lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         /* below message is checked in sanity.sh test_129 */
 499                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
 500                               osd_name(osd), parent->i_ino, PFID(fid),
 501                               rc == -ENOSPC ? "has reached" : "is approaching");
 502                 else
 503                         /* below message is checked in sanity.sh test_129 */
 504                         CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
 505                               osd_name(osd), parent->i_ino,
 506                               rc == -ENOSPC ? "has reached" : "is approaching");
 507
 508                 /* ignore such error now */
 509                 if (rc == -ENOBUFS)
 510                         rc = 0;
 511         }
 512
 513         return rc;
 514 }
 515
 516
 517 struct inode *
 518 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 519              struct osd_inode_id *id, struct lu_fid *fid)
 520 {
 521         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 522         struct inode *inode;
 523         int rc;
 524
 525         inode = osd_iget(info, dev, id);
 526         if (IS_ERR(inode))
 527                 return inode;
 528
 529         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 530         if (!rc) {
 531                 *fid = loa->loa_lma.lma_self_fid;
 532         } else if (rc == -ENODATA) {
 533                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 534                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 535                 else
 536                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 537         } else {
 538                 iput(inode);
 539                 inode = ERR_PTR(rc);
 540         }
 541         return inode;
 542 }
 543
 544 static struct inode *osd_iget_check(struct osd_thread_info *info,
 545                                     struct osd_device *dev,
 546                                     const struct lu_fid *fid,
 547                                     struct osd_inode_id *id,
 548                                     bool trusted)
 549 {
 550         struct inode *inode;
 551         int rc = 0;
 552
 553         ENTRY;
 554
 555         /*
 556          * The cached OI mapping is trustable. If we cannot locate the inode
 557          * via the cached OI mapping, then return the failure to the caller
 558          * directly without further OI checking.
 559          */
 560
 561 again:
 562         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_18_23, contains self-fid
 711  * \retval -v: other failure cases
 712  */
 713 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 714                  struct dentry *dentry, struct lu_fid *fid)
 715 {
 716         struct filter_fid *ff = &info->oti_ff;
 717         struct ost_id *ostid = &info->oti_ostid;
 718         int rc;
 719
 720         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 721         if (rc == sizeof(struct filter_fid_18_23)) {
 722                 struct filter_fid_18_23 *ff_old = (void *)ff;
 723
 724                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 725                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 726                 /*
 727                  * XXX: use 0 as the index for compatibility, the caller will
 728                  * handle index related issues when necessary.
 729                  */
 730                 if (!rc)
 731                         ostid_to_fid(fid, ostid, 0);
 732         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 733                 rc = 1;
 734         } else if (rc >= 0) {
 735                 rc = -EINVAL;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int osd_lma_self_repair(struct osd_thread_info *info,
 742                                struct osd_device *osd, struct inode *inode,
 743                                const struct lu_fid *fid, __u32 compat)
 744 {
 745         handle_t *jh;
 746         int rc;
 747
 748         LASSERT(current->journal_info == NULL);
 749
 750         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 751                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 752         if (IS_ERR(jh)) {
 753                 rc = PTR_ERR(jh);
 754                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 755                       osd_name(osd), rc);
 756                 return rc;
 757         }
 758
 759         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 760         if (rc != 0)
 761                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 762                       osd_name(osd), rc);
 763         ldiskfs_journal_stop(jh);
 764         return rc;
 765 }
 766
 767 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 768 {
 769         struct osd_thread_info *info = osd_oti_get(env);
 770         struct osd_device *osd = osd_obj2dev(obj);
 771         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 772         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 773         struct inode *inode = obj->oo_inode;
 774         struct dentry *dentry = &info->oti_obj_dentry;
 775         struct lu_fid *fid = NULL;
 776         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 777         int rc;
 778
 779         ENTRY;
 780
 781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 782                              (void *)loa, sizeof(*loa));
 783         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 784                 fid = &lma->lma_self_fid;
 785                 rc = osd_get_idif(info, inode, dentry, fid);
 786                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 787                         /*
 788                          * For the given OST-object, if it has neither LMA nor
 789                          * FID in XATTR_NAME_FID, then the given FID (which is
 790                          * contained in the @obj, from client RPC for locating
 791                          * the OST-object) is trusted. We use it to generate
 792                          * the LMA.
 793                          */
 794                         osd_lma_self_repair(info, osd, inode, rfid,
 795                                             LMAC_FID_ON_OST);
 796                         RETURN(0);
 797                 }
 798         }
 799
 800         if (rc < 0)
 801                 RETURN(rc);
 802
 803         if (rc > 0) {
 804                 rc = 0;
 805                 lustre_lma_swab(lma);
 806                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 807                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 808                               S_ISREG(inode->i_mode)))) {
 809                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 810                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 811                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 812                               PFID(rfid), inode->i_ino);
 813                         rc = -EOPNOTSUPP;
 814                 } else {
 815                         fid = &lma->lma_self_fid;
 816                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 817                             osd->od_is_ost)
 818                                 obj->oo_pfid_in_lma = 1;
 819                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 820                             !osd->od_is_ost)
 821                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 822                 }
 823         }
 824
 825         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 826                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 827                         struct ost_id   *oi   = &info->oti_ostid;
 828                         struct lu_fid   *fid1 = &info->oti_fid3;
 829                         __u32            idx  = fid_idif_ost_idx(rfid);
 830
 831                         /*
 832                          * For old IDIF, the OST index is not part of the IDIF,
 833                          * Means that different OSTs may have the same IDIFs.
 834                          * Under such case, we need to make some compatible
 835                          * check to make sure to trigger OI scrub properly.
 836                          */
 837                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 838                                 /* Given @rfid is new, LMA is old. */
 839                                 fid_to_ostid(fid, oi);
 840                                 ostid_to_fid(fid1, oi, idx);
 841                                 if (lu_fid_eq(fid1, rfid)) {
 842                                         if (osd->od_index_in_idif)
 843                                                 osd_lma_self_repair(info, osd,
 844                                                         inode, rfid,
 845                                                         LMAC_FID_ON_OST);
 846                                         RETURN(0);
 847                                 }
 848                         }
 849                 }
 850
 851                 rc = -EREMCHG;
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 struct osd_check_lmv_buf {
 858         /* please keep it as first member */
 859         struct dir_context ctx;
 860         struct osd_thread_info *oclb_info;
 861         struct osd_device *oclb_dev;
 862         struct osd_idmap_cache *oclb_oic;
 863         int oclb_items;
 864         bool oclb_found;
 865 };
 866
 867 /**
 868  * It is called internally by ->iterate*() to filter out the
 869  * local slave object's FID of the striped directory.
 870  *
 871  * \retval      1 found the local slave's FID
 872  * \retval      0 continue to check next item
 873  * \retval      -ve for failure
 874  */
 875 #ifdef HAVE_FILLDIR_USE_CTX
 876 static int osd_stripe_dir_filldir(struct dir_context *buf,
 877 #else
 878 static int osd_stripe_dir_filldir(void *buf,
 879 #endif
 880                                   const char *name, int namelen,
 881                                   loff_t offset, __u64 ino, unsigned int d_type)
 882 {
 883         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 884         struct osd_thread_info *oti = oclb->oclb_info;
 885         struct lu_fid *fid = &oti->oti_fid3;
 886         struct osd_inode_id *id = &oti->oti_id3;
 887         struct osd_device *dev = oclb->oclb_dev;
 888         struct osd_idmap_cache *oic = oclb->oclb_oic;
 889         struct inode *inode;
 890
 891         oclb->oclb_items++;
 892
 893         if (name[0] == '.')
 894                 return 0;
 895
 896         fid_zero(fid);
 897         sscanf(name + 1, SFID, RFID(fid));
 898         if (!fid_is_sane(fid))
 899                 return 0;
 900
 901         if (osd_remote_fid(oti->oti_env, dev, fid))
 902                 return 0;
 903
 904         osd_id_gen(id, ino, OSD_OII_NOGEN);
 905         inode = osd_iget(oti, dev, id);
 906         if (IS_ERR(inode))
 907                 return PTR_ERR(inode);
 908
 909         iput(inode);
 910         osd_add_oi_cache(oti, dev, id, fid);
 911         oic->oic_fid = *fid;
 912         oic->oic_lid = *id;
 913         oic->oic_dev = dev;
 914         osd_oii_insert(dev, oic, true);
 915         oclb->oclb_found = true;
 916
 917         return 1;
 918 }
 919
 920 /*
 921  * When lookup item under striped directory, we need to locate the master
 922  * MDT-object of the striped directory firstly, then the client will send
 923  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 924  * and the item's name. If the system is restored from MDT file level backup,
 925  * then before the OI scrub completely built the OI files, the OI mappings of
 926  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 927  * not a problem for the master MDT-object. Because when locate the master
 928  * MDT-object, we will do name based lookup (for the striped directory itself)
 929  * firstly, during such process we can setup the correct OI mapping for the
 930  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 931  * the client will not trigger name based lookup on the MDT to locate the slave
 932  * MDT-object before locating item under the striped directory, then when
 933  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 934  * is invalid and does not know what the right OI mapping is, then the MDT has
 935  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 936  * the OI file, related OI mapping is unknown yet, please try again later. And
 937  * then client will re-try the RPC again and again until related OI mapping has
 938  * been updated. That is quite inefficient.
 939  *
 940  * To resolve above trouble, we will handle it as the following two cases:
 941  *
 942  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 943  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 944  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 945  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 946  *    directly. Please check osd_fid_lookup().
 947  *
 948  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 949  *    Under such case, during lookup the master MDT-object, we will lookup the
 950  *    slave MDT-object via readdir against the master MDT-object, because the
 951  *    slave MDT-objects information are stored as sub-directories with the name
 952  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 953  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 954  *    the correct OI mapping for the slave MDT-object.
 955  */
 956 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 957                          struct inode *inode, struct osd_idmap_cache *oic)
 958 {
 959         struct lu_buf *buf = &oti->oti_big_buf;
 960         struct dentry *dentry = &oti->oti_obj_dentry;
 961         struct file *filp = &oti->oti_file;
 962         const struct file_operations *fops;
 963         struct lmv_mds_md_v1 *lmv1;
 964         struct osd_check_lmv_buf oclb = {
 965                 .ctx.actor = osd_stripe_dir_filldir,
 966                 .oclb_info = oti,
 967                 .oclb_dev = dev,
 968                 .oclb_oic = oic,
 969                 .oclb_found = false,
 970         };
 971         int rc = 0;
 972
 973         ENTRY;
 974
 975 again:
 976         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 977                              buf->lb_len);
 978         if (rc == -ERANGE) {
 979                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 980                 if (rc > 0) {
 981                         lu_buf_realloc(buf, rc);
 982                         if (buf->lb_buf == NULL)
 983                                 GOTO(out, rc = -ENOMEM);
 984
 985                         goto again;
 986                 }
 987         }
 988
 989         if (unlikely(rc == 0 || rc == -ENODATA))
 990                 GOTO(out, rc = 0);
 991
 992         if (rc < 0)
 993                 GOTO(out, rc);
 994
 995         if (unlikely(buf->lb_buf == NULL)) {
 996                 lu_buf_realloc(buf, rc);
 997                 if (buf->lb_buf == NULL)
 998                         GOTO(out, rc = -ENOMEM);
 999
1000                 goto again;
1001         }
1002
1003         lmv1 = buf->lb_buf;
1004         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1005                 GOTO(out, rc = 0);
1006
1007         fops = inode->i_fop;
1008         dentry->d_inode = inode;
1009         dentry->d_sb = inode->i_sb;
1010         filp->f_pos = 0;
1011         filp->f_path.dentry = dentry;
1012         filp->f_flags |= O_NOATIME;
1013         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1014         filp->f_mapping = inode->i_mapping;
1015         filp->f_op = fops;
1016         filp->private_data = NULL;
1017         set_file_inode(filp, inode);
1018         rc = osd_security_file_alloc(filp);
1019         if (rc)
1020                 goto out;
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024                 rc = iterate_dir(filp, &oclb.ctx);
1025         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1026                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1027         fops->release(inode, filp);
1028
1029 out:
1030         if (rc < 0)
1031                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1032                        DFID": rc = %d\n", osd_ino2name(inode),
1033                        inode->i_ino, inode->i_generation,
1034                        PFID(&oic->oic_fid), rc);
1035         else
1036                 rc = 0;
1037
1038         RETURN(rc);
1039 }
1040
1041 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1042                           const struct lu_fid *fid,
1043                           const struct lu_object_conf *conf)
1044 {
1045         struct osd_thread_info *info;
1046         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1047         struct osd_device *dev;
1048         struct osd_idmap_cache *oic;
1049         struct osd_inode_id *id;
1050         struct inode *inode = NULL;
1051         struct lustre_scrub *scrub;
1052         struct scrub_file *sf;
1053         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1054         __u32 saved_ino;
1055         __u32 saved_gen;
1056         int result = 0;
1057         int rc1 = 0;
1058         bool remote = false;
1059         bool trusted = true;
1060         bool updated = false;
1061         bool checked = false;
1062
1063         ENTRY;
1064
1065         LINVRNT(osd_invariant(obj));
1066         LASSERT(obj->oo_inode == NULL);
1067         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1068
1069         dev = osd_dev(ldev);
1070         scrub = &dev->od_scrub.os_scrub;
1071         sf = &scrub->os_file;
1072         info = osd_oti_get(env);
1073         LASSERT(info);
1074         oic = &info->oti_cache;
1075
1076         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1077                 RETURN(-ENOENT);
1078
1079         /*
1080          * For the object is created as locking anchor, or for the object to
1081          * be created on disk. No need to osd_oi_lookup() at here because FID
1082          * shouldn't never be re-used, if it's really a duplicate FID from
1083          * unexpected reason, we should be able to detect it later by calling
1084          * do_create->osd_oi_insert().
1085          */
1086         if (conf && conf->loc_flags & LOC_F_NEW)
1087                 GOTO(out, result = 0);
1088
1089         /* Search order: 1. per-thread cache. */
1090         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1091                 id = &oic->oic_lid;
1092                 goto iget;
1093         }
1094
1095         id = &info->oti_id;
1096         if (!list_empty(&scrub->os_inconsistent_items)) {
1097                 /* Search order: 2. OI scrub pending list. */
1098                 result = osd_oii_lookup(dev, fid, id);
1099                 if (!result)
1100                         goto iget;
1101         }
1102
1103         /*
1104          * The OI mapping in the OI file can be updated by the OI scrub
1105          * when we locate the inode via FID. So it may be not trustable.
1106          */
1107         trusted = false;
1108
1109         /* Search order: 3. OI files. */
1110         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1111         if (result == -ENOENT) {
1112                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1113                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1114                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1115                                       sf->sf_oi_bitmap))
1116                         GOTO(out, result = 0);
1117
1118                 goto trigger;
1119         }
1120
1121         /* -ESTALE is returned if inode of OST object doesn't exist */
1122         if (result == -ESTALE &&
1123             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1124                 GOTO(out, result = 0);
1125         }
1126
1127         if (result)
1128                 GOTO(out, result);
1129
1130 iget:
1131         obj->oo_inode = NULL;
1132         /* for later passes through checks, not true on first pass */
1133         if (!IS_ERR_OR_NULL(inode))
1134                 iput(inode);
1135
1136         inode = osd_iget_check(info, dev, fid, id, trusted);
1137         if (!IS_ERR(inode)) {
1138                 obj->oo_inode = inode;
1139                 result = 0;
1140                 if (remote)
1141                         goto trigger;
1142
1143                 goto check_lma;
1144         }
1145
1146         result = PTR_ERR(inode);
1147         if (result == -ENOENT || result == -ESTALE)
1148                 GOTO(out, result = 0);
1149
1150         if (result != -EREMCHG)
1151                 GOTO(out, result);
1152
1153 trigger:
1154         /*
1155          * We still have chance to get the valid inode: for the
1156          * object which is referenced by remote name entry, the
1157          * object on the local MDT will be linked under the dir
1158          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1159          *
1160          * We do not know whether the object for the given FID
1161          * is referenced by some remote name entry or not, and
1162          * especially for DNE II, a multiple-linked object may
1163          * have many name entries reside on many MDTs.
1164          *
1165          * To simplify the operation, OSD will not distinguish
1166          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1167          * only happened for the RPC from other MDT during the
1168          * OI scrub, or for the client side RPC with FID only,
1169          * such as FID to path, or from old connected client.
1170          */
1171         if (!remote) {
1172                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1173                 if (!rc1) {
1174                         remote = true;
1175                         trusted = true;
1176                         flags |= SS_AUTO_PARTIAL;
1177                         flags &= ~SS_AUTO_FULL;
1178                         goto iget;
1179                 }
1180         }
1181
1182         if (thread_is_running(&scrub->os_thread)) {
1183                 if (scrub->os_partial_scan && !scrub->os_in_join)
1184                         goto join;
1185
1186                 if (IS_ERR_OR_NULL(inode) || result)
1187                         GOTO(out, result = -EINPROGRESS);
1188
1189                 LASSERT(remote);
1190                 LASSERT(obj->oo_inode == inode);
1191
1192                 osd_add_oi_cache(info, dev, id, fid);
1193                 osd_oii_insert(dev, oic, true);
1194                 goto found;
1195         }
1196
1197         if (dev->od_auto_scrub_interval == AS_NEVER) {
1198                 if (!remote)
1199                         GOTO(out, result = -EREMCHG);
1200
1201                 LASSERT(!result);
1202                 LASSERT(obj->oo_inode == inode);
1203
1204                 osd_add_oi_cache(info, dev, id, fid);
1205                 goto found;
1206         }
1207
1208 join:
1209         rc1 = osd_scrub_start(env, dev, flags);
1210         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1211                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1212         if (rc1 && rc1 != -EALREADY)
1213                 GOTO(out, result = -EREMCHG);
1214
1215         if (IS_ERR_OR_NULL(inode) || result)
1216                 GOTO(out, result = -EINPROGRESS);
1217
1218         LASSERT(remote);
1219         LASSERT(obj->oo_inode == inode);
1220
1221         osd_add_oi_cache(info, dev, id, fid);
1222         osd_oii_insert(dev, oic, true);
1223         goto found;
1224
1225 check_lma:
1226         checked = true;
1227         if (unlikely(obj->oo_header))
1228                 goto found;
1229
1230         result = osd_check_lma(env, obj);
1231         if (!result)
1232                 goto found;
1233
1234         LASSERTF(id->oii_ino == inode->i_ino &&
1235                  id->oii_gen == inode->i_generation,
1236                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1237                  PFID(fid), id->oii_ino, id->oii_gen,
1238                  inode->i_ino, inode->i_generation);
1239
1240         saved_ino = inode->i_ino;
1241         saved_gen = inode->i_generation;
1242
1243         if (unlikely(result == -ENODATA)) {
1244                 /*
1245                  * If the OI scrub updated the OI mapping by race, it
1246                  * must be valid. Trust the inode that has no LMA EA.
1247                  */
1248                 if (updated)
1249                         goto found;
1250
1251                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1252                 if (!result) {
1253                         /*
1254                          * The OI mapping is still there, the inode is still
1255                          * valid. It is just becaues the inode has no LMA EA.
1256                          */
1257                         if (saved_ino == id->oii_ino &&
1258                             saved_gen == id->oii_gen)
1259                                 goto found;
1260
1261                         /*
1262                          * It is the OI scrub updated the OI mapping by race.
1263                          * The new OI mapping must be valid.
1264                          */
1265                         trusted = true;
1266                         updated = true;
1267                         goto iget;
1268                 }
1269
1270                 /*
1271                  * "result == -ENOENT" means that the OI mappinghas been
1272                  * removed by race, so the inode belongs to other object.
1273                  *
1274                  * Others error can be returned  directly.
1275                  */
1276                 if (result == -ENOENT) {
1277                         LASSERT(trusted);
1278
1279                         obj->oo_inode = NULL;
1280                         result = 0;
1281                 }
1282         }
1283
1284         if (result != -EREMCHG)
1285                 GOTO(out, result);
1286
1287         LASSERT(!updated);
1288
1289         /*
1290          * if two OST objects map to the same inode, and inode mode is
1291          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1292          * reserved by precreate, and not written yet, in this case, don't
1293          * set inode for the object whose FID mismatch, so that it can create
1294          * inode and not block precreate.
1295          */
1296         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1297             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1298                 obj->oo_inode = NULL;
1299                 GOTO(out, result = 0);
1300         }
1301
1302         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1303         /*
1304          * "result == -ENOENT" means the cached OI mapping has been removed
1305          * from the OI file by race, above inode belongs to other object.
1306          */
1307         if (result == -ENOENT) {
1308                 LASSERT(trusted);
1309
1310                 obj->oo_inode = NULL;
1311                 GOTO(out, result = 0);
1312         }
1313
1314         if (result)
1315                 GOTO(out, result);
1316
1317         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1318                 result = -EREMCHG;
1319                 goto trigger;
1320         }
1321
1322         /*
1323          * It is the OI scrub updated the OI mapping by race.
1324          * The new OI mapping must be valid.
1325          */
1326         trusted = true;
1327         updated = true;
1328         goto iget;
1329
1330 found:
1331         if (!checked) {
1332                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1333                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1334
1335                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1336                 if (!result) {
1337                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1338                             dev->od_is_ost)
1339                                 obj->oo_pfid_in_lma = 1;
1340                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1341                             !dev->od_is_ost)
1342                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1343                 } else if (result != -ENODATA) {
1344                         GOTO(out, result);
1345                 }
1346         }
1347
1348         obj->oo_compat_dot_created = 1;
1349         obj->oo_compat_dotdot_created = 1;
1350
1351         if (S_ISDIR(inode->i_mode) &&
1352             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1353                 osd_check_lmv(info, dev, inode, oic);
1354
1355         result = osd_attach_jinode(inode);
1356         if (result)
1357                 GOTO(out, result);
1358
1359         if (!ldiskfs_pdo)
1360                 GOTO(out, result = 0);
1361
1362         LASSERT(!obj->oo_hl_head);
1363         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1364
1365         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1366
1367 out:
1368         if (result || !obj->oo_inode) {
1369                 if (!IS_ERR_OR_NULL(inode))
1370                         iput(inode);
1371
1372                 obj->oo_inode = NULL;
1373                 if (trusted)
1374                         fid_zero(&oic->oic_fid);
1375         }
1376
1377         LINVRNT(osd_invariant(obj));
1378         return result;
1379 }
1380
1381 /*
1382  * Concurrency: shouldn't matter.
1383  */
1384 static void osd_object_init0(struct osd_object *obj)
1385 {
1386         LASSERT(obj->oo_inode != NULL);
1387         obj->oo_dt.do_body_ops = &osd_body_ops;
1388         obj->oo_dt.do_lu.lo_header->loh_attr |=
1389                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1390 }
1391
1392 /*
1393  * Concurrency: no concurrent access is possible that early in object
1394  * life-cycle.
1395  */
1396 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1397                            const struct lu_object_conf *conf)
1398 {
1399         struct osd_object *obj = osd_obj(l);
1400         int result;
1401
1402         LINVRNT(osd_invariant(obj));
1403
1404         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1405                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1406                 l->lo_header->loh_attr |= LOHA_EXISTS;
1407                 return 0;
1408         }
1409
1410         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1411         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1412         if (result == 0 && obj->oo_inode != NULL) {
1413                 struct osd_thread_info *oti = osd_oti_get(env);
1414                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1415
1416                 osd_object_init0(obj);
1417                 if (unlikely(obj->oo_header))
1418                         return 0;
1419
1420                 result = osd_get_lma(oti, obj->oo_inode,
1421                                      &oti->oti_obj_dentry, loa);
1422                 if (!result) {
1423                         /*
1424                          * Convert LMAI flags to lustre LMA flags
1425                          * and cache it to oo_lma_flags
1426                          */
1427                         obj->oo_lma_flags =
1428                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1429                 } else if (result == -ENODATA) {
1430                         result = 0;
1431                 }
1432         }
1433
1434         LINVRNT(osd_invariant(obj));
1435         return result;
1436 }
1437
1438 /*
1439  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1440  * The left part is for value, binary mode.
