)
 192                         continue;
 193
 194                 return idc + i;
 195         }
 196
 197         return NULL;
 198 }
 199
 200 static struct osd_idmap_cache *
 201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 202             const struct lu_fid *fid)
 203 {
 204         struct osd_thread_info *oti   = osd_oti_get(env);
 205         struct osd_idmap_cache *idc;
 206         int i;
 207
 208         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 209                 i = oti->oti_ins_cache_size * 2;
 210                 if (i == 0)
 211                         i = OSD_INS_CACHE_SIZE;
 212                 OBD_ALLOC_PTR_ARRAY(idc, i);
 213                 if (idc == NULL)
 214                         return ERR_PTR(-ENOMEM);
 215                 if (oti->oti_ins_cache != NULL) {
 216                         memcpy(idc, oti->oti_ins_cache,
 217                                oti->oti_ins_cache_used * sizeof(*idc));
 218                         OBD_FREE_PTR_ARRAY(oti->oti_ins_cache,
 219                                            oti->oti_ins_cache_used);
 220                 }
 221                 oti->oti_ins_cache = idc;
 222                 oti->oti_ins_cache_size = i;
 223         }
 224
 225         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 226         idc->oic_fid = *fid;
 227         idc->oic_dev = osd;
 228         idc->oic_lid.oii_ino = 0;
 229         idc->oic_lid.oii_gen = 0;
 230         idc->oic_remote = 0;
 231
 232         return idc;
 233 }
 234
 235 /*
 236  * lookup mapping for the given fid in the cache, initialize a
 237  * new one if not found. the initialization checks whether the
 238  * object is local or remote. for local objects, OI is used to
 239  * learn ino/generation. the function is used when the caller
 240  * has no information about the object, e.g. at dt_insert().
 241  */
 242 static struct osd_idmap_cache *
 243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 244                      const struct lu_fid *fid)
 245 {
 246         struct osd_idmap_cache *idc;
 247         int rc;
 248
 249         idc = osd_idc_find(env, osd, fid);
 250         LASSERT(!IS_ERR(idc));
 251         if (idc != NULL)
 252                 return idc;
 253
 254         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 255                osd->od_svname, PFID(fid));
 256
 257         /* new mapping is needed */
 258         idc = osd_idc_add(env, osd, fid);
 259         if (IS_ERR(idc)) {
 260                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 261                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 262                 return idc;
 263         }
 264
 265         /* initialize it */
 266         rc = osd_remote_fid(env, osd, fid);
 267         if (unlikely(rc < 0))
 268                 return ERR_PTR(rc);
 269
 270         if (rc == 0) {
 271                 /* the object is local, lookup in OI */
 272                 /* XXX: probably cheaper to lookup in LU first? */
 273                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 274                                    &idc->oic_lid, 0);
 275                 if (unlikely(rc < 0)) {
 276                         CERROR("can't lookup: rc = %d\n", rc);
 277                         return ERR_PTR(rc);
 278                 }
 279         } else {
 280                 /* the object is remote */
 281                 idc->oic_remote = 1;
 282         }
 283
 284         return idc;
 285 }
 286
 287 /*
 288  * lookup mapping for given FID and fill it from the given object.
 289  * the object is lolcal by definition.
 290  */
 291 static int osd_idc_find_and_init(const struct lu_env *env,
 292                                  struct osd_device *osd,
 293                                  struct osd_object *obj)
 294 {
 295         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 296         struct osd_idmap_cache *idc;
 297
 298         idc = osd_idc_find(env, osd, fid);
 299         LASSERT(!IS_ERR(idc));
 300         if (idc != NULL) {
 301                 if (obj->oo_inode == NULL)
 302                         return 0;
 303                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 304                         LASSERT(idc->oic_lid.oii_ino == 0);
 305                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 306                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 307                 }
 308                 return 0;
 309         }
 310
 311         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 312                osd->od_svname, PFID(fid));
 313
 314         /* new mapping is needed */
 315         idc = osd_idc_add(env, osd, fid);
 316         if (IS_ERR(idc)) {
 317                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 318                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 319                 return PTR_ERR(idc);
 320         }
 321
 322         if (obj->oo_inode != NULL) {
 323                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 324                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 325         }
 326         return 0;
 327 }
 328
 329 /*
 330  * OSD object methods.
 331  */
 332
 333 /*
 334  * Concurrency: no concurrent access is possible that early in object
 335  * life-cycle.
 336  */
 337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 338                                           const struct lu_object_header *hdr,
 339                                           struct lu_device *d)
 340 {
 341         struct osd_object *mo;
 342
 343         OBD_ALLOC_PTR(mo);
 344         if (mo != NULL) {
 345                 struct lu_object *l;
 346                 struct lu_object_header *h;
 347                 struct osd_device *o = osd_dev(d);
 348
 349                 l = &mo->oo_dt.do_lu;
 350                 if (unlikely(o->od_in_init)) {
 351                         OBD_ALLOC_PTR(h);
 352                         if (!h) {
 353                                 OBD_FREE_PTR(mo);
 354                                 return NULL;
 355                         }
 356
 357                         lu_object_header_init(h);
 358                         lu_object_init(l, h, d);
 359                         lu_object_add_top(h, l);
 360                         mo->oo_header = h;
 361                 } else {
 362                         dt_object_init(&mo->oo_dt, NULL, d);
 363                         mo->oo_header = NULL;
 364                 }
 365
 366                 mo->oo_dt.do_ops = &osd_obj_ops;
 367                 l->lo_ops = &osd_lu_obj_ops;
 368                 init_rwsem(&mo->oo_sem);
 369                 init_rwsem(&mo->oo_ext_idx_sem);
 370                 spin_lock_init(&mo->oo_guard);
 371                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 372                 return l;
 373         }
 374         return NULL;
 375 }
 376
 377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 378                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 379 {
 380         int rc;
 381
 382         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 383                              (void *)loa, sizeof(*loa));
 384         if (rc > 0) {
 385                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 386
 387                 if (rc < sizeof(*lma))
 388                         return -EINVAL;
 389
 390                 rc = 0;
 391                 lustre_loa_swab(loa, true);
 392                 /* Check LMA compatibility */
 393                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 394                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 395                               "for fid = "DFID", ino = %lu\n",
 396                               osd_ino2name(inode),
 397                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 398                               PFID(&lma->lma_self_fid), inode->i_ino);
 399                         rc = -EOPNOTSUPP;
 400                 }
 401         } else if (rc == 0) {
 402                 rc = -ENODATA;
 403         }
 404
 405         return rc;
 406 }
 407
 408 /*
 409  * retrieve object from backend ext fs.
 410  **/
 411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 412                        struct osd_inode_id *id)
 413 {
 414         int rc;
 415         struct inode *inode = NULL;
 416
 417         /*
 418          * if we look for an inode withing a running
 419          * transaction, then we risk to deadlock
 420          * osd_dirent_check_repair() breaks this
 421          */
 422          /* LASSERT(current->journal_info == NULL); */
 423
 424         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 425         if (IS_ERR(inode)) {
 426                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 427                        id->oii_ino, PTR_ERR(inode));
 428         } else if (id->oii_gen != OSD_OII_NOGEN &&
 429                    inode->i_generation != id->oii_gen) {
 430                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 431                        "i_generation = %u\n",
 432                        id->oii_ino, id->oii_gen, inode->i_generation);
 433                 iput(inode);
 434                 inode = ERR_PTR(-ESTALE);
 435         } else if (inode->i_nlink == 0) {
 436                 /*
 437                  * due to parallel readdir and unlink,
 438                  * we can have dead inode here.
 439                  */
 440                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 441                 iput(inode);
 442                 inode = ERR_PTR(-ESTALE);
 443         } else if (is_bad_inode(inode)) {
 444                 CWARN("%s: bad inode: ino = %u\n",
 445                       osd_dev2name(dev), id->oii_ino);
 446                 iput(inode);
 447                 inode = ERR_PTR(-ENOENT);
 448         } else if ((rc = osd_attach_jinode(inode))) {
 449                 iput(inode);
 450                 inode = ERR_PTR(rc);
 451         } else {
 452                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 453                 if (id->oii_gen == OSD_OII_NOGEN)
 454                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 455
 456                 /*
 457                  * Do not update file c/mtime in ldiskfs.
 458                  * NB: we don't have any lock to protect this because we don't
 459                  * have reference on osd_object now, but contention with
 460                  * another lookup + attr_set can't happen in the tiny window
 461                  * between if (...) and set S_NOCMTIME.
 462                  */
 463                 if (!(inode->i_flags & S_NOCMTIME))
 464                         inode->i_flags |= S_NOCMTIME;
 465         }
 466         return inode;
 467 }
 468
 469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 470                           handle_t *handle, struct dentry *child,
 471                           struct inode *inode, struct htree_lock *hlock)
 472 {
 473         int rc, rc2;
 474
 475         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 476         if (rc == -ENOBUFS || rc == -ENOSPC) {
 477                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 478                 struct inode *parent = child->d_parent->d_inode;
 479                 struct lu_fid *fid = NULL;
 480                 char fidstr[FID_LEN + 1] = "unknown";
 481
 482                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 483                 if (!rc2) {
 484                         fid = &loa->loa_lma.lma_self_fid;
 485                 } else if (rc2 == -ENODATA) {
 486                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 487                                 fid = &info->oti_fid3;
 488                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 489                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 490                                 fid = &info->oti_fid3;
 491                                 lu_igif_build(fid, parent->i_ino,
 492                                               parent->i_generation);
 493                         }
 494                 }
 495
 496                 if (fid != NULL)
 497                         snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
 498
 499                 /* below message is checked in sanity.sh test_129 */
 500                 if (rc == -ENOSPC) {
 501                         CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
 502                               osd_name(osd), parent->i_ino, fidstr);
 503                 } else {
 504                         rc = 0; /* ignore such error now */
 505                         CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
 506                               osd_name(osd), parent->i_ino, fidstr);
 507                 }
 508
 509         }
 510
 511         return rc;
 512 }
 513
 514
 515 struct inode *
 516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 517              struct osd_inode_id *id, struct lu_fid *fid)
 518 {
 519         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 520         struct inode *inode;
 521         int rc;
 522
 523         inode = osd_iget(info, dev, id);
 524         if (IS_ERR(inode))
 525                 return inode;
 526
 527         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 528         if (!rc) {
 529                 *fid = loa->loa_lma.lma_self_fid;
 530         } else if (rc == -ENODATA) {
 531                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 532                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 533                 else
 534                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 535         } else {
 536                 iput(inode);
 537                 inode = ERR_PTR(rc);
 538         }
 539         return inode;
 540 }
 541
 542 static struct inode *osd_iget_check(struct osd_thread_info *info,
 543                                     struct osd_device *dev,
 544                                     const struct lu_fid *fid,
 545                                     struct osd_inode_id *id,
 546                                     bool trusted)
 547 {
 548         struct inode *inode;
 549         int rc = 0;
 550
 551         ENTRY;
 552
 553         /*
 554          * The cached OI mapping is trustable. If we cannot locate the inode
 555          * via the cached OI mapping, then return the failure to the caller
 556          * directly without further OI checking.
 557          */
 558
 559 again:
 560         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 561         if (IS_ERR(inode)) {
 562                 rc = PTR_ERR(inode);
 563                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 564                         goto check_oi;
 565
 566                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 567                        PFID(fid), id->oii_ino, rc);
 568                 GOTO(put, rc);
 569         }
 570
 571         if (is_bad_inode(inode)) {
 572                 rc = -ENOENT;
 573                 if (!trusted)
 574                         goto check_oi;
 575
 576                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 577                        PFID(fid), id->oii_ino);
 578                 GOTO(put, rc);
 579         }
 580
 581         if (id->oii_gen != OSD_OII_NOGEN &&
 582             inode->i_generation != id->oii_gen) {
 583                 rc = -ESTALE;
 584                 if (!trusted)
 585                         goto check_oi;
 586
 587                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 588                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 589                        id->oii_ino, id->oii_gen, inode->i_generation);
 590                 GOTO(put, rc);
 591         }
 592
 593         if (inode->i_nlink == 0) {
 594                 rc = -ENOENT;
 595                 if (!trusted)
 596                         goto check_oi;
 597
 598                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 599                        PFID(fid), id->oii_ino);
 600                 GOTO(put, rc);
 601         }
 602
 603         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 604
 605 check_oi:
 606         if (rc != 0) {
 607                 __u32 saved_ino = id->oii_ino;
 608                 __u32 saved_gen = id->oii_gen;
 609
 610                 LASSERT(!trusted);
 611                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 612
 613                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 614                 /*
 615                  * XXX: There are four possible cases:
 616                  *      1. rc = 0.
 617                  *         Backup/restore caused the OI invalid.
 618                  *      2. rc = 0.
 619                  *         Someone unlinked the object but NOT removed
 620                  *         the OI mapping, such as mount target device
 621                  *         as ldiskfs, and modify something directly.
 622                  *      3. rc = -ENOENT.
 623                  *         Someone just removed the object between the
 624                  *         former oi_lookup and the iget. It is normal.
 625                  *      4. Other failure cases.
 626                  *
 627                  *      Generally, when the device is mounted, it will
 628                  *      auto check whether the system is restored from
 629                  *      file-level backup or not. We trust such detect
 630                  *      to distinguish the 1st case from the 2nd case:
 631                  *      if the OI files are consistent but may contain
 632                  *      stale OI mappings because of case 2, if iget()
 633                  *      returns -ENOENT or -ESTALE, then it should be
 634                  *      the case 2.
 635                  */
 636                 if (rc != 0)
 637                         /*
 638                          * If the OI mapping was in OI file before the
 639                          * osd_iget_check(), but now, it is disappear,
 640                          * then it must be removed by race. That is a
 641                          * normal race case.
 642                          */
 643                         GOTO(put, rc);
 644
 645                 /*
 646                  * It is the OI scrub updated the OI mapping by race.
 647                  * The new OI mapping must be valid.
 648                  */
 649                 if (saved_ino != id->oii_ino ||
 650                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 651                         if (!IS_ERR(inode))
 652                                 iput(inode);
 653
 654                         trusted = true;
 655                         goto again;
 656                 }
 657
 658                 if (IS_ERR(inode)) {
 659                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 660                             SF_INCONSISTENT)
 661                                 /*
 662                                  * It still can be the case 2, but we cannot
 663                                  * distinguish it from the case 1. So return
 664                                  * -EREMCHG to block current operation until
 665                                  *  OI scrub rebuilt the OI mappings.
 666                                  */
 667                                 rc = -EREMCHG;
 668                         else
 669                                 rc = -ENOENT;
 670
 671                         GOTO(put, rc);
 672                 }
 673
 674                 if (inode->i_generation == id->oii_gen)
 675                         rc = -ENOENT;
 676                 else
 677                         rc = -EREMCHG;
 678         } else {
 679                 if (id->oii_gen == OSD_OII_NOGEN)
 680                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 681
 682                 /*
 683                  * Do not update file c/mtime in ldiskfs.
 684                  * NB: we don't have any lock to protect this because we don't
 685                  * have reference on osd_object now, but contention with
 686                  * another lookup + attr_set can't happen in the tiny window
 687                  * between if (...) and set S_NOCMTIME.
 688                  */
 689                 if (!(inode->i_flags & S_NOCMTIME))
 690                         inode->i_flags |= S_NOCMTIME;
 691         }
 692
 693         GOTO(put, rc);
 694
 695 put:
 696         if (rc != 0) {
 697                 if (!IS_ERR(inode))
 698                         iput(inode);
 699
 700                 inode = ERR_PTR(rc);
 701         }
 702
 703         return inode;
 704 }
 705
 706 /**
 707  * \retval +v: new filter_fid does not contain self-fid
 708  * \retval 0:  filter_fid_18_23, contains self-fid
 709  * \retval -v: other failure cases
 710  */
 711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 712                  struct dentry *dentry, struct lu_fid *fid)
 713 {
 714         struct filter_fid *ff = &info->oti_ff;
 715         struct ost_id *ostid = &info->oti_ostid;
 716         int rc;
 717
 718         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 719         if (rc == sizeof(struct filter_fid_18_23)) {
 720                 struct filter_fid_18_23 *ff_old = (void *)ff;
 721
 722                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 723                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 724                 /*
 725                  * XXX: use 0 as the index for compatibility, the caller will
 726                  * handle index related issues when necessary.
 727                  */
 728                 if (!rc)
 729                         ostid_to_fid(fid, ostid, 0);
 730         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 731                 rc = 1;
 732         } else if (rc >= 0) {
 733                 rc = -EINVAL;
 734         }
 735
 736         return rc;
 737 }
 738
 739 static int osd_lma_self_repair(struct osd_thread_info *info,
 740                                struct osd_device *osd, struct inode *inode,
 741                                const struct lu_fid *fid, __u32 compat)
 742 {
 743         handle_t *jh;
 744         int rc;
 745
 746         LASSERT(current->journal_info == NULL);
 747
 748         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 749                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 750         if (IS_ERR(jh)) {
 751                 rc = PTR_ERR(jh);
 752                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 753                       osd_name(osd), rc);
 754                 return rc;
 755         }
 756
 757         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 758         if (rc != 0)
 759                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 760                       osd_name(osd), rc);
 761         ldiskfs_journal_stop(jh);
 762         return rc;
 763 }
 764
 765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 766 {
 767         struct osd_thread_info *info = osd_oti_get(env);
 768         struct osd_device *osd = osd_obj2dev(obj);
 769         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 770         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 771         struct inode *inode = obj->oo_inode;
 772         struct dentry *dentry = &info->oti_obj_dentry;
 773         struct lu_fid *fid = NULL;
 774         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 775         int rc;
 776
 777         ENTRY;
 778
 779         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 780                              (void *)loa, sizeof(*loa));
 781         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 782                 fid = &lma->lma_self_fid;
 783                 rc = osd_get_idif(info, inode, dentry, fid);
 784                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 785                         /*
 786                          * For the given OST-object, if it has neither LMA nor
 787                          * FID in XATTR_NAME_FID, then the given FID (which is
 788                          * contained in the @obj, from client RPC for locating
 789                          * the OST-object) is trusted. We use it to generate
 790                          * the LMA.
 791                          */
 792                         osd_lma_self_repair(info, osd, inode, rfid,
 793                                             LMAC_FID_ON_OST);
 794                         RETURN(0);
 795                 }
 796         }
 797
 798         if (rc < 0)
 799                 RETURN(rc);
 800
 801         if (rc > 0) {
 802                 rc = 0;
 803                 lustre_lma_swab(lma);
 804                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 805                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 806                               S_ISREG(inode->i_mode)))) {
 807                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 808                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 809                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 810                               PFID(rfid), inode->i_ino);
 811                         rc = -EOPNOTSUPP;
 812                 } else {
 813                         fid = &lma->lma_self_fid;
 814                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 815                             osd->od_is_ost)
 816                                 obj->oo_pfid_in_lma = 1;
 817                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 818                             !osd->od_is_ost)
 819                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 820                 }
 821         }
 822
 823         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 824                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 825                         struct ost_id   *oi   = &info->oti_ostid;
 826                         struct lu_fid   *fid1 = &info->oti_fid3;
 827                         __u32            idx  = fid_idif_ost_idx(rfid);
 828
 829                         /*
 830                          * For old IDIF, the OST index is not part of the IDIF,
 831                          * Means that different OSTs may have the same IDIFs.
 832                          * Under such case, we need to make some compatible
 833                          * check to make sure to trigger OI scrub properly.
 834                          */
 835                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 836                                 /* Given @rfid is new, LMA is old. */
 837                                 fid_to_ostid(fid, oi);
 838                                 ostid_to_fid(fid1, oi, idx);
 839                                 if (lu_fid_eq(fid1, rfid)) {
 840                                         if (osd->od_index_in_idif)
 841                                                 osd_lma_self_repair(info, osd,
 842                                                         inode, rfid,
 843                                                         LMAC_FID_ON_OST);
 844                                         RETURN(0);
 845                                 }
 846                         }
 847                 }
 848
 849                 rc = -EREMCHG;
 850         }
 851
 852         RETURN(rc);
 853 }
 854
 855 struct osd_check_lmv_buf {
 856         /* please keep it as first member */
 857         struct dir_context ctx;
 858         struct osd_thread_info *oclb_info;
 859         struct osd_device *oclb_dev;
 860         struct osd_idmap_cache *oclb_oic;
 861         int oclb_items;
 862         bool oclb_found;
 863 };
 864
 865 /**
 866  * It is called internally by ->iterate*() to filter out the
 867  * local slave object's FID of the striped directory.
 868  *
 869  * \retval      1 found the local slave's FID
 870  * \retval      0 continue to check next item
 871  * \retval      -ve for failure
 872  */
 873 #ifdef HAVE_FILLDIR_USE_CTX
 874 static int osd_stripe_dir_filldir(struct dir_context *buf,
 875 #else
 876 static int osd_stripe_dir_filldir(void *buf,
 877 #endif
 878                                   const char *name, int namelen,
 879                                   loff_t offset, __u64 ino, unsigned int d_type)
 880 {
 881         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 882         struct osd_thread_info *oti = oclb->oclb_info;
 883         struct lu_fid *fid = &oti->oti_fid3;
 884         struct osd_inode_id *id = &oti->oti_id3;
 885         struct osd_device *dev = oclb->oclb_dev;
 886         struct osd_idmap_cache *oic = oclb->oclb_oic;
 887         struct inode *inode;
 888
 889         oclb->oclb_items++;
 890
 891         if (name[0] == '.')
 892                 return 0;
 893
 894         fid_zero(fid);
 895         sscanf(name + 1, SFID, RFID(fid));
 896         if (!fid_is_sane(fid))
 897                 return 0;
 898
 899         if (osd_remote_fid(oti->oti_env, dev, fid))
 900                 return 0;
 901
 902         osd_id_gen(id, ino, OSD_OII_NOGEN);
 903         inode = osd_iget(oti, dev, id);
 904         if (IS_ERR(inode))
 905                 return PTR_ERR(inode);
 906
 907         iput(inode);
 908         osd_add_oi_cache(oti, dev, id, fid);
 909         oic->oic_fid = *fid;
 910         oic->oic_lid = *id;
 911         oic->oic_dev = dev;
 912         osd_oii_insert(dev, oic, true);
 913         oclb->oclb_found = true;
 914
 915         return 1;
 916 }
 917
 918 /*
 919  * When lookup item under striped directory, we need to locate the master
 920  * MDT-object of the striped directory firstly, then the client will send
 921  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 922  * and the item's name. If the system is restored from MDT file level backup,
 923  * then before the OI scrub completely built the OI files, the OI mappings of
 924  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 925  * not a problem for the master MDT-object. Because when locate the master
 926  * MDT-object, we will do name based lookup (for the striped directory itself)
 927  * firstly, during such process we can setup the correct OI mapping for the
 928  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 929  * the client will not trigger name based lookup on the MDT to locate the slave
 930  * MDT-object before locating item under the striped directory, then when
 931  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 932  * is invalid and does not know what the right OI mapping is, then the MDT has
 933  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 934  * the OI file, related OI mapping is unknown yet, please try again later. And
 935  * then client will re-try the RPC again and again until related OI mapping has
 936  * been updated. That is quite inefficient.
 937  *
 938  * To resolve above trouble, we will handle it as the following two cases:
 939  *
 940  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 941  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 942  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 943  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 944  *    directly. Please check osd_fid_lookup().
 945  *
 946  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 947  *    Under such case, during lookup the master MDT-object, we will lookup the
 948  *    slave MDT-object via readdir against the master MDT-object, because the
 949  *    slave MDT-objects information are stored as sub-directories with the name
 950  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 951  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 952  *    the correct OI mapping for the slave MDT-object.
 953  */
 954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 955                          struct inode *inode, struct osd_idmap_cache *oic)
 956 {
 957         struct lu_buf *buf = &oti->oti_big_buf;
 958         struct dentry *dentry = &oti->oti_obj_dentry;
 959         struct file *filp = &oti->oti_file;
 960         const struct file_operations *fops;
 961         struct lmv_mds_md_v1 *lmv1;
 962         struct osd_check_lmv_buf oclb = {
 963                 .ctx.actor = osd_stripe_dir_filldir,
 964                 .oclb_info = oti,
 965                 .oclb_dev = dev,
 966                 .oclb_oic = oic,
 967                 .oclb_found = false,
 968         };
 969         int rc = 0;
 970
 971         ENTRY;
 972
 973 again:
 974         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 975                              buf->lb_len);
 976         if (rc == -ERANGE) {
 977                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 978                 if (rc > 0) {
 979                         lu_buf_realloc(buf, rc);
 980                         if (buf->lb_buf == NULL)
 981                                 GOTO(out, rc = -ENOMEM);
 982
 983                         goto again;
 984                 }
 985         }
 986
 987         if (unlikely(rc == 0 || rc == -ENODATA))
 988                 GOTO(out, rc = 0);
 989
 990         if (rc < 0)
 991                 GOTO(out, rc);
 992
 993         if (unlikely(buf->lb_buf == NULL)) {
 994                 lu_buf_realloc(buf, rc);
 995                 if (buf->lb_buf == NULL)
 996                         GOTO(out, rc = -ENOMEM);
 997
 998                 goto again;
 999         }
1000
1001         lmv1 = buf->lb_buf;
1002         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1003                 GOTO(out, rc = 0);
1004
1005         fops = inode->i_fop;
1006         dentry->d_inode = inode;
1007         dentry->d_sb = inode->i_sb;
1008         filp->f_pos = 0;
1009         filp->f_path.dentry = dentry;
1010         filp->f_flags |= O_NOATIME;
1011         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012         filp->f_mapping = inode->i_mapping;
1013         filp->f_op = fops;
1014         filp->private_data = NULL;
1015         filp->f_cred = current_cred();
1016         filp->f_inode = inode;
1017         rc = osd_security_file_alloc(filp);
1018         if (rc)
1019                 goto out;
1020
1021         do {
1022                 oclb.oclb_items = 0;
1023                 rc = iterate_dir(filp, &oclb.ctx);
1024         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026         fops->release(inode, filp);
1027
1028 out:
1029         if (rc < 0)
1030                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031                        DFID": rc = %d\n", osd_ino2name(inode),
1032                        inode->i_ino, inode->i_generation,
1033                        PFID(&oic->oic_fid), rc);
1034         else
1035                 rc = 0;
1036
1037         RETURN(rc);
1038 }
1039
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041                           const struct lu_fid *fid,
1042                           const struct lu_object_conf *conf)
1043 {
1044         struct osd_thread_info *info;
1045         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046         struct osd_device *dev;
1047         struct osd_idmap_cache *oic;
1048         struct osd_inode_id *id;
1049         struct inode *inode = NULL;
1050         struct lustre_scrub *scrub;
1051         struct scrub_file *sf;
1052         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1053         __u32 saved_ino;
1054         __u32 saved_gen;
1055         int result = 0;
1056         int rc1 = 0;
1057         bool remote = false;
1058         bool trusted = true;
1059         bool updated = false;
1060         bool checked = false;
1061
1062         ENTRY;
1063
1064         LINVRNT(osd_invariant(obj));
1065         LASSERT(obj->oo_inode == NULL);
1066         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1067
1068         dev = osd_dev(ldev);
1069         scrub = &dev->od_scrub.os_scrub;
1070         sf = &scrub->os_file;
1071         info = osd_oti_get(env);
1072         LASSERT(info);
1073         oic = &info->oti_cache;
1074
1075         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1076                 RETURN(-ENOENT);
1077
1078         /*
1079          * For the object is created as locking anchor, or for the object to
1080          * be created on disk. No need to osd_oi_lookup() at here because FID
1081          * shouldn't never be re-used, if it's really a duplicate FID from
1082          * unexpected reason, we should be able to detect it later by calling
1083          * do_create->osd_oi_insert().
1084          */
1085         if (conf && conf->loc_flags & LOC_F_NEW)
1086                 GOTO(out, result = 0);
1087
1088         /* Search order: 1. per-thread cache. */
1089         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1090                 id = &oic->oic_lid;
1091                 goto iget;
1092         }
1093
1094         id = &info->oti_id;
1095         if (!list_empty(&scrub->os_inconsistent_items)) {
1096                 /* Search order: 2. OI scrub pending list. */
1097                 result = osd_oii_lookup(dev, fid, id);
1098                 if (!result)
1099                         goto iget;
1100         }
1101
1102         /*
1103          * The OI mapping in the OI file can be updated by the OI scrub
1104          * when we locate the inode via FID. So it may be not trustable.
1105          */
1106         trusted = false;
1107
1108         /* Search order: 3. OI files. */
1109         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110         if (result == -ENOENT) {
1111                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1114                                       sf->sf_oi_bitmap))
1115                         GOTO(out, result = 0);
1116
1117                 goto trigger;
1118         }
1119
1120         /* -ESTALE is returned if inode of OST object doesn't exist */
1121         if (result == -ESTALE &&
1122             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123                 GOTO(out, result = 0);
1124         }
1125
1126         if (result)
1127                 GOTO(out, result);
1128
1129 iget:
1130         obj->oo_inode = NULL;
1131         /* for later passes through checks, not true on first pass */
1132         if (!IS_ERR_OR_NULL(inode))
1133                 iput(inode);
1134
1135         inode = osd_iget_check(info, dev, fid, id, trusted);
1136         if (!IS_ERR(inode)) {
1137                 obj->oo_inode = inode;
1138                 result = 0;
1139                 if (remote)
1140                         goto trigger;
1141
1142                 goto check_lma;
1143         }
1144
1145         result = PTR_ERR(inode);
1146         if (result == -ENOENT || result == -ESTALE)
1147                 GOTO(out, result = 0);
1148
1149         if (result != -EREMCHG)
1150                 GOTO(out, result);
1151
1152 trigger:
1153         /*
1154          * We still have chance to get the valid inode: for the
1155          * object which is referenced by remote name entry, the
1156          * object on the local MDT will be linked under the dir
1157          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1158          *
1159          * We do not know whether the object for the given FID
1160          * is referenced by some remote name entry or not, and
1161          * especially for DNE II, a multiple-linked object may
1162          * have many name entries reside on many MDTs.
1163          *
1164          * To simplify the operation, OSD will not distinguish
1165          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166          * only happened for the RPC from other MDT during the
1167          * OI scrub, or for the client side RPC with FID only,
1168          * such as FID to path, or from old connected client.