1441  */
1442 struct osd_xattr_entry {
1443         struct list_head        oxe_list;
1444         size_t                  oxe_len;
1445         size_t                  oxe_namelen;
1446         bool                    oxe_exist;
1447         struct rcu_head         oxe_rcu;
1448         char                    oxe_buf[0];
1449 };
1450
1451 static int osd_oxc_get(struct osd_object *obj, const char *name,
1452                        struct lu_buf *buf)
1453 {
1454         struct osd_xattr_entry *tmp;
1455         struct osd_xattr_entry *oxe = NULL;
1456         size_t namelen = strlen(name);
1457         int rc;
1458
1459         ENTRY;
1460
1461         rcu_read_lock();
1462         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1463                 if (namelen == tmp->oxe_namelen &&
1464                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1465                         oxe = tmp;
1466                         break;
1467                 }
1468         }
1469
1470         if (oxe == NULL)
1471                 GOTO(out, rc = -ENOENT);
1472
1473         if (!oxe->oxe_exist)
1474                 GOTO(out, rc = -ENODATA);
1475
1476         /* vallen */
1477         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1478         LASSERT(rc > 0);
1479
1480         if (buf->lb_buf == NULL)
1481                 GOTO(out, rc);
1482
1483         if (buf->lb_len < rc)
1484                 GOTO(out, rc = -ERANGE);
1485
1486         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1487         EXIT;
1488 out:
1489         rcu_read_unlock();
1490
1491         return rc;
1492 }
1493
1494 static void osd_oxc_free(struct rcu_head *head)
1495 {
1496         struct osd_xattr_entry *oxe;
1497
1498         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1499         OBD_FREE(oxe, oxe->oxe_len);
1500 }
1501
1502 static void osd_oxc_add(struct osd_object *obj, const char *name,
1503                         const char *buf, int buflen)
1504 {
1505         struct osd_xattr_entry *oxe;
1506         struct osd_xattr_entry *old = NULL;
1507         struct osd_xattr_entry *tmp;
1508         size_t namelen = strlen(name);
1509         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1510
1511         OBD_ALLOC(oxe, len);
1512         if (oxe == NULL)
1513                 return;
1514
1515         INIT_LIST_HEAD(&oxe->oxe_list);
1516         oxe->oxe_len = len;
1517         oxe->oxe_namelen = namelen;
1518         memcpy(oxe->oxe_buf, name, namelen);
1519         if (buflen > 0) {
1520                 LASSERT(buf != NULL);
1521                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1522                 oxe->oxe_exist = true;
1523         } else {
1524                 oxe->oxe_exist = false;
1525         }
1526
1527         /* this should be rarely called, just remove old and add new */
1528         spin_lock(&obj->oo_guard);
1529         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1530                 if (namelen == tmp->oxe_namelen &&
1531                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1532                         old = tmp;
1533                         break;
1534                 }
1535         }
1536         if (old != NULL) {
1537                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1538                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1539         } else {
1540                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1541         }
1542         spin_unlock(&obj->oo_guard);
1543 }
1544
1545 static void osd_oxc_del(struct osd_object *obj, const char *name)
1546 {
1547         struct osd_xattr_entry *oxe;
1548         size_t namelen = strlen(name);
1549
1550         spin_lock(&obj->oo_guard);
1551         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1552                 if (namelen == oxe->oxe_namelen &&
1553                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1554                         list_del_rcu(&oxe->oxe_list);
1555                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1556                         break;
1557                 }
1558         }
1559         spin_unlock(&obj->oo_guard);
1560 }
1561
1562 static void osd_oxc_fini(struct osd_object *obj)
1563 {
1564         struct osd_xattr_entry *oxe, *next;
1565
1566         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1567                 list_del(&oxe->oxe_list);
1568                 OBD_FREE(oxe, oxe->oxe_len);
1569         }
1570 }
1571
1572 /*
1573  * Concurrency: no concurrent access is possible that late in object
1574  * life-cycle.
1575  */
1576 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1577 {
1578         struct osd_object *obj = osd_obj(l);
1579         struct lu_object_header *h = obj->oo_header;
1580
1581         LINVRNT(osd_invariant(obj));
1582
1583         osd_oxc_fini(obj);
1584         dt_object_fini(&obj->oo_dt);
1585         if (obj->oo_hl_head != NULL)
1586                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1587         OBD_FREE_PTR(obj);
1588         if (unlikely(h)) {
1589                 lu_object_header_fini(h);
1590                 OBD_FREE_PTR(h);
1591         }
1592 }
1593
1594 /*
1595  * Concurrency: no concurrent access is possible that late in object
1596  * life-cycle.
1597  */
1598 static void osd_index_fini(struct osd_object *o)
1599 {
1600         struct iam_container *bag;
1601
1602         if (o->oo_dir != NULL) {
1603                 bag = &o->oo_dir->od_container;
1604                 if (o->oo_inode != NULL) {
1605                         if (bag->ic_object == o->oo_inode)
1606                                 iam_container_fini(bag);
1607                 }
1608                 OBD_FREE_PTR(o->oo_dir);
1609                 o->oo_dir = NULL;
1610         }
1611 }
1612
1613 /*
1614  * Concurrency: no concurrent access is possible that late in object
1615  * life-cycle (for all existing callers, that is. New callers have to provide
1616  * their own locking.)
1617  */
1618 static int osd_inode_unlinked(const struct inode *inode)
1619 {
1620         return inode->i_nlink == 0;
1621 }
1622
1623 enum {
1624         OSD_TXN_OI_DELETE_CREDITS    = 20,
1625         OSD_TXN_INODE_DELETE_CREDITS = 20
1626 };
1627
1628 /*
1629  * Journal
1630  */
1631
1632 #if OSD_THANDLE_STATS
1633 /**
1634  * Set time when the handle is allocated
1635  */
1636 static void osd_th_alloced(struct osd_thandle *oth)
1637 {
1638         oth->oth_alloced = ktime_get();
1639 }
1640
1641 /**
1642  * Set time when the handle started
1643  */
1644 static void osd_th_started(struct osd_thandle *oth)
1645 {
1646         oth->oth_started = ktime_get();
1647 }
1648
1649 /**
1650  * Check whether the we deal with this handle for too long.
1651  */
1652 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1653                                 ktime_t alloced, ktime_t started,
1654                                 ktime_t closed)
1655 {
1656         ktime_t now = ktime_get();
1657
1658         LASSERT(dev != NULL);
1659
1660         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1661                             ktime_us_delta(started, alloced));
1662         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1663                             ktime_us_delta(closed, started));
1664         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1665                             ktime_us_delta(now, closed));
1666
1667         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1668                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1669                                 oth, now, alloced, started, closed);
1670                 libcfs_debug_dumpstack(NULL);
1671         }
1672 }
1673
1674 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1675 {                                                                       \
1676         ktime_t __closed = ktime_get();                                 \
1677         ktime_t __alloced = oth->oth_alloced;                           \
1678         ktime_t __started = oth->oth_started;                           \
1679                                                                         \
1680         expr;                                                           \
1681         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1682 }
1683
1684 #else /* OSD_THANDLE_STATS */
1685
1686 #define osd_th_alloced(h)                  do {} while(0)
1687 #define osd_th_started(h)                  do {} while(0)
1688 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1689
1690 #endif /* OSD_THANDLE_STATS */
1691
1692 /*
1693  * Concurrency: doesn't access mutable data.
1694  */
1695 static int osd_param_is_not_sane(const struct osd_device *dev,
1696                                  const struct thandle *th)
1697 {
1698         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1699
1700         return oh->ot_credits > osd_transaction_size(dev);
1701 }
1702
1703 /*
1704  * Concurrency: shouldn't matter.
1705  */
1706 static void osd_trans_commit_cb(struct super_block *sb,
1707                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1708 {
1709         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1710         struct thandle *th = &oh->ot_super;
1711         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1712         struct dt_txn_commit_cb *dcb, *tmp;
1713
1714         LASSERT(oh->ot_handle == NULL);
1715
1716         if (error)
1717                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1718
1719         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1720         /* call per-transaction callbacks if any */
1721         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1722                                  dcb_linkage) {
1723                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1724                          "commit callback entry: magic=%x name='%s'\n",
1725                          dcb->dcb_magic, dcb->dcb_name);
1726                 list_del_init(&dcb->dcb_linkage);
1727                 dcb->dcb_func(NULL, th, dcb, error);
1728         }
1729
1730         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1731         lu_device_put(lud);
1732         th->th_dev = NULL;
1733
1734         OBD_FREE_PTR(oh);
1735 }
1736
1737 #ifndef HAVE_SB_START_WRITE
1738 # define sb_start_write(sb) do {} while (0)
1739 # define sb_end_write(sb) do {} while (0)
1740 #endif
1741
1742 static struct thandle *osd_trans_create(const struct lu_env *env,
1743                                         struct dt_device *d)
1744 {
1745         struct osd_thread_info *oti = osd_oti_get(env);
1746         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1747         struct osd_thandle *oh;
1748         struct thandle *th;
1749
1750         ENTRY;
1751
1752         if (d->dd_rdonly) {
1753                 CERROR("%s: someone try to start transaction under "
1754                        "readonly mode, should be disabled.\n",
1755                        osd_name(osd_dt_dev(d)));
1756                 dump_stack();
1757                 RETURN(ERR_PTR(-EROFS));
1758         }
1759
1760         /* on pending IO in this thread should left from prev. request */
1761         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1762
1763         sb_start_write(osd_sb(osd_dt_dev(d)));
1764
1765         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1766         if (!oh) {
1767                 sb_end_write(osd_sb(osd_dt_dev(d)));
1768                 RETURN(ERR_PTR(-ENOMEM));
1769         }
1770
1771         oh->ot_quota_trans = &oti->oti_quota_trans;
1772         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1773         th = &oh->ot_super;
1774         th->th_dev = d;
1775         th->th_result = 0;
1776         oh->ot_credits = 0;
1777         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1778         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1779         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1780         osd_th_alloced(oh);
1781
1782         memset(oti->oti_declare_ops, 0,
1783                sizeof(oti->oti_declare_ops));
1784         memset(oti->oti_declare_ops_cred, 0,
1785                sizeof(oti->oti_declare_ops_cred));
1786         memset(oti->oti_declare_ops_used, 0,
1787                sizeof(oti->oti_declare_ops_used));
1788
1789         oti->oti_ins_cache_depth++;
1790
1791         RETURN(th);
1792 }
1793
1794 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1795 {
1796         struct osd_thread_info *oti = osd_oti_get(env);
1797         struct osd_thandle *oh;
1798
1799         oh = container_of0(th, struct osd_thandle, ot_super);
1800         LASSERT(oh != NULL);
1801
1802         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1803               oti->oti_declare_ops[OSD_OT_CREATE],
1804               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1805               oti->oti_declare_ops_used[OSD_OT_CREATE],
1806               oti->oti_declare_ops[OSD_OT_DESTROY],
1807               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1808               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1809         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1810               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1811               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1812               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1813               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1814               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1815               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1816         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1817               oti->oti_declare_ops[OSD_OT_WRITE],
1818               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1819               oti->oti_declare_ops_used[OSD_OT_WRITE],
1820               oti->oti_declare_ops[OSD_OT_PUNCH],
1821               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1822               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1823               oti->oti_declare_ops[OSD_OT_QUOTA],
1824               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1825               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1826         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1827               oti->oti_declare_ops[OSD_OT_INSERT],
1828               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1829               oti->oti_declare_ops_used[OSD_OT_INSERT],
1830               oti->oti_declare_ops[OSD_OT_DELETE],
1831               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1832               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1833         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1834               oti->oti_declare_ops[OSD_OT_REF_ADD],
1835               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1836               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1837               oti->oti_declare_ops[OSD_OT_REF_DEL],
1838               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1839               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1840 }
1841
1842 /*
1843  * Concurrency: shouldn't matter.
1844  */
1845 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1846                            struct thandle *th)
1847 {
1848         struct osd_thread_info *oti = osd_oti_get(env);
1849         struct osd_device *dev = osd_dt_dev(d);
1850         handle_t *jh;
1851         struct osd_thandle *oh;
1852         int rc;
1853
1854         ENTRY;
1855
1856         LASSERT(current->journal_info == NULL);
1857
1858         oh = container_of0(th, struct osd_thandle, ot_super);
1859         LASSERT(oh != NULL);
1860         LASSERT(oh->ot_handle == NULL);
1861
1862         rc = dt_txn_hook_start(env, d, th);
1863         if (rc != 0)
1864                 GOTO(out, rc);
1865
1866         if (unlikely(osd_param_is_not_sane(dev, th))) {
1867                 static unsigned long last_printed;
1868                 static int last_credits;
1869
1870                 /*
1871                  * don't make noise on a tiny testing systems
1872                  * actual credits misuse will be caught anyway
1873                  */
1874                 if (last_credits != oh->ot_credits &&
1875                     time_after(jiffies, last_printed +
1876                                cfs_time_seconds(60)) &&
1877                     osd_transaction_size(dev) > 512) {
1878                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1879                               oh->ot_credits, osd_transaction_size(dev));
1880                         osd_trans_dump_creds(env, th);
1881                         libcfs_debug_dumpstack(NULL);
1882                         last_credits = oh->ot_credits;
1883                         last_printed = jiffies;
1884                 }
1885                 /*
1886                  * XXX Limit the credits to 'max_transaction_buffers', and
1887                  *     let the underlying filesystem to catch the error if
1888                  *     we really need so many credits.
1889                  *
1890                  *     This should be removed when we can calculate the
1891                  *     credits precisely.
1892                  */
1893                 oh->ot_credits = osd_transaction_size(dev);
1894         } else if (ldiskfs_track_declares_assert != 0) {
1895                 /*
1896                  * reserve few credits to prevent an assertion in JBD
1897                  * our debugging mechanism will be able to detected
1898                  * overuse. this can help to debug single-update
1899                  * transactions
1900                  */
1901                 oh->ot_credits += 10;
1902                 if (unlikely(osd_param_is_not_sane(dev, th)))
1903                         oh->ot_credits = osd_transaction_size(dev);
1904         }
1905
1906         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1907                 GOTO(out, rc = -EIO);
1908
1909         /*
1910          * XXX temporary stuff. Some abstraction layer should
1911          * be used.
1912          */
1913         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1914         osd_th_started(oh);
1915         if (!IS_ERR(jh)) {
1916                 oh->ot_handle = jh;
1917                 LASSERT(oti->oti_txns == 0);
1918
1919                 lu_device_get(&d->dd_lu_dev);
1920                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1921                               "osd-tx", th);
1922                 oti->oti_txns++;
1923                 rc = 0;
1924         } else {
1925                 rc = PTR_ERR(jh);
1926         }
1927 out:
1928         RETURN(rc);
1929 }
1930
1931 static int osd_seq_exists(const struct lu_env *env,
1932                           struct osd_device *osd, u64 seq)
1933 {
1934         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1935         struct seq_server_site *ss = osd_seq_site(osd);
1936         int rc;
1937
1938         ENTRY;
1939
1940         LASSERT(ss != NULL);
1941         LASSERT(ss->ss_server_fld != NULL);
1942
1943         rc = osd_fld_lookup(env, osd, seq, range);
1944         if (rc != 0) {
1945                 if (rc != -ENOENT)
1946                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1947                                osd_name(osd), seq, rc);
1948                 RETURN(0);
1949         }
1950
1951         RETURN(ss->ss_node_id == range->lsr_index);
1952 }
1953
1954 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1955 {
1956         struct dt_txn_commit_cb *dcb;
1957         struct dt_txn_commit_cb *tmp;
1958
1959         /* call per-transaction stop callbacks if any */
1960         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1961                                  dcb_linkage) {
1962                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1963                          "commit callback entry: magic=%x name='%s'\n",
1964                          dcb->dcb_magic, dcb->dcb_name);
1965                 list_del_init(&dcb->dcb_linkage);
1966                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1967         }
1968 }
1969
1970 /*
1971  * Concurrency: shouldn't matter.
1972  */
1973 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1974                           struct thandle *th)
1975 {
1976         struct osd_thread_info *oti = osd_oti_get(env);
1977         struct osd_thandle *oh;
1978         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1979         struct osd_device *osd = osd_dt_dev(th->th_dev);
1980         struct qsd_instance *qsd = osd_def_qsd(osd);
1981         struct lquota_trans *qtrans;
1982         LIST_HEAD(truncates);
1983         int rc = 0, remove_agents = 0;
1984
1985         ENTRY;
1986
1987         oh = container_of0(th, struct osd_thandle, ot_super);
1988
1989         remove_agents = oh->ot_remove_agents;
1990
1991         qtrans = oh->ot_quota_trans;
1992         oh->ot_quota_trans = NULL;
1993
1994         /* move locks to local list, stop tx, execute truncates */
1995         list_splice(&oh->ot_trunc_locks, &truncates);
1996
1997         if (oh->ot_handle != NULL) {
1998                 int rc2;
1999
2000                 handle_t *hdl = oh->ot_handle;
2001
2002                 /*
2003                  * add commit callback
2004                  * notice we don't do this in osd_trans_start()
2005                  * as underlying transaction can change during truncate
2006                  */
2007                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2008                                              &oh->ot_jcb);
2009
2010                 LASSERT(oti->oti_txns == 1);
2011                 oti->oti_txns--;
2012
2013                 rc = dt_txn_hook_stop(env, th);
2014                 if (rc != 0)
2015                         CERROR("%s: failed in transaction hook: rc = %d\n",
2016                                osd_name(osd), rc);
2017
2018                 osd_trans_stop_cb(oh, rc);
2019                 /* hook functions might modify th_sync */
2020                 hdl->h_sync = th->th_sync;
2021
2022                 oh->ot_handle = NULL;
2023                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2024                 if (rc2 != 0)
2025                         CERROR("%s: failed to stop transaction: rc = %d\n",
2026                                osd_name(osd), rc2);
2027                 if (!rc)
2028                         rc = rc2;
2029
2030                 osd_process_truncates(&truncates);
2031         } else {
2032                 osd_trans_stop_cb(oh, th->th_result);
2033                 OBD_FREE_PTR(oh);
2034         }
2035
2036         osd_trunc_unlock_all(&truncates);
2037
2038         /* inform the quota slave device that the transaction is stopping */
2039         qsd_op_end(env, qsd, qtrans);
2040
2041         /*
2042          * as we want IO to journal and data IO be concurrent, we don't block
2043          * awaiting data IO completion in osd_do_bio(), instead we wait here
2044          * once transaction is submitted to the journal. all reqular requests
2045          * don't do direct IO (except read/write), thus this wait_event becomes
2046          * no-op for them.
2047          *
2048          * IMPORTANT: we have to wait till any IO submited by the thread is
2049          * completed otherwise iobuf may be corrupted by different request
2050          */
2051         wait_event(iobuf->dr_wait,
2052                        atomic_read(&iobuf->dr_numreqs) == 0);
2053         osd_fini_iobuf(osd, iobuf);
2054         if (!rc)
2055                 rc = iobuf->dr_error;
2056
2057         if (unlikely(remove_agents != 0))
2058                 osd_process_scheduled_agent_removals(env, osd);
2059
2060         oti->oti_ins_cache_depth--;
2061         /* reset OI cache for safety */
2062         if (oti->oti_ins_cache_depth == 0)
2063                 oti->oti_ins_cache_used = 0;
2064
2065         sb_end_write(osd_sb(osd));
2066
2067         RETURN(rc);
2068 }
2069
2070 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2071 {
2072         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2073                                                ot_super);
2074
2075         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2076         LASSERT(&dcb->dcb_func != NULL);
2077         if (dcb->dcb_flags & DCB_TRANS_STOP)
2078                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2079         else
2080                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2081
2082         return 0;
2083 }
2084
2085 /*
2086  * Called just before object is freed. Releases all resources except for
2087  * object itself (that is released by osd_object_free()).
2088  *
2089  * Concurrency: no concurrent access is possible that late in object
2090  * life-cycle.
2091  */
2092 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2093 {
2094         struct osd_object *obj = osd_obj(l);
2095         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2096         struct inode *inode = obj->oo_inode;
2097         __u64 projid;
2098         qid_t uid;
2099         qid_t gid;
2100
2101         LINVRNT(osd_invariant(obj));
2102
2103         /*
2104          * If object is unlinked remove fid->ino mapping from object index.
2105          */
2106
2107         osd_index_fini(obj);
2108
2109         if (!inode)
2110                 return;
2111
2112         uid = i_uid_read(inode);
2113         gid = i_gid_read(inode);
2114         projid = i_projid_read(inode);
2115
2116         obj->oo_inode = NULL;
2117         iput(inode);
2118
2119         /* do not rebalance quota if the caller needs to release memory
2120          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2121          * transaction and risk to deadlock - LU-12178 */
2122         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2123                 return;
2124
2125         if (!obj->oo_header && qsd) {
2126                 struct osd_thread_info *info = osd_oti_get(env);
2127                 struct lquota_id_info *qi = &info->oti_qi;
2128
2129                 /* Release granted quota to master if necessary */
2130                 qi->lqi_id.qid_uid = uid;
2131                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2132
2133                 qi->lqi_id.qid_uid = gid;
2134                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2135
2136                 qi->lqi_id.qid_uid = projid;
2137                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2138         }
2139 }
2140
2141 /*
2142  * Concurrency: ->loo_object_release() is called under site spin-lock.
2143  */
2144 static void osd_object_release(const struct lu_env *env,
2145                                struct lu_object *l)
2146 {
2147         struct osd_object *o = osd_obj(l);
2148
2149         /*
2150          * nobody should be releasing a non-destroyed object with nlink=0
2151          * the API allows this, but ldiskfs doesn't like and then report
2152          * this inode as deleted
2153          */
2154         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2155                   o->oo_inode->i_nlink == 0));
2156 }
2157
2158 /*
2159  * Concurrency: shouldn't matter.
2160  */
2161 static int osd_object_print(const struct lu_env *env, void *cookie,
2162                             lu_printer_t p, const struct lu_object *l)
2163 {
2164         struct osd_object *o = osd_obj(l);
2165         struct iam_descr *d;
2166
2167         if (o->oo_dir != NULL)
2168                 d = o->oo_dir->od_container.ic_descr;
2169         else
2170                 d = NULL;
2171         return (*p)(env, cookie,
2172                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2173                     o, o->oo_inode,
2174                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2175                     o->oo_inode ? o->oo_inode->i_generation : 0,
2176                     d ? d->id_ops->id_name : "plain");
2177 }
2178
2179 /*
2180  * Concurrency: shouldn't matter.
2181  */
2182 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2183                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2184 {
2185         struct osd_device *osd = osd_dt_dev(d);
2186         struct super_block *sb = osd_sb(osd);
2187         struct kstatfs *ksfs;
2188         __u64 reserved;
2189         int result = 0;
2190
2191         if (unlikely(osd->od_mnt == NULL))
2192                 return -EINPROGRESS;
2193
2194         /* osd_lproc.c call this without env, allocate ksfs for that case */
2195         if (unlikely(env == NULL)) {
2196                 OBD_ALLOC_PTR(ksfs);
2197                 if (ksfs == NULL)
2198                         return -ENOMEM;
2199         } else {
2200                 ksfs = &osd_oti_get(env)->oti_ksfs;
2201         }
2202
2203         result = sb->s_op->statfs(sb->s_root, ksfs);
2204         if (result)
2205                 goto out;
2206
2207         statfs_pack(sfs, ksfs);
2208         if (unlikely(sb->s_flags & SB_RDONLY))
2209                 sfs->os_state |= OS_STATE_READONLY;
2210
2211         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2212
2213         if (ldiskfs_has_feature_extents(sb))
2214                 sfs->os_maxbytes = sb->s_maxbytes;
2215         else
2216                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2217
2218         /*
2219          * Reserve some space so to avoid fragmenting the filesystem too much.
2220          * Fragmentation not only impacts performance, but can also increase
2221          * metadata overhead significantly, causing grant calculation to be
2222          * wrong.
2223          *
2224          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2225          */
2226         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2227         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2228         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2229                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2230
2231         sfs->os_blocks -= reserved;
2232         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2233         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2234
2235 out:
2236         if (unlikely(env == NULL))
2237                 OBD_FREE_PTR(ksfs);
2238         return result;
2239 }
2240
2241 /**
2242  * Estimate space needed for file creations. We assume the largest filename
2243  * which is 2^64 - 1, hence a filename of 20 chars.
2244  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2245  */
2246 #ifdef __LDISKFS_DIR_REC_LEN
2247 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2248 #else
2249 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2250 #endif
2251
2252 /*
2253  * Concurrency: doesn't access mutable data.
2254  */
2255 static void osd_conf_get(const struct lu_env *env,
2256                          const struct dt_device *dev,
2257                          struct dt_device_param *param)
2258 {
2259         struct osd_device *d = osd_dt_dev(dev);
2260         struct super_block *sb = osd_sb(d);
2261         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2262         const char *name;
2263         int ea_overhead;
2264
2265         /*
2266          * XXX should be taken from not-yet-existing fs abstraction layer.
2267          */
2268         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2269         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2270         param->ddp_symlink_max  = sb->s_blocksize;
2271         param->ddp_mount_type   = LDD_MT_LDISKFS;
2272         if (ldiskfs_has_feature_extents(sb))
2273                 param->ddp_maxbytes = sb->s_maxbytes;
2274         else
2275                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2276         /*
2277          * inode are statically allocated, so per-inode space consumption
2278          * is the space consumed by the directory entry
2279          */
2280         param->ddp_inodespace     = PER_OBJ_USAGE;
2281         /*
2282          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2283          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2284          * maximum length to not under-count the actual number of extents
2285          * needed for writing a file if there are sub-optimal block allocations.
2286          */
2287         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2288         /* worst-case extent insertion metadata overhead */
2289         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2290         param->ddp_mntopts = 0;
2291         if (test_opt(sb, XATTR_USER))
2292                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2293         if (test_opt(sb, POSIX_ACL))
2294                 param->ddp_mntopts |= MNTOPT_ACL;
2295
2296         /*
2297          * LOD might calculate the max stripe count based on max_ea_size,
2298          * so we need take account in the overhead as well,
2299          * xattr_header + magic + xattr_entry_head
2300          */
2301         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2302                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2303
2304 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2305         if (ldiskfs_has_feature_ea_inode(sb))
2306                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2307                                                                 ea_overhead;
2308         else
2309 #endif
2310                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2311
2312         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2313                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2314
2315         /*
2316          * Preferred RPC size for efficient disk IO.  4MB shows good
2317          * all-around performance for ldiskfs, but use bigalloc chunk size
2318          * by default if larger.
2319          */
2320 #if defined(LDISKFS_CLUSTER_SIZE)
2321         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2322                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2323         else
2324 #endif
2325                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2326
2327         param->ddp_t10_cksum_type = 0;
2328         if (bi) {
2329                 unsigned short interval = blk_integrity_interval(bi);
2330                 name = blk_integrity_name(bi);
2331                 /*
2332                  * Expected values:
2333                  * T10-DIF-TYPE1-CRC
2334                  * T10-DIF-TYPE3-CRC
2335                  * T10-DIF-TYPE1-IP
2336                  * T10-DIF-TYPE3-IP
2337                  */
2338                 if (strncmp(name, "T10-DIF-TYPE",
2339                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2340                         /* also skip "1/3-" at end */
2341                         const int type_off = sizeof("T10-DIF-TYPE.");
2342                         char type_number = name[type_off - 2];
2343
2344                         if (interval != 512 && interval != 4096) {
2345                                 CERROR("%s: unsupported T10PI sector size %u\n",
2346                                        d->od_svname, interval);
2347                         } else if (type_number != '1' && type_number != '3') {
2348                                 CERROR("%s: unsupported T10PI type %s\n",
2349                                        d->od_svname, name);
2350                         } else if (strcmp(name + type_off, "CRC") == 0) {
2351                                 d->od_t10_type = type_number == '1' ?
2352                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2353                                 param->ddp_t10_cksum_type = interval == 512 ?
2354                                         OBD_CKSUM_T10CRC512 :
2355                                         OBD_CKSUM_T10CRC4K;
2356                         } else if (strcmp(name + type_off, "IP") == 0) {
2357                                 d->od_t10_type = type_number == '1' ?
2358                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2359                                 param->ddp_t10_cksum_type = interval == 512 ?
2360                                         OBD_CKSUM_T10IP512 :
2361                                         OBD_CKSUM_T10IP4K;
2362                         } else {
2363                                 CERROR("%s: unsupported checksum type of "
2364                                        "T10PI type '%s'",
2365                                        d->od_svname, name);
2366                         }
2367
2368                 } else {
2369                         CERROR("%s: unsupported T10PI type '%s'",
2370                                d->od_svname, name);
2371                 }
2372         }
2373 }
2374
2375 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2376 {
2377         return osd_sb(osd_dt_dev(d));
2378 }
2379
2380 /*
2381  * Concurrency: shouldn't matter.
2382  */
2383 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2384 {
2385         int rc;
2386         struct super_block *s = osd_sb(osd_dt_dev(d));
2387         ENTRY;
2388
2389         down_read(&s->s_umount);
2390         rc = s->s_op->sync_fs(s, 1);
2391         up_read(&s->s_umount);
2392
2393         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2394                rc);
2395
2396         return rc;
2397 }
2398
2399 /**
2400  * Start commit for OSD device.
2401  *
2402  * An implementation of dt_commit_async method for OSD device.
2403  * Asychronously starts underlayng fs sync and thereby a transaction
2404  * commit.
2405  *
2406  * \param env environment
2407  * \param d dt device
2408  *
2409  * \see dt_device_operations
2410  */
2411 static int osd_commit_async(const struct lu_env *env,
2412                             struct dt_device *d)
2413 {
2414         struct super_block *s = osd_sb(osd_dt_dev(d));
2415         int rc;
2416
2417         ENTRY;
2418
2419         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2420         down_read(&s->s_umount);
2421         rc = s->s_op->sync_fs(s, 0);
2422         up_read(&s->s_umount);
2423
2424         RETURN(rc);
2425 }
2426
2427 /* Our own copy of the set readonly functions if present, or NU if not. */
2428 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2429 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2430 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
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         if (priv_dev_set_rdonly) {
2452                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2453
2454                 rc = 0;
2455                 CERROR("*** setting %s read-only ***\n",
2456                        osd_dt_dev(d)->od_svname);
2457
2458                 if (sb->s_op->freeze_fs) {
2459                         rc = sb->s_op->freeze_fs(sb);
2460                         if (rc)
2461                                 goto out;
2462                 }
2463
2464                 if (jdev && (jdev != dev)) {
2465                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2466                                (long)jdev);
2467                         priv_dev_set_rdonly(jdev);
2468                 }
2469                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2470                 priv_dev_set_rdonly(dev);
2471
2472                 if (sb->s_op->unfreeze_fs)
2473                         sb->s_op->unfreeze_fs(sb);
2474         }
2475
2476 out:
2477         if (rc)
2478                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2479                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2480
2481         RETURN(rc);
2482 }
2483
2484 /**
2485  * Note: we do not count into QUOTA here.
2486  * If we mount with --data_journal we may need more.
2487  */
2488 const int osd_dto_credits_noquota[DTO_NR] = {
2489         /**
2490          * Insert.
2491          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2492          * SINGLEDATA_TRANS_BLOCKS(8)
2493          * XXX Note: maybe iam need more, since iam have more level than
2494          *           EXT3 htree.
2495          */
2496         [DTO_INDEX_INSERT]  = 16,
2497         /**
2498          * Delete
2499          * just modify a single entry, probably merge few within a block
2500          */
2501         [DTO_INDEX_DELETE]  = 1,
2502         /**
2503          * Used for OI scrub
2504          */
2505         [DTO_INDEX_UPDATE]  = 16,
2506         /**
2507          * 4(inode, inode bits, groups, GDT)
2508          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2509          */
2510         [DTO_OBJECT_CREATE] = 4,
2511         /**
2512          * 4(inode, inode bits, groups, GDT)
2513          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2514          */
2515         [DTO_OBJECT_DELETE] = 4,
2516         /**
2517          * Attr set credits (inode)
2518          */
2519         [DTO_ATTR_SET_BASE] = 1,
2520         /**
2521          * Xattr set. The same as xattr of EXT3.
2522          * DATA_TRANS_BLOCKS(14)
2523          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2524          * are also counted in. Do not know why?
2525          */
2526         [DTO_XATTR_SET]     = 14,
2527         /**
2528          * credits for inode change during write.
2529          */
2530         [DTO_WRITE_BASE]    = 3,
2531         /**
2532          * credits for single block write.
2533          */
2534         [DTO_WRITE_BLOCK]   = 14,
2535         /**
2536          * Attr set credits for chown.
2537          * This is extra credits for setattr, and it is null without quota
2538          */
2539         [DTO_ATTR_SET_CHOWN] = 0
2540 };
2541
2542 static const struct dt_device_operations osd_dt_ops = {
2543         .dt_root_get       = osd_root_get,
2544         .dt_statfs         = osd_statfs,
2545         .dt_trans_create   = osd_trans_create,
2546         .dt_trans_start    = osd_trans_start,
2547         .dt_trans_stop     = osd_trans_stop,
2548         .dt_trans_cb_add   = osd_trans_cb_add,
2549         .dt_conf_get       = osd_conf_get,
2550         .dt_mnt_sb_get     = osd_mnt_sb_get,
2551         .dt_sync           = osd_sync,
2552         .dt_ro             = osd_ro,
2553         .dt_commit_async   = osd_commit_async,
2554 };
2555
2556 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2557                           unsigned int role)
2558 {
2559         struct osd_object *obj = osd_dt_obj(dt);
2560         struct osd_thread_info *oti = osd_oti_get(env);
2561
2562         LINVRNT(osd_invariant(obj));
2563
2564         LASSERT(obj->oo_owner != env);
2565         down_read_nested(&obj->oo_sem, role);
2566
2567         LASSERT(obj->oo_owner == NULL);
2568         oti->oti_r_locks++;
2569 }
2570
2571 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2572                            unsigned int role)
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         down_write_nested(&obj->oo_sem, role);
2581
2582         LASSERT(obj->oo_owner == NULL);
2583         obj->oo_owner = env;
2584         oti->oti_w_locks++;
2585 }
2586
2587 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2588 {
2589         struct osd_object *obj = osd_dt_obj(dt);
2590         struct osd_thread_info *oti = osd_oti_get(env);
2591
2592         LINVRNT(osd_invariant(obj));
2593
2594         LASSERT(oti->oti_r_locks > 0);
2595         oti->oti_r_locks--;
2596         up_read(&obj->oo_sem);
2597 }
2598
2599 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2600 {
2601         struct osd_object *obj = osd_dt_obj(dt);
2602         struct osd_thread_info *oti = osd_oti_get(env);
2603
2604         LINVRNT(osd_invariant(obj));
2605
2606         LASSERT(obj->oo_owner == env);
2607         LASSERT(oti->oti_w_locks > 0);
2608         oti->oti_w_locks--;
2609         obj->oo_owner = NULL;
2610         up_write(&obj->oo_sem);
2611 }
2612
2613 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2614 {
2615         struct osd_object *obj = osd_dt_obj(dt);
2616
2617         LINVRNT(osd_invariant(obj));
2618
2619         return obj->oo_owner == env;
2620 }
2621
2622 static void osd_inode_getattr(const struct lu_env *env,
2623                               struct inode *inode, struct lu_attr *attr)
2624 {
2625         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2626                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2627                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2628                            LA_BLKSIZE | LA_TYPE;
2629
2630         attr->la_atime = inode->i_atime.tv_sec;
2631         attr->la_mtime = inode->i_mtime.tv_sec;
2632         attr->la_ctime = inode->i_ctime.tv_sec;
2633         attr->la_mode    = inode->i_mode;
2634         attr->la_size    = i_size_read(inode);
2635         attr->la_blocks  = inode->i_blocks;
2636         attr->la_uid     = i_uid_read(inode);
2637         attr->la_gid     = i_gid_read(inode);
2638         attr->la_projid  = i_projid_read(inode);
2639         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2640         attr->la_nlink   = inode->i_nlink;
2641         attr->la_rdev    = inode->i_rdev;
2642         attr->la_blksize = 1 << inode->i_blkbits;
2643         attr->la_blkbits = inode->i_blkbits;
2644         /*
2645          * Ext4 did not transfer inherit flags from raw inode
2646          * to inode flags, and ext4 internally test raw inode
2647          * @i_flags directly. Instead of patching ext4, we do it here.
2648          */
2649         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2650                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2651 }
2652
2653 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2654                         struct lu_attr *attr)
2655 {
2656         struct osd_object *obj = osd_dt_obj(dt);
2657
2658         if (unlikely(!dt_object_exists(dt)))
2659                 return -ENOENT;
2660         if (unlikely(obj->oo_destroyed))
2661                 return -ENOENT;
2662
2663         LASSERT(!dt_object_remote(dt));
2664         LINVRNT(osd_invariant(obj));
2665
2666         spin_lock(&obj->oo_guard);
2667         osd_inode_getattr(env, obj->oo_inode, attr);
2668         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2669                 attr->la_valid |= LA_FLAGS;
2670                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2671         }
2672         spin_unlock(&obj->oo_guard);
2673
2674         return 0;
2675 }
2676
2677 static int osd_declare_attr_qid(const struct lu_env *env,
2678                                 struct osd_object *obj,
2679                                 struct osd_thandle *oh, long long bspace,
2680                                 qid_t old_id, qid_t new_id, bool enforce,
2681                                 unsigned int type, bool ignore_edquot)
2682 {
2683         int rc;
2684         struct osd_thread_info *info = osd_oti_get(env);
2685         struct lquota_id_info  *qi = &info->oti_qi;
2686
2687         qi->lqi_type = type;
2688         /* inode accounting */
2689         qi->lqi_is_blk = false;
2690
2691         /* one more inode for the new id ... */
2692         qi->lqi_id.qid_uid = new_id;
2693         qi->lqi_space      = 1;
2694         /* Reserve credits for the new id */
2695         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2696         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2697                 rc = 0;
2698         if (rc)
2699                 RETURN(rc);
2700
2701         /* and one less inode for the current id */
2702         qi->lqi_id.qid_uid = old_id;
2703         qi->lqi_space = -1;
2704         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2705         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2706                 rc = 0;
2707         if (rc)
2708                 RETURN(rc);
2709
2710         /* block accounting */
2711         qi->lqi_is_blk = true;
2712
2713         /* more blocks for the new id ... */
2714         qi->lqi_id.qid_uid = new_id;
2715         qi->lqi_space      = bspace;
2716         /*
2717          * Credits for the new uid has been reserved, re-use "obj"
2718          * to save credit reservation.
2719          */
2720         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2721         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2722                 rc = 0;
2723         if (rc)
2724                 RETURN(rc);
2725
2726         /* and finally less blocks for the current uid */
2727         qi->lqi_id.qid_uid = old_id;
2728         qi->lqi_space      = -bspace;
2729         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2730         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2731                 rc = 0;
2732
2733         RETURN(rc);
2734 }
2735
2736 static int osd_declare_attr_set(const struct lu_env *env,
2737                                 struct dt_object *dt,
2738                                 const struct lu_attr *attr,
2739                                 struct thandle *handle)
2740 {
2741         struct osd_thandle *oh;
2742         struct osd_object *obj;
2743         qid_t uid;
2744         qid_t gid;
2745         long long bspace;
2746         int rc = 0;
2747         bool enforce;
2748
2749         ENTRY;
2750
2751         LASSERT(dt != NULL);
2752         LASSERT(handle != NULL);
2753
2754         obj = osd_dt_obj(dt);
2755         LASSERT(osd_invariant(obj));
2756
2757         oh = container_of0(handle, struct osd_thandle, ot_super);
2758         LASSERT(oh->ot_handle == NULL);
2759
2760         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2761                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2762
2763         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2764                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2765
2766         if (attr == NULL || obj->oo_inode == NULL)
2767                 RETURN(rc);
2768
2769         bspace   = obj->oo_inode->i_blocks << 9;
2770         bspace   = toqb(bspace);
2771
2772         /*
2773          * Changing ownership is always preformed by super user, it should not
2774          * fail with EDQUOT unless required explicitly.
2775          *
2776          * We still need to call the osd_declare_qid() to calculate the journal
2777          * credits for updating quota accounting files and to trigger quota
2778          * space adjustment once the operation is completed.
2779          */
2780         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2781                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2782
2783                 if (!ignore_edquot)
2784                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2785                                "(the quota space is %lld)\n",
2786                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2787                                attr->la_gid, bspace);
2788
2789                 /* USERQUOTA */
2790                 uid = i_uid_read(obj->oo_inode);
2791                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2792                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2793                                           attr->la_uid, enforce, USRQUOTA,
2794                                           true);
2795                 if (rc)
2796                         RETURN(rc);
2797
2798                 gid = i_gid_read(obj->oo_inode);
2799                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2800                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2801                                           i_gid_read(obj->oo_inode),
2802                                           attr->la_gid, enforce, GRPQUOTA,
2803                                           ignore_edquot);
2804                 if (rc)
2805                         RETURN(rc);
2806
2807         }
2808 #ifdef HAVE_PROJECT_QUOTA
2809         if (attr->la_valid & LA_PROJID) {
2810                 __u32 projid = i_projid_read(obj->oo_inode);
2811
2812                 enforce = (attr->la_valid & LA_PROJID) &&
2813                                         (attr->la_projid != projid);
2814                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2815                                           (qid_t)projid, (qid_t)attr->la_projid,
2816                                           enforce, PRJQUOTA, true);
2817                 if (rc)
2818                         RETURN(rc);
2819         }
2820 #endif
2821         RETURN(rc);
2822 }
2823
2824 static int osd_inode_setattr(const struct lu_env *env,
2825                              struct inode *inode, const struct lu_attr *attr)
2826 {
2827         __u64 bits = attr->la_valid;
2828
2829         /* Only allow set size for regular file */
2830         if (!S_ISREG(inode->i_mode))
2831                 bits &= ~(LA_SIZE | LA_BLOCKS);
2832
2833         if (bits == 0)
2834                 return 0;
2835
2836         if (bits & LA_ATIME)
2837                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2838         if (bits & LA_CTIME)
2839                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2840         if (bits & LA_MTIME)
2841                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2842         if (bits & LA_SIZE) {
2843                 spin_lock(&inode->i_lock);
2844                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2845                 i_size_write(inode, attr->la_size);
2846                 spin_unlock(&inode->i_lock);
2847         }
2848
2849         /*
2850          * OSD should not change "i_blocks" which is used by quota.
2851          * "i_blocks" should be changed by ldiskfs only.
2852          */
2853         if (bits & LA_MODE)
2854                 inode->i_mode = (inode->i_mode & S_IFMT) |
2855                                 (attr->la_mode & ~S_IFMT);
2856         if (bits & LA_UID)
2857                 i_uid_write(inode, attr->la_uid);
2858         if (bits & LA_GID)
2859                 i_gid_write(inode, attr->la_gid);
2860         if (bits & LA_PROJID)
2861                 i_projid_write(inode, attr->la_projid);
2862         if (bits & LA_NLINK)
2863                 set_nlink(inode, attr->la_nlink);
2864         if (bits & LA_RDEV)
2865                 inode->i_rdev = attr->la_rdev;
2866
2867         if (bits & LA_FLAGS) {
2868                 /* always keep S_NOCMTIME */
2869                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2870                                  S_NOCMTIME;
2871                 /*
2872                  * Ext4 did not transfer inherit flags from
2873                  * @inode->i_flags to raw inode i_flags when writing
2874                  * flags, we do it explictly here.
2875                  */
2876                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2877                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2878                 else
2879                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2880         }
2881         return 0;
2882 }
2883
2884 #ifdef HAVE_PROJECT_QUOTA
2885 static int osd_transfer_project(struct inode *inode, __u32 projid)
2886 {
2887         struct super_block *sb = inode->i_sb;
2888         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2889         int err;
2890         kprojid_t kprojid;
2891         struct ldiskfs_iloc iloc;
2892         struct ldiskfs_inode *raw_inode;
2893         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2894
2895         if (!ldiskfs_has_feature_project(sb)) {
2896                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2897                         == LDISKFS_DEF_PROJID);
2898                 if (projid != LDISKFS_DEF_PROJID)
2899                         return -EOPNOTSUPP;
2900                 else
2901                         return 0;
2902         }
2903
2904         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2905                 return -EOPNOTSUPP;
2906
2907         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2908         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2909                 return 0;
2910
2911         err = ldiskfs_get_inode_loc(inode, &iloc);
2912         if (err)
2913                 return err;
2914
2915         raw_inode = ldiskfs_raw_inode(&iloc);
2916         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2917                 err = -EOVERFLOW;
2918                 brelse(iloc.bh);
2919                 return err;
2920         }
2921         brelse(iloc.bh);
2922
2923         dquot_initialize(inode);
2924         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2925         if (transfer_to[PRJQUOTA]) {
2926                 err = __dquot_transfer(inode, transfer_to);
2927                 dqput(transfer_to[PRJQUOTA]);
2928                 if (err)
2929                         return err;
2930         }
2931
2932         return err;
2933 }
2934 #endif
2935
2936 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2937 {
2938         int rc;
2939
2940         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2941             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2942                 struct iattr iattr;
2943
2944                 dquot_initialize(inode);
2945                 iattr.ia_valid = 0;
2946                 if (attr->la_valid & LA_UID)
2947                         iattr.ia_valid |= ATTR_UID;
2948                 if (attr->la_valid & LA_GID)
2949                         iattr.ia_valid |= ATTR_GID;
2950                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2951                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2952
2953                 rc = dquot_transfer(inode, &iattr);
2954                 if (rc) {
2955                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2956                                "enforcement enabled on the ldiskfs "
2957                                "filesystem?\n", inode->i_sb->s_id, rc);
2958                         return rc;
2959                 }
2960         }
2961
2962         /* Handle project id transfer here properly */
2963         if (attr->la_valid & LA_PROJID &&
2964             attr->la_projid != i_projid_read(inode)) {
2965 #ifdef HAVE_PROJECT_QUOTA
2966                 rc = osd_transfer_project(inode, attr->la_projid);
2967 #else
2968                 rc = -ENOTSUPP;
2969 #endif
2970                 if (rc) {
2971                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2972                                "enforcement enabled on the ldiskfs "
2973                                "filesystem?\n", inode->i_sb->s_id, rc);
2974                         return rc;
2975                 }
2976         }
2977         return 0;
2978 }
2979
2980 static int osd_attr_set(const struct lu_env *env,
2981                         struct dt_object *dt,
2982                         const struct lu_attr *attr,
2983                         struct thandle *handle)
2984 {
2985         struct osd_object *obj = osd_dt_obj(dt);
2986         struct inode *inode;
2987         int rc;
2988
2989         if (!dt_object_exists(dt))
2990                 return -ENOENT;
2991
2992         LASSERT(handle != NULL);
2993         LASSERT(!dt_object_remote(dt));
2994         LASSERT(osd_invariant(obj));
2995
2996         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2997
2998         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2999             !osd_obj2dev(obj)->od_is_ost) {
3000                 struct osd_thread_info *oti = osd_oti_get(env);
3001                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3002                 struct lu_fid *fid1 = &oti->oti_fid;
3003                 struct osd_inode_id *id = &oti->oti_id;
3004                 struct iam_path_descr *ipd;
3005                 struct iam_container *bag;
3006                 struct osd_thandle *oh;
3007                 int rc;
3008
3009                 fid_cpu_to_be(fid1, fid0);
3010                 memset(id, 1, sizeof(*id));
3011                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3012                                   fid0)->oi_dir.od_container;
3013                 ipd = osd_idx_ipd_get(env, bag);
3014                 if (unlikely(ipd == NULL))
3015                         RETURN(-ENOMEM);
3016
3017                 oh = container_of0(handle, struct osd_thandle, ot_super);
3018                 rc = iam_update(oh->ot_handle, bag,
3019                                 (const struct iam_key *)fid1,
3020                                 (const struct iam_rec *)id, ipd);
3021                 osd_ipd_put(env, bag, ipd);
3022                 return(rc > 0 ? 0 : rc);
3023         }
3024
3025         inode = obj->oo_inode;
3026
3027         rc = osd_quota_transfer(inode, attr);
3028         if (rc)
3029                 return rc;
3030
3031         spin_lock(&obj->oo_guard);
3032         rc = osd_inode_setattr(env, inode, attr);
3033         spin_unlock(&obj->oo_guard);
3034         if (rc != 0)
3035                 GOTO(out, rc);
3036
3037         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3038
3039         if (!(attr->la_valid & LA_FLAGS))
3040                 GOTO(out, rc);
3041
3042         /* Let's check if there are extra flags need to be set into LMA */
3043         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3044                 struct osd_thread_info *info = osd_oti_get(env);
3045                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3046
3047                 LASSERT(!obj->oo_pfid_in_lma);
3048
3049                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3050                                  &info->oti_ost_attrs);
3051                 if (rc)
3052                         GOTO(out, rc);
3053
3054                 lma->lma_incompat |=
3055                         lustre_to_lma_flags(attr->la_flags);
3056                 lustre_lma_swab(lma);
3057                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3058                                      lma, sizeof(*lma), XATTR_REPLACE);
3059                 if (rc != 0) {
3060                         struct osd_device *osd = osd_obj2dev(obj);
3061
3062                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3063                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3064                               lma->lma_incompat, rc);
3065                 } else {
3066                         obj->oo_lma_flags =
3067                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3068                 }
3069                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3070         }
3071 out:
3072         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3073
3074         return rc;
3075 }
3076
3077 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3078                                            struct osd_object *obj,
3079                                            const char *name, const int namelen)
3080 {
3081         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3082 }
3083
3084 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3085                       umode_t mode, struct dt_allocation_hint *hint,
3086                       struct thandle *th, struct lu_attr *attr)
3087 {
3088         int result;
3089         struct osd_device *osd = osd_obj2dev(obj);
3090         struct osd_thandle *oth;
3091         struct dt_object *parent = NULL;
3092         struct inode *inode;
3093         uid_t owner[2] = {0, 0};
3094
3095         if (attr->la_valid & LA_UID)
3096                 owner[0] = attr->la_uid;
3097         if (attr->la_valid & LA_GID)
3098                 owner[1] = attr->la_gid;
3099
3100         LINVRNT(osd_invariant(obj));
3101         LASSERT(obj->oo_inode == NULL);
3102         LASSERT(obj->oo_hl_head == NULL);
3103
3104         if (S_ISDIR(mode) && ldiskfs_pdo) {
3105                 obj->oo_hl_head =
3106                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3107                 if (obj->oo_hl_head == NULL)
3108                         return -ENOMEM;
3109         }
3110
3111         oth = container_of(th, struct osd_thandle, ot_super);
3112         LASSERT(oth->ot_handle->h_transaction != NULL);
3113
3114         if (hint != NULL && hint->dah_parent != NULL &&
3115             !dt_object_remote(hint->dah_parent))
3116                 parent = hint->dah_parent;
3117
3118         inode = ldiskfs_create_inode(oth->ot_handle,
3119                                      parent ? osd_dt_obj(parent)->oo_inode :
3120                                               osd_sb(osd)->s_root->d_inode,
3121                                      mode, owner);
3122         if (!IS_ERR(inode)) {
3123                 /* Do not update file c/mtime in ldiskfs. */
3124                 inode->i_flags |= S_NOCMTIME;
3125
3126                 /*
3127                  * For new created object, it must be consistent,
3128                  * and it is unnecessary to scrub against it.