1169          */
1170         if (!remote) {
1171                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1172                 if (!rc1) {
1173                         remote = true;
1174                         trusted = true;
1175                         flags |= SS_AUTO_PARTIAL;
1176                         flags &= ~SS_AUTO_FULL;
1177                         goto iget;
1178                 }
1179         }
1180
1181         if (thread_is_running(&scrub->os_thread)) {
1182                 if (scrub->os_partial_scan && !scrub->os_in_join)
1183                         goto join;
1184
1185                 if (IS_ERR_OR_NULL(inode) || result)
1186                         GOTO(out, result = -EINPROGRESS);
1187
1188                 LASSERT(remote);
1189                 LASSERT(obj->oo_inode == inode);
1190
1191                 osd_add_oi_cache(info, dev, id, fid);
1192                 osd_oii_insert(dev, oic, true);
1193                 goto found;
1194         }
1195
1196         if (dev->od_auto_scrub_interval == AS_NEVER) {
1197                 if (!remote)
1198                         GOTO(out, result = -EREMCHG);
1199
1200                 LASSERT(!result);
1201                 LASSERT(obj->oo_inode == inode);
1202
1203                 osd_add_oi_cache(info, dev, id, fid);
1204                 goto found;
1205         }
1206
1207 join:
1208         rc1 = osd_scrub_start(env, dev, flags);
1209         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211         if (rc1 && rc1 != -EALREADY)
1212                 GOTO(out, result = -EREMCHG);
1213
1214         if (IS_ERR_OR_NULL(inode) || result)
1215                 GOTO(out, result = -EINPROGRESS);
1216
1217         LASSERT(remote);
1218         LASSERT(obj->oo_inode == inode);
1219
1220         osd_add_oi_cache(info, dev, id, fid);
1221         osd_oii_insert(dev, oic, true);
1222         goto found;
1223
1224 check_lma:
1225         checked = true;
1226         if (unlikely(obj->oo_header))
1227                 goto found;
1228
1229         result = osd_check_lma(env, obj);
1230         if (!result)
1231                 goto found;
1232
1233         LASSERTF(id->oii_ino == inode->i_ino &&
1234                  id->oii_gen == inode->i_generation,
1235                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236                  PFID(fid), id->oii_ino, id->oii_gen,
1237                  inode->i_ino, inode->i_generation);
1238
1239         saved_ino = inode->i_ino;
1240         saved_gen = inode->i_generation;
1241
1242         if (unlikely(result == -ENODATA)) {
1243                 /*
1244                  * If the OI scrub updated the OI mapping by race, it
1245                  * must be valid. Trust the inode that has no LMA EA.
1246                  */
1247                 if (updated)
1248                         goto found;
1249
1250                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1251                 if (!result) {
1252                         /*
1253                          * The OI mapping is still there, the inode is still
1254                          * valid. It is just becaues the inode has no LMA EA.
1255                          */
1256                         if (saved_ino == id->oii_ino &&
1257                             saved_gen == id->oii_gen)
1258                                 goto found;
1259
1260                         /*
1261                          * It is the OI scrub updated the OI mapping by race.
1262                          * The new OI mapping must be valid.
1263                          */
1264                         trusted = true;
1265                         updated = true;
1266                         goto iget;
1267                 }
1268
1269                 /*
1270                  * "result == -ENOENT" means that the OI mappinghas been
1271                  * removed by race, so the inode belongs to other object.
1272                  *
1273                  * Others error can be returned  directly.
1274                  */
1275                 if (result == -ENOENT) {
1276                         LASSERT(trusted);
1277
1278                         obj->oo_inode = NULL;
1279                         result = 0;
1280                 }
1281         }
1282
1283         if (result != -EREMCHG)
1284                 GOTO(out, result);
1285
1286         LASSERT(!updated);
1287
1288         /*
1289          * if two OST objects map to the same inode, and inode mode is
1290          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291          * reserved by precreate, and not written yet, in this case, don't
1292          * set inode for the object whose FID mismatch, so that it can create
1293          * inode and not block precreate.
1294          */
1295         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297                 obj->oo_inode = NULL;
1298                 GOTO(out, result = 0);
1299         }
1300
1301         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1302         /*
1303          * "result == -ENOENT" means the cached OI mapping has been removed
1304          * from the OI file by race, above inode belongs to other object.
1305          */
1306         if (result == -ENOENT) {
1307                 LASSERT(trusted);
1308
1309                 obj->oo_inode = NULL;
1310                 GOTO(out, result = 0);
1311         }
1312
1313         if (result)
1314                 GOTO(out, result);
1315
1316         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1317                 result = -EREMCHG;
1318                 goto trigger;
1319         }
1320
1321         /*
1322          * It is the OI scrub updated the OI mapping by race.
1323          * The new OI mapping must be valid.
1324          */
1325         trusted = true;
1326         updated = true;
1327         goto iget;
1328
1329 found:
1330         if (!checked) {
1331                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1333
1334                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1335                 if (!result) {
1336                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1337                             dev->od_is_ost)
1338                                 obj->oo_pfid_in_lma = 1;
1339                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1340                             !dev->od_is_ost)
1341                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342                 } else if (result != -ENODATA) {
1343                         GOTO(out, result);
1344                 }
1345         }
1346
1347         obj->oo_compat_dot_created = 1;
1348         obj->oo_compat_dotdot_created = 1;
1349
1350         if (S_ISDIR(inode->i_mode) &&
1351             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352                 osd_check_lmv(info, dev, inode, oic);
1353
1354         result = osd_attach_jinode(inode);
1355         if (result)
1356                 GOTO(out, result);
1357
1358         if (!ldiskfs_pdo)
1359                 GOTO(out, result = 0);
1360
1361         LASSERT(!obj->oo_hl_head);
1362         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1363
1364         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1365
1366 out:
1367         if (result || !obj->oo_inode) {
1368                 if (!IS_ERR_OR_NULL(inode))
1369                         iput(inode);
1370
1371                 obj->oo_inode = NULL;
1372                 if (trusted)
1373                         fid_zero(&oic->oic_fid);
1374         }
1375
1376         LINVRNT(osd_invariant(obj));
1377         return result;
1378 }
1379
1380 /*
1381  * Concurrency: shouldn't matter.
1382  */
1383 static void osd_object_init0(struct osd_object *obj)
1384 {
1385         LASSERT(obj->oo_inode != NULL);
1386         obj->oo_dt.do_body_ops = &osd_body_ops;
1387         obj->oo_dt.do_lu.lo_header->loh_attr |=
1388                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1389 }
1390
1391 /*
1392  * Concurrency: no concurrent access is possible that early in object
1393  * life-cycle.
1394  */
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396                            const struct lu_object_conf *conf)
1397 {
1398         struct osd_object *obj = osd_obj(l);
1399         int result;
1400
1401         LINVRNT(osd_invariant(obj));
1402
1403         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1404             cfs_fail_val == 2) {
1405                 struct osd_thread_info *info = osd_oti_get(env);
1406                 struct osd_idmap_cache *oic = &info->oti_cache;
1407                 /* invalidate thread cache */
1408                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1409         }
1410         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1411                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1412                 l->lo_header->loh_attr |= LOHA_EXISTS;
1413                 return 0;
1414         }
1415
1416         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1417         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1418         if (result == 0 && obj->oo_inode != NULL) {
1419                 struct osd_thread_info *oti = osd_oti_get(env);
1420                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1421
1422                 osd_object_init0(obj);
1423                 if (unlikely(obj->oo_header))
1424                         return 0;
1425
1426                 result = osd_get_lma(oti, obj->oo_inode,
1427                                      &oti->oti_obj_dentry, loa);
1428                 if (!result) {
1429                         /*
1430                          * Convert LMAI flags to lustre LMA flags
1431                          * and cache it to oo_lma_flags
1432                          */
1433                         obj->oo_lma_flags =
1434                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1435                 } else if (result == -ENODATA) {
1436                         result = 0;
1437                 }
1438         }
1439         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1440
1441         LINVRNT(osd_invariant(obj));
1442         return result;
1443 }
1444
1445 /*
1446  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1447  * The left part is for value, binary mode.
1448  */
1449 struct osd_xattr_entry {
1450         struct list_head        oxe_list;
1451         size_t                  oxe_len;
1452         size_t                  oxe_namelen;
1453         bool                    oxe_exist;
1454         struct rcu_head         oxe_rcu;
1455         char                    oxe_buf[0];
1456 };
1457
1458 static int osd_oxc_get(struct osd_object *obj, const char *name,
1459                        struct lu_buf *buf)
1460 {
1461         struct osd_xattr_entry *tmp;
1462         struct osd_xattr_entry *oxe = NULL;
1463         size_t namelen = strlen(name);
1464         int rc;
1465
1466         rcu_read_lock();
1467         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1468                 if (namelen == tmp->oxe_namelen &&
1469                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1470                         oxe = tmp;
1471                         break;
1472                 }
1473         }
1474
1475         if (oxe == NULL)
1476                 GOTO(out, rc = -ENOENT);
1477
1478         if (!oxe->oxe_exist)
1479                 GOTO(out, rc = -ENODATA);
1480
1481         /* vallen */
1482         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1483         LASSERT(rc > 0);
1484
1485         if (buf->lb_buf == NULL)
1486                 GOTO(out, rc);
1487
1488         if (buf->lb_len < rc)
1489                 GOTO(out, rc = -ERANGE);
1490
1491         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1492 out:
1493         rcu_read_unlock();
1494
1495         return rc;
1496 }
1497
1498 static void osd_oxc_free(struct rcu_head *head)
1499 {
1500         struct osd_xattr_entry *oxe;
1501
1502         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1503         OBD_FREE(oxe, oxe->oxe_len);
1504 }
1505
1506 static void osd_oxc_add(struct osd_object *obj, const char *name,
1507                         const char *buf, int buflen)
1508 {
1509         struct osd_xattr_entry *oxe;
1510         struct osd_xattr_entry *old = NULL;
1511         struct osd_xattr_entry *tmp;
1512         size_t namelen = strlen(name);
1513         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1514
1515         OBD_ALLOC(oxe, len);
1516         if (oxe == NULL)
1517                 return;
1518
1519         INIT_LIST_HEAD(&oxe->oxe_list);
1520         oxe->oxe_len = len;
1521         oxe->oxe_namelen = namelen;
1522         memcpy(oxe->oxe_buf, name, namelen);
1523         if (buflen > 0) {
1524                 LASSERT(buf != NULL);
1525                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1526                 oxe->oxe_exist = true;
1527         } else {
1528                 oxe->oxe_exist = false;
1529         }
1530
1531         /* this should be rarely called, just remove old and add new */
1532         spin_lock(&obj->oo_guard);
1533         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1534                 if (namelen == tmp->oxe_namelen &&
1535                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1536                         old = tmp;
1537                         break;
1538                 }
1539         }
1540         if (old != NULL) {
1541                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1542                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1543         } else {
1544                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1545         }
1546         spin_unlock(&obj->oo_guard);
1547 }
1548
1549 static void osd_oxc_del(struct osd_object *obj, const char *name)
1550 {
1551         struct osd_xattr_entry *oxe;
1552         size_t namelen = strlen(name);
1553
1554         spin_lock(&obj->oo_guard);
1555         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1556                 if (namelen == oxe->oxe_namelen &&
1557                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1558                         list_del_rcu(&oxe->oxe_list);
1559                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1560                         break;
1561                 }
1562         }
1563         spin_unlock(&obj->oo_guard);
1564 }
1565
1566 static void osd_oxc_fini(struct osd_object *obj)
1567 {
1568         struct osd_xattr_entry *oxe, *next;
1569
1570         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1571                 list_del(&oxe->oxe_list);
1572                 OBD_FREE(oxe, oxe->oxe_len);
1573         }
1574 }
1575
1576 /*
1577  * Concurrency: no concurrent access is possible that late in object
1578  * life-cycle.
1579  */
1580 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1581 {
1582         struct osd_object *obj = osd_obj(l);
1583         struct lu_object_header *h = obj->oo_header;
1584
1585         LINVRNT(osd_invariant(obj));
1586
1587         osd_oxc_fini(obj);
1588         dt_object_fini(&obj->oo_dt);
1589         if (obj->oo_hl_head != NULL)
1590                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1591         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1592         OBD_FREE_PTR(obj);
1593         if (unlikely(h)) {
1594                 lu_object_header_fini(h);
1595                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1596                 kfree_rcu(h, loh_rcu);
1597         }
1598 }
1599
1600 /*
1601  * Concurrency: no concurrent access is possible that late in object
1602  * life-cycle.
1603  */
1604 static void osd_index_fini(struct osd_object *o)
1605 {
1606         struct iam_container *bag;
1607
1608         if (o->oo_dir != NULL) {
1609                 bag = &o->oo_dir->od_container;
1610                 if (o->oo_inode != NULL) {
1611                         if (bag->ic_object == o->oo_inode)
1612                                 iam_container_fini(bag);
1613                 }
1614                 OBD_FREE_PTR(o->oo_dir);
1615                 o->oo_dir = NULL;
1616         }
1617 }
1618
1619 /*
1620  * Concurrency: no concurrent access is possible that late in object
1621  * life-cycle (for all existing callers, that is. New callers have to provide
1622  * their own locking.)
1623  */
1624 static int osd_inode_unlinked(const struct inode *inode)
1625 {
1626         return inode->i_nlink == 0;
1627 }
1628
1629 enum {
1630         OSD_TXN_OI_DELETE_CREDITS    = 20,
1631         OSD_TXN_INODE_DELETE_CREDITS = 20
1632 };
1633
1634 /*
1635  * Journal
1636  */
1637
1638 #if OSD_THANDLE_STATS
1639 /**
1640  * Set time when the handle is allocated
1641  */
1642 static void osd_th_alloced(struct osd_thandle *oth)
1643 {
1644         oth->oth_alloced = ktime_get();
1645 }
1646
1647 /**
1648  * Set time when the handle started
1649  */
1650 static void osd_th_started(struct osd_thandle *oth)
1651 {
1652         oth->oth_started = ktime_get();
1653 }
1654
1655 /**
1656  * Check whether the we deal with this handle for too long.
1657  */
1658 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1659                                 ktime_t alloced, ktime_t started,
1660                                 ktime_t closed)
1661 {
1662         ktime_t now = ktime_get();
1663
1664         LASSERT(dev != NULL);
1665
1666         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1667                             ktime_us_delta(started, alloced));
1668         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1669                             ktime_us_delta(closed, started));
1670         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1671                             ktime_us_delta(now, closed));
1672
1673         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1674                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1675                                 oth, now, alloced, started, closed);
1676                 libcfs_debug_dumpstack(NULL);
1677         }
1678 }
1679
1680 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1681 {                                                                       \
1682         ktime_t __closed = ktime_get();                                 \
1683         ktime_t __alloced = oth->oth_alloced;                           \
1684         ktime_t __started = oth->oth_started;                           \
1685                                                                         \
1686         expr;                                                           \
1687         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1688 }
1689
1690 #else /* OSD_THANDLE_STATS */
1691
1692 #define osd_th_alloced(h)                  do {} while(0)
1693 #define osd_th_started(h)                  do {} while(0)
1694 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1695
1696 #endif /* OSD_THANDLE_STATS */
1697
1698 /*
1699  * Concurrency: doesn't access mutable data.
1700  */
1701 static int osd_param_is_not_sane(const struct osd_device *dev,
1702                                  const struct thandle *th)
1703 {
1704         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1705
1706         return oh->ot_credits > osd_transaction_size(dev);
1707 }
1708
1709 /*
1710  * Concurrency: shouldn't matter.
1711  */
1712 static void osd_trans_commit_cb(struct super_block *sb,
1713                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1714 {
1715         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1716         struct thandle *th = &oh->ot_super;
1717         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1718         struct dt_txn_commit_cb *dcb, *tmp;
1719
1720         LASSERT(oh->ot_handle == NULL);
1721
1722         if (error)
1723                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1724
1725         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1726         /* call per-transaction callbacks if any */
1727         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1728                                  dcb_linkage) {
1729                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1730                          "commit callback entry: magic=%x name='%s'\n",
1731                          dcb->dcb_magic, dcb->dcb_name);
1732                 list_del_init(&dcb->dcb_linkage);
1733                 dcb->dcb_func(NULL, th, dcb, error);
1734         }
1735
1736         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1737         lu_device_put(lud);
1738         th->th_dev = NULL;
1739
1740         OBD_FREE_PTR(oh);
1741 }
1742
1743 static struct thandle *osd_trans_create(const struct lu_env *env,
1744                                         struct dt_device *d)
1745 {
1746         struct osd_thread_info *oti = osd_oti_get(env);
1747         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1748         struct osd_thandle *oh;
1749         struct thandle *th;
1750
1751         ENTRY;
1752
1753         if (d->dd_rdonly) {
1754                 CERROR("%s: someone try to start transaction under "
1755                        "readonly mode, should be disabled.\n",
1756                        osd_name(osd_dt_dev(d)));
1757                 dump_stack();
1758                 RETURN(ERR_PTR(-EROFS));
1759         }
1760
1761         /* on pending IO in this thread should left from prev. request */
1762         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1763
1764         sb_start_write(osd_sb(osd_dt_dev(d)));
1765
1766         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1767         if (!oh) {
1768                 sb_end_write(osd_sb(osd_dt_dev(d)));
1769                 RETURN(ERR_PTR(-ENOMEM));
1770         }
1771
1772         oh->ot_quota_trans = &oti->oti_quota_trans;
1773         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1774         th = &oh->ot_super;
1775         th->th_dev = d;
1776         th->th_result = 0;
1777         oh->ot_credits = 0;
1778         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1779         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1780         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1781         osd_th_alloced(oh);
1782
1783         memset(oti->oti_declare_ops, 0,
1784                sizeof(oti->oti_declare_ops));
1785         memset(oti->oti_declare_ops_cred, 0,
1786                sizeof(oti->oti_declare_ops_cred));
1787         memset(oti->oti_declare_ops_used, 0,
1788                sizeof(oti->oti_declare_ops_used));
1789
1790         oti->oti_ins_cache_depth++;
1791
1792         RETURN(th);
1793 }
1794
1795 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1796 {
1797         struct osd_thread_info *oti = osd_oti_get(env);
1798         struct osd_thandle *oh;
1799
1800         oh = container_of(th, struct osd_thandle, ot_super);
1801         LASSERT(oh != NULL);
1802
1803         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1804               oti->oti_declare_ops[OSD_OT_CREATE],
1805               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1806               oti->oti_declare_ops_used[OSD_OT_CREATE],
1807               oti->oti_declare_ops[OSD_OT_DESTROY],
1808               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1809               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1810         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1811               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1812               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1813               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1814               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1815               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1816               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1817         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_WRITE],
1819               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1820               oti->oti_declare_ops_used[OSD_OT_WRITE],
1821               oti->oti_declare_ops[OSD_OT_PUNCH],
1822               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1823               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1824               oti->oti_declare_ops[OSD_OT_QUOTA],
1825               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1826               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1827         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1828               oti->oti_declare_ops[OSD_OT_INSERT],
1829               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1830               oti->oti_declare_ops_used[OSD_OT_INSERT],
1831               oti->oti_declare_ops[OSD_OT_DELETE],
1832               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1833               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1834         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1835               oti->oti_declare_ops[OSD_OT_REF_ADD],
1836               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1837               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1838               oti->oti_declare_ops[OSD_OT_REF_DEL],
1839               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1840               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1841 }
1842
1843 /*
1844  * Concurrency: shouldn't matter.
1845  */
1846 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1847                            struct thandle *th)
1848 {
1849         struct osd_thread_info *oti = osd_oti_get(env);
1850         struct osd_device *dev = osd_dt_dev(d);
1851         handle_t *jh;
1852         struct osd_thandle *oh;
1853         int rc;
1854
1855         ENTRY;
1856
1857         LASSERT(current->journal_info == NULL);
1858
1859         oh = container_of(th, struct osd_thandle, ot_super);
1860         LASSERT(oh != NULL);
1861         LASSERT(oh->ot_handle == NULL);
1862
1863         rc = dt_txn_hook_start(env, d, th);
1864         if (rc != 0)
1865                 GOTO(out, rc);
1866
1867         if (unlikely(osd_param_is_not_sane(dev, th))) {
1868                 static unsigned long last_printed;
1869                 static int last_credits;
1870
1871                 /*
1872                  * don't make noise on a tiny testing systems
1873                  * actual credits misuse will be caught anyway
1874                  */
1875                 if (last_credits != oh->ot_credits &&
1876                     time_after(jiffies, last_printed +
1877                                cfs_time_seconds(60)) &&
1878                     osd_transaction_size(dev) > 512) {
1879                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1880                               oh->ot_credits, osd_transaction_size(dev));
1881                         osd_trans_dump_creds(env, th);
1882                         libcfs_debug_dumpstack(NULL);
1883                         last_credits = oh->ot_credits;
1884                         last_printed = jiffies;
1885                 }
1886                 /*
1887                  * XXX Limit the credits to 'max_transaction_buffers', and
1888                  *     let the underlying filesystem to catch the error if
1889                  *     we really need so many credits.
1890                  *
1891                  *     This should be removed when we can calculate the
1892                  *     credits precisely.
1893                  */
1894                 oh->ot_credits = osd_transaction_size(dev);
1895         } else if (ldiskfs_track_declares_assert != 0) {
1896                 /*
1897                  * reserve few credits to prevent an assertion in JBD
1898                  * our debugging mechanism will be able to detected
1899                  * overuse. this can help to debug single-update
1900                  * transactions
1901                  */
1902                 oh->ot_credits += 10;
1903                 if (unlikely(osd_param_is_not_sane(dev, th)))
1904                         oh->ot_credits = osd_transaction_size(dev);
1905         }
1906
1907         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1908                 GOTO(out, rc = -EIO);
1909
1910         /*
1911          * XXX temporary stuff. Some abstraction layer should
1912          * be used.
1913          */
1914         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1915         osd_th_started(oh);
1916         if (!IS_ERR(jh)) {
1917                 oh->ot_handle = jh;
1918                 LASSERT(oti->oti_txns == 0);
1919
1920                 lu_device_get(&d->dd_lu_dev);
1921                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1922                               "osd-tx", th);
1923                 oti->oti_txns++;
1924                 rc = 0;
1925         } else {
1926                 rc = PTR_ERR(jh);
1927         }
1928 out:
1929         RETURN(rc);
1930 }
1931
1932 static int osd_seq_exists(const struct lu_env *env,
1933                           struct osd_device *osd, u64 seq)
1934 {
1935         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1936         struct seq_server_site *ss = osd_seq_site(osd);
1937         int rc;
1938
1939         ENTRY;
1940
1941         LASSERT(ss != NULL);
1942         LASSERT(ss->ss_server_fld != NULL);
1943
1944         rc = osd_fld_lookup(env, osd, seq, range);
1945         if (rc != 0) {
1946                 if (rc != -ENOENT)
1947                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1948                                osd_name(osd), seq, rc);
1949                 RETURN(0);
1950         }
1951
1952         RETURN(ss->ss_node_id == range->lsr_index);
1953 }
1954
1955 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1956 {
1957         struct dt_txn_commit_cb *dcb;
1958         struct dt_txn_commit_cb *tmp;
1959
1960         /* call per-transaction stop callbacks if any */
1961         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1962                                  dcb_linkage) {
1963                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1964                          "commit callback entry: magic=%x name='%s'\n",
1965                          dcb->dcb_magic, dcb->dcb_name);
1966                 list_del_init(&dcb->dcb_linkage);
1967                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1968         }
1969 }
1970
1971 /*
1972  * Concurrency: shouldn't matter.
1973  */
1974 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1975                           struct thandle *th)
1976 {
1977         struct osd_thread_info *oti = osd_oti_get(env);
1978         struct osd_thandle *oh;
1979         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1980         struct osd_device *osd = osd_dt_dev(th->th_dev);
1981         struct qsd_instance *qsd = osd_def_qsd(osd);
1982         struct lquota_trans *qtrans;
1983         LIST_HEAD(truncates);
1984         int rc = 0, remove_agents = 0;
1985
1986         ENTRY;
1987
1988         oh = container_of(th, struct osd_thandle, ot_super);
1989
1990         remove_agents = oh->ot_remove_agents;
1991
1992         qtrans = oh->ot_quota_trans;
1993         oh->ot_quota_trans = NULL;
1994
1995         /* move locks to local list, stop tx, execute truncates */
1996         list_splice(&oh->ot_trunc_locks, &truncates);
1997
1998         if (oh->ot_handle != NULL) {
1999                 int rc2;
2000
2001                 handle_t *hdl = oh->ot_handle;
2002
2003                 /*
2004                  * add commit callback
2005                  * notice we don't do this in osd_trans_start()
2006                  * as underlying transaction can change during truncate
2007                  */
2008                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2009                                              &oh->ot_jcb);
2010
2011                 LASSERT(oti->oti_txns == 1);
2012                 oti->oti_txns--;
2013
2014                 rc = dt_txn_hook_stop(env, th);
2015                 if (rc != 0)
2016                         CERROR("%s: failed in transaction hook: rc = %d\n",
2017                                osd_name(osd), rc);
2018
2019                 osd_trans_stop_cb(oh, rc);
2020                 /* hook functions might modify th_sync */
2021                 hdl->h_sync = th->th_sync;
2022
2023                 oh->ot_handle = NULL;
2024                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2025                 if (rc2 != 0)
2026                         CERROR("%s: failed to stop transaction: rc = %d\n",
2027                                osd_name(osd), rc2);
2028                 if (!rc)
2029                         rc = rc2;
2030
2031                 osd_process_truncates(&truncates);
2032         } else {
2033                 osd_trans_stop_cb(oh, th->th_result);
2034                 OBD_FREE_PTR(oh);
2035         }
2036
2037         osd_trunc_unlock_all(env, &truncates);
2038
2039         /* inform the quota slave device that the transaction is stopping */
2040         qsd_op_end(env, qsd, qtrans);
2041
2042         /*
2043          * as we want IO to journal and data IO be concurrent, we don't block
2044          * awaiting data IO completion in osd_do_bio(), instead we wait here
2045          * once transaction is submitted to the journal. all reqular requests
2046          * don't do direct IO (except read/write), thus this wait_event becomes
2047          * no-op for them.
2048          *
2049          * IMPORTANT: we have to wait till any IO submited by the thread is
2050          * completed otherwise iobuf may be corrupted by different request
2051          */
2052         wait_event(iobuf->dr_wait,
2053                        atomic_read(&iobuf->dr_numreqs) == 0);
2054         osd_fini_iobuf(osd, iobuf);
2055         if (!rc)
2056                 rc = iobuf->dr_error;
2057
2058         if (unlikely(remove_agents != 0))
2059                 osd_process_scheduled_agent_removals(env, osd);
2060
2061         oti->oti_ins_cache_depth--;
2062         /* reset OI cache for safety */
2063         if (oti->oti_ins_cache_depth == 0)
2064                 oti->oti_ins_cache_used = 0;
2065
2066         sb_end_write(osd_sb(osd));
2067
2068         RETURN(rc);
2069 }
2070
2071 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2072 {
2073         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2074                                               ot_super);
2075
2076         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2077         LASSERT(&dcb->dcb_func != NULL);
2078         if (dcb->dcb_flags & DCB_TRANS_STOP)
2079                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2080         else
2081                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2082
2083         return 0;
2084 }
2085
2086 /*
2087  * Called just before object is freed. Releases all resources except for
2088  * object itself (that is released by osd_object_free()).
2089  *
2090  * Concurrency: no concurrent access is possible that late in object
2091  * life-cycle.
2092  */
2093 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2094 {
2095         struct osd_object *obj = osd_obj(l);
2096         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2097         struct inode *inode = obj->oo_inode;
2098         __u64 projid;
2099         qid_t uid;
2100         qid_t gid;
2101
2102         LINVRNT(osd_invariant(obj));
2103
2104         /*
2105          * If object is unlinked remove fid->ino mapping from object index.
2106          */
2107
2108         osd_index_fini(obj);
2109
2110         if (!inode)
2111                 return;
2112
2113         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2114                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2115
2116         uid = i_uid_read(inode);
2117         gid = i_gid_read(inode);
2118         projid = i_projid_read(inode);
2119
2120         obj->oo_inode = NULL;
2121         iput(inode);
2122
2123         /* do not rebalance quota if the caller needs to release memory
2124          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2125          * transaction and risk to deadlock - LU-12178 */
2126         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2127                 return;
2128
2129         if (!obj->oo_header && qsd) {
2130                 struct osd_thread_info *info = osd_oti_get(env);
2131                 struct lquota_id_info *qi = &info->oti_qi;
2132
2133                 /* Release granted quota to master if necessary */
2134                 qi->lqi_id.qid_uid = uid;
2135                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2136
2137                 qi->lqi_id.qid_uid = gid;
2138                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2139
2140                 qi->lqi_id.qid_uid = projid;
2141                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2142         }
2143 }
2144
2145 /*
2146  * Concurrency: ->loo_object_release() is called under site spin-lock.
2147  */
2148 static void osd_object_release(const struct lu_env *env,
2149                                struct lu_object *l)
2150 {
2151         struct osd_object *o = osd_obj(l);
2152
2153         /*
2154          * nobody should be releasing a non-destroyed object with nlink=0
2155          * the API allows this, but ldiskfs doesn't like and then report
2156          * this inode as deleted
2157          */
2158         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2159                   o->oo_inode->i_nlink == 0));
2160 }
2161
2162 /*
2163  * Concurrency: shouldn't matter.
2164  */
2165 static int osd_object_print(const struct lu_env *env, void *cookie,
2166                             lu_printer_t p, const struct lu_object *l)
2167 {
2168         struct osd_object *o = osd_obj(l);
2169         struct iam_descr *d;
2170
2171         if (o->oo_dir != NULL)
2172                 d = o->oo_dir->od_container.ic_descr;
2173         else
2174                 d = NULL;
2175         return (*p)(env, cookie,
2176                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2177                     o, o->oo_inode,
2178                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2179                     o->oo_inode ? o->oo_inode->i_generation : 0,
2180                     d ? d->id_ops->id_name : "plain");
2181 }
2182
2183 /*
2184  * Concurrency: shouldn't matter.
2185  */
2186 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2187                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2188 {
2189         struct osd_device *osd = osd_dt_dev(d);
2190         struct super_block *sb = osd_sb(osd);
2191         struct kstatfs *ksfs;
2192         __u64 reserved;
2193         int result = 0;
2194
2195         if (unlikely(osd->od_mnt == NULL))
2196                 return -EINPROGRESS;
2197
2198         /* osd_lproc.c call this without env, allocate ksfs for that case */
2199         if (unlikely(env == NULL)) {
2200                 OBD_ALLOC_PTR(ksfs);
2201                 if (ksfs == NULL)
2202                         return -ENOMEM;
2203         } else {
2204                 ksfs = &osd_oti_get(env)->oti_ksfs;
2205         }
2206
2207         result = sb->s_op->statfs(sb->s_root, ksfs);
2208         if (result)
2209                 goto out;
2210
2211         statfs_pack(sfs, ksfs);
2212         if (unlikely(sb->s_flags & SB_RDONLY))
2213                 sfs->os_state |= OS_STATFS_READONLY;
2214
2215         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2216
2217         if (ldiskfs_has_feature_extents(sb))
2218                 sfs->os_maxbytes = sb->s_maxbytes;
2219         else
2220                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2221
2222         /*
2223          * Reserve some space so to avoid fragmenting the filesystem too much.