3129                  */
3130                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3131
3132                 obj->oo_inode = inode;
3133                 result = 0;
3134         } else {
3135                 if (obj->oo_hl_head != NULL) {
3136                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3137                         obj->oo_hl_head = NULL;
3138                 }
3139                 result = PTR_ERR(inode);
3140         }
3141         LINVRNT(osd_invariant(obj));
3142         return result;
3143 }
3144
3145 enum {
3146         OSD_NAME_LEN = 255
3147 };
3148
3149 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3150                      struct lu_attr *attr,
3151                      struct dt_allocation_hint *hint,
3152                      struct dt_object_format *dof,
3153                      struct thandle *th)
3154 {
3155         int result;
3156         struct osd_thandle *oth;
3157         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3158
3159         LASSERT(S_ISDIR(attr->la_mode));
3160
3161         oth = container_of(th, struct osd_thandle, ot_super);
3162         LASSERT(oth->ot_handle->h_transaction != NULL);
3163         result = osd_mkfile(info, obj, mode, hint, th, attr);
3164
3165         return result;
3166 }
3167
3168 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3169                         struct lu_attr *attr,
3170                         struct dt_allocation_hint *hint,
3171                         struct dt_object_format *dof,
3172                         struct thandle *th)
3173 {
3174         int result;
3175         struct osd_thandle *oth;
3176         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3177
3178         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3179
3180         LASSERT(S_ISREG(attr->la_mode));
3181
3182         oth = container_of(th, struct osd_thandle, ot_super);
3183         LASSERT(oth->ot_handle->h_transaction != NULL);
3184
3185         result = osd_mkfile(info, obj, mode, hint, th, attr);
3186         if (result == 0) {
3187                 LASSERT(obj->oo_inode != NULL);
3188                 if (feat->dif_flags & DT_IND_VARKEY)
3189                         result = iam_lvar_create(obj->oo_inode,
3190                                                  feat->dif_keysize_max,
3191                                                  feat->dif_ptrsize,
3192                                                  feat->dif_recsize_max,
3193                                                  oth->ot_handle);
3194                 else
3195                         result = iam_lfix_create(obj->oo_inode,
3196                                                  feat->dif_keysize_max,
3197                                                  feat->dif_ptrsize,
3198                                                  feat->dif_recsize_max,
3199                                                  oth->ot_handle);
3200         }
3201         return result;
3202 }
3203
3204 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3205                      struct lu_attr *attr,
3206                      struct dt_allocation_hint *hint,
3207                      struct dt_object_format *dof,
3208                      struct thandle *th)
3209 {
3210         LASSERT(S_ISREG(attr->la_mode));
3211         return osd_mkfile(info, obj, (attr->la_mode &
3212                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3213                           attr);
3214 }
3215
3216 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3217                      struct lu_attr *attr,
3218                      struct dt_allocation_hint *hint,
3219                      struct dt_object_format *dof,
3220                      struct thandle *th)
3221 {
3222         LASSERT(S_ISLNK(attr->la_mode));
3223         return osd_mkfile(info, obj, (attr->la_mode &
3224                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3225                           attr);
3226 }
3227
3228 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3229                      struct lu_attr *attr,
3230                      struct dt_allocation_hint *hint,
3231                      struct dt_object_format *dof,
3232                      struct thandle *th)
3233 {
3234         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3235         int result;
3236
3237         LINVRNT(osd_invariant(obj));
3238         LASSERT(obj->oo_inode == NULL);
3239         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3240                 S_ISFIFO(mode) || S_ISSOCK(mode));
3241
3242         result = osd_mkfile(info, obj, mode, hint, th, attr);
3243         if (result == 0) {
3244                 LASSERT(obj->oo_inode != NULL);
3245                 /*
3246                  * This inode should be marked dirty for i_rdev.  Currently
3247                  * that is done in the osd_attr_init().
3248                  */
3249                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3250                                    attr->la_rdev);
3251         }
3252         LINVRNT(osd_invariant(obj));
3253         return result;
3254 }
3255
3256 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3257                               struct lu_attr *,
3258                               struct dt_allocation_hint *hint,
3259                               struct dt_object_format *dof,
3260                               struct thandle *);
3261
3262 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3263 {
3264         osd_obj_type_f result;
3265
3266         switch (type) {
3267         case DFT_DIR:
3268                 result = osd_mkdir;
3269                 break;
3270         case DFT_REGULAR:
3271                 result = osd_mkreg;
3272                 break;
3273         case DFT_SYM:
3274                 result = osd_mksym;
3275                 break;
3276         case DFT_NODE:
3277                 result = osd_mknod;
3278                 break;
3279         case DFT_INDEX:
3280                 result = osd_mk_index;
3281                 break;
3282
3283         default:
3284                 LBUG();
3285                 break;
3286         }
3287         return result;
3288 }
3289
3290
3291 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3292                         struct dt_object *parent, struct dt_object *child,
3293                         umode_t child_mode)
3294 {
3295         LASSERT(ah);
3296
3297         ah->dah_parent = parent;
3298         ah->dah_mode = child_mode;
3299
3300         if (parent != NULL && !dt_object_remote(parent)) {
3301                 /* will help to find FID->ino at dt_insert("..") */
3302                 struct osd_object *pobj = osd_dt_obj(parent);
3303
3304                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3305         }
3306 }
3307
3308 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3309                           struct lu_attr *attr, struct dt_object_format *dof)
3310 {
3311         struct inode *inode = obj->oo_inode;
3312         __u64 valid = attr->la_valid;
3313         int result;
3314
3315         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3316
3317         if (dof->dof_type != DFT_NODE)
3318                 attr->la_valid &= ~LA_RDEV;
3319         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3320                 attr->la_valid &= ~LA_ATIME;
3321         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3322                 attr->la_valid &= ~LA_CTIME;
3323         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3324                 attr->la_valid &= ~LA_MTIME;
3325
3326         result = osd_quota_transfer(inode, attr);
3327         if (result)
3328                 return;
3329
3330         if (attr->la_valid != 0) {
3331                 result = osd_inode_setattr(info->oti_env, inode, attr);
3332                 /*
3333                  * The osd_inode_setattr() should always succeed here.  The
3334                  * only error that could be returned is EDQUOT when we are
3335                  * trying to change the UID or GID of the inode. However, this
3336                  * should not happen since quota enforcement is no longer
3337                  * enabled on ldiskfs (lquota takes care of it).
3338                  */
3339                 LASSERTF(result == 0, "%d\n", result);
3340                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3341         }
3342
3343         attr->la_valid = valid;
3344 }
3345
3346 /**
3347  * Helper function for osd_create()
3348  *
3349  * \retval 0, on success
3350  */
3351 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3352                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3353                         struct dt_object_format *dof, struct thandle *th)
3354 {
3355         int result;
3356         __u32 umask;
3357
3358         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3359
3360         /* we drop umask so that permissions we pass are not affected */
3361         umask = current->fs->umask;
3362         current->fs->umask = 0;
3363
3364         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3365                                                   th);
3366         if (likely(obj->oo_inode != NULL)) {
3367                 LASSERT(obj->oo_inode->i_state & I_NEW);
3368
3369                 /*
3370                  * Unlock the inode before attr initialization to avoid
3371                  * unnecessary dqget operations. LU-6378
3372                  */
3373                 unlock_new_inode(obj->oo_inode);
3374         }
3375
3376         if (likely(result == 0)) {
3377                 osd_attr_init(info, obj, attr, dof);
3378                 osd_object_init0(obj);
3379         }
3380
3381         /* restore previous umask value */
3382         current->fs->umask = umask;
3383
3384         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3385
3386         return result;
3387 }
3388
3389 /**
3390  * Helper function for osd_create()
3391  *
3392  * \retval 0, on success
3393  */
3394 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3395                            const struct lu_fid *fid, struct thandle *th)
3396 {
3397         struct osd_thread_info *info = osd_oti_get(env);
3398         struct osd_inode_id    *id   = &info->oti_id;
3399         struct osd_device      *osd  = osd_obj2dev(obj);
3400         struct osd_thandle     *oh;
3401         int rc;
3402
3403         LASSERT(obj->oo_inode != NULL);
3404
3405         oh = container_of0(th, struct osd_thandle, ot_super);
3406         LASSERT(oh->ot_handle);
3407         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3408
3409         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3410         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3411                            OI_CHECK_FLD, NULL);
3412         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3413                 struct lu_fid next_fid = *fid;
3414
3415                 /* insert next object in advance, and map to the same inode */
3416                 next_fid.f_oid++;
3417                 if (next_fid.f_oid != 0) {
3418                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3419                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3420                                       OI_CHECK_FLD, NULL);
3421                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3422                 }
3423         }
3424
3425         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3426
3427         return rc;
3428 }
3429
3430 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3431                    u64 seq, struct lu_seq_range *range)
3432 {
3433         struct seq_server_site *ss = osd_seq_site(osd);
3434
3435         if (fid_seq_is_idif(seq)) {
3436                 fld_range_set_ost(range);
3437                 range->lsr_index = idif_ost_idx(seq);
3438                 return 0;
3439         }
3440
3441         if (!fid_seq_in_fldb(seq)) {
3442                 fld_range_set_mdt(range);
3443                 if (ss != NULL)
3444                         /*
3445                          * FIXME: If ss is NULL, it suppose not get lsr_index
3446                          * at all
3447                          */
3448                         range->lsr_index = ss->ss_node_id;
3449                 return 0;
3450         }
3451
3452         LASSERT(ss != NULL);
3453         fld_range_set_any(range);
3454         /* OSD will only do local fld lookup */
3455         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3456 }
3457
3458 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3459                               struct lu_attr *attr,
3460                               struct dt_allocation_hint *hint,
3461                               struct dt_object_format *dof,
3462                               struct thandle *handle)
3463 {
3464         struct osd_thandle *oh;
3465         int rc;
3466
3467         ENTRY;
3468
3469         LASSERT(handle != NULL);
3470
3471         oh = container_of0(handle, struct osd_thandle, ot_super);
3472         LASSERT(oh->ot_handle == NULL);
3473
3474         /*
3475          * EA object consumes more credits than regular object: osd_mk_index
3476          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3477          * leaf_node, could involves the block, block bitmap, groups, GDT
3478          * change for each block, so add 4 * 2 credits in that case.
3479          */
3480         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3481                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3482                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3483         /*
3484          * Reuse idle OI block may cause additional one OI block
3485          * to be changed.
3486          */
3487         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3488                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3489         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3490                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3491                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3492
3493         /* will help to find FID->ino mapping at dt_insert() */
3494         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3495                                    osd_dt_obj(dt));
3496         if (rc != 0)
3497                 RETURN(rc);
3498
3499         if (!attr)
3500                 RETURN(0);
3501
3502         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3503                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3504                                    NULL, OSD_QID_INODE);
3505         if (rc != 0)
3506                 RETURN(rc);
3507
3508         RETURN(rc);
3509 }
3510
3511 /**
3512  * Called to destroy on-disk representation of the object
3513  *
3514  * Concurrency: must be locked
3515  */
3516 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3517                                struct thandle *th)
3518 {
3519         struct osd_object *obj = osd_dt_obj(dt);
3520         struct inode *inode = obj->oo_inode;
3521         struct osd_thandle *oh;
3522         int rc;
3523
3524         ENTRY;
3525
3526         if (inode == NULL)
3527                 RETURN(-ENOENT);
3528
3529         oh = container_of0(th, struct osd_thandle, ot_super);
3530         LASSERT(oh->ot_handle == NULL);
3531
3532         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3533                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3534
3535         /* For removing agent entry */
3536         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3537                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3538
3539         /*
3540          * Recycle idle OI leaf may cause additional three OI blocks
3541          * to be changed.
3542          */
3543         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3544                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3545                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3546         /* one less inode */
3547         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3548                                    i_projid_read(inode), -1, oh, obj, NULL,
3549                                    OSD_QID_INODE);
3550         if (rc)
3551                 RETURN(rc);
3552         /* data to be truncated */
3553         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3554                                    i_projid_read(inode), 0, oh, obj, NULL,
3555                                    OSD_QID_BLK);
3556         if (rc)
3557                 RETURN(rc);
3558
3559         /*
3560          * will help to find FID->ino when this object is being
3561          * added to PENDING
3562          */
3563         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3564
3565         RETURN(rc);
3566 }
3567
3568 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3569                        struct thandle *th)
3570 {
3571         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3572         struct osd_object *obj = osd_dt_obj(dt);
3573         struct inode *inode = obj->oo_inode;
3574         struct osd_device *osd = osd_obj2dev(obj);
3575         struct osd_thandle *oh;
3576         int result;
3577
3578         ENTRY;
3579
3580         oh = container_of0(th, struct osd_thandle, ot_super);
3581         LASSERT(oh->ot_handle);
3582         LASSERT(inode);
3583         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3584
3585         if (unlikely(fid_is_acct(fid)))
3586                 RETURN(-EPERM);
3587
3588         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3589                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3590                 if (result != 0)
3591                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3592                                osd_name(osd), PFID(fid), result);
3593         }
3594
3595         if (S_ISDIR(inode->i_mode)) {
3596                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3597                         inode->i_nlink == 2);
3598
3599                 spin_lock(&obj->oo_guard);
3600                 clear_nlink(inode);
3601                 spin_unlock(&obj->oo_guard);
3602                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3603         }
3604
3605         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3606
3607         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3608
3609         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3610                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3611                                        oh->ot_handle, OI_CHECK_FLD);
3612
3613         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3614         /* XXX: add to ext3 orphan list */
3615         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3616
3617         /* not needed in the cache anymore */
3618         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3619         obj->oo_destroyed = 1;
3620
3621         RETURN(0);
3622 }
3623
3624 /**
3625  * Put the fid into lustre_mdt_attrs, and then place the structure
3626  * inode's ea. This fid should not be altered during the life time
3627  * of the inode.
3628  *
3629  * \retval +ve, on success
3630  * \retval -ve, on error
3631  *
3632  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3633  */
3634 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3635                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3636 {
3637         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3638         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3639         int rc;
3640
3641         ENTRY;
3642
3643         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3644                 RETURN(0);
3645
3646         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3647                 rc = -ENOMEM;
3648
3649         lustre_loa_init(loa, fid, compat, incompat);
3650         lustre_loa_swab(loa, false);
3651
3652         /*
3653          * For the OST device with 256 bytes inode size by default,
3654          * the PFID EA will be stored together with LMA EA to avoid
3655          * performance trouble. Otherwise the PFID EA can be stored
3656          * independently. LU-8998
3657          */
3658         if ((compat & LMAC_FID_ON_OST) &&
3659             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3660                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3661                                      sizeof(*loa), XATTR_CREATE);
3662         else
3663                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3664                                      sizeof(*lma), XATTR_CREATE);
3665         /*
3666          * LMA may already exist, but we need to check that all the
3667          * desired compat/incompat flags have been added.
3668          */
3669         if (unlikely(rc == -EEXIST)) {
3670                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3671                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3672                 if (rc < 0)
3673                         RETURN(rc);
3674
3675                 if (rc < sizeof(*lma))
3676                         RETURN(-EINVAL);
3677
3678                 lustre_loa_swab(loa, true);
3679                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3680                     ((compat == 0 && incompat == 0) ||
3681                      (!(~lma->lma_compat & compat) &&
3682                       !(~lma->lma_incompat & incompat))))
3683                         RETURN(0);
3684
3685                 lma->lma_self_fid = *fid;
3686                 lma->lma_compat |= compat;
3687                 lma->lma_incompat |= incompat;
3688                 if (rc == sizeof(*lma)) {
3689                         lustre_lma_swab(lma);
3690                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3691                                              sizeof(*lma), XATTR_REPLACE);
3692                 } else {
3693                         lustre_loa_swab(loa, false);
3694                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3695                                              sizeof(*loa), XATTR_REPLACE);
3696                 }
3697         }
3698
3699         RETURN(rc);
3700 }
3701
3702 /**
3703  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3704  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3705  * To have compatilibility with 1.8 ldiskfs driver we need to have
3706  * magic number at start of fid data.
3707  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3708  * its inmemory API.
3709  */
3710 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3711                                          const struct lu_fid *fid)
3712 {
3713         if (!fid_is_namespace_visible(fid) ||
3714             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3715                 param->edp_magic = 0;
3716                 return;
3717         }
3718
3719         param->edp_magic = LDISKFS_LUFID_MAGIC;
3720         param->edp_len =  sizeof(struct lu_fid) + 1;
3721         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3722 }
3723
3724 /**
3725  * Try to read the fid from inode ea into dt_rec.
3726  *
3727  * \param fid object fid.
3728  *
3729  * \retval 0 on success
3730  */
3731 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3732                           __u32 ino, struct lu_fid *fid,
3733                           struct osd_inode_id *id)
3734 {
3735         struct osd_thread_info *info  = osd_oti_get(env);
3736         struct inode *inode;
3737
3738         ENTRY;
3739
3740         osd_id_gen(id, ino, OSD_OII_NOGEN);
3741         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3742         if (IS_ERR(inode))
3743                 RETURN(PTR_ERR(inode));
3744
3745         iput(inode);
3746         RETURN(0);
3747 }
3748
3749 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3750                                         struct inode *dir,
3751                                         struct inode *parent_dir,
3752                                         const struct lu_fid *dot_fid,
3753                                         const struct lu_fid *dot_dot_fid,
3754                                         struct osd_thandle *oth)
3755 {
3756         struct ldiskfs_dentry_param *dot_ldp;
3757         struct ldiskfs_dentry_param *dot_dot_ldp;
3758         __u32 saved_nlink = dir->i_nlink;
3759         int rc;
3760
3761         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3762         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3763
3764         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3765         dot_ldp->edp_magic = 0;
3766
3767         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3768                                     dir, dot_ldp, dot_dot_ldp);
3769         /*
3770          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3771          * the subseqent ref_add() will increase the dir->i_nlink
3772          * as 3. That is incorrect for new created directory.
3773          *
3774          * It looks like hack, because we want to make the OSD API
3775          * to be order-independent for new created directory object
3776          * between dt_insert(..) and ref_add() operations.
3777          *
3778          * Here, we only restore the in-RAM dir-inode's nlink attr,
3779          * becuase if the nlink attr is not 2, then there will be
3780          * ref_add() called following the dt_insert(..), such call
3781          * will make both the in-RAM and on-disk dir-inode's nlink
3782          * attr to be set as 2. LU-7447
3783          */
3784         set_nlink(dir, saved_nlink);
3785         return rc;
3786 }
3787
3788 /**
3789  * Create an local agent inode for remote entry
3790  */
3791 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3792                                                   struct osd_device *osd,
3793                                                   struct osd_object *pobj,
3794                                                   const struct lu_fid *fid,
3795                                                   __u32 type,
3796                                                   struct thandle *th)
3797 {
3798         struct osd_thread_info *info = osd_oti_get(env);
3799         struct inode *local;
3800         struct osd_thandle *oh;
3801         int rc;
3802
3803         ENTRY;
3804
3805         LASSERT(th);
3806         oh = container_of(th, struct osd_thandle, ot_super);
3807         LASSERT(oh->ot_handle->h_transaction != NULL);
3808
3809         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3810                                      NULL);
3811         if (IS_ERR(local)) {
3812                 CERROR("%s: create local error %d\n", osd_name(osd),
3813                        (int)PTR_ERR(local));
3814                 RETURN(local);
3815         }
3816
3817         /*
3818          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3819          * correct gid on remote file, not agent here
3820          */
3821         local->i_gid = current_fsgid();
3822         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3823
3824         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3825          * That gives e2fsck something to look at and be happy, and allows
3826          * debugging if we need to determine where this symlink came from.
3827          */
3828         if (S_ISLNK(type)) {
3829                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3830                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3831                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3832
3833                 i_size_write(local, rc);
3834                 LDISKFS_I(local)->i_disksize = rc;
3835         }
3836         unlock_new_inode(local);
3837
3838         /* Agent inode should not have project ID */
3839 #ifdef  HAVE_PROJECT_QUOTA
3840         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3841             i_projid_read(pobj->oo_inode) != 0) {
3842                 rc = osd_transfer_project(local, 0);
3843                 if (rc) {
3844                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3845                                "quota enforcement enabled on the ldiskfs "
3846                                "filesystem?\n", local->i_sb->s_id, rc);
3847                         RETURN(ERR_PTR(rc));
3848                 }
3849         }
3850 #endif
3851         /* Set special LMA flag for local agent inode */
3852         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3853         if (rc != 0) {
3854                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3855                        osd_name(osd), PFID(fid), rc);
3856                 RETURN(ERR_PTR(rc));
3857         }
3858
3859         if (!S_ISDIR(type))
3860                 RETURN(local);
3861
3862         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3863                                          lu_object_fid(&pobj->oo_dt.do_lu),
3864                                          fid, oh);
3865         if (rc != 0) {
3866                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3867                         osd_name(osd), PFID(fid), rc);
3868                 RETURN(ERR_PTR(rc));
3869         }
3870
3871         RETURN(local);
3872 }
3873
3874 /**
3875  * when direntry is deleted, we have to take care of possible agent inode
3876  * referenced by that. unfortunately we can't do this at that point:
3877  * iget() within a running transaction leads to deadlock and we better do
3878  * not call that every delete declaration to save performance. so we put
3879  * a potention agent inode on a list and process that once the transaction
3880  * is over. Notice it's not any worse in terms of real orphans as regular
3881  * object destroy doesn't put inodes on the on-disk orphan list. this should
3882  * be addressed separately
3883  */
3884 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3885                                             struct osd_thandle *oh,
3886                                             __u32 ino)
3887 {
3888         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3889         struct osd_obj_orphan *oor;
3890
3891         OBD_ALLOC_PTR(oor);
3892         if (oor == NULL)
3893                 return -ENOMEM;
3894
3895         oor->oor_ino = ino;
3896         oor->oor_env = (struct lu_env *)env;
3897         spin_lock(&osd->od_osfs_lock);
3898         list_add(&oor->oor_list, &osd->od_orphan_list);
3899         spin_unlock(&osd->od_osfs_lock);
3900
3901         oh->ot_remove_agents = 1;
3902
3903         return 0;
3904
3905 }
3906
3907 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3908                                                 struct osd_device *osd)
3909 {
3910         struct osd_thread_info *info = osd_oti_get(env);
3911         struct osd_obj_orphan *oor, *tmp;
3912         struct osd_inode_id id;
3913         struct list_head list;
3914         struct inode *inode;
3915         struct lu_fid fid;
3916         handle_t *jh;
3917         __u32 ino;
3918
3919         INIT_LIST_HEAD(&list);
3920
3921         spin_lock(&osd->od_osfs_lock);
3922         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3923                 if (oor->oor_env == env) {
3924                         list_del(&oor->oor_list);
3925                         list_add(&oor->oor_list, &list);
3926                 }
3927         }
3928         spin_unlock(&osd->od_osfs_lock);
3929
3930         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3931
3932                 ino = oor->oor_ino;
3933
3934                 list_del(&oor->oor_list);
3935                 OBD_FREE_PTR(oor);
3936
3937                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3938                 inode = osd_iget_fid(info, osd, &id, &fid);
3939                 if (IS_ERR(inode))
3940                         continue;
3941
3942                 if (!osd_remote_fid(env, osd, &fid)) {
3943                         iput(inode);
3944                         continue;
3945                 }
3946
3947                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3948                 clear_nlink(inode);
3949                 mark_inode_dirty(inode);
3950                 ldiskfs_journal_stop(jh);
3951                 iput(inode);
3952         }
3953
3954         return 0;
3955 }
3956
3957 /**
3958  * OSD layer object create function for OST objects (b=11826).
3959  *
3960  * The FID is inserted into inode xattr here.
3961  *
3962  * \retval   0, on success
3963  * \retval -ve, on error
3964  */
3965 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3966                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3967                       struct dt_object_format *dof, struct thandle *th)
3968 {
3969         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3970         struct osd_object *obj = osd_dt_obj(dt);
3971         struct osd_thread_info *info = osd_oti_get(env);
3972         int result, on_ost = 0;
3973
3974         ENTRY;
3975
3976         if (dt_object_exists(dt))
3977                 RETURN(-EEXIST);
3978
3979         LINVRNT(osd_invariant(obj));
3980         LASSERT(!dt_object_remote(dt));
3981         LASSERT(osd_is_write_locked(env, obj));
3982         LASSERT(th != NULL);
3983
3984         if (unlikely(fid_is_acct(fid)))
3985                 /*
3986                  * Quota files can't be created from the kernel any more,
3987                  * 'tune2fs -O quota' will take care of creating them
3988                  */
3989                 RETURN(-EPERM);
3990
3991         result = __osd_create(info, obj, attr, hint, dof, th);
3992         if (result == 0) {
3993                 if (fid_is_idif(fid) &&
3994                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3995                         struct lu_fid *tfid = &info->oti_fid;
3996                         struct ost_id *oi   = &info->oti_ostid;
3997
3998                         fid_to_ostid(fid, oi);
3999                         ostid_to_fid(tfid, oi, 0);
4000                         on_ost = 1;
4001                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4002                                                 LMAC_FID_ON_OST, 0);
4003                 } else {
4004                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4005                                                fid, OI_CHECK_FLD);
4006                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4007                                                 on_ost ? LMAC_FID_ON_OST : 0,
4008                                                 0);
4009                 }
4010                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4011                         obj->oo_dt.do_body_ops = &osd_body_ops;
4012         }
4013
4014         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4015                 result = __osd_oi_insert(env, obj, fid, th);
4016
4017         /*
4018          * a small optimization - dt_insert() isn't usually applied
4019          * to OST objects, so we don't need to cache OI mapping for
4020          * OST objects
4021          */
4022         if (result == 0 && on_ost == 0) {
4023                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4024
4025                 result = osd_idc_find_and_init(env, osd, obj);
4026                 LASSERT(result == 0);
4027         }
4028
4029         LASSERT(ergo(result == 0,
4030                      dt_object_exists(dt) && !dt_object_remote(dt)));
4031         LINVRNT(osd_invariant(obj));
4032         RETURN(result);
4033 }
4034
4035 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4036                                struct thandle *handle)
4037 {
4038         struct osd_thandle *oh;
4039
4040         /* it's possible that object doesn't exist yet */
4041         LASSERT(handle != NULL);
4042
4043         oh = container_of0(handle, struct osd_thandle, ot_super);
4044         LASSERT(oh->ot_handle == NULL);
4045
4046         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4047                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4048
4049         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4050
4051         return 0;
4052 }
4053
4054 /*
4055  * Concurrency: @dt is write locked.
4056  */
4057 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4058                        struct thandle *th)
4059 {
4060         struct osd_object *obj = osd_dt_obj(dt);
4061         struct inode *inode = obj->oo_inode;
4062         struct osd_thandle *oh;
4063         int rc = 0;
4064
4065         if (!dt_object_exists(dt) || obj->oo_destroyed)
4066                 return -ENOENT;
4067
4068         LINVRNT(osd_invariant(obj));
4069         LASSERT(!dt_object_remote(dt));
4070         LASSERT(osd_is_write_locked(env, obj));
4071         LASSERT(th != NULL);
4072
4073         oh = container_of0(th, struct osd_thandle, ot_super);
4074         LASSERT(oh->ot_handle != NULL);
4075
4076         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4077
4078         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4079                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4080         /*
4081          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4082          * (65000) subdirectories by storing "1" in i_nlink if the link count
4083          * would otherwise overflow. Directory tranversal tools understand
4084          * that (st_nlink == 1) indicates that the filesystem dose not track
4085          * hard links count on the directory, and will not abort subdirectory
4086          * scanning early once (st_nlink - 2) subdirs have been found.
4087          *
4088          * This also has to properly handle the case of inodes with nlink == 0
4089          * in case they are being linked into the PENDING directory
4090          */
4091         spin_lock(&obj->oo_guard);
4092         if (unlikely(inode->i_nlink == 0))
4093                 /* inc_nlink from 0 may cause WARN_ON */
4094                 set_nlink(inode, 1);
4095         else {
4096                 ldiskfs_inc_count(oh->ot_handle, inode);
4097                 if (!S_ISDIR(inode->i_mode))
4098                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4099         }
4100         spin_unlock(&obj->oo_guard);
4101
4102         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4103         LINVRNT(osd_invariant(obj));
4104
4105         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4106
4107         return rc;
4108 }
4109
4110 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4111                                struct thandle *handle)
4112 {
4113         struct osd_thandle *oh;
4114
4115         if (!dt_object_exists(dt))
4116                 return -ENOENT;
4117
4118         LASSERT(!dt_object_remote(dt));
4119         LASSERT(handle != NULL);
4120
4121         oh = container_of0(handle, struct osd_thandle, ot_super);
4122         LASSERT(oh->ot_handle == NULL);
4123
4124         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4125                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4126
4127         return 0;
4128 }
4129
4130 /*
4131  * Concurrency: @dt is write locked.
4132  */
4133 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4134                        struct thandle *th)
4135 {
4136         struct osd_object *obj = osd_dt_obj(dt);
4137         struct inode *inode = obj->oo_inode;
4138         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4139         struct osd_thandle *oh;
4140
4141         if (!dt_object_exists(dt))
4142                 return -ENOENT;
4143
4144         LINVRNT(osd_invariant(obj));
4145         LASSERT(!dt_object_remote(dt));
4146         LASSERT(osd_is_write_locked(env, obj));
4147         LASSERT(th != NULL);
4148
4149         oh = container_of0(th, struct osd_thandle, ot_super);
4150         LASSERT(oh->ot_handle != NULL);
4151
4152         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4153
4154         spin_lock(&obj->oo_guard);
4155         /*
4156          * That can be result of upgrade from old Lustre version and
4157          * applied only to local files.  Just skip this ref_del call.
4158          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4159          */
4160         if (inode->i_nlink == 0) {
4161                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4162                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4163                              ", maybe an upgraded file? (LU-3915)\n",
4164                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4165                 spin_unlock(&obj->oo_guard);
4166                 return 0;
4167         }
4168
4169         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4170                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4171
4172         ldiskfs_dec_count(oh->ot_handle, inode);
4173         spin_unlock(&obj->oo_guard);
4174
4175         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4176         LINVRNT(osd_invariant(obj));
4177
4178         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4179
4180         return 0;
4181 }
4182
4183 /*
4184  * Concurrency: @dt is read locked.
4185  */
4186 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4187                          struct lu_buf *buf, const char *name)
4188 {
4189         struct osd_object      *obj    = osd_dt_obj(dt);
4190         struct inode           *inode  = obj->oo_inode;
4191         struct osd_thread_info *info   = osd_oti_get(env);
4192         struct dentry          *dentry = &info->oti_obj_dentry;
4193         bool cache_xattr = false;
4194         int rc;
4195
4196         LASSERT(buf);
4197
4198         /* version get is not real XATTR but uses xattr API */
4199         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4200                 dt_obj_version_t *ver = buf->lb_buf;
4201
4202                 /*
4203                  * for version we are just using xattr API but change inode
4204                  * field instead
4205                  */
4206                 if (buf->lb_len == 0)
4207                         return sizeof(dt_obj_version_t);
4208
4209                 if (buf->lb_len < sizeof(dt_obj_version_t))
4210                         return -ERANGE;
4211
4212                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4213                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4214
4215                 *ver = LDISKFS_I(inode)->i_fs_version;
4216
4217                 return sizeof(dt_obj_version_t);
4218         }
4219
4220         if (!dt_object_exists(dt))
4221                 return -ENOENT;
4222
4223         LASSERT(!dt_object_remote(dt));
4224         LASSERT(inode->i_op != NULL);
4225 #ifdef HAVE_IOP_XATTR
4226         LASSERT(inode->i_op->getxattr != NULL);
4227 #endif
4228
4229         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4230             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4231                 cache_xattr = true;
4232
4233         if (cache_xattr) {
4234                 rc = osd_oxc_get(obj, name, buf);
4235                 if (rc != -ENOENT)
4236                         return rc;
4237         }
4238
4239         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4240                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4241                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4242                 struct filter_fid *ff;
4243                 struct ost_layout *ol;
4244
4245                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4246                 if (rc)
4247                         return rc;
4248
4249                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4250
4251                 rc = sizeof(*ff);
4252                 if (buf->lb_len == 0 || !buf->lb_buf)
4253                         return rc;
4254
4255                 if (buf->lb_len < rc)
4256                         return -ERANGE;
4257
4258                 ff = buf->lb_buf;
4259                 ol = &ff->ff_layout;
4260                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4261                                                   PFID_STRIPE_IDX_BITS);
4262                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4263                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4264                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4265                 if (lma->lma_compat & LMAC_COMP_INFO) {
4266                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4267                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4268                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4269                 } else {
4270                         ol->ol_comp_start = 0;
4271                         ol->ol_comp_end = 0;
4272                         ol->ol_comp_id = 0;
4273                 }
4274         } else {
4275                 rc = __osd_xattr_get(inode, dentry, name,
4276                                      buf->lb_buf, buf->lb_len);
4277         }
4278
4279         if (cache_xattr) {
4280                 if (rc == -ENOENT || rc == -ENODATA)
4281                         osd_oxc_add(obj, name, NULL, 0);
4282                 else if (rc > 0 && buf->lb_buf != NULL)
4283                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4284         }
4285
4286         return rc;
4287 }
4288
4289 static int osd_declare_xattr_set(const struct lu_env *env,
4290                                  struct dt_object *dt,
4291                                  const struct lu_buf *buf, const char *name,
4292                                  int fl, struct thandle *handle)
4293 {
4294         struct osd_thandle *oh;
4295         int credits = 0;
4296         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4297
4298         LASSERT(handle != NULL);
4299
4300         oh = container_of0(handle, struct osd_thandle, ot_super);
4301         LASSERT(oh->ot_handle == NULL);
4302
4303         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4304                 /*
4305                  * For non-upgrading case, the LMA is set first and
4306                  * usually fit inode. But for upgrade case, the LMA
4307                  * may be in another separated EA block.
4308                  */
4309                 if (dt_object_exists(dt)) {
4310                         if (fl == LU_XATTR_REPLACE)
4311                                 credits = 1;
4312                         else
4313                                 goto upgrade;
4314                 }
4315         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4316                 credits = 1;
4317         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4318                 /* We may need to delete the old PFID EA. */
4319                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4320                 if (fl == LU_XATTR_REPLACE)
4321                         credits += 1;
4322                 else
4323                         goto upgrade;
4324         } else {
4325                 /*
4326                  * If some name entry resides on remote MDT, then will create
4327                  * agent entry under remote parent. On the other hand, if the
4328                  * remote entry will be removed, then related agent entry may
4329                  * need to be removed from the remote parent. So there may be
4330                  * kinds of cases, let's declare enough credits. The credits
4331                  * for create agent entry is enough for remove case.
4332                  */
4333                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4334                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4335                         if (dt_object_exists(dt))
4336                                 credits += 1; /* For updating LMA */
4337                 }
4338
4339 upgrade:
4340                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4341
4342                 if (buf != NULL) {
4343                         ssize_t buflen;
4344
4345                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4346                                 /*
4347                                  * learn xattr size from osd_xattr_get if
4348                                  * attribute has not been read yet
4349                                  */
4350                                 buflen = __osd_xattr_get(
4351                                     osd_dt_obj(dt)->oo_inode,
4352                                     &osd_oti_get(env)->oti_obj_dentry,
4353                                     name, NULL, 0);
4354                                 if (buflen < 0)
4355                                         buflen = 0;
4356                         } else {
4357                                 buflen = buf->lb_len;
4358                         }
4359
4360                         if (buflen > sb->s_blocksize) {
4361                                 credits += osd_calc_bkmap_credits(
4362                                     sb, NULL, 0, -1,
4363                                     (buflen + sb->s_blocksize - 1) >>
4364                                     sb->s_blocksize_bits);
4365                         }
4366                 }
4367                 /*
4368                  * xattr set may involve inode quota change, reserve credits for
4369                  * dquot_initialize()
4370                  */
4371                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4372         }
4373
4374         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4375
4376         return 0;
4377 }
4378
4379 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4380                               const struct lu_buf *buf, int fl,
4381                               struct thandle *handle)
4382 {
4383         struct osd_thread_info *info = osd_oti_get(env);
4384         struct dentry *dentry = &info->oti_obj_dentry;
4385         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4386         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4387         struct inode *inode = obj->oo_inode;
4388         struct filter_fid *ff = buf->lb_buf;
4389         struct ost_layout *ol = &ff->ff_layout;
4390         int flags = XATTR_REPLACE;
4391         int rc;
4392
4393         ENTRY;
4394
4395         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4396                 RETURN(-EINVAL);
4397
4398         rc = osd_get_lma(info, inode, dentry, loa);
4399         if (rc == -ENODATA) {
4400                 /* Usually for upgarding from old device */
4401                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4402                                 LMAC_FID_ON_OST, 0);
4403                 flags = XATTR_CREATE;
4404         } else if (rc) {
4405                 RETURN(rc);
4406         }
4407
4408         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4409                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4410                         RETURN(-EEXIST);
4411
4412                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4413                         /* Separate PFID EA from LMA */
4414                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4415                         lustre_lma_swab(lma);
4416                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4417                                              sizeof(*lma), XATTR_REPLACE);
4418                         if (!rc) {
4419                                 obj->oo_pfid_in_lma = 0;
4420                                 rc = LU_XATTR_CREATE;
4421                         }
4422
4423                         RETURN(rc);
4424                 }
4425         } else {
4426                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4427                         RETURN(fl);
4428
4429                 /*
4430                  * Old client does not send stripe information,
4431                  * then store the PFID EA on disk separatedly.
4432                  */
4433                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4434                              ol->ol_stripe_size == 0))
4435                         RETURN(fl);
4436
4437                 /* Remove old PFID EA entry firstly. */
4438                 dquot_initialize(inode);
4439                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4440                 if (rc == -ENODATA) {
4441                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4442                                 RETURN(rc);
4443                 } else if (rc) {
4444                         RETURN(rc);
4445                 }
4446         }
4447
4448         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4449         if (likely(ol->ol_stripe_size != 0)) {
4450                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4451                                              PFID_STRIPE_IDX_BITS;
4452                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4453                 lma->lma_compat |= LMAC_STRIPE_INFO;
4454                 if (ol->ol_comp_id != 0) {
4455                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4456                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4457                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4458                         lma->lma_compat |= LMAC_COMP_INFO;
4459                 }
4460         }
4461
4462         lustre_loa_swab(loa, false);
4463
4464         /* Store the PFID EA inside LMA. */
4465         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4466                              flags);
4467         if (!rc)
4468                 obj->oo_pfid_in_lma = 1;
4469
4470         RETURN(rc);
4471 }
4472
4473 /*
4474  * In DNE environment, the object (in spite of regular file or directory)
4475  * and its name entry may reside on different MDTs. Under such case, we will
4476  * create an agent entry on the MDT where the object resides. The agent entry
4477  * references the object locally, that makes the object to be visible to the
4478  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4479  * such as 'tar' can handle the object properly.
4480  *
4481  * We handle the agent entry during set linkEA that is the common interface
4482  * for both regular file and directroy, can handle kinds of cases, such as
4483  * create/link/unlink/rename, and so on.
4484  *
4485  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4486  *
4487  * XXX: There are two known issues:
4488  * 1. For one object, we will create at most one agent entry even if there
4489  *    may be more than one cross-MDTs hard links on the object. So the local
4490  *    e2fsck may claim that the object's nlink is larger than the name entries
4491  *    that reference such inode. And in further, the e2fsck will fix the nlink
4492  *    attribute to match the local references. Then it will cause the object's
4493  *    nlink attribute to be inconsistent with the global references. it is bad
4494  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4495  *    other hand, the global namespace LFSCK can repair the object's attribute
4496  *    according to the linkEA.
4497  * 2. There may be too many hard links on the object as to its linkEA overflow,
4498  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4499  *    case happened, then at this point, we do not know whether there are some
4500  *    cross-MDTs reference. But there are local references, it guarantees that
4501  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4502  */
4503 static int osd_xattr_handle_linkea(const struct lu_env *env,
4504                                    struct osd_device *osd,
4505                                    struct osd_object *obj,
4506                                    const struct lu_buf *buf,
4507                                    struct thandle *handle)
4508 {
4509         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4510         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4511         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4512         struct lu_name tmpname;
4513         struct osd_thandle *oh;
4514         int rc;
4515         bool remote = false;
4516
4517         ENTRY;
4518
4519         oh = container_of0(handle, struct osd_thandle, ot_super);
4520         LASSERT(oh->ot_handle != NULL);
4521
4522         rc = linkea_init_with_rec(&ldata);
4523         if (!rc) {
4524                 linkea_first_entry(&ldata);
4525                 while (ldata.ld_lee != NULL && !remote) {
4526                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4527                                             &tmpname, tfid);
4528                         if (osd_remote_fid(env, osd, tfid) > 0)
4529                                 remote = true;
4530                         else
4531                                 linkea_next_entry(&ldata);
4532                 }
4533         } else if (rc == -ENODATA) {
4534                 rc = 0;
4535         } else {
4536                 RETURN(rc);
4537         }
4538
4539         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4540                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4541                 if (rc)
4542                         CERROR("%s: failed to remove agent entry for "DFID
4543                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4544         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4545                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4546                 if (rc)
4547                         CERROR("%s: failed to create agent entry for "DFID
4548                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4549         }
4550
4551         RETURN(rc);
4552 }
4553
4554 /*
4555  * Concurrency: @dt is write locked.
4556  */
4557 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4558                          const struct lu_buf *buf, const char *name, int fl,
4559                          struct thandle *handle)
4560 {
4561         struct osd_object *obj = osd_dt_obj(dt);
4562         struct osd_device *osd = osd_obj2dev(obj);
4563         struct inode *inode = obj->oo_inode;
4564         struct osd_thread_info *info = osd_oti_get(env);
4565         int fs_flags = 0;
4566         int len;
4567         int rc;
4568
4569         ENTRY;
4570
4571         LASSERT(handle);
4572         LASSERT(buf);
4573
4574         /* version set is not real XATTR */
4575         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4576                 dt_obj_version_t *version = buf->lb_buf;
4577
4578                 /*
4579                  * for version we are just using xattr API but change inode
4580                  * field instead
4581                  */
4582                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4583
4584                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4585                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4586
4587                 LDISKFS_I(inode)->i_fs_version = *version;
4588                 /*
4589                  * Version is set after all inode operations are finished,
4590                  * so we should mark it dirty here
4591                  */
4592                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4593
4594                 RETURN(0);
4595         }
4596
4597         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4598                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4599
4600         len = buf->lb_len;
4601         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4602
4603         /*
4604          * For the OST device with 256 bytes inode size by default,
4605          * the PFID EA will be stored together with LMA EA to avoid
4606          * performance trouble. Otherwise the PFID EA can be stored
4607          * independently. LU-8998
4608          */
4609         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4610             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4611                 LASSERT(buf->lb_buf);
4612
4613                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4614                 if (fl <= 0)
4615                         RETURN(fl);
4616         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4617                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4618                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4619
4620                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4621                 if (rc)
4622                         RETURN(rc);
4623
4624                 lma->lma_incompat |= LMAI_STRIPED;
4625                 lustre_lma_swab(lma);
4626                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4627                                      sizeof(*lma), XATTR_REPLACE);
4628                 if (rc != 0)
4629                         RETURN(rc);
4630         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4631                 LASSERT(!osd->od_is_ost);
4632
4633                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4634                 if (rc)
4635                         RETURN(rc);
4636         }
4637
4638         if (fl & LU_XATTR_REPLACE)
4639                 fs_flags |= XATTR_REPLACE;
4640
4641         if (fl & LU_XATTR_CREATE)
4642                 fs_flags |= XATTR_CREATE;
4643
4644         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4645         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4646
4647         if (rc == 0 &&
4648             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4649              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4650                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4651
4652         return rc;
4653 }
4654
4655 /*
4656  * Concurrency: @dt is read locked.
4657  */
4658 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4659                           const struct lu_buf *buf)
4660 {
4661         struct osd_object *obj = osd_dt_obj(dt);
4662         struct inode *inode = obj->oo_inode;
4663         struct osd_thread_info *info = osd_oti_get(env);
4664         struct dentry *dentry = &info->oti_obj_dentry;
4665
4666         if (!dt_object_exists(dt))
4667                 return -ENOENT;
4668
4669         LASSERT(!dt_object_remote(dt));
4670         LASSERT(inode->i_op != NULL);
4671         LASSERT(inode->i_op->listxattr != NULL);
4672
4673         dentry->d_inode = inode;
4674         dentry->d_sb = inode->i_sb;
4675         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4676 }
4677
4678 static int osd_declare_xattr_del(const struct lu_env *env,
4679                                  struct dt_object *dt, const char *name,
4680                                  struct thandle *handle)
4681 {
4682         struct osd_thandle *oh;
4683         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4684
4685         LASSERT(!dt_object_remote(dt));
4686         LASSERT(handle != NULL);
4687
4688         oh = container_of0(handle, struct osd_thandle, ot_super);
4689         LASSERT(oh->ot_handle == NULL);
4690
4691         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4692                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4693         /*
4694          * xattr del may involve inode quota change, reserve credits for
4695          * dquot_initialize()
4696          */
4697         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4698
4699         return 0;
4700 }
4701
4702 /*
4703  * Concurrency: @dt is write locked.