2224          * Fragmentation not only impacts performance, but can also increase
2225          * metadata overhead significantly, causing grant calculation to be
2226          * wrong.
2227          *
2228          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2229          */
2230         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2231         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2232         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2233                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2234
2235         sfs->os_blocks -= reserved;
2236         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2237         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2238
2239 out:
2240         if (unlikely(env == NULL))
2241                 OBD_FREE_PTR(ksfs);
2242         return result;
2243 }
2244
2245 /**
2246  * Estimate space needed for file creations. We assume the largest filename
2247  * which is 2^64 - 1, hence a filename of 20 chars.
2248  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2249  */
2250 #ifdef __LDISKFS_DIR_REC_LEN
2251 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2252 #else
2253 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2254 #endif
2255
2256 /*
2257  * Concurrency: doesn't access mutable data.
2258  */
2259 static void osd_conf_get(const struct lu_env *env,
2260                          const struct dt_device *dev,
2261                          struct dt_device_param *param)
2262 {
2263         struct osd_device *d = osd_dt_dev(dev);
2264         struct super_block *sb = osd_sb(d);
2265         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2266         const char *name;
2267         int ea_overhead;
2268
2269         /*
2270          * XXX should be taken from not-yet-existing fs abstraction layer.
2271          */
2272         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2273         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2274         param->ddp_symlink_max  = sb->s_blocksize;
2275         param->ddp_mount_type   = LDD_MT_LDISKFS;
2276         if (ldiskfs_has_feature_extents(sb))
2277                 param->ddp_maxbytes = sb->s_maxbytes;
2278         else
2279                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2280         /*
2281          * inode are statically allocated, so per-inode space consumption
2282          * is the space consumed by the directory entry
2283          */
2284         param->ddp_inodespace     = PER_OBJ_USAGE;
2285         /*
2286          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2287          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2288          * maximum length to not under-count the actual number of extents
2289          * needed for writing a file if there are sub-optimal block allocations.
2290          */
2291         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2292         /* worst-case extent insertion metadata overhead */
2293         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2294         param->ddp_mntopts = 0;
2295         if (test_opt(sb, XATTR_USER))
2296                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2297         if (test_opt(sb, POSIX_ACL))
2298                 param->ddp_mntopts |= MNTOPT_ACL;
2299
2300         /*
2301          * LOD might calculate the max stripe count based on max_ea_size,
2302          * so we need take account in the overhead as well,
2303          * xattr_header + magic + xattr_entry_head
2304          */
2305         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2306                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2307
2308 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2309         if (ldiskfs_has_feature_ea_inode(sb))
2310                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2311                                                                 ea_overhead;
2312         else
2313 #endif
2314                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2315
2316         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2317                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2318
2319         /*
2320          * Preferred RPC size for efficient disk IO.  4MB shows good
2321          * all-around performance for ldiskfs, but use bigalloc chunk size
2322          * by default if larger.
2323          */
2324 #if defined(LDISKFS_CLUSTER_SIZE)
2325         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2326                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2327         else
2328 #endif
2329                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2330
2331         param->ddp_t10_cksum_type = 0;
2332         if (bi) {
2333                 unsigned short interval = blk_integrity_interval(bi);
2334                 name = blk_integrity_name(bi);
2335                 /*
2336                  * Expected values:
2337                  * T10-DIF-TYPE1-CRC
2338                  * T10-DIF-TYPE3-CRC
2339                  * T10-DIF-TYPE1-IP
2340                  * T10-DIF-TYPE3-IP
2341                  */
2342                 if (strncmp(name, "T10-DIF-TYPE",
2343                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2344                         /* also skip "1/3-" at end */
2345                         const int type_off = sizeof("T10-DIF-TYPE.");
2346                         char type_number = name[type_off - 2];
2347
2348                         if (interval != 512 && interval != 4096) {
2349                                 CERROR("%s: unsupported T10PI sector size %u\n",
2350                                        d->od_svname, interval);
2351                         } else if (type_number != '1' && type_number != '3') {
2352                                 CERROR("%s: unsupported T10PI type %s\n",
2353                                        d->od_svname, name);
2354                         } else if (strcmp(name + type_off, "CRC") == 0) {
2355                                 d->od_t10_type = type_number == '1' ?
2356                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2357                                 param->ddp_t10_cksum_type = interval == 512 ?
2358                                         OBD_CKSUM_T10CRC512 :
2359                                         OBD_CKSUM_T10CRC4K;
2360                         } else if (strcmp(name + type_off, "IP") == 0) {
2361                                 d->od_t10_type = type_number == '1' ?
2362                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2363                                 param->ddp_t10_cksum_type = interval == 512 ?
2364                                         OBD_CKSUM_T10IP512 :
2365                                         OBD_CKSUM_T10IP4K;
2366                         } else {
2367                                 CERROR("%s: unsupported checksum type of "
2368                                        "T10PI type '%s'",
2369                                        d->od_svname, name);
2370                         }
2371
2372                 } else {
2373                         CERROR("%s: unsupported T10PI type '%s'",
2374                                d->od_svname, name);
2375                 }
2376         }
2377 }
2378
2379 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2380 {
2381         return osd_sb(osd_dt_dev(d));
2382 }
2383
2384 /*
2385  * Concurrency: shouldn't matter.
2386  */
2387 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2388 {
2389         int rc;
2390         struct super_block *s = osd_sb(osd_dt_dev(d));
2391         ENTRY;
2392
2393         down_read(&s->s_umount);
2394         rc = s->s_op->sync_fs(s, 1);
2395         up_read(&s->s_umount);
2396
2397         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2398                rc);
2399
2400         return rc;
2401 }
2402
2403 /**
2404  * Start commit for OSD device.
2405  *
2406  * An implementation of dt_commit_async method for OSD device.
2407  * Asychronously starts underlayng fs sync and thereby a transaction
2408  * commit.
2409  *
2410  * \param env environment
2411  * \param d dt device
2412  *
2413  * \see dt_device_operations
2414  */
2415 static int osd_commit_async(const struct lu_env *env,
2416                             struct dt_device *d)
2417 {
2418         struct super_block *s = osd_sb(osd_dt_dev(d));
2419         int rc;
2420
2421         ENTRY;
2422
2423         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2424         down_read(&s->s_umount);
2425         rc = s->s_op->sync_fs(s, 0);
2426         up_read(&s->s_umount);
2427
2428         RETURN(rc);
2429 }
2430
2431 static int (*priv_security_file_alloc)(struct file *file);
2432
2433 int osd_security_file_alloc(struct file *file)
2434 {
2435         if (priv_security_file_alloc)
2436                 return priv_security_file_alloc(file);
2437         return 0;
2438 }
2439
2440 /*
2441  * Concurrency: shouldn't matter.
2442  */
2443 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2444 {
2445         struct super_block *sb = osd_sb(osd_dt_dev(d));
2446         struct block_device *dev = sb->s_bdev;
2447         int rc = -EOPNOTSUPP;
2448
2449         ENTRY;
2450
2451         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2452                osd_dt_dev(d)->od_svname, (long)dev, rc);
2453
2454         RETURN(rc);
2455 }
2456
2457 /**
2458  * Note: we do not count into QUOTA here.
2459  * If we mount with --data_journal we may need more.
2460  */
2461 const int osd_dto_credits_noquota[DTO_NR] = {
2462         /**
2463          * Insert.
2464          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2465          * SINGLEDATA_TRANS_BLOCKS(8)
2466          * XXX Note: maybe iam need more, since iam have more level than
2467          *           EXT3 htree.
2468          */
2469         [DTO_INDEX_INSERT]  = 16,
2470         /**
2471          * Delete
2472          * just modify a single entry, probably merge few within a block
2473          */
2474         [DTO_INDEX_DELETE]  = 1,
2475         /**
2476          * Used for OI scrub
2477          */
2478         [DTO_INDEX_UPDATE]  = 16,
2479         /**
2480          * 4(inode, inode bits, groups, GDT)
2481          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2482          */
2483         [DTO_OBJECT_CREATE] = 4,
2484         /**
2485          * 4(inode, inode bits, groups, GDT)
2486          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2487          */
2488         [DTO_OBJECT_DELETE] = 4,
2489         /**
2490          * Attr set credits (inode)
2491          */
2492         [DTO_ATTR_SET_BASE] = 1,
2493         /**
2494          * Xattr set. The same as xattr of EXT3.
2495          * DATA_TRANS_BLOCKS(14)
2496          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2497          * are also counted in. Do not know why?
2498          */
2499         [DTO_XATTR_SET]     = 14,
2500         /**
2501          * credits for inode change during write.
2502          */
2503         [DTO_WRITE_BASE]    = 3,
2504         /**
2505          * credits for single block write.
2506          */
2507         [DTO_WRITE_BLOCK]   = 14,
2508         /**
2509          * Attr set credits for chown.
2510          * This is extra credits for setattr, and it is null without quota
2511          */
2512         [DTO_ATTR_SET_CHOWN] = 0
2513 };
2514
2515 static const struct dt_device_operations osd_dt_ops = {
2516         .dt_root_get       = osd_root_get,
2517         .dt_statfs         = osd_statfs,
2518         .dt_trans_create   = osd_trans_create,
2519         .dt_trans_start    = osd_trans_start,
2520         .dt_trans_stop     = osd_trans_stop,
2521         .dt_trans_cb_add   = osd_trans_cb_add,
2522         .dt_conf_get       = osd_conf_get,
2523         .dt_mnt_sb_get     = osd_mnt_sb_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 void osd_inode_getattr(const struct lu_env *env,
2596                               struct inode *inode, struct lu_attr *attr)
2597 {
2598         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2599                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2600                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2601                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2602
2603         attr->la_atime = inode->i_atime.tv_sec;
2604         attr->la_mtime = inode->i_mtime.tv_sec;
2605         attr->la_ctime = inode->i_ctime.tv_sec;
2606         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2607         attr->la_mode    = inode->i_mode;
2608         attr->la_size    = i_size_read(inode);
2609         attr->la_blocks  = inode->i_blocks;
2610         attr->la_uid     = i_uid_read(inode);
2611         attr->la_gid     = i_gid_read(inode);
2612         attr->la_projid  = i_projid_read(inode);
2613         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2614         attr->la_nlink   = inode->i_nlink;
2615         attr->la_rdev    = inode->i_rdev;
2616         attr->la_blksize = 1 << inode->i_blkbits;
2617         attr->la_blkbits = inode->i_blkbits;
2618         /*
2619          * Ext4 did not transfer inherit flags from raw inode
2620          * to inode flags, and ext4 internally test raw inode
2621          * @i_flags directly. Instead of patching ext4, we do it here.
2622          */
2623         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2624                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2625 }
2626
2627 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2628                             u64 *count)
2629 {
2630         struct osd_object *obj = osd_dt_obj(dt);
2631         const struct dt_it_ops *iops;
2632         struct dt_it *it;
2633         int rc;
2634
2635         ENTRY;
2636
2637         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2638         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2639
2640         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2641                 *count = obj->oo_dirent_count;
2642                 RETURN(0);
2643         }
2644
2645         /* directory not initialized yet */
2646         if (!dt->do_index_ops) {
2647                 *count = 0;
2648                 RETURN(0);
2649         }
2650
2651         iops = &dt->do_index_ops->dio_it;
2652         it = iops->init(env, dt, LUDA_64BITHASH);
2653         if (IS_ERR(it))
2654                 RETURN(PTR_ERR(it));
2655
2656         rc = iops->load(env, it, 0);
2657         if (rc < 0) {
2658                 if (rc == -ENODATA) {
2659                         rc = 0;
2660                         *count = 0;
2661                 }
2662                 GOTO(out, rc);
2663         }
2664         if (rc > 0)
2665                 rc = iops->next(env, it);
2666
2667         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2668                 if (rc == -ESTALE)
2669                         continue;
2670
2671                 if (iops->key_size(env, it) == 0)
2672                         continue;
2673
2674                 (*count)++;
2675         }
2676         if (rc == 1) {
2677                 obj->oo_dirent_count = *count;
2678                 rc = 0;
2679         }
2680 out:
2681         iops->put(env, it);
2682         iops->fini(env, it);
2683
2684         RETURN(rc);
2685 }
2686
2687 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2688                         struct lu_attr *attr)
2689 {
2690         struct osd_object *obj = osd_dt_obj(dt);
2691         int rc = 0;
2692
2693         if (unlikely(!dt_object_exists(dt)))
2694                 return -ENOENT;
2695         if (unlikely(obj->oo_destroyed))
2696                 return -ENOENT;
2697
2698         LASSERT(!dt_object_remote(dt));
2699         LINVRNT(osd_invariant(obj));
2700
2701         spin_lock(&obj->oo_guard);
2702         osd_inode_getattr(env, obj->oo_inode, attr);
2703         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2704                 attr->la_valid |= LA_FLAGS;
2705                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2706         }
2707         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2708                 attr->la_valid |= LA_FLAGS;
2709                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2710         }
2711         spin_unlock(&obj->oo_guard);
2712
2713         if (S_ISDIR(obj->oo_inode->i_mode) &&
2714             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2715                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2716
2717         return rc;
2718 }
2719
2720 static int osd_declare_attr_qid(const struct lu_env *env,
2721                                 struct osd_object *obj,
2722                                 struct osd_thandle *oh, long long bspace,
2723                                 qid_t old_id, qid_t new_id, bool enforce,
2724                                 unsigned int type, bool ignore_edquot)
2725 {
2726         int rc;
2727         struct osd_thread_info *info = osd_oti_get(env);
2728         struct lquota_id_info  *qi = &info->oti_qi;
2729
2730         qi->lqi_type = type;
2731         /* inode accounting */
2732         qi->lqi_is_blk = false;
2733
2734         /* one more inode for the new id ... */
2735         qi->lqi_id.qid_uid = new_id;
2736         qi->lqi_space      = 1;
2737         /* Reserve credits for the new id */
2738         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2739         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2740                 rc = 0;
2741         if (rc)
2742                 RETURN(rc);
2743
2744         /* and one less inode for the current id */
2745         qi->lqi_id.qid_uid = old_id;
2746         qi->lqi_space = -1;
2747         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2748         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2749                 rc = 0;
2750         if (rc)
2751                 RETURN(rc);
2752
2753         /* block accounting */
2754         qi->lqi_is_blk = true;
2755
2756         /* more blocks for the new id ... */
2757         qi->lqi_id.qid_uid = new_id;
2758         qi->lqi_space      = bspace;
2759         /*
2760          * Credits for the new uid has been reserved, re-use "obj"
2761          * to save credit reservation.
2762          */
2763         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2764         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2765                 rc = 0;
2766         if (rc)
2767                 RETURN(rc);
2768
2769         /* and finally less blocks for the current uid */
2770         qi->lqi_id.qid_uid = old_id;
2771         qi->lqi_space      = -bspace;
2772         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2773         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2774                 rc = 0;
2775
2776         RETURN(rc);
2777 }
2778
2779 static int osd_declare_attr_set(const struct lu_env *env,
2780                                 struct dt_object *dt,
2781                                 const struct lu_attr *attr,
2782                                 struct thandle *handle)
2783 {
2784         struct osd_thandle *oh;
2785         struct osd_object *obj;
2786         qid_t uid;
2787         qid_t gid;
2788         long long bspace;
2789         int rc = 0;
2790         bool enforce;
2791
2792         ENTRY;
2793
2794         LASSERT(dt != NULL);
2795         LASSERT(handle != NULL);
2796
2797         obj = osd_dt_obj(dt);
2798         LASSERT(osd_invariant(obj));
2799
2800         oh = container_of(handle, struct osd_thandle, ot_super);
2801         LASSERT(oh->ot_handle == NULL);
2802
2803         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2804                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2805
2806         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2807                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2808
2809         if (attr == NULL || obj->oo_inode == NULL)
2810                 RETURN(rc);
2811
2812         bspace   = obj->oo_inode->i_blocks << 9;
2813         bspace   = toqb(bspace);
2814
2815         /*
2816          * Changing ownership is always preformed by super user, it should not
2817          * fail with EDQUOT unless required explicitly.
2818          *
2819          * We still need to call the osd_declare_qid() to calculate the journal
2820          * credits for updating quota accounting files and to trigger quota
2821          * space adjustment once the operation is completed.
2822          */
2823         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2824                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2825
2826                 if (!ignore_edquot)
2827                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2828                                "(the quota space is %lld)\n",
2829                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2830                                attr->la_gid, bspace);
2831
2832                 /* USERQUOTA */
2833                 uid = i_uid_read(obj->oo_inode);
2834                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2835                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2836                                           attr->la_uid, enforce, USRQUOTA,
2837                                           true);
2838                 if (rc)
2839                         RETURN(rc);
2840
2841                 gid = i_gid_read(obj->oo_inode);
2842                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2843                        attr->la_uid, gid, attr->la_gid);
2844                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2845                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2846                                           attr->la_gid, enforce, GRPQUOTA,
2847                                           ignore_edquot);
2848                 if (rc)
2849                         RETURN(rc);
2850
2851         }
2852 #ifdef HAVE_PROJECT_QUOTA
2853         if (attr->la_valid & LA_PROJID) {
2854                 __u32 projid = i_projid_read(obj->oo_inode);
2855
2856                 enforce = (attr->la_valid & LA_PROJID) &&
2857                                         (attr->la_projid != projid);
2858                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2859                                           (qid_t)projid, (qid_t)attr->la_projid,
2860                                           enforce, PRJQUOTA, true);
2861                 if (rc)
2862                         RETURN(rc);
2863         }
2864 #endif
2865         RETURN(rc);
2866 }
2867
2868 static int osd_inode_setattr(const struct lu_env *env,
2869                              struct inode *inode, const struct lu_attr *attr)
2870 {
2871         __u64 bits = attr->la_valid;
2872
2873         /* Only allow set size for regular file */
2874         if (!S_ISREG(inode->i_mode))
2875                 bits &= ~(LA_SIZE | LA_BLOCKS);
2876
2877         if (bits == 0)
2878                 return 0;
2879
2880         if (bits & LA_ATIME)
2881                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2882         if (bits & LA_CTIME)
2883                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2884         if (bits & LA_MTIME)
2885                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2886         if (bits & LA_SIZE) {
2887                 spin_lock(&inode->i_lock);
2888                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2889                 i_size_write(inode, attr->la_size);
2890                 spin_unlock(&inode->i_lock);
2891         }
2892
2893         /*
2894          * OSD should not change "i_blocks" which is used by quota.
2895          * "i_blocks" should be changed by ldiskfs only.
2896          */
2897         if (bits & LA_MODE)
2898                 inode->i_mode = (inode->i_mode & S_IFMT) |
2899                                 (attr->la_mode & ~S_IFMT);
2900         if (bits & LA_UID)
2901                 i_uid_write(inode, attr->la_uid);
2902         if (bits & LA_GID)
2903                 i_gid_write(inode, attr->la_gid);
2904         if (bits & LA_PROJID)
2905                 i_projid_write(inode, attr->la_projid);
2906         if (bits & LA_NLINK)
2907                 set_nlink(inode, attr->la_nlink);
2908         if (bits & LA_RDEV)
2909                 inode->i_rdev = attr->la_rdev;
2910
2911         if (bits & LA_FLAGS) {
2912                 /* always keep S_NOCMTIME */
2913                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2914                                  S_NOCMTIME;
2915                 /*
2916                  * Ext4 did not transfer inherit flags from
2917                  * @inode->i_flags to raw inode i_flags when writing
2918                  * flags, we do it explictly here.
2919                  */
2920                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2921                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2922                 else
2923                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2924         }
2925         return 0;
2926 }
2927
2928 #ifdef HAVE_PROJECT_QUOTA
2929 static int osd_transfer_project(struct inode *inode, __u32 projid)
2930 {
2931         struct super_block *sb = inode->i_sb;
2932         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2933         int err;
2934         kprojid_t kprojid;
2935         struct ldiskfs_iloc iloc;
2936         struct ldiskfs_inode *raw_inode;
2937         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2938
2939         if (!ldiskfs_has_feature_project(sb)) {
2940                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2941                         == LDISKFS_DEF_PROJID);
2942                 if (projid != LDISKFS_DEF_PROJID)
2943                         return -EOPNOTSUPP;
2944                 else
2945                         return 0;
2946         }
2947
2948         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2949                 return -EOPNOTSUPP;
2950
2951         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2952         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2953                 return 0;
2954
2955         err = ldiskfs_get_inode_loc(inode, &iloc);
2956         if (err)
2957                 return err;
2958
2959         raw_inode = ldiskfs_raw_inode(&iloc);
2960         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2961                 err = -EOVERFLOW;
2962                 brelse(iloc.bh);
2963                 return err;
2964         }
2965         brelse(iloc.bh);
2966
2967         dquot_initialize(inode);
2968         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2969         if (transfer_to[PRJQUOTA]) {
2970                 err = __dquot_transfer(inode, transfer_to);
2971                 dqput(transfer_to[PRJQUOTA]);
2972                 if (err)
2973                         return err;
2974         }
2975
2976         return err;
2977 }
2978 #endif
2979
2980 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2981 {
2982         int rc;
2983
2984         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2985             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2986                 struct iattr iattr;
2987
2988                 CDEBUG(D_QUOTA,
2989                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2990                        inode->i_ino, i_uid_read(inode), attr->la_uid,
2991                        i_gid_read(inode), attr->la_gid);
2992
2993                 dquot_initialize(inode);
2994                 iattr.ia_valid = 0;
2995                 if (attr->la_valid & LA_UID)
2996                         iattr.ia_valid |= ATTR_UID;
2997                 if (attr->la_valid & LA_GID)
2998                         iattr.ia_valid |= ATTR_GID;
2999                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3000                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3001
3002                 rc = dquot_transfer(inode, &iattr);
3003                 if (rc) {
3004                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
3005                                "enforcement enabled on the ldiskfs "
3006                                "filesystem?\n", inode->i_sb->s_id, rc);
3007                         return rc;
3008                 }
3009         }
3010
3011         /* Handle project id transfer here properly */
3012         if (attr->la_valid & LA_PROJID &&
3013             attr->la_projid != i_projid_read(inode)) {
3014 #ifdef HAVE_PROJECT_QUOTA
3015                 rc = osd_transfer_project(inode, attr->la_projid);
3016 #else
3017                 rc = -ENOTSUPP;
3018 #endif
3019                 if (rc) {
3020                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3021                                "enforcement enabled on the ldiskfs "
3022                                "filesystem?\n", inode->i_sb->s_id, rc);
3023                         return rc;
3024                 }
3025         }
3026         return 0;
3027 }
3028
3029 static int osd_attr_set(const struct lu_env *env,
3030                         struct dt_object *dt,
3031                         const struct lu_attr *attr,
3032                         struct thandle *handle)
3033 {
3034         struct osd_object *obj = osd_dt_obj(dt);
3035         struct inode *inode;
3036         int rc;
3037
3038         if (!dt_object_exists(dt))
3039                 return -ENOENT;
3040
3041         LASSERT(handle != NULL);
3042         LASSERT(!dt_object_remote(dt));
3043         LASSERT(osd_invariant(obj));
3044
3045         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3046
3047         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3048             !osd_obj2dev(obj)->od_is_ost) {
3049                 struct osd_thread_info *oti = osd_oti_get(env);
3050                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3051                 struct lu_fid *fid1 = &oti->oti_fid;
3052                 struct osd_inode_id *id = &oti->oti_id;
3053                 struct iam_path_descr *ipd;
3054                 struct iam_container *bag;
3055                 struct osd_thandle *oh;
3056                 int rc;
3057
3058                 fid_cpu_to_be(fid1, fid0);
3059                 memset(id, 1, sizeof(*id));
3060                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3061                                   fid0)->oi_dir.od_container;
3062                 ipd = osd_idx_ipd_get(env, bag);
3063                 if (unlikely(ipd == NULL))
3064                         RETURN(-ENOMEM);
3065
3066                 oh = container_of(handle, struct osd_thandle, ot_super);
3067                 rc = iam_update(oh->ot_handle, bag,
3068                                 (const struct iam_key *)fid1,
3069                                 (const struct iam_rec *)id, ipd);
3070                 osd_ipd_put(env, bag, ipd);
3071                 return(rc > 0 ? 0 : rc);
3072         }
3073
3074         inode = obj->oo_inode;
3075
3076         rc = osd_quota_transfer(inode, attr);
3077         if (rc)
3078                 return rc;
3079
3080         spin_lock(&obj->oo_guard);
3081         rc = osd_inode_setattr(env, inode, attr);
3082         spin_unlock(&obj->oo_guard);
3083         if (rc != 0)
3084                 GOTO(out, rc);
3085
3086         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3087
3088         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3089
3090         if (!(attr->la_valid & LA_FLAGS))
3091                 GOTO(out, rc);
3092
3093         /* Let's check if there are extra flags need to be set into LMA */
3094         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3095                 struct osd_thread_info *info = osd_oti_get(env);
3096                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3097
3098                 LASSERT(!obj->oo_pfid_in_lma);
3099
3100                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3101                                  &info->oti_ost_attrs);
3102                 if (rc)
3103                         GOTO(out, rc);
3104
3105                 lma->lma_incompat |=
3106                         lustre_to_lma_flags(attr->la_flags);
3107                 lustre_lma_swab(lma);
3108
3109                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3110
3111                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3112                                      lma, sizeof(*lma), XATTR_REPLACE);
3113                 if (rc != 0) {
3114                         struct osd_device *osd = osd_obj2dev(obj);
3115
3116                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3117                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3118                               lma->lma_incompat, rc);
3119                 } else {
3120                         obj->oo_lma_flags =
3121                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3122                 }
3123                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3124         }
3125 out:
3126
3127         return rc;
3128 }
3129
3130 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3131                                            struct osd_object *obj,
3132                                            const char *name, const int namelen)
3133 {
3134         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3135 }
3136
3137 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3138                       umode_t mode, struct dt_allocation_hint *hint,
3139                       struct thandle *th, struct lu_attr *attr)
3140 {
3141         int result;
3142         struct osd_device *osd = osd_obj2dev(obj);
3143         struct osd_thandle *oth;
3144         struct dt_object *parent = NULL;
3145         struct inode *inode;
3146         uid_t owner[2] = {0, 0};
3147
3148         if (attr->la_valid & LA_UID)
3149                 owner[0] = attr->la_uid;
3150         if (attr->la_valid & LA_GID)
3151                 owner[1] = attr->la_gid;
3152
3153         LINVRNT(osd_invariant(obj));
3154         LASSERT(obj->oo_inode == NULL);
3155         LASSERT(obj->oo_hl_head == NULL);
3156
3157         if (S_ISDIR(mode) && ldiskfs_pdo) {
3158                 obj->oo_hl_head =
3159                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3160                 if (obj->oo_hl_head == NULL)
3161                         return -ENOMEM;
3162         }
3163
3164         oth = container_of(th, struct osd_thandle, ot_super);
3165         LASSERT(oth->ot_handle->h_transaction != NULL);
3166
3167         if (hint != NULL && hint->dah_parent != NULL &&
3168             !dt_object_remote(hint->dah_parent))
3169                 parent = hint->dah_parent;
3170
3171         inode = ldiskfs_create_inode(oth->ot_handle,
3172                                      parent ? osd_dt_obj(parent)->oo_inode :
3173                                               osd_sb(osd)->s_root->d_inode,
3174                                      mode, owner);
3175         if (!IS_ERR(inode)) {
3176                 /* Do not update file c/mtime in ldiskfs. */
3177                 inode->i_flags |= S_NOCMTIME;
3178
3179                 /*
3180                  * For new created object, it must be consistent,
3181                  * and it is unnecessary to scrub against it.
3182                  */
3183                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3184
3185                 obj->oo_inode = inode;
3186                 result = 0;
3187         } else {
3188                 if (obj->oo_hl_head != NULL) {
3189                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3190                         obj->oo_hl_head = NULL;
3191                 }
3192                 result = PTR_ERR(inode);
3193         }
3194         LINVRNT(osd_invariant(obj));
3195         return result;
3196 }
3197
3198 enum {
3199         OSD_NAME_LEN = 255
3200 };
3201
3202 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3203                      struct lu_attr *attr,
3204                      struct dt_allocation_hint *hint,
3205                      struct dt_object_format *dof,
3206                      struct thandle *th)
3207 {
3208         int result;
3209         struct osd_thandle *oth;
3210         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3211
3212         LASSERT(S_ISDIR(attr->la_mode));
3213
3214         oth = container_of(th, struct osd_thandle, ot_super);
3215         LASSERT(oth->ot_handle->h_transaction != NULL);
3216         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3217                 obj->oo_dirent_count = 0;
3218         result = osd_mkfile(info, obj, mode, hint, th, attr);
3219
3220         return result;
3221 }
3222
3223 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3224                         struct lu_attr *attr,
3225                         struct dt_allocation_hint *hint,
3226                         struct dt_object_format *dof,
3227                         struct thandle *th)
3228 {
3229         int result;
3230         struct osd_thandle *oth;
3231         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3232
3233         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3234
3235         LASSERT(S_ISREG(attr->la_mode));
3236
3237         oth = container_of(th, struct osd_thandle, ot_super);
3238         LASSERT(oth->ot_handle->h_transaction != NULL);
3239
3240         result = osd_mkfile(info, obj, mode, hint, th, attr);
3241         if (result == 0) {
3242                 LASSERT(obj->oo_inode != NULL);
3243                 if (feat->dif_flags & DT_IND_VARKEY)
3244                         result = iam_lvar_create(obj->oo_inode,
3245                                                  feat->dif_keysize_max,
3246                                                  feat->dif_ptrsize,
3247                                                  feat->dif_recsize_max,
3248                                                  oth->ot_handle);
3249                 else
3250                         result = iam_lfix_create(obj->oo_inode,
3251                                                  feat->dif_keysize_max,
3252                                                  feat->dif_ptrsize,
3253                                                  feat->dif_recsize_max,
3254                                                  oth->ot_handle);
3255         }
3256         return result;
3257 }
3258
3259 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3260                      struct lu_attr *attr,
3261                      struct dt_allocation_hint *hint,
3262                      struct dt_object_format *dof,
3263                      struct thandle *th)
3264 {
3265         LASSERT(S_ISREG(attr->la_mode));
3266         return osd_mkfile(info, obj, (attr->la_mode &
3267                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3268                           attr);
3269 }
3270
3271 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3272                      struct lu_attr *attr,
3273                      struct dt_allocation_hint *hint,
3274                      struct dt_object_format *dof,
3275                      struct thandle *th)
3276 {
3277         LASSERT(S_ISLNK(attr->la_mode));
3278         return osd_mkfile(info, obj, (attr->la_mode &
3279                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3280                           attr);
3281 }
3282
3283 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3284                      struct lu_attr *attr,
3285                      struct dt_allocation_hint *hint,
3286                      struct dt_object_format *dof,
3287                      struct thandle *th)
3288 {
3289         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3290         int result;
3291
3292         LINVRNT(osd_invariant(obj));
3293         LASSERT(obj->oo_inode == NULL);
3294         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3295                 S_ISFIFO(mode) || S_ISSOCK(mode));
3296
3297         result = osd_mkfile(info, obj, mode, hint, th, attr);
3298         if (result == 0) {
3299                 LASSERT(obj->oo_inode != NULL);
3300                 /*
3301                  * This inode should be marked dirty for i_rdev.  Currently
3302                  * that is done in the osd_attr_init().