4704  */
4705 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4706                          const char *name, struct thandle *handle)
4707 {
4708         struct osd_object *obj = osd_dt_obj(dt);
4709         struct inode *inode = obj->oo_inode;
4710         struct osd_thread_info *info = osd_oti_get(env);
4711         struct dentry *dentry = &info->oti_obj_dentry;
4712         int rc;
4713
4714         if (!dt_object_exists(dt))
4715                 return -ENOENT;
4716
4717         LASSERT(!dt_object_remote(dt));
4718         LASSERT(inode->i_op != NULL);
4719         LASSERT(handle != NULL);
4720 #ifdef HAVE_IOP_XATTR
4721         LASSERT(inode->i_op->removexattr != NULL);
4722 #endif
4723
4724         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4725
4726         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4727                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4728
4729                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4730                                  &info->oti_ost_attrs);
4731                 if (!rc) {
4732                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4733
4734                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4735                         lustre_lma_swab(lma);
4736                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4737                                              sizeof(*lma), XATTR_REPLACE);
4738                         if (!rc)
4739                                 obj->oo_pfid_in_lma = 0;
4740                 }
4741         } else {
4742                 dquot_initialize(inode);
4743                 dentry->d_inode = inode;
4744                 dentry->d_sb = inode->i_sb;
4745                 rc = osd_removexattr(dentry, inode, name);
4746         }
4747
4748         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4749
4750         if (rc == 0 &&
4751             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4752              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4753                 osd_oxc_del(obj, name);
4754
4755         return rc;
4756 }
4757
4758 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4759                            __u64 start, __u64 end)
4760 {
4761         struct osd_object *obj = osd_dt_obj(dt);
4762         struct inode *inode = obj->oo_inode;
4763         struct osd_thread_info *info = osd_oti_get(env);
4764         struct dentry *dentry = &info->oti_obj_dentry;
4765         struct file *file = &info->oti_file;
4766         int rc;
4767
4768         ENTRY;
4769
4770         dentry->d_inode = inode;
4771         dentry->d_sb = inode->i_sb;
4772         file->f_path.dentry = dentry;
4773         file->f_mapping = inode->i_mapping;
4774         file->f_op = inode->i_fop;
4775         set_file_inode(file, inode);
4776
4777         rc = vfs_fsync_range(file, start, end, 0);
4778
4779         RETURN(rc);
4780 }
4781
4782 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4783 {
4784         return 0;
4785 }
4786
4787 /*
4788  * Index operations.
4789  */
4790
4791 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4792                                const struct dt_index_features *feat)
4793 {
4794         struct iam_descr *descr;
4795
4796         if (osd_object_is_root(o))
4797                 return feat == &dt_directory_features;
4798
4799         LASSERT(o->oo_dir != NULL);
4800
4801         descr = o->oo_dir->od_container.ic_descr;
4802         if (feat == &dt_directory_features) {
4803                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4804                         return 1;
4805                 else
4806                         return 0;
4807         } else {
4808                 return feat->dif_keysize_min <= descr->id_key_size &&
4809                        descr->id_key_size <= feat->dif_keysize_max &&
4810                        feat->dif_recsize_min <= descr->id_rec_size &&
4811                        descr->id_rec_size <= feat->dif_recsize_max &&
4812                        !(feat->dif_flags & (DT_IND_VARKEY |
4813                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4814                        ergo(feat->dif_flags & DT_IND_UPDATE,
4815                             1 /* XXX check that object (and fs) is writable */);
4816         }
4817 }
4818
4819 static int osd_iam_container_init(const struct lu_env *env,
4820                                   struct osd_object *obj,
4821                                   struct osd_directory *dir)
4822 {
4823         struct iam_container *bag = &dir->od_container;
4824         int result;
4825
4826         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4827         if (result != 0)
4828                 return result;
4829
4830         result = iam_container_setup(bag);
4831         if (result == 0)
4832                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4833         else
4834                 iam_container_fini(bag);
4835
4836         return result;
4837 }
4838
4839
4840 /*
4841  * Concurrency: no external locking is necessary.
4842  */
4843 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4844                          const struct dt_index_features *feat)
4845 {
4846         int result;
4847         int skip_iam = 0;
4848         struct osd_object *obj = osd_dt_obj(dt);
4849
4850         LINVRNT(osd_invariant(obj));
4851
4852         if (osd_object_is_root(obj)) {
4853                 dt->do_index_ops = &osd_index_ea_ops;
4854                 result = 0;
4855         } else if (feat == &dt_directory_features) {
4856                 dt->do_index_ops = &osd_index_ea_ops;
4857                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4858                         result = 0;
4859                 else
4860                         result = -ENOTDIR;
4861                 skip_iam = 1;
4862         } else if (unlikely(feat == &dt_otable_features)) {
4863                 dt->do_index_ops = &osd_otable_ops;
4864                 return 0;
4865         } else if (unlikely(feat == &dt_acct_features)) {
4866                 dt->do_index_ops = &osd_acct_index_ops;
4867                 result = 0;
4868                 skip_iam = 1;
4869         } else if (!osd_has_index(obj)) {
4870                 struct osd_directory *dir;
4871                 struct osd_device *osd = osd_obj2dev(obj);
4872                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4873
4874                 OBD_ALLOC_PTR(dir);
4875                 if (dir) {
4876
4877                         spin_lock(&obj->oo_guard);
4878                         if (obj->oo_dir == NULL)
4879                                 obj->oo_dir = dir;
4880                         else
4881                                 /*
4882                                  * Concurrent thread allocated container data.
4883                                  */
4884                                 OBD_FREE_PTR(dir);
4885                         spin_unlock(&obj->oo_guard);
4886                         /*
4887                          * Now, that we have container data, serialize its
4888                          * initialization.
4889                          */
4890                         down_write(&obj->oo_ext_idx_sem);
4891                         /*
4892                          * recheck under lock.
4893                          */
4894
4895                         if (osd_has_index(obj)) {
4896                                 result = 0;
4897                                 goto unlock;
4898                         }
4899
4900                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4901                         if (result || feat == &dt_lfsck_namespace_features ||
4902                             feat == &dt_lfsck_layout_orphan_features ||
4903                             feat == &dt_lfsck_layout_dangling_features)
4904                                 goto unlock;
4905
4906                         result = osd_index_register(osd, fid,
4907                                                     feat->dif_keysize_max,
4908                                                     feat->dif_recsize_max);
4909                         if (result < 0)
4910                                 CWARN("%s: failed to register index "
4911                                       DFID": rc = %d\n",
4912                                       osd_name(osd), PFID(fid), result);
4913                         else if (result > 0)
4914                                 result = 0;
4915                         else
4916                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4917                                        " (%d/%d) registered\n",
4918                                        osd_name(osd), PFID(fid),
4919                                        (int)feat->dif_keysize_max,
4920                                        (int)feat->dif_recsize_max);
4921
4922 unlock:
4923                         up_write(&obj->oo_ext_idx_sem);
4924                 } else {
4925                         result = -ENOMEM;
4926                 }
4927         } else {
4928                 result = 0;
4929         }
4930
4931         if (result == 0 && skip_iam == 0) {
4932                 if (!osd_iam_index_probe(env, obj, feat))
4933                         result = -ENOTDIR;
4934         }
4935         LINVRNT(osd_invariant(obj));
4936
4937         return result;
4938 }
4939
4940 static int osd_otable_it_attr_get(const struct lu_env *env,
4941                                  struct dt_object *dt,
4942                                  struct lu_attr *attr)
4943 {
4944         attr->la_valid = 0;
4945         return 0;
4946 }
4947
4948 static const struct dt_object_operations osd_obj_ops = {
4949         .do_read_lock           = osd_read_lock,
4950         .do_write_lock          = osd_write_lock,
4951         .do_read_unlock         = osd_read_unlock,
4952         .do_write_unlock        = osd_write_unlock,
4953         .do_write_locked        = osd_write_locked,
4954         .do_attr_get            = osd_attr_get,
4955         .do_declare_attr_set    = osd_declare_attr_set,
4956         .do_attr_set            = osd_attr_set,
4957         .do_ah_init             = osd_ah_init,
4958         .do_declare_create      = osd_declare_create,
4959         .do_create              = osd_create,
4960         .do_declare_destroy     = osd_declare_destroy,
4961         .do_destroy             = osd_destroy,
4962         .do_index_try           = osd_index_try,
4963         .do_declare_ref_add     = osd_declare_ref_add,
4964         .do_ref_add             = osd_ref_add,
4965         .do_declare_ref_del     = osd_declare_ref_del,
4966         .do_ref_del             = osd_ref_del,
4967         .do_xattr_get           = osd_xattr_get,
4968         .do_declare_xattr_set   = osd_declare_xattr_set,
4969         .do_xattr_set           = osd_xattr_set,
4970         .do_declare_xattr_del   = osd_declare_xattr_del,
4971         .do_xattr_del           = osd_xattr_del,
4972         .do_xattr_list          = osd_xattr_list,
4973         .do_object_sync         = osd_object_sync,
4974         .do_invalidate          = osd_invalidate,
4975 };
4976
4977 static const struct dt_object_operations osd_obj_otable_it_ops = {
4978         .do_attr_get    = osd_otable_it_attr_get,
4979         .do_index_try   = osd_index_try,
4980 };
4981
4982 static int osd_index_declare_iam_delete(const struct lu_env *env,
4983                                         struct dt_object *dt,
4984                                         const struct dt_key *key,
4985                                         struct thandle *handle)
4986 {
4987         struct osd_thandle *oh;
4988
4989         oh = container_of0(handle, struct osd_thandle, ot_super);
4990         LASSERT(oh->ot_handle == NULL);
4991
4992         /* Recycle  may cause additional three blocks to be changed. */
4993         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4994                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4995
4996         return 0;
4997 }
4998
4999 /**
5000  *      delete a (key, value) pair from index \a dt specified by \a key
5001  *
5002  *      \param  dt      osd index object
5003  *      \param  key     key for index
5004  *      \param  rec     record reference
5005  *      \param  handle  transaction handler
5006  *
5007  *      \retval  0  success
5008  *      \retval -ve   failure
5009  */
5010 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5011                                 const struct dt_key *key,
5012                                 struct thandle *handle)
5013 {
5014         struct osd_thread_info *oti = osd_oti_get(env);
5015         struct osd_object *obj = osd_dt_obj(dt);
5016         struct osd_thandle *oh;
5017         struct iam_path_descr *ipd;
5018         struct iam_container *bag = &obj->oo_dir->od_container;
5019         int rc;
5020
5021         ENTRY;
5022
5023         if (!dt_object_exists(dt))
5024                 RETURN(-ENOENT);
5025
5026         LINVRNT(osd_invariant(obj));
5027         LASSERT(!dt_object_remote(dt));
5028         LASSERT(bag->ic_object == obj->oo_inode);
5029         LASSERT(handle != NULL);
5030
5031         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5032
5033         ipd = osd_idx_ipd_get(env, bag);
5034         if (unlikely(ipd == NULL))
5035                 RETURN(-ENOMEM);
5036
5037         oh = container_of0(handle, struct osd_thandle, ot_super);
5038         LASSERT(oh->ot_handle != NULL);
5039         LASSERT(oh->ot_handle->h_transaction != NULL);
5040
5041         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5042                 /* swab quota uid/gid provided by caller */
5043                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5044                 key = (const struct dt_key *)&oti->oti_quota_id;
5045         }
5046
5047         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5048         osd_ipd_put(env, bag, ipd);
5049         LINVRNT(osd_invariant(obj));
5050         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5051         RETURN(rc);
5052 }
5053
5054 static int osd_index_declare_ea_delete(const struct lu_env *env,
5055                                        struct dt_object *dt,
5056                                        const struct dt_key *key,
5057                                        struct thandle *handle)
5058 {
5059         struct osd_thandle *oh;
5060         struct inode *inode;
5061         int rc, credits;
5062
5063         ENTRY;
5064
5065         LASSERT(!dt_object_remote(dt));
5066         LASSERT(handle != NULL);
5067
5068         oh = container_of0(handle, struct osd_thandle, ot_super);
5069         LASSERT(oh->ot_handle == NULL);
5070
5071         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5072         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5073
5074         inode = osd_dt_obj(dt)->oo_inode;
5075         if (inode == NULL)
5076                 RETURN(-ENOENT);
5077
5078         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5079                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5080                                    NULL, OSD_QID_BLK);
5081         RETURN(rc);
5082 }
5083
5084 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5085                                           struct dt_rec *fid)
5086 {
5087         struct osd_fid_pack *rec;
5088         int rc = -ENODATA;
5089
5090         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5091                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5092                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5093                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5094                         rc = -EINVAL;
5095         }
5096         return rc;
5097 }
5098
5099 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5100                           const struct lu_fid *fid)
5101 {
5102         struct seq_server_site *ss = osd_seq_site(osd);
5103
5104         ENTRY;
5105
5106         /* FID seqs not in FLDB, must be local seq */
5107         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5108                 RETURN(0);
5109
5110         /*
5111          * If FLD is not being initialized yet, it only happens during the
5112          * initialization, likely during mgs initialization, and we assume
5113          * this is local FID.
5114          */
5115         if (ss == NULL || ss->ss_server_fld == NULL)
5116                 RETURN(0);
5117
5118         /* Only check the local FLDB here */
5119         if (osd_seq_exists(env, osd, fid_seq(fid)))
5120                 RETURN(0);
5121
5122         RETURN(1);
5123 }
5124
5125 static void osd_take_care_of_agent(const struct lu_env *env,
5126                                    struct osd_device *osd,
5127                                    struct osd_thandle *oh,
5128                                    struct ldiskfs_dir_entry_2 *de)
5129 {
5130         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5131         struct osd_idmap_cache *idc;
5132         int rc, schedule = 0;
5133
5134         LASSERT(de != NULL);
5135
5136         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5137         if (likely(rc == 0)) {
5138                 idc = osd_idc_find_or_init(env, osd, fid);
5139                 if (IS_ERR(idc) || idc->oic_remote)
5140                         schedule = 1;
5141         } else if (rc == -ENODATA) {
5142                 /*
5143                  * can't get FID, postpone to the end of the
5144                  * transaction when iget() is safe
5145                  */
5146                 schedule = 1;
5147         } else {
5148                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5149         }
5150         if (schedule)
5151                 osd_schedule_agent_inode_removal(env, oh,
5152                                                  le32_to_cpu(de->inode));
5153 }
5154
5155 /**
5156  * Index delete function for interoperability mode (b11826).
5157  * It will remove the directory entry added by osd_index_ea_insert().
5158  * This entry is needed to maintain name->fid mapping.
5159  *
5160  * \param key,  key i.e. file entry to be deleted
5161  *
5162  * \retval   0, on success
5163  * \retval -ve, on error
5164  */
5165 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5166                                const struct dt_key *key, struct thandle *handle)
5167 {
5168         struct osd_object *obj = osd_dt_obj(dt);
5169         struct inode *dir = obj->oo_inode;
5170         struct dentry *dentry;
5171         struct osd_thandle *oh;
5172         struct ldiskfs_dir_entry_2 *de = NULL;
5173         struct buffer_head *bh;
5174         struct htree_lock *hlock = NULL;
5175         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5176         int rc;
5177
5178         ENTRY;
5179
5180         if (!dt_object_exists(dt))
5181                 RETURN(-ENOENT);
5182
5183         LINVRNT(osd_invariant(obj));
5184         LASSERT(!dt_object_remote(dt));
5185         LASSERT(handle != NULL);
5186
5187         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5188
5189         oh = container_of(handle, struct osd_thandle, ot_super);
5190         LASSERT(oh->ot_handle != NULL);
5191         LASSERT(oh->ot_handle->h_transaction != NULL);
5192
5193         dquot_initialize(dir);
5194         dentry = osd_child_dentry_get(env, obj,
5195                                       (char *)key, strlen((char *)key));
5196
5197         if (obj->oo_hl_head != NULL) {
5198                 hlock = osd_oti_get(env)->oti_hlock;
5199                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5200                                    dir, LDISKFS_HLOCK_DEL);
5201         } else {
5202                 down_write(&obj->oo_ext_idx_sem);
5203         }
5204
5205         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5206         if (!IS_ERR(bh)) {
5207                 /*
5208                  * If this is not the ".." entry, it might be a remote DNE
5209                  * entry and  we need to check if the FID is for a remote
5210                  * MDT.  If the FID is  not in the directory entry (e.g.
5211                  * upgraded 1.8 filesystem without dirdata enabled) then
5212                  * we need to get the FID from the LMA. For a remote directory
5213                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5214                  * case.
5215                  *
5216                  * Delete the entry before the agent inode in order to
5217                  * simplify error handling.  At worst an error after deleting
5218                  * the entry first might leak the agent inode afterward. The
5219                  * reverse would need filesystem abort in case of error deleting
5220                  * the entry after the agent had been removed, or leave a
5221                  * dangling entry pointing at a random inode.
5222                  */
5223                 if (strcmp((char *)key, dotdot) != 0)
5224                         osd_take_care_of_agent(env, osd, oh, de);
5225                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5226                 brelse(bh);
5227         } else {
5228                 rc = PTR_ERR(bh);
5229         }
5230         if (hlock != NULL)
5231                 ldiskfs_htree_unlock(hlock);
5232         else
5233                 up_write(&obj->oo_ext_idx_sem);
5234
5235         GOTO(out, rc);
5236 out:
5237         LASSERT(osd_invariant(obj));
5238         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5239         RETURN(rc);
5240 }
5241
5242 /**
5243  *      Lookup index for \a key and copy record to \a rec.
5244  *
5245  *      \param  dt      osd index object
5246  *      \param  key     key for index
5247  *      \param  rec     record reference
5248  *
5249  *      \retval  +ve  success : exact mach
5250  *      \retval  0    return record with key not greater than \a key
5251  *      \retval -ve   failure
5252  */
5253 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5254                                 struct dt_rec *rec, const struct dt_key *key)
5255 {
5256         struct osd_object *obj = osd_dt_obj(dt);
5257         struct iam_path_descr *ipd;
5258         struct iam_container *bag = &obj->oo_dir->od_container;
5259         struct osd_thread_info *oti = osd_oti_get(env);
5260         struct iam_iterator *it = &oti->oti_idx_it;
5261         struct iam_rec *iam_rec;
5262         int rc;
5263
5264         ENTRY;
5265
5266         if (!dt_object_exists(dt))
5267                 RETURN(-ENOENT);
5268
5269         LASSERT(osd_invariant(obj));
5270         LASSERT(!dt_object_remote(dt));
5271         LASSERT(bag->ic_object == obj->oo_inode);
5272
5273         ipd = osd_idx_ipd_get(env, bag);
5274         if (IS_ERR(ipd))
5275                 RETURN(-ENOMEM);
5276
5277         /* got ipd now we can start iterator. */
5278         iam_it_init(it, bag, 0, ipd);
5279
5280         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5281                 /* swab quota uid/gid provided by caller */
5282                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5283                 key = (const struct dt_key *)&oti->oti_quota_id;
5284         }
5285
5286         rc = iam_it_get(it, (struct iam_key *)key);
5287         if (rc >= 0) {
5288                 if (S_ISDIR(obj->oo_inode->i_mode))
5289                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5290                 else
5291                         iam_rec = (struct iam_rec *)rec;
5292
5293                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5294
5295                 if (S_ISDIR(obj->oo_inode->i_mode))
5296                         osd_fid_unpack((struct lu_fid *)rec,
5297                                        (struct osd_fid_pack *)iam_rec);
5298                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5299                         osd_quota_unpack(obj, rec);
5300         }
5301
5302         iam_it_put(it);
5303         iam_it_fini(it);
5304         osd_ipd_put(env, bag, ipd);
5305
5306         LINVRNT(osd_invariant(obj));
5307
5308         RETURN(rc);
5309 }
5310
5311 static int osd_index_declare_iam_insert(const struct lu_env *env,
5312                                         struct dt_object *dt,
5313                                         const struct dt_rec *rec,
5314                                         const struct dt_key *key,
5315                                         struct thandle *handle)
5316 {
5317         struct osd_thandle *oh;
5318
5319         LASSERT(handle != NULL);
5320
5321         oh = container_of0(handle, struct osd_thandle, ot_super);
5322         LASSERT(oh->ot_handle == NULL);
5323
5324         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5325                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5326
5327         return 0;
5328 }
5329
5330 /**
5331  *      Inserts (key, value) pair in \a dt index object.
5332  *
5333  *      \param  dt      osd index object
5334  *      \param  key     key for index
5335  *      \param  rec     record reference
5336  *      \param  th      transaction handler
5337  *
5338  *      \retval  0  success
5339  *      \retval -ve failure
5340  */
5341 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5342                                 const struct dt_rec *rec,
5343                                 const struct dt_key *key, struct thandle *th)
5344 {
5345         struct osd_object *obj = osd_dt_obj(dt);
5346         struct iam_path_descr *ipd;
5347         struct osd_thandle *oh;
5348         struct iam_container *bag;
5349         struct osd_thread_info *oti = osd_oti_get(env);
5350         struct iam_rec *iam_rec;
5351         int rc;
5352
5353         ENTRY;
5354
5355         if (!dt_object_exists(dt))
5356                 RETURN(-ENOENT);
5357
5358         LINVRNT(osd_invariant(obj));
5359         LASSERT(!dt_object_remote(dt));
5360
5361         bag = &obj->oo_dir->od_container;
5362         LASSERT(bag->ic_object == obj->oo_inode);
5363         LASSERT(th != NULL);
5364
5365         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5366
5367         ipd = osd_idx_ipd_get(env, bag);
5368         if (unlikely(ipd == NULL))
5369                 RETURN(-ENOMEM);
5370
5371         oh = container_of0(th, struct osd_thandle, ot_super);
5372         LASSERT(oh->ot_handle != NULL);
5373         LASSERT(oh->ot_handle->h_transaction != NULL);
5374         if (S_ISDIR(obj->oo_inode->i_mode)) {
5375                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5376                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5377                              &oti->oti_fid);
5378         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5379                 /* pack quota uid/gid */
5380                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5381                 key = (const struct dt_key *)&oti->oti_quota_id;
5382                 /* pack quota record */
5383                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5384                 iam_rec = (struct iam_rec *)rec;
5385         } else {
5386                 iam_rec = (struct iam_rec *)rec;
5387         }
5388
5389         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5390                         iam_rec, ipd);
5391         osd_ipd_put(env, bag, ipd);
5392         LINVRNT(osd_invariant(obj));
5393         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5394         RETURN(rc);
5395 }
5396
5397 /**
5398  * Calls ldiskfs_add_entry() to add directory entry
5399  * into the directory. This is required for
5400  * interoperability mode (b11826)
5401  *
5402  * \retval   0, on success
5403  * \retval -ve, on error
5404  */
5405 static int __osd_ea_add_rec(struct osd_thread_info *info,
5406                             struct osd_object *pobj, struct inode  *cinode,
5407                             const char *name, const struct lu_fid *fid,
5408                             struct htree_lock *hlock, struct thandle *th)
5409 {
5410         struct ldiskfs_dentry_param *ldp;
5411         struct dentry *child;
5412         struct osd_thandle *oth;
5413         int rc;
5414
5415         oth = container_of(th, struct osd_thandle, ot_super);
5416         LASSERT(oth->ot_handle != NULL);
5417         LASSERT(oth->ot_handle->h_transaction != NULL);
5418         LASSERT(pobj->oo_inode);
5419
5420         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5421         if (unlikely(pobj->oo_inode ==
5422                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5423                 ldp->edp_magic = 0;
5424         else
5425                 osd_get_ldiskfs_dirent_param(ldp, fid);
5426         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5427         child->d_fsdata = (void *)ldp;
5428         dquot_initialize(pobj->oo_inode);
5429         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5430                                    child, cinode, hlock);
5431         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5432                 struct ldiskfs_dir_entry_2      *de;
5433                 struct buffer_head              *bh;
5434                 int                              rc1;
5435
5436                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5437                                             NULL, hlock);
5438                 if (!IS_ERR(bh)) {
5439                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5440                                                                bh);
5441                         if (rc1 == 0) {
5442                                 if (S_ISDIR(cinode->i_mode))
5443                                         de->file_type = LDISKFS_DIRENT_LUFID |
5444                                                         LDISKFS_FT_REG_FILE;
5445                                 else
5446                                         de->file_type = LDISKFS_DIRENT_LUFID |
5447                                                         LDISKFS_FT_DIR;
5448                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5449                                                               NULL, bh);
5450                         }
5451                         brelse(bh);
5452                 }
5453         }
5454
5455         RETURN(rc);
5456 }
5457
5458 /**
5459  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5460  * into the directory.Also sets flags into osd object to
5461  * indicate dot and dotdot are created. This is required for
5462  * interoperability mode (b11826)
5463  *
5464  * \param dir   directory for dot and dotdot fixup.
5465  * \param obj   child object for linking
5466  *
5467  * \retval   0, on success
5468  * \retval -ve, on error
5469  */
5470 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5471                               struct osd_object *dir,
5472                               struct inode *parent_dir, const char *name,
5473                               const struct lu_fid *dot_fid,
5474                               const struct lu_fid *dot_dot_fid,
5475                               struct thandle *th)
5476 {
5477         struct inode *inode = dir->oo_inode;
5478         struct osd_thandle *oth;
5479         int result = 0;
5480
5481         oth = container_of(th, struct osd_thandle, ot_super);
5482         LASSERT(oth->ot_handle->h_transaction != NULL);
5483         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5484
5485         if (strcmp(name, dot) == 0) {
5486                 if (dir->oo_compat_dot_created) {
5487                         result = -EEXIST;
5488                 } else {
5489                         LASSERT(inode->i_ino == parent_dir->i_ino);
5490                         dir->oo_compat_dot_created = 1;
5491                         result = 0;
5492                 }
5493         } else if (strcmp(name, dotdot) == 0) {
5494                 if (!dir->oo_compat_dot_created)
5495                         return -EINVAL;
5496                 /* in case of rename, dotdot is already created */
5497                 if (dir->oo_compat_dotdot_created) {
5498                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5499                                                 dot_dot_fid, NULL, th);
5500                 }
5501
5502                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5503                         struct lu_fid tfid = *dot_dot_fid;
5504
5505                         tfid.f_oid--;
5506                         result = osd_add_dot_dotdot_internal(info,
5507                                         dir->oo_inode, parent_dir, dot_fid,
5508                                         &tfid, oth);
5509                 } else {
5510                         result = osd_add_dot_dotdot_internal(info,
5511                                         dir->oo_inode, parent_dir, dot_fid,
5512                                         dot_dot_fid, oth);
5513                 }
5514
5515                 if (result == 0)
5516                         dir->oo_compat_dotdot_created = 1;
5517         }
5518
5519         return result;
5520 }
5521
5522
5523 /**
5524  * It will call the appropriate osd_add* function and return the
5525  * value, return by respective functions.
5526  */
5527 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5528                           struct inode *cinode, const char *name,
5529                           const struct lu_fid *fid, struct thandle *th)
5530 {
5531         struct osd_thread_info *info = osd_oti_get(env);
5532         struct htree_lock *hlock;
5533         int rc;
5534
5535         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5536
5537         if (name[0] == '.' && (name[1] == '\0' ||
5538                                (name[1] == '.' && name[2] == '\0'))) {
5539                 if (hlock != NULL) {
5540                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5541                                            pobj->oo_inode, 0);
5542                 } else {
5543                         down_write(&pobj->oo_ext_idx_sem);
5544                 }
5545
5546                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5547                                         lu_object_fid(&pobj->oo_dt.do_lu),
5548                                         fid, th);
5549         } else {
5550                 if (hlock != NULL) {
5551                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5552                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5553                 } else {
5554                         down_write(&pobj->oo_ext_idx_sem);
5555                 }
5556
5557                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5558                         struct lu_fid *tfid = &info->oti_fid;
5559
5560                         *tfid = *fid;
5561                         tfid->f_ver = ~0;
5562                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5563                                               tfid, hlock, th);
5564                 } else {
5565                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5566                                               hlock, th);
5567                 }
5568         }
5569         if (hlock != NULL)
5570                 ldiskfs_htree_unlock(hlock);
5571         else
5572                 up_write(&pobj->oo_ext_idx_sem);
5573
5574         return rc;
5575 }
5576
5577 static int
5578 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5579                       struct osd_idmap_cache *oic)
5580 {
5581         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5582         struct lu_fid *fid = &oic->oic_fid;
5583         struct osd_inode_id *id = &oic->oic_lid;
5584         struct inode *inode = NULL;
5585         int once = 0;
5586         bool insert;
5587         int rc;
5588
5589         ENTRY;
5590
5591         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5592                 RETURN(0);
5593
5594         if (thread_is_running(&scrub->os_thread) &&
5595             scrub->os_pos_current > id->oii_ino)
5596                 RETURN(0);
5597
5598         if (dev->od_auto_scrub_interval == AS_NEVER ||
5599             ktime_get_real_seconds() <
5600             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5601                 RETURN(0);
5602
5603 again:
5604         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5605         if (rc == -ENOENT) {
5606                 __u32 gen = id->oii_gen;
5607
5608                 insert = true;
5609                 if (inode != NULL)
5610                         goto trigger;
5611
5612                 inode = osd_iget(oti, dev, id);
5613                 /* The inode has been removed (by race maybe). */
5614                 if (IS_ERR(inode)) {
5615                         rc = PTR_ERR(inode);
5616
5617                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5618                 }
5619
5620                 /* The OI mapping is lost. */
5621                 if (gen != OSD_OII_NOGEN)
5622                         goto trigger;
5623
5624                 /*
5625                  * The inode may has been reused by others, we do not know,
5626                  * leave it to be handled by subsequent osd_fid_lookup().
5627                  */
5628                 GOTO(out, rc = 0);
5629         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5630                 GOTO(out, rc);
5631         }
5632
5633         insert = false;
5634
5635 trigger:
5636         if (thread_is_running(&scrub->os_thread)) {
5637                 if (inode == NULL) {
5638                         inode = osd_iget(oti, dev, id);
5639                         /* The inode has been removed (by race maybe). */
5640                         if (IS_ERR(inode)) {
5641                                 rc = PTR_ERR(inode);
5642
5643                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5644                         }
5645                 }
5646
5647                 rc = osd_oii_insert(dev, oic, insert);
5648                 /*
5649                  * There is race condition between osd_oi_lookup and OI scrub.
5650                  * The OI scrub finished just after osd_oi_lookup() failure.
5651                  * Under such case, it is unnecessary to trigger OI scrub again,
5652                  * but try to call osd_oi_lookup() again.
5653                  */
5654                 if (unlikely(rc == -EAGAIN))
5655                         goto again;
5656
5657                 if (!S_ISDIR(inode->i_mode))
5658                         rc = 0;
5659                 else
5660                         rc = osd_check_lmv(oti, dev, inode, oic);
5661
5662                 GOTO(out, rc);
5663         }
5664
5665         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5666                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5667                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5668                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5669                        "%s: trigger partial OI scrub for RPC inconsistency "
5670                        "checking FID "DFID": rc = %d\n",
5671                        osd_dev2name(dev), PFID(fid), rc);
5672                 if (rc == 0 || rc == -EALREADY)
5673                         goto again;
5674         }
5675
5676         GOTO(out, rc);
5677
5678 out:
5679         if (inode)
5680                 iput(inode);
5681
5682         RETURN(rc);
5683 }
5684
5685 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5686                                struct osd_device *dev,
5687                                struct osd_idmap_cache *oic,
5688                                struct lu_fid *fid, __u32 ino)
5689 {
5690         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5691         struct inode *inode;
5692         int rc;
5693
5694         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5695         inode = osd_iget(oti, dev, &oic->oic_lid);
5696         if (IS_ERR(inode)) {
5697                 fid_zero(&oic->oic_fid);
5698                 return PTR_ERR(inode);
5699         }
5700
5701         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5702         iput(inode);
5703         if (rc != 0)
5704                 fid_zero(&oic->oic_fid);
5705         else
5706                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5707         return rc;
5708 }
5709
5710 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5711                       struct osd_inode_id *id, const struct lu_fid *fid)
5712 {
5713         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5714                id->oii_ino, id->oii_gen, info);
5715         info->oti_cache.oic_lid = *id;
5716         info->oti_cache.oic_fid = *fid;
5717         info->oti_cache.oic_dev = osd;
5718 }
5719
5720 /**
5721  * Get parent FID from the linkEA.
5722  *
5723  * For a directory which parent resides on remote MDT, to satisfy the
5724  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5725  * the other hand, to make the lookup(..) on the directory can return
5726  * the real parent FID, we append the real parent FID after its ".."
5727  * name entry in the /REMOTE_PARENT_DIR.
5728  *
5729  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5730  * backup. So after the restore, we cannot get the right parent FID from
5731  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5732  * we have stored the real parent FID in the directory object's linkEA,
5733  * we can parse the linkEA for the real parent FID.
5734  *
5735  * \param[in] env       pointer to the thread context
5736  * \param[in] obj       pointer to the object to be handled
5737  * \param[out]fid       pointer to the buffer to hold the parent FID
5738  *
5739  * \retval              0 for getting the real parent FID successfully
5740  * \retval              negative error number on failure
5741  */
5742 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5743                                     struct osd_object *obj,
5744                                     struct lu_fid *fid)
5745 {
5746         struct osd_thread_info *oti = osd_oti_get(env);
5747         struct lu_buf *buf = &oti->oti_big_buf;
5748         struct dentry *dentry = &oti->oti_obj_dentry;
5749         struct inode *inode = obj->oo_inode;
5750         struct linkea_data ldata = { NULL };
5751         int rc;
5752
5753         ENTRY;
5754
5755         fid_zero(fid);
5756         if (!S_ISDIR(inode->i_mode))
5757                 RETURN(-EIO);
5758
5759 again:
5760         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5761                              buf->lb_buf, buf->lb_len);
5762         if (rc == -ERANGE) {
5763                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5764                                      NULL, 0);
5765                 if (rc > 0) {
5766                         lu_buf_realloc(buf, rc);
5767                         if (buf->lb_buf == NULL)
5768                                 RETURN(-ENOMEM);
5769
5770                         goto again;
5771                 }
5772         }
5773
5774         if (unlikely(rc == 0))
5775                 RETURN(-ENODATA);
5776
5777         if (rc < 0)
5778                 RETURN(rc);
5779
5780         if (unlikely(buf->lb_buf == NULL)) {
5781                 lu_buf_realloc(buf, rc);
5782                 if (buf->lb_buf == NULL)
5783                         RETURN(-ENOMEM);
5784
5785                 goto again;
5786         }
5787
5788         ldata.ld_buf = buf;
5789         rc = linkea_init_with_rec(&ldata);
5790         if (!rc) {
5791                 linkea_first_entry(&ldata);
5792                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5793         }
5794
5795         RETURN(rc);
5796 }
5797
5798 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5799                                     struct inode *inode,
5800                                     const struct lu_fid *pfid,
5801                                     const char *name, const int namelen)
5802 {
5803         struct osd_thread_info *oti = osd_oti_get(env);
5804         struct lu_buf *buf = &oti->oti_big_buf;
5805         struct dentry *dentry = &oti->oti_obj_dentry;
5806         struct linkea_data ldata = { NULL };
5807         struct lu_name cname = { .ln_name = name,
5808                                  .ln_namelen = namelen };
5809         int rc;
5810
5811         ENTRY;
5812
5813 again:
5814         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5815                              buf->lb_buf, buf->lb_len);
5816         if (rc == -ERANGE)
5817                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5818
5819         if (rc < 0)
5820                 RETURN(rc);
5821
5822         if (unlikely(rc == 0))
5823                 RETURN(-ENODATA);
5824
5825         if (buf->lb_len < rc) {
5826                 lu_buf_realloc(buf, rc);
5827                 if (buf->lb_buf == NULL)
5828                         RETURN(-ENOMEM);
5829
5830                 goto again;
5831         }
5832
5833         ldata.ld_buf = buf;
5834         rc = linkea_init_with_rec(&ldata);
5835         if (!rc)
5836                 rc = linkea_links_find(&ldata, &cname, pfid);
5837
5838         RETURN(rc);
5839 }
5840
5841 /**
5842  * Calls ->lookup() to find dentry. From dentry get inode and
5843  * read inode's ea to get fid. This is required for  interoperability
5844  * mode (b11826)
5845  *
5846  * \retval   0, on success
5847  * \retval -ve, on error
5848  */
5849 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5850                              struct dt_rec *rec, const struct dt_key *key)
5851 {
5852         struct inode *dir = obj->oo_inode;
5853         struct dentry *dentry;
5854         struct ldiskfs_dir_entry_2 *de;
5855         struct buffer_head *bh;
5856         struct lu_fid *fid = (struct lu_fid *)rec;
5857         struct htree_lock *hlock = NULL;
5858         int ino;
5859         int rc;
5860
5861         ENTRY;
5862
5863         LASSERT(dir->i_op != NULL);
5864         LASSERT(dir->i_op->lookup != NULL);
5865
5866         dentry = osd_child_dentry_get(env, obj,
5867                                       (char *)key, strlen((char *)key));
5868
5869         if (obj->oo_hl_head != NULL) {
5870                 hlock = osd_oti_get(env)->oti_hlock;
5871                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5872                                    dir, LDISKFS_HLOCK_LOOKUP);
5873         } else {
5874                 down_read(&obj->oo_ext_idx_sem);
5875         }
5876
5877         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5878         if (!IS_ERR(bh)) {
5879                 struct osd_thread_info *oti = osd_oti_get(env);
5880                 struct osd_inode_id *id = &oti->oti_id;
5881                 struct osd_idmap_cache *oic = &oti->oti_cache;
5882                 struct osd_device *dev = osd_obj2dev(obj);
5883
5884                 ino = le32_to_cpu(de->inode);
5885                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5886                         brelse(bh);
5887                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5888                         GOTO(out, rc);
5889                 }
5890
5891                 rc = osd_get_fid_from_dentry(de, rec);
5892
5893                 /* done with de, release bh */
5894                 brelse(bh);
5895                 if (rc != 0) {
5896                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5897                                 const char *name = (const char *)key;
5898
5899                                 /*
5900                                  * If the parent is on remote MDT, and there
5901                                  * is no FID-in-dirent, then we have to get
5902                                  * the parent FID from the linkEA.
5903                                  */
5904                                 if (likely(strlen(name) == 2 &&
5905                                            name[0] == '.' && name[1] == '.'))
5906                                         rc = osd_get_pfid_from_linkea(env, obj,
5907                                                                       fid);
5908                         } else {
5909                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5910                         }
5911                 } else {
5912                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5913                 }
5914
5915                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5916                         fid_zero(&oic->oic_fid);
5917
5918                         GOTO(out, rc);
5919                 }
5920
5921                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5922                 rc = osd_consistency_check(oti, dev, oic);
5923                 if (rc == -ENOENT)
5924                         fid_zero(&oic->oic_fid);
5925                 else
5926                         /* Other error should not affect lookup result. */
5927                         rc = 0;
5928         } else {
5929                 rc = PTR_ERR(bh);
5930         }
5931
5932         GOTO(out, rc);
5933
5934 out:
5935         if (hlock != NULL)
5936                 ldiskfs_htree_unlock(hlock);
5937         else
5938                 up_read(&obj->oo_ext_idx_sem);
5939         return rc;
5940 }
5941
5942 static int osd_index_declare_ea_insert(const struct lu_env *env,
5943                                        struct dt_object *dt,
5944                                        const struct dt_rec *rec,
5945                                        const struct dt_key *key,
5946                                        struct thandle *handle)
5947 {
5948         struct osd_thandle *oh;
5949         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5950         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5951         const struct lu_fid *fid = rec1->rec_fid;
5952         int credits, rc = 0;
5953         struct osd_idmap_cache *idc;
5954
5955         ENTRY;
5956
5957         LASSERT(!dt_object_remote(dt));
5958         LASSERT(handle != NULL);
5959         LASSERT(fid != NULL);
5960         LASSERT(rec1->rec_type != 0);
5961
5962         oh = container_of0(handle, struct osd_thandle, ot_super);
5963         LASSERT(oh->ot_handle == NULL);
5964
5965         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5966
5967         /*
5968          * we can't call iget() while a transactions is running
5969          * (this can lead to a deadlock), but we need to know
5970          * inum and object type. so we find this information at
5971          * declaration and cache in per-thread info
5972          */
5973         idc = osd_idc_find_or_init(env, osd, fid);
5974         if (IS_ERR(idc))
5975                 RETURN(PTR_ERR(idc));
5976         if (idc->oic_remote) {
5977                 /*
5978                  * a reference to remote inode is represented by an
5979                  * agent inode which we have to create
5980                  */
5981                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5982                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5983         }
5984
5985         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5986
5987         if (osd_dt_obj(dt)->oo_inode != NULL) {
5988                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5989
5990                 /*
5991                  * We ignore block quota on meta pool (MDTs), so needn't
5992                  * calculate how many blocks will be consumed by this index
5993                  * insert
5994                  */
5995                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5996                                            i_gid_read(inode),
5997                                            i_projid_read(inode), 0,
5998                                            oh, osd_dt_obj(dt), NULL,
5999                                            OSD_QID_BLK);
6000                 if (rc)
6001                         RETURN(rc);
6002
6003 #ifdef HAVE_PROJECT_QUOTA
6004                 /*
6005                  * Reserve credits for local agent inode to transfer
6006                  * to 0, quota enforcement is ignored in this case.
6007                  */
6008                 if (idc->oic_remote &&
6009                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6010                     i_projid_read(inode) != 0)
6011                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6012                                                   0, i_projid_read(inode),
6013                                                   0, false, PRJQUOTA, true);
6014 #endif
6015         }
6016
6017         RETURN(rc);
6018 }
6019
6020 /**
6021  * Index add function for interoperability mode (b11826).
6022  * It will add the directory entry.This entry is needed to
6023  * maintain name->fid mapping.
6024  *
6025  * \param key it is key i.e. file entry to be inserted
6026  * \param rec it is value of given key i.e. fid
6027  *
6028  * \retval   0, on success
6029  * \retval -ve, on error
6030  */
6031 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6032                                const struct dt_rec *rec,
6033                                const struct dt_key *key, struct thandle *th)
6034 {
6035         struct osd_object *obj = osd_dt_obj(dt);
6036         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6037         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6038         const struct lu_fid *fid = rec1->rec_fid;
6039         const char *name = (const char *)key;
6040         struct osd_thread_info *oti = osd_oti_get(env);
6041         struct inode *child_inode = NULL;
6042         struct osd_idmap_cache *idc;
6043         int rc;
6044
6045         ENTRY;
6046
6047         if (!dt_object_exists(dt))
6048                 RETURN(-ENOENT);
6049
6050         LASSERT(osd_invariant(obj));
6051         LASSERT(!dt_object_remote(dt));
6052         LASSERT(th != NULL);
6053
6054         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6055
6056         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6057
6058         idc = osd_idc_find(env, osd, fid);
6059         if (unlikely(idc == NULL)) {
6060                 idc = osd_idc_find_or_init(env, osd, fid);
6061                 if (IS_ERR(idc)) {
6062                         /*
6063                          * this dt_insert() wasn't declared properly, so
6064                          * FID is missing in OI cache. we better do not
6065                          * lookup FID in FLDB/OI and don't risk to deadlock,
6066                          * but in some special cases (lfsck testing, etc)
6067                          * it's much simpler than fixing a caller.
6068                          *
6069                          * normally this error should be placed after the first
6070                          * find, but migrate may attach source stripes to
6071                          * target, which doesn't create stripes.
6072                          */
6073                         CERROR("%s: "DFID" wasn't declared for insert\n",
6074                                osd_name(osd), PFID(fid));
6075                         dump_stack();
6076                         RETURN(PTR_ERR(idc));
6077                 }
6078         }
6079
6080         if (idc->oic_remote) {
6081                 /* Insert remote entry */
6082                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6083                         child_inode =
6084                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6085                 } else {
6086                         child_inode = osd_create_local_agent_inode(env, osd,
6087                                         obj, fid, rec1->rec_type & S_IFMT, th);
6088                         if (IS_ERR(child_inode))
6089                                 RETURN(PTR_ERR(child_inode));
6090                 }
6091         } else {
6092                 /* Insert local entry */
6093                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6094                         /* for a reason OI cache wasn't filled properly */
6095                         CERROR("%s: OIC for "DFID" isn't filled\n",
6096                                osd_name(osd), PFID(fid));
6097                         RETURN(-EINVAL);
6098                 }
6099                 child_inode = oti->oti_inode;
6100                 if (unlikely(child_inode == NULL)) {
6101                         struct ldiskfs_inode_info *lii;
6102
6103                         OBD_ALLOC_PTR(lii);
6104                         if (lii == NULL)
6105                                 RETURN(-ENOMEM);
6106                         child_inode = oti->oti_inode = &lii->vfs_inode;
6107                 }
6108                 child_inode->i_sb = osd_sb(osd);
6109                 child_inode->i_ino = idc->oic_lid.oii_ino;
6110                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6111         }
6112
6113         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6114
6115         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6116                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6117
6118         if (child_inode && child_inode != oti->oti_inode)
6119                 iput(child_inode);
6120         LASSERT(osd_invariant(obj));
6121         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6122         RETURN(rc);
6123 }
6124
6125 /**
6126  *  Initialize osd Iterator for given osd index object.
6127  *
6128  *  \param  dt      osd index object
6129  */
6130
6131 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6132                                      struct dt_object *dt,
6133                                      __u32 unused)
6134 {
6135         struct osd_it_iam *it;
6136         struct osd_object *obj = osd_dt_obj(dt);
6137         struct lu_object *lo = &dt->do_lu;
6138         struct iam_path_descr *ipd;
6139         struct iam_container *bag = &obj->oo_dir->od_container;
6140
6141         if (!dt_object_exists(dt))
6142                 return ERR_PTR(-ENOENT);
6143
6144         OBD_ALLOC_PTR(it);
6145         if (it == NULL)
6146                 return ERR_PTR(-ENOMEM);
6147
6148         ipd = osd_it_ipd_get(env, bag);
6149         if (likely(ipd != NULL)) {
6150                 it->oi_obj = obj;
6151                 it->oi_ipd = ipd;
6152                 lu_object_get(lo);
6153                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6154                 return (struct dt_it *)it;
6155         } else {
6156                 OBD_FREE_PTR(it);
6157                 return ERR_PTR(-ENOMEM);
6158         }
6159 }
6160
6161 /**
6162  * free given Iterator.
6163  */
6164 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6165 {
6166         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6167         struct osd_object *obj = it->oi_obj;
6168
6169         iam_it_fini(&it->oi_it);
6170         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6171         osd_object_put(env, obj);
6172         OBD_FREE_PTR(it);
6173 }
6174
6175 /**
6176  *  Move Iterator to record specified by \a key
6177  *
6178  *  \param  di      osd iterator
6179  *  \param  key     key for index
6180  *
6181  *  \retval +ve  di points to record with least key not larger than key
6182  *  \retval  0   di points to exact matched key
6183  *  \retval -ve  failure
6184  */
6185
6186 static int osd_it_iam_get(const struct lu_env *env,
6187                           struct dt_it *di, const struct dt_key *key)
6188 {
6189         struct osd_thread_info *oti = osd_oti_get(env);
6190         struct osd_it_iam *it = (struct osd_it_iam *)di;
6191
6192         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6193                 /* swab quota uid/gid */
6194                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6195                 key = (struct dt_key *)&oti->oti_quota_id;
6196         }
6197
6198         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6199 }
6200
6201 /**
6202  *  Release Iterator
6203  *
6204  *  \param  di      osd iterator
6205  */
6206 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6207 {
6208         struct osd_it_iam *it = (struct osd_it_iam *)di;
6209
6210         iam_it_put(&it->oi_it);
6211 }
6212
6213 /**
6214  *  Move iterator by one record
6215  *
6216  *  \param  di      osd iterator
6217  *
6218  *  \retval +1   end of container reached
6219  *  \retval  0   success
6220  *  \retval -ve  failure
6221  */
6222
6223 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6224 {
6225         struct osd_it_iam *it = (struct osd_it_iam *)di;
6226
6227         return iam_it_next(&it->oi_it);
6228 }
6229
6230 /**
6231  * Return pointer to the key under iterator.
6232  */
6233
6234 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6235                                      const struct dt_it *di)
6236 {
6237         struct osd_thread_info *oti = osd_oti_get(env);
6238         struct osd_it_iam *it = (struct osd_it_iam *)di;
6239         struct osd_object *obj = it->oi_obj;
6240         struct dt_key *key;
6241
6242         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6243
6244         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6245                 /* swab quota uid/gid */
6246                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6247                 key = (struct dt_key *)&oti->oti_quota_id;
6248         }
6249
6250         return key;
6251 }
6252
6253 /**
6254  * Return size of key under iterator (in bytes)
6255  */
6256
6257 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6258 {
6259         struct osd_it_iam *it = (struct osd_it_iam *)di;
6260
6261         return iam_it_key_size(&it->oi_it);
6262 }
6263
6264 static inline void
6265 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6266 {
6267         /* check if file type is required */
6268         if (ent->lde_attrs & LUDA_TYPE) {
6269                 struct luda_type *lt;
6270                 int align = sizeof(*lt) - 1;
6271
6272                 len = (len + align) & ~align;
6273                 lt = (struct luda_type *)(ent->lde_name + len);
6274                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6275         }
6276
6277         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6278 }
6279
6280 /**
6281  * build lu direct from backend fs dirent.