3303                  */
3304                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3305                                    attr->la_rdev);
3306         }
3307         LINVRNT(osd_invariant(obj));
3308         return result;
3309 }
3310
3311 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3312                               struct lu_attr *,
3313                               struct dt_allocation_hint *hint,
3314                               struct dt_object_format *dof,
3315                               struct thandle *);
3316
3317 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3318 {
3319         osd_obj_type_f result;
3320
3321         switch (type) {
3322         case DFT_DIR:
3323                 result = osd_mkdir;
3324                 break;
3325         case DFT_REGULAR:
3326                 result = osd_mkreg;
3327                 break;
3328         case DFT_SYM:
3329                 result = osd_mksym;
3330                 break;
3331         case DFT_NODE:
3332                 result = osd_mknod;
3333                 break;
3334         case DFT_INDEX:
3335                 result = osd_mk_index;
3336                 break;
3337
3338         default:
3339                 LBUG();
3340                 break;
3341         }
3342         return result;
3343 }
3344
3345
3346 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3347                         struct dt_object *parent, struct dt_object *child,
3348                         umode_t child_mode)
3349 {
3350         LASSERT(ah);
3351
3352         ah->dah_parent = parent;
3353         ah->dah_mode = child_mode;
3354
3355         if (parent != NULL && !dt_object_remote(parent)) {
3356                 /* will help to find FID->ino at dt_insert("..") */
3357                 struct osd_object *pobj = osd_dt_obj(parent);
3358
3359                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3360         }
3361 }
3362
3363 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3364                           struct lu_attr *attr, struct dt_object_format *dof)
3365 {
3366         struct inode *inode = obj->oo_inode;
3367         __u64 valid = attr->la_valid;
3368         int result;
3369
3370         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3371
3372         if (dof->dof_type != DFT_NODE)
3373                 attr->la_valid &= ~LA_RDEV;
3374         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3375                 attr->la_valid &= ~LA_ATIME;
3376         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3377                 attr->la_valid &= ~LA_CTIME;
3378         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3379                 attr->la_valid &= ~LA_MTIME;
3380
3381         result = osd_quota_transfer(inode, attr);
3382         if (result)
3383                 return;
3384
3385         if (attr->la_valid != 0) {
3386                 result = osd_inode_setattr(info->oti_env, inode, attr);
3387                 /*
3388                  * The osd_inode_setattr() should always succeed here.  The
3389                  * only error that could be returned is EDQUOT when we are
3390                  * trying to change the UID or GID of the inode. However, this
3391                  * should not happen since quota enforcement is no longer
3392                  * enabled on ldiskfs (lquota takes care of it).
3393                  */
3394                 LASSERTF(result == 0, "%d\n", result);
3395                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3396         }
3397
3398         attr->la_valid = valid;
3399 }
3400
3401 /**
3402  * Helper function for osd_create()
3403  *
3404  * \retval 0, on success
3405  */
3406 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3407                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3408                         struct dt_object_format *dof, struct thandle *th)
3409 {
3410         int result;
3411         __u32 umask;
3412
3413         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3414
3415         /* we drop umask so that permissions we pass are not affected */
3416         umask = current->fs->umask;
3417         current->fs->umask = 0;
3418
3419         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3420                                                   th);
3421         if (likely(obj->oo_inode != NULL)) {
3422                 LASSERT(obj->oo_inode->i_state & I_NEW);
3423
3424                 /*
3425                  * Unlock the inode before attr initialization to avoid
3426                  * unnecessary dqget operations. LU-6378
3427                  */
3428                 unlock_new_inode(obj->oo_inode);
3429         }
3430
3431         if (likely(result == 0)) {
3432                 osd_attr_init(info, obj, attr, dof);
3433                 osd_object_init0(obj);
3434         }
3435
3436         /* restore previous umask value */
3437         current->fs->umask = umask;
3438
3439         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3440
3441         return result;
3442 }
3443
3444 /**
3445  * Helper function for osd_create()
3446  *
3447  * \retval 0, on success
3448  */
3449 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3450                            const struct lu_fid *fid, struct thandle *th)
3451 {
3452         struct osd_thread_info *info = osd_oti_get(env);
3453         struct osd_inode_id    *id   = &info->oti_id;
3454         struct osd_device      *osd  = osd_obj2dev(obj);
3455         struct osd_thandle     *oh;
3456         int rc;
3457
3458         LASSERT(obj->oo_inode != NULL);
3459
3460         oh = container_of(th, struct osd_thandle, ot_super);
3461         LASSERT(oh->ot_handle);
3462         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3463
3464         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3465         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3466                            OI_CHECK_FLD, NULL);
3467         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3468                 struct lu_fid next_fid = *fid;
3469
3470                 /* insert next object in advance, and map to the same inode */
3471                 next_fid.f_oid++;
3472                 if (next_fid.f_oid != 0) {
3473                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3474                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3475                                       OI_CHECK_FLD, NULL);
3476                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3477                 }
3478         }
3479
3480         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3481
3482         return rc;
3483 }
3484
3485 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3486                    u64 seq, struct lu_seq_range *range)
3487 {
3488         struct seq_server_site *ss = osd_seq_site(osd);
3489
3490         if (fid_seq_is_idif(seq)) {
3491                 fld_range_set_ost(range);
3492                 range->lsr_index = idif_ost_idx(seq);
3493                 return 0;
3494         }
3495
3496         if (!fid_seq_in_fldb(seq)) {
3497                 fld_range_set_mdt(range);
3498                 if (ss != NULL)
3499                         /*
3500                          * FIXME: If ss is NULL, it suppose not get lsr_index
3501                          * at all
3502                          */
3503                         range->lsr_index = ss->ss_node_id;
3504                 return 0;
3505         }
3506
3507         LASSERT(ss != NULL);
3508         fld_range_set_any(range);
3509         /* OSD will only do local fld lookup */
3510         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3511 }
3512
3513 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3514                               struct lu_attr *attr,
3515                               struct dt_allocation_hint *hint,
3516                               struct dt_object_format *dof,
3517                               struct thandle *handle)
3518 {
3519         struct osd_thandle *oh;
3520         int rc;
3521
3522         ENTRY;
3523
3524         LASSERT(handle != NULL);
3525
3526         oh = container_of(handle, struct osd_thandle, ot_super);
3527         LASSERT(oh->ot_handle == NULL);
3528
3529         /*
3530          * EA object consumes more credits than regular object: osd_mk_index
3531          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3532          * leaf_node, could involves the block, block bitmap, groups, GDT
3533          * change for each block, so add 4 * 2 credits in that case.
3534          */
3535         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3536                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3537                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3538         /*
3539          * Reuse idle OI block may cause additional one OI block
3540          * to be changed.
3541          */
3542         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3543                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3544         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3545                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3546                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3547
3548         /* will help to find FID->ino mapping at dt_insert() */
3549         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3550                                    osd_dt_obj(dt));
3551         if (rc != 0)
3552                 RETURN(rc);
3553
3554         if (!attr)
3555                 RETURN(0);
3556
3557         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3558                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3559                                    NULL, OSD_QID_INODE);
3560         if (rc != 0)
3561                 RETURN(rc);
3562
3563         RETURN(rc);
3564 }
3565
3566 /**
3567  * Called to destroy on-disk representation of the object
3568  *
3569  * Concurrency: must be locked
3570  */
3571 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3572                                struct thandle *th)
3573 {
3574         struct osd_object *obj = osd_dt_obj(dt);
3575         struct inode *inode = obj->oo_inode;
3576         struct osd_thandle *oh;
3577         int rc;
3578
3579         ENTRY;
3580
3581         if (inode == NULL)
3582                 RETURN(-ENOENT);
3583
3584         oh = container_of(th, struct osd_thandle, ot_super);
3585         LASSERT(oh->ot_handle == NULL);
3586
3587         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3588                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3589
3590         /* For removing agent entry */
3591         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3592                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3593
3594         /*
3595          * Recycle idle OI leaf may cause additional three OI blocks
3596          * to be changed.
3597          */
3598         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3599                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3600                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3601         /* one less inode */
3602         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3603                                    i_projid_read(inode), -1, oh, obj, NULL,
3604                                    OSD_QID_INODE);
3605         if (rc)
3606                 RETURN(rc);
3607         /* data to be truncated */
3608         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3609                                    i_projid_read(inode), 0, oh, obj, NULL,
3610                                    OSD_QID_BLK);
3611         if (rc)
3612                 RETURN(rc);
3613
3614         /*
3615          * will help to find FID->ino when this object is being
3616          * added to PENDING
3617          */
3618         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3619
3620         RETURN(rc);
3621 }
3622
3623 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3624                        struct thandle *th)
3625 {
3626         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3627         struct osd_object *obj = osd_dt_obj(dt);
3628         struct inode *inode = obj->oo_inode;
3629         struct osd_device *osd = osd_obj2dev(obj);
3630         struct osd_thandle *oh;
3631         int result;
3632
3633         ENTRY;
3634
3635         oh = container_of(th, struct osd_thandle, ot_super);
3636         LASSERT(oh->ot_handle);
3637         LASSERT(inode);
3638         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3639
3640         if (unlikely(fid_is_acct(fid)))
3641                 RETURN(-EPERM);
3642
3643         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3644                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3645                 if (result != 0)
3646                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3647                                osd_name(osd), PFID(fid), result);
3648         }
3649
3650         if (S_ISDIR(inode->i_mode)) {
3651                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3652                         inode->i_nlink == 2);
3653
3654                 spin_lock(&obj->oo_guard);
3655                 clear_nlink(inode);
3656                 spin_unlock(&obj->oo_guard);
3657                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3658         }
3659
3660         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3661
3662         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3663
3664         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3665                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3666                                        oh->ot_handle, OI_CHECK_FLD);
3667
3668         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3669         /* XXX: add to ext3 orphan list */
3670         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3671
3672         /* not needed in the cache anymore */
3673         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3674         obj->oo_destroyed = 1;
3675
3676         RETURN(0);
3677 }
3678
3679 /**
3680  * Put the fid into lustre_mdt_attrs, and then place the structure
3681  * inode's ea. This fid should not be altered during the life time
3682  * of the inode.
3683  *
3684  * \retval +ve, on success
3685  * \retval -ve, on error
3686  *
3687  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3688  */
3689 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3690                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3691 {
3692         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3693         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3694         int rc;
3695
3696         ENTRY;
3697
3698         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3699                 RETURN(0);
3700
3701         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3702                 rc = -ENOMEM;
3703
3704         lustre_loa_init(loa, fid, compat, incompat);
3705         lustre_loa_swab(loa, false);
3706
3707         /*
3708          * For the OST device with 256 bytes inode size by default,
3709          * the PFID EA will be stored together with LMA EA to avoid
3710          * performance trouble. Otherwise the PFID EA can be stored
3711          * independently. LU-8998
3712          */
3713         if ((compat & LMAC_FID_ON_OST) &&
3714             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3715                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3716                                      sizeof(*loa), XATTR_CREATE);
3717         else
3718                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3719                                      sizeof(*lma), XATTR_CREATE);
3720         /*
3721          * LMA may already exist, but we need to check that all the
3722          * desired compat/incompat flags have been added.
3723          */
3724         if (unlikely(rc == -EEXIST)) {
3725                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3726                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3727                 if (rc < 0)
3728                         RETURN(rc);
3729
3730                 if (rc < sizeof(*lma))
3731                         RETURN(-EINVAL);
3732
3733                 lustre_loa_swab(loa, true);
3734                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3735                     ((compat == 0 && incompat == 0) ||
3736                      (!(~lma->lma_compat & compat) &&
3737                       !(~lma->lma_incompat & incompat))))
3738                         RETURN(0);
3739
3740                 lma->lma_self_fid = *fid;
3741                 lma->lma_compat |= compat;
3742                 lma->lma_incompat |= incompat;
3743                 if (rc == sizeof(*lma)) {
3744                         lustre_lma_swab(lma);
3745                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3746                                              sizeof(*lma), XATTR_REPLACE);
3747                 } else {
3748                         lustre_loa_swab(loa, false);
3749                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3750                                              sizeof(*loa), XATTR_REPLACE);
3751                 }
3752         }
3753
3754         RETURN(rc);
3755 }
3756
3757 /**
3758  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3759  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3760  * To have compatilibility with 1.8 ldiskfs driver we need to have
3761  * magic number at start of fid data.
3762  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3763  * its inmemory API.
3764  */
3765 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3766                                          const struct lu_fid *fid)
3767 {
3768         if (!fid_is_namespace_visible(fid) ||
3769             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3770                 param->edp_magic = 0;
3771                 return;
3772         }
3773
3774         param->edp_magic = LDISKFS_LUFID_MAGIC;
3775         param->edp_len =  sizeof(struct lu_fid) + 1;
3776         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3777 }
3778
3779 /**
3780  * Try to read the fid from inode ea into dt_rec.
3781  *
3782  * \param fid object fid.
3783  *
3784  * \retval 0 on success
3785  */
3786 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3787                           __u32 ino, struct lu_fid *fid,
3788                           struct osd_inode_id *id)
3789 {
3790         struct osd_thread_info *info  = osd_oti_get(env);
3791         struct inode *inode;
3792
3793         ENTRY;
3794
3795         osd_id_gen(id, ino, OSD_OII_NOGEN);
3796         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3797         if (IS_ERR(inode))
3798                 RETURN(PTR_ERR(inode));
3799
3800         iput(inode);
3801         RETURN(0);
3802 }
3803
3804 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3805                                         struct inode *dir,
3806                                         struct inode *parent_dir,
3807                                         const struct lu_fid *dot_fid,
3808                                         const struct lu_fid *dot_dot_fid,
3809                                         struct osd_thandle *oth)
3810 {
3811         struct ldiskfs_dentry_param *dot_ldp;
3812         struct ldiskfs_dentry_param *dot_dot_ldp;
3813         __u32 saved_nlink = dir->i_nlink;
3814         int rc;
3815
3816         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3817         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3818
3819         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3820         dot_ldp->edp_magic = 0;
3821
3822         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3823                                     dir, dot_ldp, dot_dot_ldp);
3824         /*
3825          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3826          * the subseqent ref_add() will increase the dir->i_nlink
3827          * as 3. That is incorrect for new created directory.
3828          *
3829          * It looks like hack, because we want to make the OSD API
3830          * to be order-independent for new created directory object
3831          * between dt_insert(..) and ref_add() operations.
3832          *
3833          * Here, we only restore the in-RAM dir-inode's nlink attr,
3834          * becuase if the nlink attr is not 2, then there will be
3835          * ref_add() called following the dt_insert(..), such call
3836          * will make both the in-RAM and on-disk dir-inode's nlink
3837          * attr to be set as 2. LU-7447
3838          */
3839         set_nlink(dir, saved_nlink);
3840         return rc;
3841 }
3842
3843 /**
3844  * Create an local agent inode for remote entry
3845  */
3846 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3847                                                   struct osd_device *osd,
3848                                                   struct osd_object *pobj,
3849                                                   const struct lu_fid *fid,
3850                                                   __u32 type,
3851                                                   struct thandle *th)
3852 {
3853         struct osd_thread_info *info = osd_oti_get(env);
3854         struct inode *local;
3855         struct osd_thandle *oh;
3856         int rc;
3857
3858         ENTRY;
3859
3860         LASSERT(th);
3861         oh = container_of(th, struct osd_thandle, ot_super);
3862         LASSERT(oh->ot_handle->h_transaction != NULL);
3863
3864         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3865                                      NULL);
3866         if (IS_ERR(local)) {
3867                 CERROR("%s: create local error %d\n", osd_name(osd),
3868                        (int)PTR_ERR(local));
3869                 RETURN(local);
3870         }
3871
3872         /*
3873          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3874          * correct gid on remote file, not agent here
3875          */
3876         local->i_gid = current_fsgid();
3877         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3878
3879         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3880          * That gives e2fsck something to look at and be happy, and allows
3881          * debugging if we need to determine where this symlink came from.
3882          */
3883         if (S_ISLNK(type)) {
3884                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3885                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3886                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3887
3888                 i_size_write(local, rc);
3889                 LDISKFS_I(local)->i_disksize = rc;
3890         }
3891         unlock_new_inode(local);
3892
3893         /* Agent inode should not have project ID */
3894 #ifdef  HAVE_PROJECT_QUOTA
3895         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3896             i_projid_read(pobj->oo_inode) != 0) {
3897                 rc = osd_transfer_project(local, 0);
3898                 if (rc) {
3899                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3900                                "quota enforcement enabled on the ldiskfs "
3901                                "filesystem?\n", local->i_sb->s_id, rc);
3902                         RETURN(ERR_PTR(rc));
3903                 }
3904         }
3905 #endif
3906         /* Set special LMA flag for local agent inode */
3907         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3908         if (rc != 0) {
3909                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3910                        osd_name(osd), PFID(fid), rc);
3911                 RETURN(ERR_PTR(rc));
3912         }
3913
3914         if (!S_ISDIR(type))
3915                 RETURN(local);
3916
3917         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3918                                          lu_object_fid(&pobj->oo_dt.do_lu),
3919                                          fid, oh);
3920         if (rc != 0) {
3921                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3922                         osd_name(osd), PFID(fid), rc);
3923                 RETURN(ERR_PTR(rc));
3924         }
3925
3926         RETURN(local);
3927 }
3928
3929 /**
3930  * when direntry is deleted, we have to take care of possible agent inode
3931  * referenced by that. unfortunately we can't do this at that point:
3932  * iget() within a running transaction leads to deadlock and we better do
3933  * not call that every delete declaration to save performance. so we put
3934  * a potention agent inode on a list and process that once the transaction
3935  * is over. Notice it's not any worse in terms of real orphans as regular
3936  * object destroy doesn't put inodes on the on-disk orphan list. this should
3937  * be addressed separately
3938  */
3939 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3940                                             struct osd_thandle *oh,
3941                                             __u32 ino)
3942 {
3943         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3944         struct osd_obj_orphan *oor;
3945
3946         OBD_ALLOC_PTR(oor);
3947         if (oor == NULL)
3948                 return -ENOMEM;
3949
3950         oor->oor_ino = ino;
3951         oor->oor_env = (struct lu_env *)env;
3952         spin_lock(&osd->od_osfs_lock);
3953         list_add(&oor->oor_list, &osd->od_orphan_list);
3954         spin_unlock(&osd->od_osfs_lock);
3955
3956         oh->ot_remove_agents = 1;
3957
3958         return 0;
3959
3960 }
3961
3962 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3963                                                 struct osd_device *osd)
3964 {
3965         struct osd_thread_info *info = osd_oti_get(env);
3966         struct osd_obj_orphan *oor, *tmp;
3967         struct osd_inode_id id;
3968         LIST_HEAD(list);
3969         struct inode *inode;
3970         struct lu_fid fid;
3971         handle_t *jh;
3972         __u32 ino;
3973
3974         spin_lock(&osd->od_osfs_lock);
3975         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3976                 if (oor->oor_env == env)
3977                         list_move(&oor->oor_list, &list);
3978         }
3979         spin_unlock(&osd->od_osfs_lock);
3980
3981         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3982
3983                 ino = oor->oor_ino;
3984
3985                 list_del(&oor->oor_list);
3986                 OBD_FREE_PTR(oor);
3987
3988                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3989                 inode = osd_iget_fid(info, osd, &id, &fid);
3990                 if (IS_ERR(inode))
3991                         continue;
3992
3993                 if (!osd_remote_fid(env, osd, &fid)) {
3994                         iput(inode);
3995                         continue;
3996                 }
3997
3998                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3999                 clear_nlink(inode);
4000                 mark_inode_dirty(inode);
4001                 ldiskfs_journal_stop(jh);
4002                 iput(inode);
4003         }
4004
4005         return 0;
4006 }
4007
4008 /**
4009  * OSD layer object create function for OST objects (b=11826).
4010  *
4011  * The FID is inserted into inode xattr here.
4012  *
4013  * \retval   0, on success
4014  * \retval -ve, on error
4015  */
4016 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4017                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4018                       struct dt_object_format *dof, struct thandle *th)
4019 {
4020         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4021         struct osd_object *obj = osd_dt_obj(dt);
4022         struct osd_thread_info *info = osd_oti_get(env);
4023         int result, on_ost = 0;
4024
4025         ENTRY;
4026
4027         if (dt_object_exists(dt))
4028                 RETURN(-EEXIST);
4029
4030         LINVRNT(osd_invariant(obj));
4031         LASSERT(!dt_object_remote(dt));
4032         LASSERT(osd_is_write_locked(env, obj));
4033         LASSERT(th != NULL);
4034
4035         if (unlikely(fid_is_acct(fid)))
4036                 /*
4037                  * Quota files can't be created from the kernel any more,
4038                  * 'tune2fs -O quota' will take care of creating them
4039                  */
4040                 RETURN(-EPERM);
4041
4042         result = __osd_create(info, obj, attr, hint, dof, th);
4043         if (result == 0) {
4044                 if (fid_is_idif(fid) &&
4045                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4046                         struct lu_fid *tfid = &info->oti_fid;
4047                         struct ost_id *oi   = &info->oti_ostid;
4048
4049                         fid_to_ostid(fid, oi);
4050                         ostid_to_fid(tfid, oi, 0);
4051                         on_ost = 1;
4052                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4053                                                 LMAC_FID_ON_OST, 0);
4054                 } else {
4055                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4056                                                fid, OI_CHECK_FLD);
4057                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4058                                                 on_ost ? LMAC_FID_ON_OST : 0,
4059                                                 0);
4060                 }
4061                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4062                         obj->oo_dt.do_body_ops = &osd_body_ops;
4063         }
4064
4065         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4066                 result = __osd_oi_insert(env, obj, fid, th);
4067
4068         /*
4069          * a small optimization - dt_insert() isn't usually applied
4070          * to OST objects, so we don't need to cache OI mapping for
4071          * OST objects
4072          */
4073         if (result == 0 && on_ost == 0) {
4074                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4075
4076                 result = osd_idc_find_and_init(env, osd, obj);
4077                 LASSERT(result == 0);
4078         }
4079
4080         LASSERT(ergo(result == 0,
4081                      dt_object_exists(dt) && !dt_object_remote(dt)));
4082         LINVRNT(osd_invariant(obj));
4083         RETURN(result);
4084 }
4085
4086 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4087                                struct thandle *handle)
4088 {
4089         struct osd_thandle *oh;
4090
4091         /* it's possible that object doesn't exist yet */
4092         LASSERT(handle != NULL);
4093
4094         oh = container_of(handle, struct osd_thandle, ot_super);
4095         LASSERT(oh->ot_handle == NULL);
4096
4097         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4098                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4099
4100         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4101
4102         return 0;
4103 }
4104
4105 /*
4106  * Concurrency: @dt is write locked.
4107  */
4108 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4109                        struct thandle *th)
4110 {
4111         struct osd_object *obj = osd_dt_obj(dt);
4112         struct inode *inode = obj->oo_inode;
4113         struct osd_thandle *oh;
4114         int rc = 0;
4115
4116         if (!dt_object_exists(dt) || obj->oo_destroyed)
4117                 return -ENOENT;
4118
4119         LINVRNT(osd_invariant(obj));
4120         LASSERT(!dt_object_remote(dt));
4121         LASSERT(osd_is_write_locked(env, obj));
4122         LASSERT(th != NULL);
4123
4124         oh = container_of(th, struct osd_thandle, ot_super);
4125         LASSERT(oh->ot_handle != NULL);
4126
4127         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4128
4129         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4130                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4131         /*
4132          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4133          * (65000) subdirectories by storing "1" in i_nlink if the link count
4134          * would otherwise overflow. Directory tranversal tools understand
4135          * that (st_nlink == 1) indicates that the filesystem dose not track
4136          * hard links count on the directory, and will not abort subdirectory
4137          * scanning early once (st_nlink - 2) subdirs have been found.
4138          *
4139          * This also has to properly handle the case of inodes with nlink == 0
4140          * in case they are being linked into the PENDING directory
4141          */
4142         spin_lock(&obj->oo_guard);
4143         if (unlikely(inode->i_nlink == 0))
4144                 /* inc_nlink from 0 may cause WARN_ON */
4145                 set_nlink(inode, 1);
4146         else {
4147                 ldiskfs_inc_count(oh->ot_handle, inode);
4148                 if (!S_ISDIR(inode->i_mode))
4149                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4150         }
4151         spin_unlock(&obj->oo_guard);
4152
4153         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4154         LINVRNT(osd_invariant(obj));
4155
4156         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4157
4158         return rc;
4159 }
4160
4161 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4162                                struct thandle *handle)
4163 {
4164         struct osd_thandle *oh;
4165
4166         if (!dt_object_exists(dt))
4167                 return -ENOENT;
4168
4169         LASSERT(!dt_object_remote(dt));
4170         LASSERT(handle != NULL);
4171
4172         oh = container_of(handle, struct osd_thandle, ot_super);
4173         LASSERT(oh->ot_handle == NULL);
4174
4175         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4176                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4177
4178         return 0;
4179 }
4180
4181 /*
4182  * Concurrency: @dt is write locked.
4183  */
4184 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4185                        struct thandle *th)
4186 {
4187         struct osd_object *obj = osd_dt_obj(dt);
4188         struct inode *inode = obj->oo_inode;
4189         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4190         struct osd_thandle *oh;
4191
4192         if (!dt_object_exists(dt))
4193                 return -ENOENT;
4194
4195         LINVRNT(osd_invariant(obj));
4196         LASSERT(!dt_object_remote(dt));
4197         LASSERT(osd_is_write_locked(env, obj));
4198         LASSERT(th != NULL);
4199
4200         oh = container_of(th, struct osd_thandle, ot_super);
4201         LASSERT(oh->ot_handle != NULL);
4202
4203         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4204
4205         spin_lock(&obj->oo_guard);
4206         /*
4207          * That can be result of upgrade from old Lustre version and
4208          * applied only to local files.  Just skip this ref_del call.
4209          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4210          */
4211         if (inode->i_nlink == 0) {
4212                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4213                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4214                              ", maybe an upgraded file? (LU-3915)\n",
4215                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4216                 spin_unlock(&obj->oo_guard);
4217                 return 0;
4218         }
4219
4220         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4221                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4222
4223         ldiskfs_dec_count(oh->ot_handle, inode);
4224         spin_unlock(&obj->oo_guard);
4225
4226         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4227         LINVRNT(osd_invariant(obj));
4228
4229         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4230
4231         return 0;
4232 }
4233
4234 /*
4235  * Concurrency: @dt is read locked.
4236  */
4237 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4238                          struct lu_buf *buf, const char *name)
4239 {
4240         struct osd_object      *obj    = osd_dt_obj(dt);
4241         struct inode           *inode  = obj->oo_inode;
4242         struct osd_thread_info *info   = osd_oti_get(env);
4243         struct dentry          *dentry = &info->oti_obj_dentry;
4244         bool cache_xattr = false;
4245         int rc;
4246
4247         LASSERT(buf);
4248
4249         /* version get is not real XATTR but uses xattr API */
4250         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4251                 dt_obj_version_t *ver = buf->lb_buf;
4252
4253                 /*
4254                  * for version we are just using xattr API but change inode
4255                  * field instead
4256                  */
4257                 if (buf->lb_len == 0)
4258                         return sizeof(dt_obj_version_t);
4259
4260                 if (buf->lb_len < sizeof(dt_obj_version_t))
4261                         return -ERANGE;
4262
4263                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4264                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4265
4266                 *ver = LDISKFS_I(inode)->i_fs_version;
4267
4268                 return sizeof(dt_obj_version_t);
4269         }
4270
4271         if (!dt_object_exists(dt))
4272                 return -ENOENT;
4273
4274         LASSERT(!dt_object_remote(dt));
4275         LASSERT(inode->i_op != NULL);
4276 #ifdef HAVE_IOP_XATTR
4277         LASSERT(inode->i_op->getxattr != NULL);
4278 #endif
4279
4280         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4281             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4282                 cache_xattr = true;
4283
4284         if (cache_xattr) {
4285                 rc = osd_oxc_get(obj, name, buf);
4286                 if (rc != -ENOENT)
4287                         return rc;
4288         }
4289
4290         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4291                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4292                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4293                 struct filter_fid *ff;
4294                 struct ost_layout *ol;
4295
4296                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4297                 if (rc)
4298                         return rc;
4299
4300                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4301
4302                 rc = sizeof(*ff);
4303                 if (buf->lb_len == 0 || !buf->lb_buf)
4304                         return rc;
4305
4306                 if (buf->lb_len < rc)
4307                         return -ERANGE;
4308
4309                 ff = buf->lb_buf;
4310                 ol = &ff->ff_layout;
4311                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4312                                                   PFID_STRIPE_IDX_BITS);
4313                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4314                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4315                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4316                 if (lma->lma_compat & LMAC_COMP_INFO) {
4317                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4318                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4319                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4320                 } else {
4321                         ol->ol_comp_start = 0;
4322                         ol->ol_comp_end = 0;
4323                         ol->ol_comp_id = 0;
4324                 }
4325         } else {
4326                 rc = __osd_xattr_get(inode, dentry, name,
4327                                      buf->lb_buf, buf->lb_len);
4328         }
4329
4330         if (cache_xattr) {
4331                 if (rc == -ENOENT || rc == -ENODATA)
4332                         osd_oxc_add(obj, name, NULL, 0);
4333                 else if (rc > 0 && buf->lb_buf != NULL)
4334                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4335         }
4336
4337         return rc;
4338 }
4339
4340 static int osd_declare_xattr_set(const struct lu_env *env,
4341                                  struct dt_object *dt,
4342                                  const struct lu_buf *buf, const char *name,
4343                                  int fl, struct thandle *handle)
4344 {
4345         struct osd_thandle *oh;
4346         int credits = 0;
4347         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4348
4349         LASSERT(handle != NULL);
4350
4351         oh = container_of(handle, struct osd_thandle, ot_super);
4352         LASSERT(oh->ot_handle == NULL);
4353
4354         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4355                 /*
4356                  * For non-upgrading case, the LMA is set first and
4357                  * usually fit inode. But for upgrade case, the LMA
4358                  * may be in another separated EA block.
4359                  */
4360                 if (dt_object_exists(dt)) {
4361                         if (fl == LU_XATTR_REPLACE)
4362                                 credits = 1;
4363                         else
4364                                 goto upgrade;
4365                 }
4366         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4367                 credits = 1;
4368         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4369                 /* We may need to delete the old PFID EA. */
4370                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4371                 if (fl == LU_XATTR_REPLACE)
4372                         credits += 1;
4373                 else
4374                         goto upgrade;
4375         } else {
4376                 /*
4377                  * If some name entry resides on remote MDT, then will create
4378                  * agent entry under remote parent. On the other hand, if the
4379                  * remote entry will be removed, then related agent entry may
4380                  * need to be removed from the remote parent. So there may be
4381                  * kinds of cases, let's declare enough credits. The credits
4382                  * for create agent entry is enough for remove case.
4383                  */
4384                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4385                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4386                         if (dt_object_exists(dt))
4387                                 credits += 1; /* For updating LMA */
4388                 }
4389
4390 upgrade:
4391                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4392
4393                 if (buf != NULL) {
4394                         ssize_t buflen;
4395
4396                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4397                                 /*
4398                                  * learn xattr size from osd_xattr_get if
4399                                  * attribute has not been read yet
4400                                  */
4401                                 buflen = __osd_xattr_get(
4402                                     osd_dt_obj(dt)->oo_inode,
4403                                     &osd_oti_get(env)->oti_obj_dentry,
4404                                     name, NULL, 0);
4405                                 if (buflen < 0)
4406                                         buflen = 0;
4407                         } else {
4408                                 buflen = buf->lb_len;
4409                         }
4410
4411                         if (buflen > sb->s_blocksize) {
4412                                 credits += osd_calc_bkmap_credits(
4413                                     sb, NULL, 0, -1,
4414                                     (buflen + sb->s_blocksize - 1) >>
4415                                     sb->s_blocksize_bits);
4416                         }
4417                 }
4418                 /*
4419                  * xattr set may involve inode quota change, reserve credits for
4420                  * dquot_initialize()
4421                  */
4422                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4423         }
4424
4425         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4426
4427         return 0;
4428 }
4429
4430 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4431                               const struct lu_buf *buf, int fl,
4432                               struct thandle *handle)
4433 {
4434         struct osd_thread_info *info = osd_oti_get(env);
4435         struct dentry *dentry = &info->oti_obj_dentry;
4436         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4437         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4438         struct inode *inode = obj->oo_inode;
4439         struct filter_fid *ff = buf->lb_buf;
4440         struct ost_layout *ol = &ff->ff_layout;
4441         int flags = XATTR_REPLACE;
4442         int rc;
4443
4444         ENTRY;
4445
4446         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4447                 RETURN(-EINVAL);
4448
4449         rc = osd_get_lma(info, inode, dentry, loa);
4450         if (rc == -ENODATA) {
4451                 /* Usually for upgarding from old device */
4452                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4453                                 LMAC_FID_ON_OST, 0);
4454                 flags = XATTR_CREATE;
4455         } else if (rc) {
4456                 RETURN(rc);
4457         }
4458
4459         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4460                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4461                         RETURN(-EEXIST);
4462
4463                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4464                         /* Separate PFID EA from LMA */
4465                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4466                         lustre_lma_swab(lma);
4467                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4468                                              sizeof(*lma), XATTR_REPLACE);
4469                         if (!rc) {
4470                                 obj->oo_pfid_in_lma = 0;
4471                                 rc = LU_XATTR_CREATE;
4472                         }
4473
4474                         RETURN(rc);
4475                 }
4476         } else {
4477                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4478                         RETURN(fl);
4479
4480                 /*
4481                  * Old client does not send stripe information,
4482                  * then store the PFID EA on disk separatedly.
4483                  */
4484                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4485                              ol->ol_stripe_size == 0))
4486                         RETURN(fl);
4487
4488                 /* Remove old PFID EA entry firstly. */
4489                 dquot_initialize(inode);
4490                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4491                 if (rc == -ENODATA) {
4492                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4493                                 RETURN(rc);
4494                 } else if (rc) {
4495                         RETURN(rc);
4496                 }
4497         }
4498
4499         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4500         if (likely(ol->ol_stripe_size != 0)) {
4501                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4502                                              PFID_STRIPE_IDX_BITS;
4503                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4504                 lma->lma_compat |= LMAC_STRIPE_INFO;
4505                 if (ol->ol_comp_id != 0) {
4506                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4507                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4508                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4509                         lma->lma_compat |= LMAC_COMP_INFO;
4510                 }
4511         }
4512
4513         lustre_loa_swab(loa, false);
4514
4515         /* Store the PFID EA inside LMA. */
4516         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4517                              flags);
4518         if (!rc)
4519                 obj->oo_pfid_in_lma = 1;
4520
4521         RETURN(rc);
4522 }
4523
4524 /*
4525  * In DNE environment, the object (in spite of regular file or directory)
4526  * and its name entry may reside on different MDTs. Under such case, we will
4527  * create an agent entry on the MDT where the object resides. The agent entry
4528  * references the object locally, that makes the object to be visible to the
4529  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4530  * such as 'tar' can handle the object properly.
4531  *
4532  * We handle the agent entry during set linkEA that is the common interface
4533  * for both regular file and directroy, can handle kinds of cases, such as
4534  * create/link/unlink/rename, and so on.
4535  *
4536  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4537  *
4538  * XXX: There are two known issues:
4539  * 1. For one object, we will create at most one agent entry even if there
4540  *    may be more than one cross-MDTs hard links on the object. So the local
4541  *    e2fsck may claim that the object's nlink is larger than the name entries
4542  *    that reference such inode. And in further, the e2fsck will fix the nlink
4543  *    attribute to match the local references. Then it will cause the object's
4544  *    nlink attribute to be inconsistent with the global references. it is bad
4545  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4546  *    other hand, the global namespace LFSCK can repair the object's attribute
4547  *    according to the linkEA.
4548  * 2. There may be too many hard links on the object as to its linkEA overflow,
4549  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4550  *    case happened, then at this point, we do not know whether there are some
4551  *    cross-MDTs reference. But there are local references, it guarantees that
4552  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4553  */
4554 static int osd_xattr_handle_linkea(const struct lu_env *env,
4555                                    struct osd_device *osd,
4556                                    struct osd_object *obj,
4557                                    const struct lu_buf *buf,
4558                                    struct thandle *handle)
4559 {
4560         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4561         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4562         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4563         struct lu_name tmpname;
4564         struct osd_thandle *oh;
4565         int rc;
4566         bool remote = false;
4567
4568         ENTRY;
4569
4570         oh = container_of(handle, struct osd_thandle, ot_super);
4571         LASSERT(oh->ot_handle != NULL);
4572
4573         rc = linkea_init_with_rec(&ldata);
4574         if (!rc) {
4575                 linkea_first_entry(&ldata);
4576                 while (ldata.ld_lee != NULL && !remote) {
4577                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4578                                             &tmpname, tfid);
4579                         if (osd_remote_fid(env, osd, tfid) > 0)
4580                                 remote = true;
4581                         else
4582                                 linkea_next_entry(&ldata);
4583                 }
4584         } else if (rc == -ENODATA) {
4585                 rc = 0;
4586         } else {
4587                 RETURN(rc);
4588         }
4589
4590         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4591                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4592                 if (rc)
4593                         CERROR("%s: failed to remove agent entry for "DFID
4594                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4595         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4596                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4597                 if (rc)
4598                         CERROR("%s: failed to create agent entry for "DFID
4599                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4600         }
4601
4602         RETURN(rc);
4603 }
4604
4605 /*
4606  * Concurrency: @dt is write locked.
4607  */
4608 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4609                          const struct lu_buf *buf, const char *name, int fl,
4610                          struct thandle *handle)
4611 {
4612         struct osd_object *obj = osd_dt_obj(dt);
4613         struct osd_device *osd = osd_obj2dev(obj);
4614         struct inode *inode = obj->oo_inode;
4615         struct osd_thread_info *info = osd_oti_get(env);
4616         int fs_flags = 0;
4617         int len;
4618         int rc;
4619
4620         ENTRY;
4621
4622         LASSERT(handle);
4623         LASSERT(buf);
4624
4625         /* version set is not real XATTR */
4626         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4627                 dt_obj_version_t *version = buf->lb_buf;
4628
4629                 /*
4630                  * for version we are just using xattr API but change inode
4631                  * field instead
4632                  */
4633                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4634
4635                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4636                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4637
4638                 LDISKFS_I(inode)->i_fs_version = *version;
4639                 /*
4640                  * Version is set after all inode operations are finished,
4641                  * so we should mark it dirty here
4642                  */
4643                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4644
4645                 RETURN(0);
4646         }
4647
4648         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4649                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4650
4651         len = buf->lb_len;
4652         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4653
4654         /*
4655          * For the OST device with 256 bytes inode size by default,
4656          * the PFID EA will be stored together with LMA EA to avoid
4657          * performance trouble. Otherwise the PFID EA can be stored
4658          * independently. LU-8998
4659          */
4660         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4661             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4662                 LASSERT(buf->lb_buf);
4663
4664                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4665                 if (fl <= 0)
4666                         RETURN(fl);
4667         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4668                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4669                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4670
4671                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4672                 if (rc)
4673                         RETURN(rc);
4674
4675                 lma->lma_incompat |= LMAI_STRIPED;
4676                 lustre_lma_swab(lma);
4677                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4678                                      sizeof(*lma), XATTR_REPLACE);
4679                 if (rc != 0)
4680                         RETURN(rc);
4681         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4682                 LASSERT(!osd->od_is_ost);
4683
4684                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4685                 if (rc)
4686                         RETURN(rc);
4687         }
4688
4689         if (fl & LU_XATTR_REPLACE)
4690                 fs_flags |= XATTR_REPLACE;
4691
4692         if (fl & LU_XATTR_CREATE)
4693                 fs_flags |= XATTR_CREATE;
4694
4695         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4696         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4697
4698         if (rc == 0 &&
4699             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4700              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4701                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4702
4703         return rc;
4704 }
4705
4706 /*
4707  * Concurrency: @dt is read locked.
4708  */
4709 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4710                           const struct lu_buf *buf)
4711 {
4712         struct osd_object *obj = osd_dt_obj(dt);
4713         struct inode *inode = obj->oo_inode;
4714         struct osd_thread_info *info = osd_oti_get(env);
4715         struct dentry *dentry = &info->oti_obj_dentry;
4716
4717         if (!dt_object_exists(dt))
4718                 return -ENOENT;
4719
4720         LASSERT(!dt_object_remote(dt));
4721         LASSERT(inode->i_op != NULL);
4722         LASSERT(inode->i_op->listxattr != NULL);
4723
4724         dentry->d_inode = inode;
4725         dentry->d_sb = inode->i_sb;
4726         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4727 }
4728
4729 static int osd_declare_xattr_del(const struct lu_env *env,
4730                                  struct dt_object *dt, const char *name,
4731                                  struct thandle *handle)
4732 {
4733         struct osd_thandle *oh;
4734         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4735
4736         LASSERT(!dt_object_remote(dt));
4737         LASSERT(handle != NULL);
4738
4739         oh = container_of(handle, struct osd_thandle, ot_super);
4740         LASSERT(oh->ot_handle == NULL);
4741
4742         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4743                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4744         /*
4745          * xattr del may involve inode quota change, reserve credits for
4746          * dquot_initialize()
4747          */
4748         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4749
4750         return 0;
4751 }
4752
4753 /*
4754  * Concurrency: @dt is write locked.
4755  */
4756 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4757                          const char *name, struct thandle *handle)
4758 {
4759         struct osd_object *obj = osd_dt_obj(dt);
4760         struct inode *inode = obj->oo_inode;
4761         struct osd_thread_info *info = osd_oti_get(env);
4762         struct dentry *dentry = &info->oti_obj_dentry;
4763         int rc;
4764
4765         if (!dt_object_exists(dt))
4766                 return -ENOENT;
4767
4768         LASSERT(!dt_object_remote(dt));
4769         LASSERT(inode->i_op != NULL);
4770         LASSERT(handle != NULL);
4771 #ifdef HAVE_IOP_XATTR
4772         LASSERT(inode->i_op->removexattr != NULL);
4773 #endif
4774
4775         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4776
4777         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4778                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4779
4780                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4781                                  &info->oti_ost_attrs);
4782                 if (!rc) {
4783                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4784
4785                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4786                         lustre_lma_swab(lma);
4787                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4788                                              sizeof(*lma), XATTR_REPLACE);
4789                         if (!rc)
4790                                 obj->oo_pfid_in_lma = 0;
4791                 }
4792         } else {
4793                 dquot_initialize(inode);
4794                 dentry->d_inode = inode;
4795                 dentry->d_sb = inode->i_sb;
4796                 rc = ll_vfs_removexattr(dentry, inode, name);
4797         }
4798
4799         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4800
4801         if (rc == 0 &&
4802             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4803              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4804                 osd_oxc_del(obj, name);
4805
4806         return rc;
4807 }
4808
4809 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4810                            __u64 start, __u64 end)
4811 {
4812         struct osd_object *obj = osd_dt_obj(dt);
4813         struct inode *inode = obj->oo_inode;
4814         struct osd_thread_info *info = osd_oti_get(env);
4815         struct dentry *dentry = &info->oti_obj_dentry;
4816         struct file *file = &info->oti_file;
4817         int rc;
4818
4819         ENTRY;
4820
4821         dentry->d_inode = inode;
4822         dentry->d_sb = inode->i_sb;
4823         file->f_path.dentry = dentry;
4824         file->f_mapping = inode->i_mapping;
4825         file->f_op = inode->i_fop;
4826         file->f_inode = inode;
4827
4828         rc = vfs_fsync_range(file, start, end, 0);
4829
4830         RETURN(rc);
4831 }
4832
4833 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4834 {
4835         return 0;
4836 }
4837
4838 /*
4839  * Index operations.
4840  */
4841
4842 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4843                                const struct dt_index_features *feat)
4844 {
4845         struct iam_descr *descr;
4846
4847         if (osd_object_is_root(o))
4848                 return feat == &dt_directory_features;
4849
4850         LASSERT(o->oo_dir != NULL);
4851
4852         descr = o->oo_dir->od_container.ic_descr;
4853         if (feat == &dt_directory_features) {
4854                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4855                         return 1;
4856                 else
4857                         return 0;
4858         } else {
4859                 return feat->dif_keysize_min <= descr->id_key_size &&
4860                        descr->id_key_size <= feat->dif_keysize_max &&
4861                        feat->dif_recsize_min <= descr->id_rec_size &&
4862                        descr->id_rec_size <= feat->dif_recsize_max &&
4863                        !(feat->dif_flags & (DT_IND_VARKEY |
4864                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4865                        ergo(feat->dif_flags & DT_IND_UPDATE,
4866                             1 /* XXX check that object (and fs) is writable */);
4867         }
4868 }
4869
4870 static int osd_iam_container_init(const struct lu_env *env,
4871                                   struct osd_object *obj,
4872                                   struct osd_directory *dir)
4873 {
4874         struct iam_container *bag = &dir->od_container;
4875         int result;
4876
4877         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4878         if (result != 0)
4879                 return result;
4880
4881         result = iam_container_setup(bag);
4882         if (result == 0)
4883                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4884         else
4885                 iam_container_fini(bag);
4886
4887         return result;
4888 }
4889
4890
4891 /*
4892  * Concurrency: no external locking is necessary.
4893  */
4894 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4895                          const struct dt_index_features *feat)
4896 {
4897         int result;
4898         int skip_iam = 0;
4899         struct osd_object *obj = osd_dt_obj(dt);
4900
4901         LINVRNT(osd_invariant(obj));
4902
4903         if (osd_object_is_root(obj)) {
4904                 dt->do_index_ops = &osd_index_ea_ops;
4905                 result = 0;
4906         } else if (feat == &dt_directory_features) {
4907                 dt->do_index_ops = &osd_index_ea_ops;
4908                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4909                         result = 0;
4910                 else
4911                         result = -ENOTDIR;
4912                 skip_iam = 1;
4913         } else if (unlikely(feat == &dt_otable_features)) {
4914                 dt->do_index_ops = &osd_otable_ops;
4915                 return 0;
4916         } else if (unlikely(feat == &dt_acct_features)) {
4917                 dt->do_index_ops = &osd_acct_index_ops;
4918                 result = 0;
4919                 skip_iam = 1;
4920         } else if (!osd_has_index(obj)) {
4921                 struct osd_directory *dir;
4922                 struct osd_device *osd = osd_obj2dev(obj);
4923                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4924
4925                 OBD_ALLOC_PTR(dir);
4926                 if (dir) {
4927
4928                         spin_lock(&obj->oo_guard);
4929                         if (obj->oo_dir == NULL)
4930                                 obj->oo_dir = dir;
4931                         else
4932                                 /*
4933                                  * Concurrent thread allocated container data.
4934                                  */
4935                                 OBD_FREE_PTR(dir);
4936                         spin_unlock(&obj->oo_guard);
4937                         /*
4938                          * Now, that we have container data, serialize its
4939                          * initialization.
4940                          */
4941                         down_write(&obj->oo_ext_idx_sem);
4942                         /*
4943                          * recheck under lock.
4944                          */
4945
4946                         if (osd_has_index(obj)) {
4947                                 result = 0;
4948                                 goto unlock;
4949                         }
4950
4951                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4952                         if (result || feat == &dt_lfsck_namespace_features ||
4953                             feat == &dt_lfsck_layout_orphan_features ||
4954                             feat == &dt_lfsck_layout_dangling_features)
4955                                 goto unlock;
4956
4957                         result = osd_index_register(osd, fid,
4958                                                     feat->dif_keysize_max,
4959                                                     feat->dif_recsize_max);
4960                         if (result < 0)
4961                                 CWARN("%s: failed to register index "
4962                                       DFID": rc = %d\n",
4963                                       osd_name(osd), PFID(fid), result);
4964                         else if (result > 0)
4965                                 result = 0;
4966                         else
4967                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4968                                        " (%d/%d) registered\n",
4969                                        osd_name(osd), PFID(fid),
4970                                        (int)feat->dif_keysize_max,
4971                                        (int)feat->dif_recsize_max);
4972
4973 unlock:
4974                         up_write(&obj->oo_ext_idx_sem);
4975                 } else {
4976                         result = -ENOMEM;
4977                 }
4978         } else {
4979                 result = 0;
4980         }
4981
4982         if (result == 0 && skip_iam == 0) {
4983                 if (!osd_iam_index_probe(env, obj, feat))
4984                         result = -ENOTDIR;
4985         }
4986         LINVRNT(osd_invariant(obj));
4987
4988         return result;
4989 }
4990
4991 static int osd_otable_it_attr_get(const struct lu_env *env,
4992                                  struct dt_object *dt,
4993                                  struct lu_attr *attr)
4994 {
4995         attr->la_valid = 0;
4996         return 0;
4997 }
4998
4999 static const struct dt_object_operations osd_obj_ops = {
5000         .do_read_lock           = osd_read_lock,
5001         .do_write_lock          = osd_write_lock,
5002         .do_read_unlock         = osd_read_unlock,
5003         .do_write_unlock        = osd_write_unlock,
5004         .do_write_locked        = osd_write_locked,
5005         .do_attr_get            = osd_attr_get,
5006         .do_declare_attr_set    = osd_declare_attr_set,
5007         .do_attr_set            = osd_attr_set,
5008         .do_ah_init             = osd_ah_init,
5009         .do_declare_create      = osd_declare_create,
5010         .do_create              = osd_create,
5011         .do_declare_destroy     = osd_declare_destroy,
5012         .do_destroy             = osd_destroy,
5013         .do_index_try           = osd_index_try,
5014         .do_declare_ref_add     = osd_declare_ref_add,
5015         .do_ref_add             = osd_ref_add,
5016         .do_declare_ref_del     = osd_declare_ref_del,
5017         .do_ref_del             = osd_ref_del,
5018         .do_xattr_get           = osd_xattr_get,
5019         .do_declare_xattr_set   = osd_declare_xattr_set,
5020         .do_xattr_set           = osd_xattr_set,
5021         .do_declare_xattr_del   = osd_declare_xattr_del,
5022         .do_xattr_del           = osd_xattr_del,
5023         .do_xattr_list          = osd_xattr_list,
5024         .do_object_sync         = osd_object_sync,
5025         .do_invalidate          = osd_invalidate,
5026 };
5027
5028 static const struct dt_object_operations osd_obj_otable_it_ops = {
5029         .do_attr_get    = osd_otable_it_attr_get,
5030         .do_index_try   = osd_index_try,
5031 };
5032
5033 static int osd_index_declare_iam_delete(const struct lu_env *env,
5034                                         struct dt_object *dt,
5035                                         const struct dt_key *key,
5036                                         struct thandle *handle)
5037 {
5038         struct osd_thandle *oh;
5039
5040         oh = container_of(handle, struct osd_thandle, ot_super);
5041         LASSERT(oh->ot_handle == NULL);
5042
5043         /* Recycle  may cause additional three blocks to be changed. */
5044         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5045                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5046
5047         return 0;
5048 }
5049
5050 /**
5051  *      delete a (key, value) pair from index \a dt specified by \a key
5052  *
5053  *      \param  dt      osd index object
5054  *      \param  key     key for index
5055  *      \param  rec     record reference
5056  *      \param  handle  transaction handler
5057  *
5058  *      \retval  0  success
5059  *      \retval -ve   failure
5060  */
5061 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5062                                 const struct dt_key *key,
5063                                 struct thandle *handle)
5064 {
5065         struct osd_thread_info *oti = osd_oti_get(env);
5066         struct osd_object *obj = osd_dt_obj(dt);
5067         struct osd_thandle *oh;
5068         struct iam_path_descr *ipd;
5069         struct iam_container *bag = &obj->oo_dir->od_container;
5070         int rc;
5071
5072         ENTRY;
5073
5074         if (!dt_object_exists(dt))
5075                 RETURN(-ENOENT);
5076
5077         LINVRNT(osd_invariant(obj));
5078         LASSERT(!dt_object_remote(dt));
5079         LASSERT(bag->ic_object == obj->oo_inode);
5080         LASSERT(handle != NULL);
5081
5082         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5083
5084         ipd = osd_idx_ipd_get(env, bag);
5085         if (unlikely(ipd == NULL))
5086                 RETURN(-ENOMEM);
5087
5088         oh = container_of(handle, struct osd_thandle, ot_super);
5089         LASSERT(oh->ot_handle != NULL);
5090         LASSERT(oh->ot_handle->h_transaction != NULL);
5091
5092         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5093                 /* swab quota uid/gid provided by caller */
5094                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5095                 key = (const struct dt_key *)&oti->oti_quota_id;
5096         }
5097
5098         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5099         osd_ipd_put(env, bag, ipd);
5100         LINVRNT(osd_invariant(obj));
5101         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5102         RETURN(rc);
5103 }
5104
5105 static int osd_index_declare_ea_delete(const struct lu_env *env,
5106                                        struct dt_object *dt,
5107                                        const struct dt_key *key,
5108                                        struct thandle *handle)
5109 {
5110         struct osd_thandle *oh;
5111         struct inode *inode;
5112         int rc, credits;
5113
5114         ENTRY;
5115
5116         LASSERT(!dt_object_remote(dt));
5117         LASSERT(handle != NULL);
5118
5119         oh = container_of(handle, struct osd_thandle, ot_super);
5120         LASSERT(oh->ot_handle == NULL);
5121
5122         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5123         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5124
5125         inode = osd_dt_obj(dt)->oo_inode;
5126         if (inode == NULL)
5127                 RETURN(-ENOENT);
5128
5129         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5130                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5131                                    NULL, OSD_QID_BLK);
5132         RETURN(rc);
5133 }
5134
5135 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5136                                           struct dt_rec *fid)
5137 {
5138         struct osd_fid_pack *rec;
5139         int rc = -ENODATA;
5140
5141         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5142                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5143                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5144                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5145                         rc = -EINVAL;
5146         }
5147         return rc;
5148 }
5149
5150 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5151                           const struct lu_fid *fid)
5152 {
5153         struct seq_server_site *ss = osd_seq_site(osd);
5154
5155         ENTRY;
5156
5157         /* FID seqs not in FLDB, must be local seq */
5158         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5159                 RETURN(0);
5160
5161         /*
5162          * If FLD is not being initialized yet, it only happens during the
5163          * initialization, likely during mgs initialization, and we assume
5164          * this is local FID.
5165          */
5166         if (ss == NULL || ss->ss_server_fld == NULL)
5167                 RETURN(0);
5168
5169         /* Only check the local FLDB here */
5170         if (osd_seq_exists(env, osd, fid_seq(fid)))
5171                 RETURN(0);
5172
5173         RETURN(1);
5174 }
5175
5176 static void osd_take_care_of_agent(const struct lu_env *env,
5177                                    struct osd_device *osd,
5178                                    struct osd_thandle *oh,
5179                                    struct ldiskfs_dir_entry_2 *de)
5180 {
5181         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5182         struct osd_idmap_cache *idc;
5183         int rc, schedule = 0;
5184
5185         LASSERT(de != NULL);
5186
5187         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5188         if (likely(rc == 0)) {
5189                 idc = osd_idc_find_or_init(env, osd, fid);
5190                 if (IS_ERR(idc) || idc->oic_remote)
5191                         schedule = 1;
5192         } else if (rc == -ENODATA) {
5193                 /*
5194                  * can't get FID, postpone to the end of the
5195                  * transaction when iget() is safe
5196                  */
5197                 schedule = 1;
5198         } else {
5199                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5200         }
5201         if (schedule)
5202                 osd_schedule_agent_inode_removal(env, oh,
5203                                                  le32_to_cpu(de->inode));
5204 }
5205
5206 /**
5207  * Index delete function for interoperability mode (b11826).
5208  * It will remove the directory entry added by osd_index_ea_insert().
5209  * This entry is needed to maintain name->fid mapping.
5210  *
5211  * \param key,  key i.e. file entry to be deleted
5212  *
5213  * \retval   0, on success
5214  * \retval -ve, on error
5215  */
5216 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5217                                const struct dt_key *key, struct thandle *handle)
5218 {
5219         struct osd_object *obj = osd_dt_obj(dt);
5220         struct inode *dir = obj->oo_inode;
5221         struct dentry *dentry;
5222         struct osd_thandle *oh;
5223         struct ldiskfs_dir_entry_2 *de = NULL;
5224         struct buffer_head *bh;
5225         struct htree_lock *hlock = NULL;
5226         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5227         int rc;
5228
5229         ENTRY;
5230
5231         if (!dt_object_exists(dt))
5232                 RETURN(-ENOENT);
5233
5234         LINVRNT(osd_invariant(obj));
5235         LASSERT(!dt_object_remote(dt));
5236         LASSERT(handle != NULL);
5237
5238         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5239
5240         oh = container_of(handle, struct osd_thandle, ot_super);
5241         LASSERT(oh->ot_handle != NULL);
5242         LASSERT(oh->ot_handle->h_transaction != NULL);
5243
5244         dquot_initialize(dir);
5245         dentry = osd_child_dentry_get(env, obj,
5246                                       (char *)key, strlen((char *)key));
5247
5248         if (obj->oo_hl_head != NULL) {
5249                 hlock = osd_oti_get(env)->oti_hlock;
5250                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5251                                    dir, LDISKFS_HLOCK_DEL);
5252         } else {
5253                 down_write(&obj->oo_ext_idx_sem);
5254         }
5255
5256         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5257         if (!IS_ERR(bh)) {
5258                 /*
5259                  * If this is not the ".." entry, it might be a remote DNE
5260                  * entry and  we need to check if the FID is for a remote
5261                  * MDT.  If the FID is  not in the directory entry (e.g.
5262                  * upgraded 1.8 filesystem without dirdata enabled) then
5263                  * we need to get the FID from the LMA. For a remote directory
5264                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5265                  * case.
5266                  *
5267                  * Delete the entry before the agent inode in order to
5268                  * simplify error handling.  At worst an error after deleting
5269                  * the entry first might leak the agent inode afterward. The
5270                  * reverse would need filesystem abort in case of error deleting
5271                  * the entry after the agent had been removed, or leave a
5272                  * dangling entry pointing at a random inode.
5273                  */
5274                 if (strcmp((char *)key, dotdot) != 0)
5275                         osd_take_care_of_agent(env, osd, oh, de);
5276                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5277                 brelse(bh);
5278         } else {
5279                 rc = PTR_ERR(bh);
5280         }
5281
5282         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5283             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5284                 /* NB, dirent count may not be accurate, because it's counted
5285                  * without lock.
5286                  */
5287                 if (obj->oo_dirent_count)
5288                         obj->oo_dirent_count--;
5289                 else
5290                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5291         }
5292         if (hlock != NULL)
5293                 ldiskfs_htree_unlock(hlock);
5294         else
5295                 up_write(&obj->oo_ext_idx_sem);
5296         GOTO(out, rc);
5297 out:
5298         LASSERT(osd_invariant(obj));
5299         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5300         RETURN(rc);
5301 }
5302
5303 /**
5304  *      Lookup index for \a key and copy record to \a rec.
5305  *
5306  *      \param  dt      osd index object
5307  *      \param  key     key for index
5308  *      \param  rec     record reference
5309  *
5310  *      \retval  +ve  success : exact mach
5311  *      \retval  0    return record with key not greater than \a key
5312  *      \retval -ve   failure
5313  */
5314 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5315                                 struct dt_rec *rec, const struct dt_key *key)
5316 {
5317         struct osd_object *obj = osd_dt_obj(dt);
5318         struct iam_path_descr *ipd;
5319         struct iam_container *bag = &obj->oo_dir->od_container;
5320         struct osd_thread_info *oti = osd_oti_get(env);
5321         struct iam_iterator *it = &oti->oti_idx_it;
5322         struct iam_rec *iam_rec;
5323         int rc;
5324
5325         ENTRY;
5326
5327         if (!dt_object_exists(dt))
5328                 RETURN(-ENOENT);
5329
5330         LASSERT(osd_invariant(obj));
5331         LASSERT(!dt_object_remote(dt));
5332         LASSERT(bag->ic_object == obj->oo_inode);
5333
5334         ipd = osd_idx_ipd_get(env, bag);
5335         if (IS_ERR(ipd))
5336                 RETURN(-ENOMEM);
5337
5338         /* got ipd now we can start iterator. */
5339         iam_it_init(it, bag, 0, ipd);
5340
5341         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5342                 /* swab quota uid/gid provided by caller */
5343                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5344                 key = (const struct dt_key *)&oti->oti_quota_id;
5345         }
5346
5347         rc = iam_it_get(it, (struct iam_key *)key);
5348         if (rc >= 0) {
5349                 if (S_ISDIR(obj->oo_inode->i_mode))
5350                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5351                 else
5352                         iam_rec = (struct iam_rec *)rec;
5353
5354                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5355
5356                 if (S_ISDIR(obj->oo_inode->i_mode))
5357                         osd_fid_unpack((struct lu_fid *)rec,
5358                                        (struct osd_fid_pack *)iam_rec);
5359                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5360                         osd_quota_unpack(obj, rec);
5361         }
5362
5363         iam_it_put(it);
5364         iam_it_fini(it);
5365         osd_ipd_put(env, bag, ipd);
5366
5367         LINVRNT(osd_invariant(obj));
5368
5369         RETURN(rc);
5370 }
5371
5372 static int osd_index_declare_iam_insert(const struct lu_env *env,
5373                                         struct dt_object *dt,
5374                                         const struct dt_rec *rec,
5375                                         const struct dt_key *key,
5376                                         struct thandle *handle)
5377 {
5378         struct osd_thandle *oh;
5379
5380         LASSERT(handle != NULL);
5381
5382         oh = container_of(handle, struct osd_thandle, ot_super);
5383         LASSERT(oh->ot_handle == NULL);
5384
5385         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5386                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5387
5388         return 0;
5389 }
5390
5391 /**
5392  *      Inserts (key, value) pair in \a dt index object.
5393  *
5394  *      \param  dt      osd index object
5395  *      \param  key     key for index
5396  *      \param  rec     record reference
5397  *      \param  th      transaction handler
5398  *
5399  *      \retval  0  success
5400  *      \retval -ve failure
5401  */
5402 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5403                                 const struct dt_rec *rec,
5404                                 const struct dt_key *key, struct thandle *th)
5405 {
5406         struct osd_object *obj = osd_dt_obj(dt);
5407         struct iam_path_descr *ipd;
5408         struct osd_thandle *oh;
5409         struct iam_container *bag;
5410         struct osd_thread_info *oti = osd_oti_get(env);
5411         struct iam_rec *iam_rec;
5412         int rc;
5413
5414         ENTRY;
5415
5416         if (!dt_object_exists(dt))
5417                 RETURN(-ENOENT);
5418
5419         LINVRNT(osd_invariant(obj));
5420         LASSERT(!dt_object_remote(dt));
5421
5422         bag = &obj->oo_dir->od_container;
5423         LASSERT(bag->ic_object == obj->oo_inode);
5424         LASSERT(th != NULL);
5425
5426         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5427
5428         ipd = osd_idx_ipd_get(env, bag);
5429         if (unlikely(ipd == NULL))
5430                 RETURN(-ENOMEM);
5431
5432         oh = container_of(th, struct osd_thandle, ot_super);
5433         LASSERT(oh->ot_handle != NULL);
5434         LASSERT(oh->ot_handle->h_transaction != NULL);
5435         if (S_ISDIR(obj->oo_inode->i_mode)) {
5436                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5437                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5438                              &oti->oti_fid);
5439         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5440                 /* pack quota uid/gid */
5441                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5442                 key = (const struct dt_key *)&oti->oti_quota_id;
5443                 /* pack quota record */
5444                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5445                 iam_rec = (struct iam_rec *)rec;
5446         } else {
5447                 iam_rec = (struct iam_rec *)rec;
5448         }
5449
5450         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5451                         iam_rec, ipd);
5452         osd_ipd_put(env, bag, ipd);
5453         LINVRNT(osd_invariant(obj));
5454         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5455         RETURN(rc);
5456 }
5457
5458 /**
5459  * Calls ldiskfs_add_entry() to add directory entry
5460  * into the directory. This is required for
5461  * interoperability mode (b11826)
5462  *
5463  * \retval   0, on success
5464  * \retval -ve, on error
5465  */
5466 static int __osd_ea_add_rec(struct osd_thread_info *info,
5467                             struct osd_object *pobj, struct inode  *cinode,
5468                             const char *name, const struct lu_fid *fid,
5469                             struct htree_lock *hlock, struct thandle *th)
5470 {
5471         struct ldiskfs_dentry_param *ldp;
5472         struct dentry *child;
5473         struct osd_thandle *oth;
5474         int rc;
5475
5476         oth = container_of(th, struct osd_thandle, ot_super);
5477         LASSERT(oth->ot_handle != NULL);
5478         LASSERT(oth->ot_handle->h_transaction != NULL);
5479         LASSERT(pobj->oo_inode);
5480
5481         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5482         if (unlikely(pobj->oo_inode ==
5483                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5484                 ldp->edp_magic = 0;
5485         else
5486                 osd_get_ldiskfs_dirent_param(ldp, fid);
5487         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5488         child->d_fsdata = (void *)ldp;
5489         dquot_initialize(pobj->oo_inode);
5490         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5491                                    child, cinode, hlock);
5492         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5493                 struct ldiskfs_dir_entry_2      *de;
5494                 struct buffer_head              *bh;
5495                 int                              rc1;
5496
5497                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5498                                             NULL, hlock);
5499                 if (!IS_ERR(bh)) {
5500                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5501                                                                bh);
5502                         if (rc1 == 0) {
5503                                 if (S_ISDIR(cinode->i_mode))
5504                                         de->file_type = LDISKFS_DIRENT_LUFID |
5505                                                         LDISKFS_FT_REG_FILE;
5506                                 else
5507                                         de->file_type = LDISKFS_DIRENT_LUFID |
5508                                                         LDISKFS_FT_DIR;
5509                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5510                                                               NULL, bh);
5511                         }
5512                         brelse(bh);
5513                 }
5514         }
5515
5516         RETURN(rc);
5517 }
5518
5519 /**
5520  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5521  * into the directory.Also sets flags into osd object to
5522  * indicate dot and dotdot are created. This is required for
5523  * interoperability mode (b11826)
5524  *
5525  * \param dir   directory for dot and dotdot fixup.
5526  * \param obj   child object for linking
5527  *
5528  * \retval   0, on success
5529  * \retval -ve, on error
5530  */
5531 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5532                               struct osd_object *dir,
5533                               struct inode *parent_dir, const char *name,
5534                               const struct lu_fid *dot_fid,
5535                               const struct lu_fid *dot_dot_fid,
5536                               struct thandle *th)
5537 {
5538         struct inode *inode = dir->oo_inode;
5539         struct osd_thandle *oth;
5540         int result = 0;
5541
5542         oth = container_of(th, struct osd_thandle, ot_super);
5543         LASSERT(oth->ot_handle->h_transaction != NULL);
5544         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5545
5546         if (strcmp(name, dot) == 0) {
5547                 if (dir->oo_compat_dot_created) {
5548                         result = -EEXIST;
5549                 } else {
5550                         LASSERT(inode->i_ino == parent_dir->i_ino);
5551                         dir->oo_compat_dot_created = 1;
5552                         result = 0;
5553                 }
5554         } else if (strcmp(name, dotdot) == 0) {
5555                 if (!dir->oo_compat_dot_created)
5556                         return -EINVAL;
5557                 /* in case of rename, dotdot is already created */
5558                 if (dir->oo_compat_dotdot_created) {
5559                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5560                                                 dot_dot_fid, NULL, th);
5561                 }
5562
5563                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5564                         struct lu_fid tfid = *dot_dot_fid;
5565
5566                         tfid.f_oid--;
5567                         result = osd_add_dot_dotdot_internal(info,
5568                                         dir->oo_inode, parent_dir, dot_fid,
5569                                         &tfid, oth);
5570                 } else {
5571                         result = osd_add_dot_dotdot_internal(info,
5572                                         dir->oo_inode, parent_dir, dot_fid,
5573                                         dot_dot_fid, oth);
5574                 }
5575
5576                 if (result == 0)
5577                         dir->oo_compat_dotdot_created = 1;
5578         }
5579
5580         return result;
5581 }
5582
5583
5584 /**
5585  * It will call the appropriate osd_add* function and return the
5586  * value, return by respective functions.
5587  */
5588 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5589                           struct inode *cinode, const char *name,
5590                           const struct lu_fid *fid, struct thandle *th)
5591 {
5592         struct osd_thread_info *info = osd_oti_get(env);
5593         struct htree_lock *hlock;
5594         int rc;
5595
5596         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5597
5598         if (name[0] == '.' && (name[1] == '\0' ||
5599                                (name[1] == '.' && name[2] == '\0'))) {
5600                 if (hlock != NULL) {
5601                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5602                                            pobj->oo_inode, 0);
5603                 } else {
5604                         down_write(&pobj->oo_ext_idx_sem);
5605                 }
5606
5607                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5608                                         lu_object_fid(&pobj->oo_dt.do_lu),
5609                                         fid, th);
5610         } else {
5611                 if (hlock != NULL) {
5612                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5613                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5614                 } else {
5615                         down_write(&pobj->oo_ext_idx_sem);
5616                 }
5617
5618                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5619                         struct lu_fid *tfid = &info->oti_fid;
5620
5621                         *tfid = *fid;
5622                         tfid->f_ver = ~0;
5623                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5624                                               tfid, hlock, th);
5625                 } else {
5626                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5627                                               hlock, th);
5628                 }
5629         }
5630         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5631             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5632                 pobj->oo_dirent_count++;
5633
5634         if (hlock != NULL)
5635                 ldiskfs_htree_unlock(hlock);
5636         else
5637                 up_write(&pobj->oo_ext_idx_sem);
5638
5639         return rc;
5640 }
5641
5642 static int
5643 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5644                       struct osd_idmap_cache *oic)
5645 {
5646         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5647         struct lu_fid *fid = &oic->oic_fid;
5648         struct osd_inode_id *id = &oic->oic_lid;
5649         struct inode *inode = NULL;
5650         int once = 0;
5651         bool insert;
5652         int rc;
5653
5654         ENTRY;
5655
5656         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5657                 RETURN(0);
5658
5659         if (thread_is_running(&scrub->os_thread) &&
5660             scrub->os_pos_current > id->oii_ino)
5661                 RETURN(0);
5662
5663         if (dev->od_auto_scrub_interval == AS_NEVER ||
5664             ktime_get_real_seconds() <
5665             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5666                 RETURN(0);
5667
5668 again:
5669         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5670         if (rc == -ENOENT) {
5671                 __u32 gen = id->oii_gen;
5672
5673                 insert = true;
5674                 if (inode != NULL)
5675                         goto trigger;
5676
5677                 inode = osd_iget(oti, dev, id);
5678                 /* The inode has been removed (by race maybe). */
5679                 if (IS_ERR(inode)) {
5680                         rc = PTR_ERR(inode);
5681
5682                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5683                 }
5684
5685                 /* The OI mapping is lost. */
5686                 if (gen != OSD_OII_NOGEN)
5687                         goto trigger;
5688
5689                 /*
5690                  * The inode may has been reused by others, we do not know,
5691                  * leave it to be handled by subsequent osd_fid_lookup().
5692                  */
5693                 GOTO(out, rc = 0);
5694         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5695                 GOTO(out, rc);
5696         }
5697
5698         insert = false;
5699
5700 trigger:
5701         if (thread_is_running(&scrub->os_thread)) {
5702                 if (inode == NULL) {
5703                         inode = osd_iget(oti, dev, id);
5704                         /* The inode has been removed (by race maybe). */
5705                         if (IS_ERR(inode)) {
5706                                 rc = PTR_ERR(inode);
5707
5708                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5709                         }
5710                 }
5711
5712                 rc = osd_oii_insert(dev, oic, insert);
5713                 /*
5714                  * There is race condition between osd_oi_lookup and OI scrub.
5715                  * The OI scrub finished just after osd_oi_lookup() failure.
5716                  * Under such case, it is unnecessary to trigger OI scrub again,
5717                  * but try to call osd_oi_lookup() again.
5718                  */
5719                 if (unlikely(rc == -EAGAIN))
5720                         goto again;
5721
5722                 if (!S_ISDIR(inode->i_mode))
5723                         rc = 0;
5724                 else
5725                         rc = osd_check_lmv(oti, dev, inode, oic);
5726
5727                 GOTO(out, rc);
5728         }
5729
5730         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5731                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5732                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5733                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5734                        "%s: trigger partial OI scrub for RPC inconsistency "
5735                        "checking FID "DFID": rc = %d\n",
5736                        osd_dev2name(dev), PFID(fid), rc);
5737                 if (rc == 0 || rc == -EALREADY)
5738                         goto again;
5739         }
5740
5741         GOTO(out, rc);
5742
5743 out:
5744         if (inode)
5745                 iput(inode);
5746
5747         RETURN(rc);
5748 }
5749
5750 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5751                                struct osd_device *dev,
5752                                struct osd_idmap_cache *oic,
5753                                struct lu_fid *fid, __u32 ino)
5754 {
5755         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5756         struct inode *inode;
5757         int rc;
5758
5759         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5760         inode = osd_iget(oti, dev, &oic->oic_lid);
5761         if (IS_ERR(inode)) {
5762                 fid_zero(&oic->oic_fid);
5763                 return PTR_ERR(inode);
5764         }
5765
5766         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5767         iput(inode);
5768         if (rc != 0)
5769                 fid_zero(&oic->oic_fid);
5770         else
5771                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5772         return rc;
5773 }
5774
5775 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5776                       struct osd_inode_id *id, const struct lu_fid *fid)
5777 {
5778         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5779                id->oii_ino, id->oii_gen, info);
5780         info->oti_cache.oic_lid = *id;
5781         info->oti_cache.oic_fid = *fid;
5782         info->oti_cache.oic_dev = osd;
5783 }
5784
5785 /**
5786  * Get parent FID from the linkEA.
5787  *
5788  * For a directory which parent resides on remote MDT, to satisfy the
5789  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5790  * the other hand, to make the lookup(..) on the directory can return
5791  * the real parent FID, we append the real parent FID after its ".."
5792  * name entry in the /REMOTE_PARENT_DIR.
5793  *
5794  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5795  * backup. So after the restore, we cannot get the right parent FID from
5796  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5797  * we have stored the real parent FID in the directory object's linkEA,
5798  * we can parse the linkEA for the real parent FID.
5799  *
5800  * \param[in] env       pointer to the thread context
5801  * \param[in] obj       pointer to the object to be handled
5802  * \param[out]fid       pointer to the buffer to hold the parent FID
5803  *
5804  * \retval              0 for getting the real parent FID successfully
5805  * \retval              negative error number on failure
5806  */
5807 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5808                                     struct osd_object *obj,
5809                                     struct lu_fid *fid)
5810 {
5811         struct osd_thread_info *oti = osd_oti_get(env);
5812         struct lu_buf *buf = &oti->oti_big_buf;
5813         struct dentry *dentry = &oti->oti_obj_dentry;
5814         struct inode *inode = obj->oo_inode;
5815         struct linkea_data ldata = { NULL };
5816         int rc;
5817
5818         ENTRY;
5819
5820         fid_zero(fid);
5821         if (!S_ISDIR(inode->i_mode))
5822                 RETURN(-EIO);
5823
5824 again:
5825         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5826                              buf->lb_buf, buf->lb_len);
5827         if (rc == -ERANGE) {
5828                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5829                                      NULL, 0);
5830                 if (rc > 0) {
5831                         lu_buf_realloc(buf, rc);
5832                         if (buf->lb_buf == NULL)
5833                                 RETURN(-ENOMEM);
5834
5835                         goto again;
5836                 }
5837         }
5838
5839         if (unlikely(rc == 0))
5840                 RETURN(-ENODATA);
5841
5842         if (rc < 0)
5843                 RETURN(rc);
5844
5845         if (unlikely(buf->lb_buf == NULL)) {
5846                 lu_buf_realloc(buf, rc);
5847                 if (buf->lb_buf == NULL)
5848                         RETURN(-ENOMEM);
5849
5850                 goto again;
5851         }
5852
5853         ldata.ld_buf = buf;
5854         rc = linkea_init_with_rec(&ldata);
5855         if (!rc) {
5856                 linkea_first_entry(&ldata);
5857                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5858         }
5859
5860         RETURN(rc);
5861 }
5862
5863 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5864                                     struct inode *inode,
5865                                     const struct lu_fid *pfid,
5866                                     const char *name, const int namelen)
5867 {
5868         struct osd_thread_info *oti = osd_oti_get(env);
5869         struct lu_buf *buf = &oti->oti_big_buf;
5870         struct dentry *dentry = &oti->oti_obj_dentry;
5871         struct linkea_data ldata = { NULL };
5872         struct lu_name cname = { .ln_name = name,
5873                                  .ln_namelen = namelen };
5874         int rc;
5875
5876         ENTRY;
5877
5878 again:
5879         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5880                              buf->lb_buf, buf->lb_len);
5881         if (rc == -ERANGE)
5882                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5883
5884         if (rc < 0)
5885                 RETURN(rc);
5886
5887         if (unlikely(rc == 0))
5888                 RETURN(-ENODATA);
5889
5890         if (buf->lb_len < rc) {
5891                 lu_buf_realloc(buf, rc);
5892                 if (buf->lb_buf == NULL)
5893                         RETURN(-ENOMEM);
5894
5895                 goto again;
5896         }
5897
5898         ldata.ld_buf = buf;
5899         rc = linkea_init_with_rec(&ldata);
5900         if (!rc)
5901                 rc = linkea_links_find(&ldata, &cname, pfid);
5902
5903         RETURN(rc);
5904 }
5905
5906 /**
5907  * Calls ->lookup() to find dentry. From dentry get inode and
5908  * read inode's ea to get fid. This is required for  interoperability
5909  * mode (b11826)
5910  *
5911  * \retval   0, on success
5912  * \retval -ve, on error
5913  */
5914 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5915                              struct dt_rec *rec, const struct dt_key *key)
5916 {
5917         struct inode *dir = obj->oo_inode;
5918         struct dentry *dentry;
5919         struct ldiskfs_dir_entry_2 *de;
5920         struct buffer_head *bh;
5921         struct lu_fid *fid = (struct lu_fid *)rec;
5922         struct htree_lock *hlock = NULL;
5923         int ino;
5924         int rc;
5925
5926         ENTRY;
5927
5928         LASSERT(dir->i_op != NULL);
5929         LASSERT(dir->i_op->lookup != NULL);
5930
5931         dentry = osd_child_dentry_get(env, obj,
5932                                       (char *)key, strlen((char *)key));
5933
5934         if (obj->oo_hl_head != NULL) {
5935                 hlock = osd_oti_get(env)->oti_hlock;
5936                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5937                                    dir, LDISKFS_HLOCK_LOOKUP);
5938         } else {
5939                 down_read(&obj->oo_ext_idx_sem);
5940         }
5941
5942         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5943         if (!IS_ERR(bh)) {
5944                 struct osd_thread_info *oti = osd_oti_get(env);
5945                 struct osd_inode_id *id = &oti->oti_id;
5946                 struct osd_idmap_cache *oic = &oti->oti_cache;
5947                 struct osd_device *dev = osd_obj2dev(obj);
5948
5949                 ino = le32_to_cpu(de->inode);
5950                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5951                         brelse(bh);
5952                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5953                         GOTO(out, rc);
5954                 }
5955
5956                 rc = osd_get_fid_from_dentry(de, rec);
5957
5958                 /* done with de, release bh */
5959                 brelse(bh);
5960                 if (rc != 0) {
5961                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5962                                 const char *name = (const char *)key;
5963
5964                                 /*
5965                                  * If the parent is on remote MDT, and there
5966                                  * is no FID-in-dirent, then we have to get
5967                                  * the parent FID from the linkEA.
5968                                  */
5969                                 if (likely(strlen(name) == 2 &&
5970                                            name[0] == '.' && name[1] == '.'))
5971                                         rc = osd_get_pfid_from_linkea(env, obj,
5972                                                                       fid);
5973                         } else {
5974                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5975                         }
5976                 } else {
5977                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5978                 }
5979
5980                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5981                         fid_zero(&oic->oic_fid);
5982
5983                         GOTO(out, rc);
5984                 }
5985
5986                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5987                 rc = osd_consistency_check(oti, dev, oic);
5988                 if (rc == -ENOENT)
5989                         fid_zero(&oic->oic_fid);
5990                 else
5991                         /* Other error should not affect lookup result. */
5992                         rc = 0;
5993         } else {
5994                 rc = PTR_ERR(bh);
5995         }
5996
5997         GOTO(out, rc);
5998
5999 out:
6000         if (hlock != NULL)
6001                 ldiskfs_htree_unlock(hlock);
6002         else
6003                 up_read(&obj->oo_ext_idx_sem);
6004         return rc;
6005 }
6006
6007 static int osd_index_declare_ea_insert(const struct lu_env *env,
6008                                        struct dt_object *dt,
6009                                        const struct dt_rec *rec,
6010                                        const struct dt_key *key,
6011                                        struct thandle *handle)
6012 {
6013         struct osd_thandle *oh;
6014         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6015         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6016         const struct lu_fid *fid = rec1->rec_fid;
6017         int credits, rc = 0;
6018         struct osd_idmap_cache *idc;
6019
6020         ENTRY;
6021
6022         LASSERT(!dt_object_remote(dt));
6023         LASSERT(handle != NULL);
6024         LASSERT(fid != NULL);
6025         LASSERT(rec1->rec_type != 0);
6026
6027         oh = container_of(handle, struct osd_thandle, ot_super);
6028         LASSERT(oh->ot_handle == NULL);
6029
6030         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6031
6032         /*
6033          * we can't call iget() while a transactions is running
6034          * (this can lead to a deadlock), but we need to know
6035          * inum and object type. so we find this information at
6036          * declaration and cache in per-thread info
6037          */
6038         idc = osd_idc_find_or_init(env, osd, fid);
6039         if (IS_ERR(idc))
6040                 RETURN(PTR_ERR(idc));
6041         if (idc->oic_remote) {
6042                 /*
6043                  * a reference to remote inode is represented by an
6044                  * agent inode which we have to create
6045                  */
6046                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6047                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6048         }
6049
6050         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6051
6052         if (osd_dt_obj(dt)->oo_inode != NULL) {
6053                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6054
6055                 /*
6056                  * We ignore block quota on meta pool (MDTs), so needn't
6057                  * calculate how many blocks will be consumed by this index
6058                  * insert
6059                  */
6060                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6061                                            i_gid_read(inode),
6062                                            i_projid_read(inode), 0,
6063                                            oh, osd_dt_obj(dt), NULL,
6064                                            OSD_QID_BLK);
6065                 if (rc)
6066                         RETURN(rc);
6067
6068 #ifdef HAVE_PROJECT_QUOTA
6069                 /*
6070                  * Reserve credits for local agent inode to transfer
6071                  * to 0, quota enforcement is ignored in this case.
6072                  */
6073                 if (idc->oic_remote &&
6074                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6075                     i_projid_read(inode) != 0)
6076                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6077                                                   0, i_projid_read(inode),
6078                                                   0, false, PRJQUOTA, true);
6079 #endif
6080         }
6081
6082         RETURN(rc);
6083 }
6084
6085 /**
6086  * Index add function for interoperability mode (b11826).
6087  * It will add the directory entry.This entry is needed to
6088  * maintain name->fid mapping.
6089  *
6090  * \param key it is key i.e. file entry to be inserted
6091  * \param rec it is value of given key i.e. fid
6092  *
6093  * \retval   0, on success
6094  * \retval -ve, on error
6095  */
6096 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6097                                const struct dt_rec *rec,
6098                                const struct dt_key *key, struct thandle *th)
6099 {
6100         struct osd_object *obj = osd_dt_obj(dt);
6101         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6102         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6103         const struct lu_fid *fid = rec1->rec_fid;
6104         const char *name = (const char *)key;
6105         struct osd_thread_info *oti = osd_oti_get(env);
6106         struct inode *child_inode = NULL;
6107         struct osd_idmap_cache *idc;
6108         int rc;
6109
6110         ENTRY;
6111
6112         if (!dt_object_exists(dt))
6113                 RETURN(-ENOENT);
6114
6115         LASSERT(osd_invariant(obj));
6116         LASSERT(!dt_object_remote(dt));
6117         LASSERT(th != NULL);
6118
6119         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6120
6121         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6122
6123         idc = osd_idc_find(env, osd, fid);
6124         if (unlikely(idc == NULL)) {
6125                 idc = osd_idc_find_or_init(env, osd, fid);
6126                 if (IS_ERR(idc)) {
6127                         /*
6128                          * this dt_insert() wasn't declared properly, so
6129                          * FID is missing in OI cache. we better do not
6130                          * lookup FID in FLDB/OI and don't risk to deadlock,
6131                          * but in some special cases (lfsck testing, etc)
6132                          * it's much simpler than fixing a caller.
6133                          *
6134                          * normally this error should be placed after the first
6135                          * find, but migrate may attach source stripes to
6136                          * target, which doesn't create stripes.
6137                          */
6138                         CERROR("%s: "DFID" wasn't declared for insert\n",
6139                                osd_name(osd), PFID(fid));
6140                         dump_stack();
6141                         RETURN(PTR_ERR(idc));
6142                 }
6143         }
6144
6145         if (idc->oic_remote) {
6146                 /* Insert remote entry */
6147                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6148                         child_inode =
6149                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6150                 } else {
6151                         child_inode = osd_create_local_agent_inode(env, osd,
6152                                         obj, fid, rec1->rec_type & S_IFMT, th);
6153                         if (IS_ERR(child_inode))
6154                                 RETURN(PTR_ERR(child_inode));
6155                 }
6156         } else {
6157                 /* Insert local entry */
6158                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6159                         /* for a reason OI cache wasn't filled properly */
6160                         CERROR("%s: OIC for "DFID" isn't filled\n",
6161                                osd_name(osd), PFID(fid));
6162                         RETURN(-EINVAL);
6163                 }
6164                 child_inode = oti->oti_inode;
6165                 if (unlikely(child_inode == NULL)) {
6166                         struct ldiskfs_inode_info *lii;
6167
6168                         OBD_ALLOC_PTR(lii);
6169                         if (lii == NULL)
6170                                 RETURN(-ENOMEM);
6171                         child_inode = oti->oti_inode = &lii->vfs_inode;
6172                 }
6173                 child_inode->i_sb = osd_sb(osd);
6174                 child_inode->i_ino = idc->oic_lid.oii_ino;
6175                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6176         }
6177
6178         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6179
6180         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6181                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6182
6183         if (child_inode && child_inode != oti->oti_inode)
6184                 iput(child_inode);
6185         LASSERT(osd_invariant(obj));
6186         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6187
6188         RETURN(rc);
6189 }
6190
6191 /**
6192  *  Initialize osd Iterator for given osd index object.
6193  *
6194  *  \param  dt      osd index object
6195  */
6196
6197 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6198                                      struct dt_object *dt,
6199                                      __u32 unused)
6200 {
6201         struct osd_it_iam *it;
6202         struct osd_object *obj = osd_dt_obj(dt);
6203         struct lu_object *lo = &dt->do_lu;
6204         struct iam_path_descr *ipd;
6205         struct iam_container *bag = &obj->oo_dir->od_container;
6206
6207         if (!dt_object_exists(dt))
6208                 return ERR_PTR(-ENOENT);
6209
6210         OBD_ALLOC_PTR(it);
6211         if (it == NULL)
6212                 return ERR_PTR(-ENOMEM);
6213
6214         ipd = osd_it_ipd_get(env, bag);
6215         if (likely(ipd != NULL)) {
6216                 it->oi_obj = obj;
6217                 it->oi_ipd = ipd;
6218                 lu_object_get(lo);
6219                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6220                 return (struct dt_it *)it;
6221         } else {
6222                 OBD_FREE_PTR(it);
6223                 return ERR_PTR(-ENOMEM);
6224         }
6225 }
6226
6227 /**
6228  * free given Iterator.
6229  */
6230 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6231 {
6232         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6233         struct osd_object *obj = it->oi_obj;
6234
6235         iam_it_fini(&it->oi_it);
6236         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6237         osd_object_put(env, obj);
6238         OBD_FREE_PTR(it);
6239 }
6240
6241 /**
6242  *  Move Iterator to record specified by \a key
6243  *
6244  *  \param  di      osd iterator
6245  *  \param  key     key for index
6246  *
6247  *  \retval +ve  di points to record with least key not larger than key
6248  *  \retval  0   di points to exact matched key
6249  *  \retval -ve  failure
6250  */
6251
6252 static int osd_it_iam_get(const struct lu_env *env,
6253                           struct dt_it *di, const struct dt_key *key)
6254 {
6255         struct osd_thread_info *oti = osd_oti_get(env);
6256         struct osd_it_iam *it = (struct osd_it_iam *)di;
6257
6258         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6259                 /* swab quota uid/gid */
6260                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6261                 key = (struct dt_key *)&oti->oti_quota_id;
6262         }
6263
6264         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6265 }
6266
6267 /**
6268  *  Release Iterator
6269  *
6270  *  \param  di      osd iterator
6271  */
6272 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6273 {
6274         struct osd_it_iam *it = (struct osd_it_iam *)di;
6275
6276         iam_it_put(&it->oi_it);
6277 }
6278
6279 /**
6280  *  Move iterator by one record
6281  *
6282  *  \param  di      osd iterator
6283  *
6284  *  \retval +1   end of container reached
6285  *  \retval  0   success
6286  *  \retval -ve  failure
6287  */
6288
6289 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6290 {
6291         struct osd_it_iam *it = (struct osd_it_iam *)di;
6292
6293         return iam_it_next(&it->oi_it);
6294 }
6295
6296 /**
6297  * Return pointer to the key under iterator.
6298  */
6299
6300 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6301                                      const struct dt_it *di)
6302 {
6303         struct osd_thread_info *oti = osd_oti_get(env);
6304         struct osd_it_iam *it = (struct osd_it_iam *)di;
6305         struct osd_object *obj = it->oi_obj;
6306         struct dt_key *key;
6307
6308         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6309
6310         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6311                 /* swab quota uid/gid */
6312                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6313                 key = (struct dt_key *)&oti->oti_quota_id;
6314         }
6315
6316         return key;
6317 }
6318
6319 /**
6320  * Return size of key under iterator (in bytes)
6321  */
6322
6323 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6324 {
6325         struct osd_it_iam *it = (struct osd_it_iam *)di;
6326
6327         return iam_it_key_size(&it->oi_it);
6328 }
6329
6330 static inline void
6331 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6332 {
6333         /* check if file type is required */
6334         if (ent->lde_attrs & LUDA_TYPE) {
6335                 struct luda_type *lt;
6336                 int align = sizeof(*lt) - 1;
6337
6338                 len = (len + align) & ~align;
6339                 lt = (struct luda_type *)(ent->lde_name + len);
6340                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6341         }
6342
6343         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6344 }
6345
6346 /**
6347  * build lu direct from backend fs dirent.
6348  */
6349
6350 static inline void
6351 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6352                    char *name, __u16 namelen, __u16 type, __u32 attr)
6353 {
6354         ent->lde_attrs = attr | LUDA_FID;
6355         fid_cpu_to_le(&ent->lde_fid, fid);
6356
6357         ent->lde_hash = cpu_to_le64(offset);
6358         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6359
6360         strncpy(ent->lde_name, name, namelen);
6361         ent->lde_name[namelen] = '\0';
6362         ent->lde_namelen = cpu_to_le16(namelen);
6363
6364         /* append lustre attributes */
6365         osd_it_append_attrs(ent, namelen, type);
6366 }
6367
6368 /**
6369  * Return pointer to the record under iterator.
6370  */
6371 static int osd_it_iam_rec(const struct lu_env *env,
6372                           const struct dt_it *di,
6373                           struct dt_rec *dtrec, __u32 attr)
6374 {
6375         struct osd_it_iam *it = (struct osd_it_iam *)di;
6376         struct osd_thread_info *info = osd_oti_get(env);
6377
6378         ENTRY;
6379
6380         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6381                 const struct osd_fid_pack *rec;
6382                 struct lu_fid *fid = &info->oti_fid;
6383                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6384                 char *name;
6385                 int namelen;
6386                 __u64 hash;
6387                 int rc;
6388
6389                 name = (char *)iam_it_key_get(&it->oi_it);
6390                 if (IS_ERR(name))
6391                         RETURN(PTR_ERR(name));
6392
6393                 namelen = iam_it_key_size(&it->oi_it);
6394
6395                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6396                 if (IS_ERR(rec))
6397                         RETURN(PTR_ERR(rec));
6398
6399                 rc = osd_fid_unpack(fid, rec);
6400                 if (rc)
6401                         RETURN(rc);
6402
6403                 hash = iam_it_store(&it->oi_it);
6404
6405                 /* IAM does not store object type in IAM index (dir) */
6406                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6407                                    0, LUDA_FID);
6408         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6409                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6410                            (struct iam_rec *)dtrec);
6411                 osd_quota_unpack(it->oi_obj, dtrec);
6412         } else {
6413                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6414                            (struct iam_rec *)dtrec);
6415         }
6416
6417         RETURN(0);
6418 }
6419
6420 /**
6421  * Returns cookie for current Iterator position.
6422  */
6423 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6424 {
6425         struct osd_it_iam *it = (struct osd_it_iam *)di;
6426
6427         return iam_it_store(&it->oi_it);
6428 }
6429
6430 /**
6431  * Restore iterator from cookie.
6432  *
6433  * \param  di      osd iterator
6434  * \param  hash    Iterator location cookie
6435  *
6436  * \retval +ve  di points to record with least key not larger than key.
6437  * \retval  0   di points to exact matched key
6438  * \retval -ve  failure
6439  */
6440
6441 static int osd_it_iam_load(const struct lu_env *env,
6442                            const struct dt_it *di, __u64 hash)
6443 {
6444         struct osd_it_iam *it = (struct osd_it_iam *)di;
6445
6446         return iam_it_load(&it->oi_it, hash);
6447 }
6448
6449 static const struct dt_index_operations osd_index_iam_ops = {
6450         .dio_lookup         = osd_index_iam_lookup,
6451         .dio_declare_insert = osd_index_declare_iam_insert,
6452         .dio_insert         = osd_index_iam_insert,
6453         .dio_declare_delete = osd_index_declare_iam_delete,
6454         .dio_delete         = osd_index_iam_delete,
6455         .dio_it     = {
6456                 .init     = osd_it_iam_init,
6457                 .fini     = osd_it_iam_fini,
6458                 .get      = osd_it_iam_get,
6459                 .put      = osd_it_iam_put,
6460                 .next     = osd_it_iam_next,
6461                 .key      = osd_it_iam_key,
6462                 .key_size = osd_it_iam_key_size,
6463                 .rec      = osd_it_iam_rec,
6464                 .store    = osd_it_iam_store,
6465                 .load     = osd_it_iam_load
6466         }
6467 };
6468
6469
6470 /**
6471  * Creates or initializes iterator context.
6472  *
6473  * \retval struct osd_it_ea, iterator structure on success
6474  *
6475  */
6476 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6477                                     struct dt_object *dt,
6478                                     __u32 attr)
6479 {
6480         struct osd_object *obj = osd_dt_obj(dt);
6481         struct osd_thread_info *info = osd_oti_get(env);
6482         struct osd_it_ea *oie;
6483         struct file *file;
6484         struct lu_object *lo = &dt->do_lu;
6485         struct dentry *obj_dentry;
6486
6487         ENTRY;
6488
6489         if (!dt_object_exists(dt) || obj->oo_destroyed)
6490                 RETURN(ERR_PTR(-ENOENT));
6491
6492         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6493         if (oie == NULL)
6494                 RETURN(ERR_PTR(-ENOMEM));
6495         obj_dentry = &oie->oie_dentry;
6496
6497         obj_dentry->d_inode = obj->oo_inode;
6498         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6499         obj_dentry->d_name.hash = 0;
6500
6501         oie->oie_rd_dirent       = 0;
6502         oie->oie_it_dirent       = 0;
6503         oie->oie_dirent          = NULL;
6504         if (unlikely(!info->oti_it_ea_buf_used)) {
6505                 oie->oie_buf = info->oti_it_ea_buf;
6506                 info->oti_it_ea_buf_used = 1;
6507         } else {
6508                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6509                 if (oie->oie_buf == NULL)
6510                         RETURN(ERR_PTR(-ENOMEM));
6511         }
6512         oie->oie_obj = obj;
6513
6514         file = &oie->oie_file;
6515
6516         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6517         if (attr & LUDA_64BITHASH)
6518                 file->f_mode    = FMODE_64BITHASH;
6519         else
6520                 file->f_mode    = FMODE_32BITHASH;
6521         file->f_path.dentry     = obj_dentry;
6522         file->f_mapping         = obj->oo_inode->i_mapping;
6523         file->f_op              = obj->oo_inode->i_fop;
6524         file->f_inode = obj->oo_inode;
6525
6526         lu_object_get(lo);
6527         RETURN((struct dt_it *)oie);
6528 }
6529
6530 /**
6531  * Destroy or finishes iterator context.
6532  *
6533  * \param di iterator structure to be destroyed
6534  */
6535 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6536 {
6537         struct osd_thread_info *info = osd_oti_get(env);
6538         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6539         struct osd_object *obj = oie->oie_obj;
6540         struct inode *inode = obj->oo_inode;
6541
6542         ENTRY;
6543         oie->oie_file.f_op->release(inode, &oie->oie_file);
6544         osd_object_put(env, obj);
6545         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6546                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6547         else
6548                 info->oti_it_ea_buf_used = 0;
6549         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6550         EXIT;
6551 }
6552
6553 /**
6554  * It position the iterator at given key, so that next lookup continues from
6555  * that key Or it is similar to dio_it->load() but based on a key,
6556  * rather than file position.
6557  *
6558  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6559  * to the beginning.
6560  *
6561  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6562  */
6563 static int osd_it_ea_get(const struct lu_env *env,
6564                          struct dt_it *di, const struct dt_key *key)
6565 {
6566         struct osd_it_ea *it = (struct osd_it_ea *)di;
6567
6568         ENTRY;
6569         LASSERT(((const char *)key)[0] == '\0');
6570         it->oie_file.f_pos = 0;
6571         it->oie_rd_dirent = 0;
6572         it->oie_it_dirent = 0;
6573         it->oie_dirent = NULL;
6574
6575         RETURN(+1);
6576 }
6577
6578 /**
6579  * Does nothing
6580  */
6581 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6582 {
6583 }
6584
6585 struct osd_filldir_cbs {
6586         struct dir_context ctx;
6587         struct osd_it_ea  *it;
6588 };
6589 /**
6590  * It is called internally by ->iterate*(). It fills the
6591  * iterator's in-memory data structure with required
6592  * information i.e. name, namelen, rec_size etc.
6593  *
6594  * \param buf in which information to be filled in.
6595  * \param name name of the file in given dir
6596  *
6597  * \retval 0 on success
6598  * \retval 1 on buffer full
6599  */
6600 #ifdef HAVE_FILLDIR_USE_CTX
6601 static int osd_ldiskfs_filldir(struct dir_context *buf,
6602 #else
6603 static int osd_ldiskfs_filldir(void *buf,
6604 #endif
6605                                const char *name, int namelen,
6606                                loff_t offset, __u64 ino, unsigned int d_type)
6607 {
6608         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6609         struct osd_object *obj = it->oie_obj;
6610         struct osd_it_ea_dirent *ent = it->oie_dirent;
6611         struct lu_fid *fid = &ent->oied_fid;
6612         struct osd_fid_pack *rec;
6613
6614         ENTRY;
6615
6616 /* this should never happen */
6617         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6618                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6619                 RETURN(-EIO);
6620         }
6621
6622         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6623             OSD_IT_EA_BUFSIZE)
6624                 RETURN(1);
6625
6626         /* "." is just the object itself. */
6627         if (namelen == 1 && name[0] == '.') {
6628                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6629         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6630                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6631                 if (osd_fid_unpack(fid, rec) != 0)
6632                         fid_zero(fid);
6633         } else {
6634                 fid_zero(fid);
6635         }
6636         d_type &= ~LDISKFS_DIRENT_LUFID;
6637
6638         /* NOT export local root. */
6639         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6640                 ino = obj->oo_inode->i_ino;
6641                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6642         }
6643
6644         ent->oied_ino     = ino;
6645         ent->oied_off     = offset;
6646         ent->oied_namelen = namelen;
6647         ent->oied_type    = d_type;
6648
6649         memcpy(ent->oied_name, name, namelen);
6650
6651         it->oie_rd_dirent++;
6652         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6653         RETURN(0);
6654 }
6655
6656 /**
6657  * Calls ->iterate*() to load a directory entry at a time
6658  * and stored it in iterator's in-memory data structure.
6659  *
6660  * \param di iterator's in memory structure
6661  *
6662  * \retval   0 on success
6663  * \retval -ve on error
6664  * \retval +1 reach the end of entry
6665  */
6666 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6667                                const struct dt_it *di)
6668 {
6669         struct osd_it_ea *it = (struct osd_it_ea *)di;
6670         struct osd_object *obj = it->oie_obj;
6671         struct inode *inode = obj->oo_inode;
6672         struct htree_lock *hlock = NULL;
6673         struct file *filp = &it->oie_file;
6674         int rc = 0;
6675         struct osd_filldir_cbs buf = {
6676                 .ctx.actor = osd_ldiskfs_filldir,
6677                 .it = it
6678         };
6679
6680         ENTRY;
6681         it->oie_dirent = it->oie_buf;
6682         it->oie_rd_dirent = 0;
6683
6684         if (obj->oo_hl_head != NULL) {
6685                 hlock = osd_oti_get(env)->oti_hlock;
6686                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6687                                    inode, LDISKFS_HLOCK_READDIR);
6688         } else {
6689                 down_read(&obj->oo_ext_idx_sem);
6690         }
6691
6692         filp->f_cred = current_cred();
6693         rc = osd_security_file_alloc(filp);
6694         if (rc)
6695                 RETURN(rc);
6696
6697         filp->f_flags |= O_NOATIME;
6698         filp->f_mode |= FMODE_NONOTIFY;
6699         rc = iterate_dir(filp, &buf.ctx);
6700         if (rc)
6701                 RETURN(rc);
6702
6703         if (hlock != NULL)
6704                 ldiskfs_htree_unlock(hlock);
6705         else
6706                 up_read(&obj->oo_ext_idx_sem);
6707
6708         if (it->oie_rd_dirent == 0) {
6709                 /*
6710                  * If it does not get any dirent, it means it has been reached
6711                  * to the end of the dir
6712                  */
6713                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6714                 if (rc == 0)
6715                         rc = 1;
6716         } else {
6717                 it->oie_dirent = it->oie_buf;
6718                 it->oie_it_dirent = 1;
6719         }
6720
6721         RETURN(rc);
6722 }
6723
6724 /**
6725  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6726  * to load a directory entry at a time and stored it in
6727  * iterator's in-memory data structure.
6728  *
6729  * \param di iterator's in memory structure
6730  *
6731  * \retval +ve iterator reached to end
6732  * \retval   0 iterator not reached to end
6733  * \retval -ve on error
6734  */
6735 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6736 {
6737         struct osd_it_ea *it = (struct osd_it_ea *)di;
6738         int rc;
6739
6740         ENTRY;
6741
6742         if (it->oie_it_dirent < it->oie_rd_dirent) {
6743                 it->oie_dirent =
6744                         (void *)it->oie_dirent +
6745                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6746                                        it->oie_dirent->oied_namelen);
6747                 it->oie_it_dirent++;
6748                 rc = 0;
6749         } else {
6750                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6751                         rc = 1;
6752                 else
6753                         rc = osd_ldiskfs_it_fill(env, di);
6754         }
6755
6756         RETURN(rc);
6757 }
6758
6759 /**
6760  * Returns the key at current position from iterator's in memory structure.
6761  *
6762  * \param di iterator's in memory structure
6763  *
6764  * \retval key i.e. struct dt_key on success
6765  */
6766 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6767                                     const struct dt_it *di)
6768 {
6769         struct osd_it_ea *it = (struct osd_it_ea *)di;
6770
6771         return (struct dt_key *)it->oie_dirent->oied_name;
6772 }
6773
6774 /**
6775  * Returns key's size at current position from iterator's in memory structure.
6776  *
6777  * \param di iterator's in memory structure
6778  *
6779  * \retval key_size i.e. struct dt_key on success
6780  */
6781 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6782 {
6783         struct osd_it_ea *it = (struct osd_it_ea *)di;
6784
6785         return it->oie_dirent->oied_namelen;
6786 }
6787
6788 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6789 #undef LDISKFS_DIR_REC_LEN
6790 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6791 #endif
6792
6793 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6794 {
6795         if (LDISKFS_DIR_REC_LEN(de) >=
6796             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6797                 return true;
6798
6799         return false;
6800 }
6801
6802 static inline bool
6803 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6804                      unsigned int blocksize, bool dotdot)
6805 {
6806         if (dotdot)
6807                 return osd_dotdot_has_space(de);
6808
6809         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6810             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6811                 return true;
6812
6813         return false;
6814 }
6815
6816 static int
6817 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6818                     handle_t *jh, struct dentry *dentry,
6819                     const struct lu_fid *fid, struct buffer_head *bh,
6820                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6821                     bool dotdot)
6822 {
6823         struct inode *dir = dentry->d_parent->d_inode;
6824         struct inode *inode = dentry->d_inode;
6825         struct osd_fid_pack *rec;
6826         struct ldiskfs_dentry_param *ldp;
6827         int namelen = dentry->d_name.len;
6828         int rc;
6829         struct osd_thread_info *info = osd_oti_get(env);
6830
6831         ENTRY;
6832
6833         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6834                 RETURN(0);
6835
6836         /* There is enough space to hold the FID-in-dirent. */
6837         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6838                 rc = ldiskfs_journal_get_write_access(jh, bh);
6839                 if (rc != 0)
6840                         RETURN(rc);
6841
6842                 de->name[namelen] = 0;
6843                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6844                 rec->fp_len = sizeof(struct lu_fid) + 1;
6845                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6846                 de->file_type |= LDISKFS_DIRENT_LUFID;
6847                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6848
6849                 RETURN(rc);
6850         }
6851
6852         LASSERT(!dotdot);
6853
6854         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6855         if (rc != 0)
6856                 RETURN(rc);
6857
6858         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6859         osd_get_ldiskfs_dirent_param(ldp, fid);
6860         dentry->d_fsdata = (void *)ldp;
6861         dquot_initialize(dir);
6862         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6863         /*
6864          * It is too bad, we cannot reinsert the name entry back.
6865          * That means we lose it!
6866          */
6867         if (rc != 0)
6868                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6869                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6870                        osd_ino2name(inode),
6871                        dir->i_ino, dir->i_generation, namelen,
6872                        dentry->d_name.name, PFID(fid), rc);
6873
6874         RETURN(rc);
6875 }
6876
6877 static int
6878 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6879                         struct osd_it_ea *it, struct lu_fid *fid,
6880                         struct osd_inode_id *id, __u32 *attr)
6881 {
6882         struct osd_thread_info *info = osd_oti_get(env);
6883         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6884         struct osd_device *dev = osd_obj2dev(obj);
6885         struct super_block *sb = osd_sb(dev);
6886         const char *devname = osd_name(dev);
6887         struct osd_it_ea_dirent *ent = it->oie_dirent;
6888         struct inode *dir = obj->oo_inode;
6889         struct htree_lock *hlock = NULL;
6890         struct buffer_head *bh = NULL;
6891         handle_t *jh = NULL;
6892         struct ldiskfs_dir_entry_2 *de;
6893         struct dentry *dentry;
6894         struct inode *inode;
6895         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6896         int credits;
6897         int rc;
6898         bool dotdot = false;
6899         bool dirty = false;
6900
6901         ENTRY;
6902
6903         if (ent->oied_name[0] == '.') {
6904                 if (ent->oied_namelen == 1)
6905                         RETURN(0);
6906
6907                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6908                         dotdot = true;
6909         }
6910
6911         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6912         inode = osd_iget(info, dev, id);
6913         if (IS_ERR(inode)) {
6914                 rc = PTR_ERR(inode);
6915                 if (rc == -ENOENT || rc == -ESTALE) {
6916                         /*
6917                          * Maybe dangling name entry, or
6918                          * corrupted directory entry.
6919                          */
6920                         *attr |= LUDA_UNKNOWN;
6921                         rc = 0;
6922                 } else {
6923                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6924                                "check_repair, dir = %lu/%u, name = %.*s, "
6925                                "ino = %llu, rc = %d\n",
6926                                devname, dir->i_ino, dir->i_generation,
6927                                ent->oied_namelen, ent->oied_name,
6928                                ent->oied_ino, rc);
6929                 }
6930
6931                 RETURN(rc);
6932         }
6933
6934         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6935                                            ent->oied_namelen);
6936         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6937         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6938                 lma = NULL;
6939         else if (rc != 0)
6940                 GOTO(out, rc);
6941
6942         /*
6943          * We need to ensure that the name entry is still valid.
6944          * Because it may be removed or renamed by other already.
6945          *
6946          * The unlink or rename operation will start journal before PDO lock,
6947          * so to avoid deadlock, here we need to start journal handle before
6948          * related PDO lock also. But because we do not know whether there
6949          * will be something to be repaired before PDO lock, we just start
6950          * journal without conditions.
6951          *
6952          * We may need to remove the name entry firstly, then insert back.
6953          * One credit is for user quota file update.
6954          * One credit is for group quota file update.
6955          * Two credits are for dirty inode.
6956          */
6957         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6958                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6959
6960         if (dev->od_dirent_journal != 0) {
6961
6962 again:
6963                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6964                 if (IS_ERR(jh)) {
6965                         rc = PTR_ERR(jh);
6966                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6967                                "check_repair, dir = %lu/%u, credits = %d, "
6968                                "name = %.*s, ino = %llu: rc = %d\n",
6969                                devname, dir->i_ino, dir->i_generation, credits,
6970                                ent->oied_namelen, ent->oied_name,
6971                                ent->oied_ino, rc);
6972
6973                         GOTO(out_inode, rc);
6974                 }
6975
6976                 if (obj->oo_hl_head != NULL) {
6977                         hlock = osd_oti_get(env)->oti_hlock;
6978                         /*
6979                          * "0" means exclusive lock for the whole directory.
6980                          * We need to prevent others access such name entry
6981                          * during the delete + insert. Neither HLOCK_ADD nor
6982                          * HLOCK_DEL cannot guarantee the atomicity.
6983                          */
6984                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6985                 } else {
6986                         down_write(&obj->oo_ext_idx_sem);
6987                 }
6988         } else {
6989                 if (obj->oo_hl_head != NULL) {
6990                         hlock = osd_oti_get(env)->oti_hlock;
6991                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6992                                            LDISKFS_HLOCK_LOOKUP);
6993                 } else {
6994                         down_read(&obj->oo_ext_idx_sem);
6995                 }
6996         }
6997
6998         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6999         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
7000                 *attr |= LUDA_IGNORE;
7001
7002                 GOTO(out, rc = 0);
7003         }
7004
7005         /*
7006          * For dotdot entry, if there is not enough space to hold the
7007          * FID-in-dirent, just keep them there. It only happens when the
7008          * device upgraded from 1.8 or restored from MDT file-level backup.
7009          * For the whole directory, only dotdot entry have no FID-in-dirent
7010          * and needs to get FID from LMA when readdir, it will not affect the
7011          * performance much.
7012          */
7013         if (dotdot && !osd_dotdot_has_space(de)) {
7014                 *attr |= LUDA_UNKNOWN;
7015
7016                 GOTO(out, rc = 0);
7017         }
7018
7019         if (lma != NULL) {
7020                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7021                         GOTO(out, rc = 0);
7022
7023                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7024                         struct lu_fid *tfid = &lma->lma_self_fid;
7025
7026                         if (likely(dotdot &&
7027                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7028                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7029                                 /*
7030                                  * It must be REMOTE_PARENT_DIR and as the
7031                                  * 'dotdot' entry of remote directory
7032                                  */
7033                                 *attr |= LUDA_IGNORE;
7034                         } else {
7035                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7036                                        "parent directory, but got %.*s under "
7037                                        "dir = %lu/%u with the FID "DFID"\n",
7038                                        devname, ent->oied_namelen,
7039                                        ent->oied_name, dir->i_ino,
7040                                        dir->i_generation, PFID(tfid));
7041
7042                                 *attr |= LUDA_UNKNOWN;
7043                         }
7044
7045                         GOTO(out, rc = 0);
7046                 }
7047         }
7048
7049         if (!fid_is_zero(fid)) {
7050                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7051                                               ent->oied_namelen);
7052                 if (rc == -ENOENT ||
7053                     (rc == -ENODATA &&
7054                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7055                         /*
7056                          * linkEA does not recognize the dirent entry,
7057                          * it may because the dirent entry corruption
7058                          * and points to other's inode.
7059                          */
7060                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7061                                "recognize the dirent, dir = %lu/%u, "
7062                                " name = %.*s, ino = %llu, "
7063                                DFID": rc = %d\n", devname, dir->i_ino,
7064                                dir->i_generation, ent->oied_namelen,
7065                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7066                         *attr |= LUDA_UNKNOWN;
7067
7068                         GOTO(out, rc = 0);
7069                 }
7070
7071                 if (rc && rc != -ENODATA) {
7072                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7073                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7074                                DFID": rc = %d\n", devname, dir->i_ino,
7075                                dir->i_generation, ent->oied_namelen,
7076                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7077                         *attr |= LUDA_UNKNOWN;
7078
7079                         GOTO(out, rc = 0);
7080                 }
7081         }
7082
7083         if (lma != NULL) {
7084                 /*
7085                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7086                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7087                  */
7088                 if (*attr & LUDA_VERIFY_DRYRUN) {
7089                         *fid = lma->lma_self_fid;
7090                         *attr |= LUDA_REPAIR;
7091
7092                         GOTO(out, rc = 0);
7093                 }
7094
7095                 if (jh == NULL) {
7096                         brelse(bh);
7097                         dev->od_dirent_journal = 1;
7098                         if (hlock != NULL) {
7099                                 ldiskfs_htree_unlock(hlock);
7100                                 hlock = NULL;
7101                         } else {
7102                                 up_read(&obj->oo_ext_idx_sem);
7103                         }
7104
7105                         goto again;
7106                 }
7107
7108                 *fid = lma->lma_self_fid;
7109                 dirty = true;
7110                 /* Update or append the FID-in-dirent. */
7111                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7112                                          bh, de, hlock, dotdot);
7113                 if (rc == 0)
7114                         *attr |= LUDA_REPAIR;
7115                 else
7116                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7117                                "the dirent, dir = %lu/%u, name = %.*s, "
7118                                "ino = %llu, "DFID": rc = %d\n",
7119                                devname, dir->i_ino, dir->i_generation,
7120                                ent->oied_namelen, ent->oied_name,
7121                                ent->oied_ino, PFID(fid), rc);
7122         } else {
7123                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7124                 if (*attr & LUDA_VERIFY_DRYRUN) {
7125                         if (fid_is_sane(fid)) {
7126                                 *attr |= LUDA_REPAIR;
7127                         } else if (dev->od_index == 0) {
7128                                 lu_igif_build(fid, inode->i_ino,
7129                                               inode->i_generation);
7130                                 *attr |= LUDA_UPGRADE;
7131                         }
7132
7133                         GOTO(out, rc = 0);
7134                 }
7135
7136                 if (jh == NULL) {
7137                         brelse(bh);
7138                         dev->od_dirent_journal = 1;
7139                         if (hlock != NULL) {
7140                                 ldiskfs_htree_unlock(hlock);
7141                                 hlock = NULL;
7142                         } else {
7143                                 up_read(&obj->oo_ext_idx_sem);
7144                         }
7145
7146                         goto again;
7147                 }
7148
7149                 dirty = true;
7150                 if (unlikely(fid_is_sane(fid))) {
7151                         /*
7152                          * FID-in-dirent exists, but FID-in-LMA is lost.
7153                          * Trust the FID-in-dirent, and add FID-in-LMA.
7154                          */
7155                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7156                         if (rc == 0)
7157                                 *attr |= LUDA_REPAIR;
7158                         else
7159                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7160                                        "update dirent, dir = %lu/%u, "
7161                                        "name = %.*s, ino = %llu, "
7162                                        DFID": rc = %d\n",
7163                                        devname, dir->i_ino, dir->i_generation,
7164                                        ent->oied_namelen, ent->oied_name,
7165                                        ent->oied_ino, PFID(fid), rc);
7166                 } else if (dev->od_index == 0) {
7167                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7168                         /*
7169                          * It is probably IGIF object. Only aappend the
7170                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7171                          */
7172                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7173                                                  bh, de, hlock, dotdot);
7174                         if (rc == 0)
7175                                 *attr |= LUDA_UPGRADE;
7176                         else
7177                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7178                                        "after the dirent, dir = %lu/%u, "
7179                                        "name = %.*s, ino = %llu, "
7180                                        DFID": rc = %d\n",
7181                                        devname, dir->i_ino, dir->i_generation,
7182                                        ent->oied_namelen, ent->oied_name,
7183                                        ent->oied_ino, PFID(fid), rc);
7184                 }
7185         }
7186
7187         GOTO(out, rc);
7188
7189 out:
7190         if (!IS_ERR(bh))
7191                 brelse(bh);
7192         if (hlock != NULL) {
7193                 ldiskfs_htree_unlock(hlock);
7194         } else {
7195                 if (dev->od_dirent_journal != 0)
7196                         up_write(&obj->oo_ext_idx_sem);
7197                 else
7198                         up_read(&obj->oo_ext_idx_sem);
7199         }
7200
7201         if (jh != NULL)
7202                 ldiskfs_journal_stop(jh);
7203
7204 out_inode:
7205         iput(inode);
7206         if (rc >= 0 && !dirty)
7207                 dev->od_dirent_journal = 0;
7208
7209         return rc;
7210 }
7211
7212 /**
7213  * Returns the value at current position from iterator's in memory structure.
7214  *
7215  * \param di struct osd_it_ea, iterator's in memory structure
7216  * \param attr attr requested for dirent.
7217  * \param lde lustre dirent
7218  *
7219  * \retval   0 no error and \param lde has correct lustre dirent.
7220  * \retval -ve on error
7221  */
7222 static inline int osd_it_ea_rec(const struct lu_env *env,
7223                                 const struct dt_it *di,
7224                                 struct dt_rec *dtrec, __u32 attr)
7225 {
7226         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7227         struct osd_object      *obj   = it->oie_obj;
7228         struct osd_device      *dev   = osd_obj2dev(obj);
7229         struct osd_thread_info *oti   = osd_oti_get(env);
7230         struct osd_inode_id    *id    = &oti->oti_id;
7231         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7232         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7233         __u32 ino = it->oie_dirent->oied_ino;
7234         int rc = 0;
7235
7236         ENTRY;
7237
7238         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7239
7240         if (attr & LUDA_VERIFY) {
7241                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7242                         attr |= LUDA_IGNORE;
7243                         /*
7244                          * If the parent is on remote MDT, and there
7245                          * is no FID-in-dirent, then we have to get
7246                          * the parent FID from the linkEA.
7247                          */
7248                         if (!fid_is_sane(fid) &&
7249                             it->oie_dirent->oied_namelen == 2 &&
7250                             it->oie_dirent->oied_name[0] == '.' &&
7251                             it->oie_dirent->oied_name[1] == '.')
7252                                 osd_get_pfid_from_linkea(env, obj, fid);
7253                 } else {
7254                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7255                                                      &attr);
7256                 }
7257
7258                 if (!fid_is_sane(fid))
7259                         attr |= LUDA_UNKNOWN;
7260         } else {
7261                 attr &= ~LU_DIRENT_ATTRS_MASK;
7262                 if (!fid_is_sane(fid)) {
7263                         bool is_dotdot = false;
7264
7265                         if (it->oie_dirent->oied_namelen == 2 &&
7266                             it->oie_dirent->oied_name[0] == '.' &&
7267                             it->oie_dirent->oied_name[1] == '.')
7268                                 is_dotdot = true;
7269                         /*
7270                          * If the parent is on remote MDT, and there
7271                          * is no FID-in-dirent, then we have to get
7272                          * the parent FID from the linkEA.
7273                          */
7274                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7275                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7276                         } else {
7277                                 if (is_dotdot == false &&
7278                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7279                                         RETURN(-ENOENT);
7280
7281                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7282                         }
7283                 } else {
7284                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7285                 }
7286         }
7287
7288         /* Pack the entry anyway, at least the offset is right. */
7289         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7290                            it->oie_dirent->oied_name,
7291                            it->oie_dirent->oied_namelen,
7292                            it->oie_dirent->oied_type, attr);
7293
7294         if (rc < 0)
7295                 RETURN(rc);
7296
7297         if (osd_remote_fid(env, dev, fid))
7298                 RETURN(0);
7299
7300         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7301                 osd_add_oi_cache(oti, dev, id, fid);
7302
7303         RETURN(rc > 0 ? 0 : rc);
7304 }
7305
7306 /**
7307  * Returns the record size size at current position.
7308  *
7309  * This function will return record(lu_dirent) size in bytes.
7310  *
7311  * \param[in] env       execution environment
7312  * \param[in] di        iterator's in memory structure
7313  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7314  *                      calculate the lu_dirent size.
7315  *
7316  * \retval      record size(in bytes & in memory) of the current lu_dirent
7317  *              entry.
7318  */
7319 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7320                               __u32 attr)
7321 {
7322         struct osd_it_ea *it = (struct osd_it_ea *)di;
7323
7324         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7325 }
7326
7327 /**
7328  * Returns a cookie for current position of the iterator head, so that
7329  * user can use this cookie to load/start the iterator next time.
7330  *
7331  * \param di iterator's in memory structure
7332  *
7333  * \retval cookie for current position, on success
7334  */
7335 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7336 {
7337         struct osd_it_ea *it = (struct osd_it_ea *)di;
7338
7339         return it->oie_dirent->oied_off;
7340 }
7341
7342 /**
7343  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7344  * to load a directory entry at a time and stored it i inn,
7345  * in iterator's in-memory data structure.
7346  *
7347  * \param di struct osd_it_ea, iterator's in memory structure
7348  *
7349  * \retval +ve on success
7350  * \retval -ve on error
7351  */
7352 static int osd_it_ea_load(const struct lu_env *env,
7353                           const struct dt_it *di, __u64 hash)
7354 {
7355         struct osd_it_ea *it = (struct osd_it_ea *)di;
7356         int rc;
7357
7358         ENTRY;
7359         it->oie_file.f_pos = hash;
7360
7361         rc =  osd_ldiskfs_it_fill(env, di);
7362         if (rc > 0)
7363                 rc = -ENODATA;
7364
7365         if (rc == 0)
7366                 rc = 1;
7367
7368         RETURN(rc);
7369 }
7370
7371 /**
7372  * Index lookup function for interoperability mode (b11826).
7373  *
7374  * \param key,  key i.e. file name to be searched
7375  *
7376  * \retval +ve, on success
7377  * \retval -ve, on error
7378  */
7379 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7380                                struct dt_rec *rec, const struct dt_key *key)
7381 {
7382         struct osd_object *obj = osd_dt_obj(dt);
7383         int rc = 0;
7384
7385         ENTRY;
7386
7387         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7388         LINVRNT(osd_invariant(obj));
7389
7390         rc = osd_ea_lookup_rec(env, obj, rec, key);
7391         if (rc == 0)
7392                 rc = 1;
7393         RETURN(rc);
7394 }
7395
7396 /**
7397  * Index and Iterator operations for interoperability
7398  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7399  */
7400 static const struct dt_index_operations osd_index_ea_ops = {
7401         .dio_lookup         = osd_index_ea_lookup,
7402         .dio_declare_insert = osd_index_declare_ea_insert,
7403         .dio_insert         = osd_index_ea_insert,
7404         .dio_declare_delete = osd_index_declare_ea_delete,
7405         .dio_delete         = osd_index_ea_delete,
7406         .dio_it     = {
7407                 .init     = osd_it_ea_init,
7408                 .fini     = osd_it_ea_fini,
7409                 .get      = osd_it_ea_get,
7410                 .put      = osd_it_ea_put,
7411                 .next     = osd_it_ea_next,
7412                 .key      = osd_it_ea_key,
7413                 .key_size = osd_it_ea_key_size,
7414                 .rec      = osd_it_ea_rec,
7415                 .rec_size = osd_it_ea_rec_size,
7416                 .store    = osd_it_ea_store,
7417                 .load     = osd_it_ea_load
7418         }
7419 };
7420
7421 static void *osd_key_init(const struct lu_context *ctx,
7422                           struct lu_context_key *key)
7423 {
7424         struct osd_thread_info *info;
7425
7426         OBD_ALLOC_PTR(info);
7427         if (info == NULL)
7428                 return ERR_PTR(-ENOMEM);
7429
7430         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7431         if (info->oti_it_ea_buf == NULL)
7432                 goto out_free_info;
7433
7434         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7435
7436         info->oti_hlock = ldiskfs_htree_lock_alloc();
7437         if (info->oti_hlock == NULL)
7438                 goto out_free_ea;
7439
7440         return info;
7441
7442 out_free_ea:
7443         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7444 out_free_info:
7445         OBD_FREE_PTR(info);
7446         return ERR_PTR(-ENOMEM);
7447 }
7448
7449 static void osd_key_fini(const struct lu_context *ctx,
7450                          struct lu_context_key *key, void *data)
7451 {
7452         struct osd_thread_info *info = data;
7453         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7454         struct osd_idmap_cache *idc = info->oti_ins_cache;
7455
7456         if (info->oti_dio_pages) {
7457                 int i;
7458                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7459                         struct page *page = info->oti_dio_pages[i];
7460                         if (page) {
7461                                 LASSERT(PagePrivate2(page));
7462                                 LASSERT(PageLocked(page));
7463                                 ClearPagePrivate2(page);
7464                                 unlock_page(page);
7465                                 __free_page(page);
7466                         }
7467                 }
7468                 OBD_FREE_PTR_ARRAY(info->oti_dio_pages, PTLRPC_MAX_BRW_PAGES);
7469         }
7470
7471         if (info->oti_inode != NULL)
7472                 OBD_FREE_PTR(lli);
7473         if (info->oti_hlock != NULL)
7474                 ldiskfs_htree_lock_free(info->oti_hlock);
7475         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7476         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7477         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7478         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7479         lu_buf_free(&info->oti_big_buf);
7480         if (idc != NULL) {
7481                 LASSERT(info->oti_ins_cache_size > 0);
7482                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7483                 info->oti_ins_cache = NULL;
7484                 info->oti_ins_cache_size = 0;
7485         }
7486         OBD_FREE_PTR(info);
7487 }
7488
7489 static void osd_key_exit(const struct lu_context *ctx,
7490                          struct lu_context_key *key, void *data)
7491 {
7492         struct osd_thread_info *info = data;
7493
7494         LASSERT(info->oti_r_locks == 0);
7495         LASSERT(info->oti_w_locks == 0);
7496         LASSERT(info->oti_txns    == 0);
7497 }
7498
7499 /* type constructor/destructor: osd_type_init, osd_type_fini */
7500 LU_TYPE_INIT_FINI(osd, &osd_key);
7501
7502 struct lu_context_key osd_key = {
7503         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7504         .lct_init = osd_key_init,
7505         .lct_fini = osd_key_fini,
7506         .lct_exit = osd_key_exit
7507 };
7508
7509
7510 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7511                            const char *name, struct lu_device *next)
7512 {
7513         struct osd_device *osd = osd_dev(d);
7514
7515         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7516             sizeof(osd->od_svname))
7517                 return -E2BIG;
7518         return osd_procfs_init(osd, name);
7519 }
7520
7521 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7522 {
7523         struct seq_server_site *ss = osd_seq_site(osd);
7524         int rc = 0;
7525
7526         ENTRY;
7527
7528         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7529                 RETURN(0);
7530
7531         if (unlikely(ss == NULL))
7532                 RETURN(-ENODEV);
7533
7534         OBD_ALLOC_PTR(osd->od_cl_seq);
7535         if (osd->od_cl_seq == NULL)
7536                 RETURN(-ENOMEM);
7537
7538         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7539                         osd->od_svname, ss->ss_server_seq);
7540
7541         if (ss->ss_node_id == 0) {
7542                 /*
7543                  * If the OSD on the sequence controller(MDT0), then allocate
7544                  * sequence here, otherwise allocate sequence after connected
7545                  * to MDT0 (see mdt_register_lwp_callback()).
7546                  */
7547                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7548                                    &osd->od_cl_seq->lcs_space, env);
7549         }
7550
7551         RETURN(rc);
7552 }
7553
7554 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7555 {
7556         if (osd->od_cl_seq == NULL)
7557                 return;
7558
7559         seq_client_fini(osd->od_cl_seq);
7560         OBD_FREE_PTR(osd->od_cl_seq);
7561         osd->od_cl_seq = NULL;
7562 }
7563
7564 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7565 {
7566         ENTRY;
7567
7568         /* shutdown quota slave instance associated with the device */
7569         if (o->od_quota_slave_md != NULL) {
7570                 struct qsd_instance *qsd = o->od_quota_slave_md;
7571
7572                 o->od_quota_slave_md = NULL;
7573                 qsd_fini(env, qsd);
7574         }
7575
7576         if (o->od_quota_slave_dt != NULL) {
7577                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7578
7579                 o->od_quota_slave_dt = NULL;
7580                 qsd_fini(env, qsd);
7581         }
7582
7583         osd_fid_fini(env, o);
7584         osd_scrub_cleanup(env, o);
7585
7586         RETURN(0);
7587 }
7588
7589 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7590 {
7591         ENTRY;
7592
7593         if (o->od_mnt != NULL) {
7594                 shrink_dcache_sb(osd_sb(o));
7595                 osd_sync(env, &o->od_dt_dev);
7596
7597                 mntput(o->od_mnt);
7598                 o->od_mnt = NULL;
7599         }
7600
7601         EXIT;
7602 }
7603
7604 static int osd_mount(const struct lu_env *env,
7605                      struct osd_device *o, struct lustre_cfg *cfg)
7606 {
7607         const char *name = lustre_cfg_string(cfg, 0);
7608         const char *dev = lustre_cfg_string(cfg, 1);
7609         const char *opts;
7610         unsigned long page, s_flags, lmd_flags = 0;
7611         struct page *__page;
7612         struct file_system_type *type;
7613         char *options = NULL;
7614         char *str;
7615         struct osd_thread_info *info = osd_oti_get(env);
7616         struct lu_fid *fid = &info->oti_fid;
7617         struct inode *inode;
7618         int rc = 0, force_over_1024tb = 0;
7619
7620         ENTRY;
7621
7622         if (o->od_mnt != NULL)
7623                 RETURN(0);
7624
7625         if (strlen(dev) >= sizeof(o->od_mntdev))
7626                 RETURN(-E2BIG);
7627         strcpy(o->od_mntdev, dev);
7628
7629         str = lustre_cfg_string(cfg, 2);
7630         s_flags = simple_strtoul(str, NULL, 0);
7631         str = strstr(str, ":");
7632         if (str)
7633                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7634         opts = lustre_cfg_string(cfg, 3);
7635 #ifdef __BIG_ENDIAN
7636         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7637                 CERROR("%s: device %s extents feature is not guaranteed to "
7638                        "work on big-endian systems. Use \"bigendian_extents\" "
7639                        "mount option to override.\n", name, dev);
7640                 RETURN(-EINVAL);
7641         }
7642 #endif
7643 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7644         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7645                 CWARN("force_over_128tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7646         }
7647 #endif
7648 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7649         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7650                 CWARN("force_over_256tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7651         }
7652 #endif
7653 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7654         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7655                 CWARN("force_over_512tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7656         }
7657 #endif
7658
7659         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7660                 force_over_1024tb = 1;
7661
7662         __page = alloc_page(GFP_KERNEL);
7663         if (__page == NULL)
7664                 GOTO(out, rc = -ENOMEM);
7665         page = (unsigned long)page_address(__page);
7666         options = (char *)page;
7667         *options = '\0';
7668         if (opts != NULL) {
7669                 /* strip out the options for back compatiblity */
7670                 static const char * const sout[] = {
7671                         "mballoc",
7672                         "iopen",
7673                         "noiopen",
7674                         "iopen_nopriv",
7675                         "extents",
7676                         "noextents",
7677                         /* strip out option we processed in osd */
7678                         "bigendian_extents",
7679                         "force_over_128tb",
7680                         "force_over_256tb",
7681                         "force_over_512tb",
7682                         "force_over_1024tb",
7683                         NULL
7684                 };
7685                 strncat(options, opts, PAGE_SIZE);
7686                 for (rc = 0, str = options; sout[rc]; ) {
7687                         char *op = strstr(str, sout[rc]);
7688
7689                         if (op == NULL) {
7690                                 rc++;
7691                                 str = options;
7692                                 continue;
7693                         }
7694                         if (op == options || *(op - 1) == ',') {
7695                                 str = op + strlen(sout[rc]);
7696                                 if (*str == ',' || *str == '\0') {
7697                                         *str == ',' ? str++ : str;
7698                                         memmove(op, str, strlen(str) + 1);
7699                                 }
7700                         }
7701                         for (str = op; *str != ',' && *str != '\0'; str++)
7702                                 ;
7703                 }
7704         } else {
7705                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7706         }
7707
7708         /* Glom up mount options */
7709         if (*options != '\0')
7710                 strncat(options, ",", PAGE_SIZE);
7711         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7712
7713         type = get_fs_type("ldiskfs");
7714         if (!type) {
7715                 CERROR("%s: cannot find ldiskfs module\n", name);
7716                 GOTO(out, rc = -ENODEV);
7717         }
7718
7719         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7720         module_put(type->owner);
7721
7722         if (IS_ERR(o->od_mnt)) {
7723                 rc = PTR_ERR(o->od_mnt);
7724                 o->od_mnt = NULL;
7725                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7726                 GOTO(out, rc);
7727         }
7728
7729         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7730                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7731                                  force_over_1024tb == 0) {
7732                 CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
7733                        name, dev);
7734                 GOTO(out_mnt, rc = -EINVAL);
7735         }
7736
7737         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7738                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7739                               name);
7740
7741                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7742         }
7743
7744         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7745                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7746                 GOTO(out_mnt, rc = -EINVAL);
7747         }
7748
7749 #ifdef LDISKFS_MOUNT_DIRDATA
7750         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7751                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7752         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7753                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7754                       "enabling the dirdata feature. If you do not want to "
7755                       "downgrade to Lustre-1.x again, you can enable it via "
7756                       "'tune2fs -O dirdata device'\n", name, dev);
7757 #endif
7758         inode = osd_sb(o)->s_root->d_inode;
7759         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7760         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7761         if (rc != 0) {
7762                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7763                 GOTO(out_mnt, rc);
7764         }
7765
7766         if (lmd_flags & LMD_FLG_NOSCRUB)
7767                 o->od_auto_scrub_interval = AS_NEVER;
7768
7769         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7770                 /* do not use pagecache with flash-backed storage */
7771                 o->od_writethrough_cache = 0;
7772                 o->od_read_cache = 0;
7773         }
7774
7775         GOTO(out, rc = 0);
7776
7777 out_mnt:
7778         mntput(o->od_mnt);
7779         o->od_mnt = NULL;
7780
7781 out:
7782         if (__page)
7783                 __free_page(__page);
7784
7785         return rc;
7786 }
7787
7788 static struct lu_device *osd_device_fini(const struct lu_env *env,
7789                                          struct lu_device *d)
7790 {
7791         struct osd_device *o = osd_dev(d);
7792
7793         ENTRY;
7794
7795         osd_index_backup(env, o, false);
7796         osd_shutdown(env, o);
7797         osd_procfs_fini(o);
7798         if (o->od_oi_table != NULL)
7799                 osd_oi_fini(osd_oti_get(env), o);
7800         osd_obj_map_fini(o);
7801         osd_umount(env, o);
7802
7803         RETURN(NULL);
7804 }
7805
7806 static int osd_device_init0(const struct lu_env *env,
7807                             struct osd_device *o,
7808                             struct lustre_cfg *cfg)
7809 {
7810         struct lu_device *l = osd2lu_dev(o);
7811         struct osd_thread_info *info;
7812         int rc;
7813         int cplen = 0;
7814
7815         /* if the module was re-loaded, env can loose its keys */
7816         rc = lu_env_refill((struct lu_env *)env);
7817         if (rc)
7818                 GOTO(out, rc);
7819         info = osd_oti_get(env);
7820         LASSERT(info);
7821
7822         l->ld_ops = &osd_lu_ops;
7823         o->od_dt_dev.dd_ops = &osd_dt_ops;
7824
7825         spin_lock_init(&o->od_osfs_lock);
7826         mutex_init(&o->od_otable_mutex);
7827         INIT_LIST_HEAD(&o->od_orphan_list);
7828         INIT_LIST_HEAD(&o->od_index_backup_list);
7829         INIT_LIST_HEAD(&o->od_index_restore_list);
7830         spin_lock_init(&o->od_lock);
7831         o->od_index_backup_policy = LIBP_NONE;
7832         o->od_t10_type = 0;
7833
7834         o->od_read_cache = 1;
7835         o->od_writethrough_cache = 1;
7836         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7837         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7838         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7839         o->od_auto_scrub_interval = AS_DEFAULT;
7840
7841         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7842                         sizeof(o->od_svname));
7843         if (cplen >= sizeof(o->od_svname)) {
7844                 rc = -E2BIG;
7845                 GOTO(out, rc);
7846         }
7847
7848         o->od_index_backup_stop = 0;
7849         o->od_index = -1; /* -1 means index is invalid */
7850         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7851         if (rc == LDD_F_SV_TYPE_OST)
7852                 o->od_is_ost = 1;
7853
7854         o->od_full_scrub_ratio = OFSR_DEFAULT;
7855         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7856         rc = osd_mount(env, o, cfg);
7857         if (rc != 0)
7858                 GOTO(out, rc);
7859
7860         /* Can only check block device after mount */
7861         o->od_nonrotational =
7862                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7863
7864         rc = osd_obj_map_init(env, o);
7865         if (rc != 0)
7866                 GOTO(out_mnt, rc);
7867
7868         rc = lu_site_init(&o->od_site, l);
7869         if (rc != 0)
7870                 GOTO(out_compat, rc);
7871         o->od_site.ls_bottom_dev = l;
7872
7873         rc = lu_site_init_finish(&o->od_site);
7874         if (rc != 0)
7875                 GOTO(out_site, rc);
7876
7877         INIT_LIST_HEAD(&o->od_ios_list);
7878         /* setup scrub, including OI files initialization */
7879         o->od_in_init = 1;
7880         rc = osd_scrub_setup(env, o);
7881         o->od_in_init = 0;
7882         if (rc < 0)
7883                 GOTO(out_site, rc);
7884
7885         rc = osd_procfs_init(o, o->od_svname);
7886         if (rc != 0) {
7887                 CERROR("%s: can't initialize procfs: rc = %d\n",
7888                        o->od_svname, rc);
7889                 GOTO(out_scrub, rc);
7890         }
7891
7892         LASSERT(l->ld_site->ls_linkage.next != NULL);
7893         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7894
7895         /* initialize quota slave instance */
7896         /* currently it's no need to prepare qsd_instance_md for OST */
7897         if (!o->od_is_ost) {
7898                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7899                                                 &o->od_dt_dev,
7900                                                 o->od_proc_entry, true);
7901                 if (IS_ERR(o->od_quota_slave_md)) {
7902                         rc = PTR_ERR(o->od_quota_slave_md);
7903                         o->od_quota_slave_md = NULL;
7904                         GOTO(out_procfs, rc);
7905                 }
7906         }
7907
7908         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7909                                         o->od_proc_entry, false);
7910
7911         if (IS_ERR(o->od_quota_slave_dt)) {
7912                 if (o->od_quota_slave_md != NULL) {
7913                         qsd_fini(env, o->od_quota_slave_md);
7914                         o->od_quota_slave_md = NULL;
7915                 }
7916
7917                 rc = PTR_ERR(o->od_quota_slave_dt);
7918                 o->od_quota_slave_dt = NULL;
7919                 GOTO(out_procfs, rc);
7920         }
7921
7922         RETURN(0);
7923
7924 out_procfs:
7925         osd_procfs_fini(o);
7926 out_scrub:
7927         osd_scrub_cleanup(env, o);
7928 out_site:
7929         lu_site_fini(&o->od_site);
7930 out_compat:
7931         osd_obj_map_fini(o);
7932 out_mnt:
7933         osd_umount(env, o);
7934 out:
7935         return rc;
7936 }
7937
7938 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7939                                           struct lu_device_type *t,
7940                                           struct lustre_cfg *cfg)
7941 {
7942         struct osd_device *o;
7943         int rc;
7944
7945         OBD_ALLOC_PTR(o);
7946         if (o == NULL)
7947                 return ERR_PTR(-ENOMEM);
7948
7949         rc = dt_device_init(&o->od_dt_dev, t);
7950         if (rc == 0) {
7951                 /*
7952                  * Because the ctx might be revived in dt_device_init,
7953                  * refill the env here
7954                  */
7955                 lu_env_refill((struct lu_env *)env);
7956                 rc = osd_device_init0(env, o, cfg);
7957                 if (rc)
7958                         dt_device_fini(&o->od_dt_dev);
7959         }
7960
7961         if (unlikely(rc != 0))
7962                 OBD_FREE_PTR(o);
7963
7964         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7965 }
7966
7967 static struct lu_device *osd_device_free(const struct lu_env *env,
7968                                          struct lu_device *d)
7969 {
7970         struct osd_device *o = osd_dev(d);
7971
7972         ENTRY;
7973
7974         /* XXX: make osd top device in order to release reference */
7975         d->ld_site->ls_top_dev = d;
7976         lu_site_purge(env, d->ld_site, -1);
7977         lu_site_print(env, d->ld_site, &d->ld_site->ls_obj_hash.nelems,
7978                       D_ERROR, lu_cdebug_printer);
7979         lu_site_fini(&o->od_site);
7980         dt_device_fini(&o->od_dt_dev);
7981         OBD_FREE_PTR(o);
7982         RETURN(NULL);
7983 }
7984
7985 static int osd_process_config(const struct lu_env *env,
7986                               struct lu_device *d, struct lustre_cfg *cfg)
7987 {
7988         struct osd_device *o = osd_dev(d);
7989         ssize_t count;
7990         int rc;
7991
7992         ENTRY;
7993
7994         switch (cfg->lcfg_command) {
7995         case LCFG_SETUP:
7996                 rc = osd_mount(env, o, cfg);
7997                 break;
7998         case LCFG_CLEANUP:
7999                 /*
8000                  * For the case LCFG_PRE_CLEANUP is not called in advance,
8001                  * that may happend if hit failure during mount process.
8002                  */
8003                 osd_index_backup(env, o, false);
8004                 lu_dev_del_linkage(d->ld_site, d);
8005                 rc = osd_shutdown(env, o);
8006                 break;
8007         case LCFG_PARAM:
8008                 LASSERT(&o->od_dt_dev);
8009                 count  = class_modify_config(cfg, PARAM_OSD,
8010                                              &o->od_dt_dev.dd_kobj);
8011                 if (count < 0)
8012                         count = class_modify_config(cfg, PARAM_OST,
8013                                                     &o->od_dt_dev.dd_kobj);
8014                 rc = count > 0 ? 0 : count;
8015                 break;
8016         case LCFG_PRE_CLEANUP:
8017                 osd_scrub_stop(o);
8018                 osd_index_backup(env, o,
8019                                  o->od_index_backup_policy != LIBP_NONE);
8020                 rc = 0;
8021                 break;
8022         default:
8023                 rc = -ENOSYS;
8024         }
8025
8026         RETURN(rc);
8027 }
8028
8029 static int osd_recovery_complete(const struct lu_env *env,
8030                                  struct lu_device *d)
8031 {
8032         struct osd_device *osd = osd_dev(d);
8033         int rc = 0;
8034
8035         ENTRY;
8036
8037         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8038                 RETURN(0);
8039
8040         /*
8041          * start qsd instance on recovery completion, this notifies the quota
8042          * slave code that we are about to process new requests now
8043          */
8044         rc = qsd_start(env, osd->od_quota_slave_dt);
8045         if (rc == 0 && osd->od_quota_slave_md != NULL)
8046                 rc = qsd_start(env, osd->od_quota_slave_md);
8047
8048         RETURN(rc);
8049 }
8050
8051 /*
8052  * we use exports to track all osd users
8053  */
8054 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8055                            struct obd_device *obd, struct obd_uuid *cluuid,
8056                            struct obd_connect_data *data, void *localdata)
8057 {
8058         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8059         struct lustre_handle conn;
8060         int rc;
8061
8062         ENTRY;
8063
8064         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8065
8066         rc = class_connect(&conn, obd, cluuid);
8067         if (rc)
8068                 RETURN(rc);
8069
8070         *exp = class_conn2export(&conn);
8071
8072         spin_lock(&osd->od_osfs_lock);
8073         osd->od_connects++;
8074         spin_unlock(&osd->od_osfs_lock);
8075
8076         RETURN(0);
8077 }
8078
8079 /*
8080  * once last export (we don't count self-export) disappeared
8081  * osd can be released
8082  */
8083 static int osd_obd_disconnect(struct obd_export *exp)
8084 {
8085         struct obd_device *obd = exp->exp_obd;
8086         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8087         int rc, release = 0;
8088
8089         ENTRY;
8090
8091         /* Only disconnect the underlying layers on the final disconnect. */
8092         spin_lock(&osd->od_osfs_lock);
8093         osd->od_connects--;
8094         if (osd->od_connects == 0)
8095                 release = 1;
8096         spin_unlock(&osd->od_osfs_lock);
8097
8098         rc = class_disconnect(exp); /* bz 9811 */
8099
8100         if (rc == 0 && release)
8101                 class_manual_cleanup(obd);
8102         RETURN(rc);
8103 }
8104
8105 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8106                        struct lu_device *dev)
8107 {
8108         struct osd_device *osd = osd_dev(dev);
8109         struct lr_server_data *lsd =
8110                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8111         int result = 0;
8112
8113         ENTRY;
8114
8115         if (osd->od_quota_slave_md != NULL) {
8116                 /* set up quota slave objects for inode */
8117                 result = qsd_prepare(env, osd->od_quota_slave_md);
8118                 if (result != 0)
8119                         RETURN(result);
8120         }
8121
8122         if (osd->od_quota_slave_dt != NULL) {
8123                 /* set up quota slave objects for block */
8124                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8125                 if (result != 0)
8126                         RETURN(result);
8127         }
8128
8129
8130         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8131 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8132                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8133                         osd->od_index_in_idif = 1;
8134                 } else {
8135                         osd->od_index_in_idif = 0;
8136                         result = osd_register_proc_index_in_idif(osd);
8137                         if (result != 0)
8138                                 RETURN(result);
8139                 }
8140 #else
8141                 osd->od_index_in_idif = 1;
8142 #endif
8143         }
8144
8145         result = osd_fid_init(env, osd);
8146
8147         RETURN(result);
8148 }
8149
8150 /**
8151  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8152  *
8153  * Allocate FID.
8154  *
8155  * see include/lu_object.h for the details.
8156  */
8157 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8158                          struct lu_fid *fid, struct lu_object *parent,
8159                          const struct lu_name *name)
8160 {
8161         struct osd_device *osd = osd_dev(d);
8162
8163         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8164 }
8165
8166 static const struct lu_object_operations osd_lu_obj_ops = {
8167         .loo_object_init      = osd_object_init,
8168         .loo_object_delete    = osd_object_delete,
8169         .loo_object_release   = osd_object_release,
8170         .loo_object_free      = osd_object_free,
8171         .loo_object_print     = osd_object_print,
8172         .loo_object_invariant = osd_object_invariant
8173 };
8174
8175 const struct lu_device_operations osd_lu_ops = {
8176         .ldo_object_alloc      = osd_object_alloc,
8177         .ldo_process_config    = osd_process_config,
8178         .ldo_recovery_complete = osd_recovery_complete,
8179         .ldo_prepare           = osd_prepare,
8180         .ldo_fid_alloc         = osd_fid_alloc,
8181 };
8182
8183 static const struct lu_device_type_operations osd_device_type_ops = {
8184         .ldto_init = osd_type_init,
8185         .ldto_fini = osd_type_fini,
8186
8187         .ldto_start = osd_type_start,
8188         .ldto_stop  = osd_type_stop,
8189
8190         .ldto_device_alloc = osd_device_alloc,
8191         .ldto_device_free  = osd_device_free,
8192
8193         .ldto_device_init = osd_device_init,
8194         .ldto_device_fini = osd_device_fini
8195 };
8196
8197 static struct lu_device_type osd_device_type = {
8198         .ldt_tags     = LU_DEVICE_DT,
8199         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8200         .ldt_ops      = &osd_device_type_ops,
8201         .ldt_ctx_tags = LCT_LOCAL,
8202 };
8203
8204 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8205 {
8206         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8207         struct super_block *sb = osd_sb(osd);
8208
8209         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8210 }
8211
8212 /*
8213  * lprocfs legacy support.
8214  */
8215 static const struct obd_ops osd_obd_device_ops = {
8216         .o_owner = THIS_MODULE,
8217         .o_connect      = osd_obd_connect,
8218         .o_disconnect   = osd_obd_disconnect,
8219         .o_health_check = osd_health_check,
8220 };
8221
8222 static ssize_t track_declares_assert_show(struct kobject *kobj,
8223                                    struct attribute *attr,
8224                                    char *buf)
8225 {
8226         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8227 }
8228
8229 static ssize_t track_declares_assert_store(struct kobject *kobj,
8230                                            struct attribute *attr,
8231                                            const char *buffer, size_t count)
8232 {
8233         bool track_declares_assert;
8234         int rc;
8235
8236         rc = kstrtobool(buffer, &track_declares_assert);
8237         if (rc)
8238                 return rc;
8239
8240         ldiskfs_track_declares_assert = track_declares_assert;
8241
8242         return count;
8243 }
8244 LUSTRE_RW_ATTR(track_declares_assert);
8245
8246 static int __init osd_init(void)
8247 {
8248         struct kobject *kobj;
8249         int rc;
8250
8251         BUILD_BUG_ON(BH_DXLock >=
8252                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8253 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8254         /* please, try to keep osd_thread_info smaller than a page */
8255         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8256 #endif
8257
8258         osd_oi_mod_init();
8259
8260         rc = lu_kmem_init(ldiskfs_caches);
8261         if (rc)
8262                 return rc;
8263
8264 #ifdef CONFIG_KALLSYMS
8265         priv_security_file_alloc =
8266                 (void *)kallsyms_lookup_name("security_file_alloc");
8267 #endif
8268
8269         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8270                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8271         if (rc) {
8272                 lu_kmem_fini(ldiskfs_caches);
8273                 return rc;
8274         }
8275
8276         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8277         if (kobj) {
8278                 rc = sysfs_create_file(kobj,
8279                                        &lustre_attr_track_declares_assert.attr);
8280                 kobject_put(kobj);
8281                 if (rc) {
8282                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8283                         rc = 0;
8284                 }
8285         }
8286         return rc;
8287 }
8288
8289 static void __exit osd_exit(void)
8290 {
8291         struct kobject *kobj;
8292
8293         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8294         if (kobj) {
8295                 sysfs_remove_file(kobj,
8296                                   &lustre_attr_track_declares_assert.attr);
8297                 kobject_put(kobj);
8298         }
8299         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8300         lu_kmem_fini(ldiskfs_caches);
8301 }
8302
8303 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8304 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8305 MODULE_VERSION(LUSTRE_VERSION_STRING);
8306 MODULE_LICENSE("GPL");
8307
8308 module_init(osd_init);
8309 module_exit(osd_exit);