6282  */
6283
6284 static inline void
6285 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6286                    char *name, __u16 namelen, __u16 type, __u32 attr)
6287 {
6288         ent->lde_attrs = attr | LUDA_FID;
6289         fid_cpu_to_le(&ent->lde_fid, fid);
6290
6291         ent->lde_hash = cpu_to_le64(offset);
6292         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6293
6294         strncpy(ent->lde_name, name, namelen);
6295         ent->lde_name[namelen] = '\0';
6296         ent->lde_namelen = cpu_to_le16(namelen);
6297
6298         /* append lustre attributes */
6299         osd_it_append_attrs(ent, namelen, type);
6300 }
6301
6302 /**
6303  * Return pointer to the record under iterator.
6304  */
6305 static int osd_it_iam_rec(const struct lu_env *env,
6306                           const struct dt_it *di,
6307                           struct dt_rec *dtrec, __u32 attr)
6308 {
6309         struct osd_it_iam *it = (struct osd_it_iam *)di;
6310         struct osd_thread_info *info = osd_oti_get(env);
6311
6312         ENTRY;
6313
6314         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6315                 const struct osd_fid_pack *rec;
6316                 struct lu_fid *fid = &info->oti_fid;
6317                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6318                 char *name;
6319                 int namelen;
6320                 __u64 hash;
6321                 int rc;
6322
6323                 name = (char *)iam_it_key_get(&it->oi_it);
6324                 if (IS_ERR(name))
6325                         RETURN(PTR_ERR(name));
6326
6327                 namelen = iam_it_key_size(&it->oi_it);
6328
6329                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6330                 if (IS_ERR(rec))
6331                         RETURN(PTR_ERR(rec));
6332
6333                 rc = osd_fid_unpack(fid, rec);
6334                 if (rc)
6335                         RETURN(rc);
6336
6337                 hash = iam_it_store(&it->oi_it);
6338
6339                 /* IAM does not store object type in IAM index (dir) */
6340                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6341                                    0, LUDA_FID);
6342         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6343                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6344                            (struct iam_rec *)dtrec);
6345                 osd_quota_unpack(it->oi_obj, dtrec);
6346         } else {
6347                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6348                            (struct iam_rec *)dtrec);
6349         }
6350
6351         RETURN(0);
6352 }
6353
6354 /**
6355  * Returns cookie for current Iterator position.
6356  */
6357 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6358 {
6359         struct osd_it_iam *it = (struct osd_it_iam *)di;
6360
6361         return iam_it_store(&it->oi_it);
6362 }
6363
6364 /**
6365  * Restore iterator from cookie.
6366  *
6367  * \param  di      osd iterator
6368  * \param  hash    Iterator location cookie
6369  *
6370  * \retval +ve  di points to record with least key not larger than key.
6371  * \retval  0   di points to exact matched key
6372  * \retval -ve  failure
6373  */
6374
6375 static int osd_it_iam_load(const struct lu_env *env,
6376                            const struct dt_it *di, __u64 hash)
6377 {
6378         struct osd_it_iam *it = (struct osd_it_iam *)di;
6379
6380         return iam_it_load(&it->oi_it, hash);
6381 }
6382
6383 static const struct dt_index_operations osd_index_iam_ops = {
6384         .dio_lookup         = osd_index_iam_lookup,
6385         .dio_declare_insert = osd_index_declare_iam_insert,
6386         .dio_insert         = osd_index_iam_insert,
6387         .dio_declare_delete = osd_index_declare_iam_delete,
6388         .dio_delete         = osd_index_iam_delete,
6389         .dio_it     = {
6390                 .init     = osd_it_iam_init,
6391                 .fini     = osd_it_iam_fini,
6392                 .get      = osd_it_iam_get,
6393                 .put      = osd_it_iam_put,
6394                 .next     = osd_it_iam_next,
6395                 .key      = osd_it_iam_key,
6396                 .key_size = osd_it_iam_key_size,
6397                 .rec      = osd_it_iam_rec,
6398                 .store    = osd_it_iam_store,
6399                 .load     = osd_it_iam_load
6400         }
6401 };
6402
6403
6404 /**
6405  * Creates or initializes iterator context.
6406  *
6407  * \retval struct osd_it_ea, iterator structure on success
6408  *
6409  */
6410 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6411                                     struct dt_object *dt,
6412                                     __u32 attr)
6413 {
6414         struct osd_object *obj = osd_dt_obj(dt);
6415         struct osd_thread_info *info = osd_oti_get(env);
6416         struct osd_it_ea *oie;
6417         struct file *file;
6418         struct lu_object *lo = &dt->do_lu;
6419         struct dentry *obj_dentry;
6420
6421         ENTRY;
6422
6423         if (!dt_object_exists(dt) || obj->oo_destroyed)
6424                 RETURN(ERR_PTR(-ENOENT));
6425
6426         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6427         if (oie == NULL)
6428                 RETURN(ERR_PTR(-ENOMEM));
6429         obj_dentry = &oie->oie_dentry;
6430
6431         obj_dentry->d_inode = obj->oo_inode;
6432         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6433         obj_dentry->d_name.hash = 0;
6434
6435         oie->oie_rd_dirent       = 0;
6436         oie->oie_it_dirent       = 0;
6437         oie->oie_dirent          = NULL;
6438         if (unlikely(!info->oti_it_ea_buf_used)) {
6439                 oie->oie_buf = info->oti_it_ea_buf;
6440                 info->oti_it_ea_buf_used = 1;
6441         } else {
6442                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6443                 if (oie->oie_buf == NULL)
6444                         RETURN(ERR_PTR(-ENOMEM));
6445         }
6446         oie->oie_obj = obj;
6447
6448         file = &oie->oie_file;
6449
6450         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6451         if (attr & LUDA_64BITHASH)
6452                 file->f_mode    = FMODE_64BITHASH;
6453         else
6454                 file->f_mode    = FMODE_32BITHASH;
6455         file->f_path.dentry     = obj_dentry;
6456         file->f_mapping         = obj->oo_inode->i_mapping;
6457         file->f_op              = obj->oo_inode->i_fop;
6458         set_file_inode(file, obj->oo_inode);
6459
6460         lu_object_get(lo);
6461         RETURN((struct dt_it *)oie);
6462 }
6463
6464 /**
6465  * Destroy or finishes iterator context.
6466  *
6467  * \param di iterator structure to be destroyed
6468  */
6469 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6470 {
6471         struct osd_thread_info *info = osd_oti_get(env);
6472         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6473         struct osd_object *obj = oie->oie_obj;
6474         struct inode *inode = obj->oo_inode;
6475
6476         ENTRY;
6477         oie->oie_file.f_op->release(inode, &oie->oie_file);
6478         osd_object_put(env, obj);
6479         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6480                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6481         else
6482                 info->oti_it_ea_buf_used = 0;
6483         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6484         EXIT;
6485 }
6486
6487 /**
6488  * It position the iterator at given key, so that next lookup continues from
6489  * that key Or it is similar to dio_it->load() but based on a key,
6490  * rather than file position.
6491  *
6492  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6493  * to the beginning.
6494  *
6495  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6496  */
6497 static int osd_it_ea_get(const struct lu_env *env,
6498                          struct dt_it *di, const struct dt_key *key)
6499 {
6500         struct osd_it_ea *it = (struct osd_it_ea *)di;
6501
6502         ENTRY;
6503         LASSERT(((const char *)key)[0] == '\0');
6504         it->oie_file.f_pos = 0;
6505         it->oie_rd_dirent = 0;
6506         it->oie_it_dirent = 0;
6507         it->oie_dirent = NULL;
6508
6509         RETURN(+1);
6510 }
6511
6512 /**
6513  * Does nothing
6514  */
6515 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6516 {
6517 }
6518
6519 struct osd_filldir_cbs {
6520         struct dir_context ctx;
6521         struct osd_it_ea  *it;
6522 };
6523 /**
6524  * It is called internally by ->iterate*(). It fills the
6525  * iterator's in-memory data structure with required
6526  * information i.e. name, namelen, rec_size etc.
6527  *
6528  * \param buf in which information to be filled in.
6529  * \param name name of the file in given dir
6530  *
6531  * \retval 0 on success
6532  * \retval 1 on buffer full
6533  */
6534 #ifdef HAVE_FILLDIR_USE_CTX
6535 static int osd_ldiskfs_filldir(struct dir_context *buf,
6536 #else
6537 static int osd_ldiskfs_filldir(void *buf,
6538 #endif
6539                                const char *name, int namelen,
6540                                loff_t offset, __u64 ino, unsigned int d_type)
6541 {
6542         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6543         struct osd_object *obj = it->oie_obj;
6544         struct osd_it_ea_dirent *ent = it->oie_dirent;
6545         struct lu_fid *fid = &ent->oied_fid;
6546         struct osd_fid_pack *rec;
6547
6548         ENTRY;
6549
6550 /* this should never happen */
6551         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6552                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6553                 RETURN(-EIO);
6554         }
6555
6556         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6557             OSD_IT_EA_BUFSIZE)
6558                 RETURN(1);
6559
6560         /* "." is just the object itself. */
6561         if (namelen == 1 && name[0] == '.') {
6562                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6563         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6564                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6565                 if (osd_fid_unpack(fid, rec) != 0)
6566                         fid_zero(fid);
6567         } else {
6568                 fid_zero(fid);
6569         }
6570         d_type &= ~LDISKFS_DIRENT_LUFID;
6571
6572         /* NOT export local root. */
6573         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6574                 ino = obj->oo_inode->i_ino;
6575                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6576         }
6577
6578         ent->oied_ino     = ino;
6579         ent->oied_off     = offset;
6580         ent->oied_namelen = namelen;
6581         ent->oied_type    = d_type;
6582
6583         memcpy(ent->oied_name, name, namelen);
6584
6585         it->oie_rd_dirent++;
6586         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6587         RETURN(0);
6588 }
6589
6590 /**
6591  * Calls ->iterate*() to load a directory entry at a time
6592  * and stored it in iterator's in-memory data structure.
6593  *
6594  * \param di iterator's in memory structure
6595  *
6596  * \retval   0 on success
6597  * \retval -ve on error
6598  * \retval +1 reach the end of entry
6599  */
6600 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6601                                const struct dt_it *di)
6602 {
6603         struct osd_it_ea *it = (struct osd_it_ea *)di;
6604         struct osd_object *obj = it->oie_obj;
6605         struct inode *inode = obj->oo_inode;
6606         struct htree_lock *hlock = NULL;
6607         struct file *filp = &it->oie_file;
6608         int rc = 0;
6609         struct osd_filldir_cbs buf = {
6610                 .ctx.actor = osd_ldiskfs_filldir,
6611                 .it = it
6612         };
6613
6614         ENTRY;
6615         it->oie_dirent = it->oie_buf;
6616         it->oie_rd_dirent = 0;
6617
6618         if (obj->oo_hl_head != NULL) {
6619                 hlock = osd_oti_get(env)->oti_hlock;
6620                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6621                                    inode, LDISKFS_HLOCK_READDIR);
6622         } else {
6623                 down_read(&obj->oo_ext_idx_sem);
6624         }
6625
6626         rc = osd_security_file_alloc(filp);
6627         if (rc)
6628                 RETURN(rc);
6629
6630         filp->f_flags |= O_NOATIME;
6631         filp->f_mode |= FMODE_NONOTIFY;
6632         rc = iterate_dir(filp, &buf.ctx);
6633         if (rc)
6634                 RETURN(rc);
6635
6636         if (hlock != NULL)
6637                 ldiskfs_htree_unlock(hlock);
6638         else
6639                 up_read(&obj->oo_ext_idx_sem);
6640
6641         if (it->oie_rd_dirent == 0) {
6642                 /*
6643                  * If it does not get any dirent, it means it has been reached
6644                  * to the end of the dir
6645                  */
6646                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6647                 if (rc == 0)
6648                         rc = 1;
6649         } else {
6650                 it->oie_dirent = it->oie_buf;
6651                 it->oie_it_dirent = 1;
6652         }
6653
6654         RETURN(rc);
6655 }
6656
6657 /**
6658  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6659  * to load a directory entry at a time and stored it in
6660  * iterator's in-memory data structure.
6661  *
6662  * \param di iterator's in memory structure
6663  *
6664  * \retval +ve iterator reached to end
6665  * \retval   0 iterator not reached to end
6666  * \retval -ve on error
6667  */
6668 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6669 {
6670         struct osd_it_ea *it = (struct osd_it_ea *)di;
6671         int rc;
6672
6673         ENTRY;
6674
6675         if (it->oie_it_dirent < it->oie_rd_dirent) {
6676                 it->oie_dirent =
6677                         (void *)it->oie_dirent +
6678                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6679                                        it->oie_dirent->oied_namelen);
6680                 it->oie_it_dirent++;
6681                 rc = 0;
6682         } else {
6683                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6684                         rc = 1;
6685                 else
6686                         rc = osd_ldiskfs_it_fill(env, di);
6687         }
6688
6689         RETURN(rc);
6690 }
6691
6692 /**
6693  * Returns the key at current position from iterator's in memory structure.
6694  *
6695  * \param di iterator's in memory structure
6696  *
6697  * \retval key i.e. struct dt_key on success
6698  */
6699 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6700                                     const struct dt_it *di)
6701 {
6702         struct osd_it_ea *it = (struct osd_it_ea *)di;
6703
6704         return (struct dt_key *)it->oie_dirent->oied_name;
6705 }
6706
6707 /**
6708  * Returns key's size at current position from iterator's in memory structure.
6709  *
6710  * \param di iterator's in memory structure
6711  *
6712  * \retval key_size i.e. struct dt_key on success
6713  */
6714 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6715 {
6716         struct osd_it_ea *it = (struct osd_it_ea *)di;
6717
6718         return it->oie_dirent->oied_namelen;
6719 }
6720
6721 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6722 {
6723         if (LDISKFS_DIR_REC_LEN(de) >=
6724             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6725                 return true;
6726
6727         return false;
6728 }
6729
6730 static inline bool
6731 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6732                      unsigned int blocksize, bool dotdot)
6733 {
6734         if (dotdot)
6735                 return osd_dotdot_has_space(de);
6736
6737         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6738             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6739                 return true;
6740
6741         return false;
6742 }
6743
6744 static int
6745 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6746                     handle_t *jh, struct dentry *dentry,
6747                     const struct lu_fid *fid, struct buffer_head *bh,
6748                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6749                     bool dotdot)
6750 {
6751         struct inode *dir = dentry->d_parent->d_inode;
6752         struct inode *inode = dentry->d_inode;
6753         struct osd_fid_pack *rec;
6754         struct ldiskfs_dentry_param *ldp;
6755         int namelen = dentry->d_name.len;
6756         int rc;
6757         struct osd_thread_info *info = osd_oti_get(env);
6758
6759         ENTRY;
6760
6761         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6762                 RETURN(0);
6763
6764         /* There is enough space to hold the FID-in-dirent. */
6765         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6766                 rc = ldiskfs_journal_get_write_access(jh, bh);
6767                 if (rc != 0)
6768                         RETURN(rc);
6769
6770                 de->name[namelen] = 0;
6771                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6772                 rec->fp_len = sizeof(struct lu_fid) + 1;
6773                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6774                 de->file_type |= LDISKFS_DIRENT_LUFID;
6775                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6776
6777                 RETURN(rc);
6778         }
6779
6780         LASSERT(!dotdot);
6781
6782         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6783         if (rc != 0)
6784                 RETURN(rc);
6785
6786         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6787         osd_get_ldiskfs_dirent_param(ldp, fid);
6788         dentry->d_fsdata = (void *)ldp;
6789         dquot_initialize(dir);
6790         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6791         /*
6792          * It is too bad, we cannot reinsert the name entry back.
6793          * That means we lose it!
6794          */
6795         if (rc != 0)
6796                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6797                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6798                        osd_ino2name(inode),
6799                        dir->i_ino, dir->i_generation, namelen,
6800                        dentry->d_name.name, PFID(fid), rc);
6801
6802         RETURN(rc);
6803 }
6804
6805 static int
6806 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6807                         struct osd_it_ea *it, struct lu_fid *fid,
6808                         struct osd_inode_id *id, __u32 *attr)
6809 {
6810         struct osd_thread_info *info = osd_oti_get(env);
6811         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6812         struct osd_device *dev = osd_obj2dev(obj);
6813         struct super_block *sb = osd_sb(dev);
6814         const char *devname = osd_name(dev);
6815         struct osd_it_ea_dirent *ent = it->oie_dirent;
6816         struct inode *dir = obj->oo_inode;
6817         struct htree_lock *hlock = NULL;
6818         struct buffer_head *bh = NULL;
6819         handle_t *jh = NULL;
6820         struct ldiskfs_dir_entry_2 *de;
6821         struct dentry *dentry;
6822         struct inode *inode;
6823         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6824         int credits;
6825         int rc;
6826         bool dotdot = false;
6827         bool dirty = false;
6828
6829         ENTRY;
6830
6831         if (ent->oied_name[0] == '.') {
6832                 if (ent->oied_namelen == 1)
6833                         RETURN(0);
6834
6835                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6836                         dotdot = true;
6837         }
6838
6839         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6840         inode = osd_iget(info, dev, id);
6841         if (IS_ERR(inode)) {
6842                 rc = PTR_ERR(inode);
6843                 if (rc == -ENOENT || rc == -ESTALE) {
6844                         /*
6845                          * Maybe dangling name entry, or
6846                          * corrupted directory entry.
6847                          */
6848                         *attr |= LUDA_UNKNOWN;
6849                         rc = 0;
6850                 } else {
6851                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6852                                "check_repair, dir = %lu/%u, name = %.*s, "
6853                                "ino = %llu, rc = %d\n",
6854                                devname, dir->i_ino, dir->i_generation,
6855                                ent->oied_namelen, ent->oied_name,
6856                                ent->oied_ino, rc);
6857                 }
6858
6859                 RETURN(rc);
6860         }
6861
6862         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6863                                            ent->oied_namelen);
6864         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6865         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6866                 lma = NULL;
6867         else if (rc != 0)
6868                 GOTO(out, rc);
6869
6870         /*
6871          * We need to ensure that the name entry is still valid.
6872          * Because it may be removed or renamed by other already.
6873          *
6874          * The unlink or rename operation will start journal before PDO lock,
6875          * so to avoid deadlock, here we need to start journal handle before
6876          * related PDO lock also. But because we do not know whether there
6877          * will be something to be repaired before PDO lock, we just start
6878          * journal without conditions.
6879          *
6880          * We may need to remove the name entry firstly, then insert back.
6881          * One credit is for user quota file update.
6882          * One credit is for group quota file update.
6883          * Two credits are for dirty inode.
6884          */
6885         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6886                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6887
6888         if (dev->od_dirent_journal != 0) {
6889
6890 again:
6891                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6892                 if (IS_ERR(jh)) {
6893                         rc = PTR_ERR(jh);
6894                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6895                                "check_repair, dir = %lu/%u, credits = %d, "
6896                                "name = %.*s, ino = %llu: rc = %d\n",
6897                                devname, dir->i_ino, dir->i_generation, credits,
6898                                ent->oied_namelen, ent->oied_name,
6899                                ent->oied_ino, rc);
6900
6901                         GOTO(out_inode, rc);
6902                 }
6903
6904                 if (obj->oo_hl_head != NULL) {
6905                         hlock = osd_oti_get(env)->oti_hlock;
6906                         /*
6907                          * "0" means exclusive lock for the whole directory.
6908                          * We need to prevent others access such name entry
6909                          * during the delete + insert. Neither HLOCK_ADD nor
6910                          * HLOCK_DEL cannot guarantee the atomicity.
6911                          */
6912                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6913                 } else {
6914                         down_write(&obj->oo_ext_idx_sem);
6915                 }
6916         } else {
6917                 if (obj->oo_hl_head != NULL) {
6918                         hlock = osd_oti_get(env)->oti_hlock;
6919                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6920                                            LDISKFS_HLOCK_LOOKUP);
6921                 } else {
6922                         down_read(&obj->oo_ext_idx_sem);
6923                 }
6924         }
6925
6926         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6927         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6928                 *attr |= LUDA_IGNORE;
6929
6930                 GOTO(out, rc = 0);
6931         }
6932
6933         /*
6934          * For dotdot entry, if there is not enough space to hold the
6935          * FID-in-dirent, just keep them there. It only happens when the
6936          * device upgraded from 1.8 or restored from MDT file-level backup.
6937          * For the whole directory, only dotdot entry have no FID-in-dirent
6938          * and needs to get FID from LMA when readdir, it will not affect the
6939          * performance much.
6940          */
6941         if (dotdot && !osd_dotdot_has_space(de)) {
6942                 *attr |= LUDA_UNKNOWN;
6943
6944                 GOTO(out, rc = 0);
6945         }
6946
6947         if (lma != NULL) {
6948                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6949                         GOTO(out, rc = 0);
6950
6951                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6952                         struct lu_fid *tfid = &lma->lma_self_fid;
6953
6954                         if (likely(dotdot &&
6955                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6956                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6957                                 /*
6958                                  * It must be REMOTE_PARENT_DIR and as the
6959                                  * 'dotdot' entry of remote directory
6960                                  */
6961                                 *attr |= LUDA_IGNORE;
6962                         } else {
6963                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6964                                        "parent directory, but got %.*s under "
6965                                        "dir = %lu/%u with the FID "DFID"\n",
6966                                        devname, ent->oied_namelen,
6967                                        ent->oied_name, dir->i_ino,
6968                                        dir->i_generation, PFID(tfid));
6969
6970                                 *attr |= LUDA_UNKNOWN;
6971                         }
6972
6973                         GOTO(out, rc = 0);
6974                 }
6975         }
6976
6977         if (!fid_is_zero(fid)) {
6978                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6979                                               ent->oied_namelen);
6980                 if (rc == -ENOENT ||
6981                     (rc == -ENODATA &&
6982                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6983                         /*
6984                          * linkEA does not recognize the dirent entry,
6985                          * it may because the dirent entry corruption
6986                          * and points to other's inode.
6987                          */
6988                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6989                                "recognize the dirent, dir = %lu/%u, "
6990                                " name = %.*s, ino = %llu, "
6991                                DFID": rc = %d\n", devname, dir->i_ino,
6992                                dir->i_generation, ent->oied_namelen,
6993                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6994                         *attr |= LUDA_UNKNOWN;
6995
6996                         GOTO(out, rc = 0);
6997                 }
6998
6999                 if (rc && rc != -ENODATA) {
7000                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7001                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7002                                DFID": rc = %d\n", devname, dir->i_ino,
7003                                dir->i_generation, ent->oied_namelen,
7004                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7005                         *attr |= LUDA_UNKNOWN;
7006
7007                         GOTO(out, rc = 0);
7008                 }
7009         }
7010
7011         if (lma != NULL) {
7012                 /*
7013                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7014                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7015                  */
7016                 if (*attr & LUDA_VERIFY_DRYRUN) {
7017                         *fid = lma->lma_self_fid;
7018                         *attr |= LUDA_REPAIR;
7019
7020                         GOTO(out, rc = 0);
7021                 }
7022
7023                 if (jh == NULL) {
7024                         brelse(bh);
7025                         dev->od_dirent_journal = 1;
7026                         if (hlock != NULL) {
7027                                 ldiskfs_htree_unlock(hlock);
7028                                 hlock = NULL;
7029                         } else {
7030                                 up_read(&obj->oo_ext_idx_sem);
7031                         }
7032
7033                         goto again;
7034                 }
7035
7036                 *fid = lma->lma_self_fid;
7037                 dirty = true;
7038                 /* Update or append the FID-in-dirent. */
7039                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7040                                          bh, de, hlock, dotdot);
7041                 if (rc == 0)
7042                         *attr |= LUDA_REPAIR;
7043                 else
7044                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7045                                "the dirent, dir = %lu/%u, name = %.*s, "
7046                                "ino = %llu, "DFID": rc = %d\n",
7047                                devname, dir->i_ino, dir->i_generation,
7048                                ent->oied_namelen, ent->oied_name,
7049                                ent->oied_ino, PFID(fid), rc);
7050         } else {
7051                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7052                 if (*attr & LUDA_VERIFY_DRYRUN) {
7053                         if (fid_is_sane(fid)) {
7054                                 *attr |= LUDA_REPAIR;
7055                         } else if (dev->od_index == 0) {
7056                                 lu_igif_build(fid, inode->i_ino,
7057                                               inode->i_generation);
7058                                 *attr |= LUDA_UPGRADE;
7059                         }
7060
7061                         GOTO(out, rc = 0);
7062                 }
7063
7064                 if (jh == NULL) {
7065                         brelse(bh);
7066                         dev->od_dirent_journal = 1;
7067                         if (hlock != NULL) {
7068                                 ldiskfs_htree_unlock(hlock);
7069                                 hlock = NULL;
7070                         } else {
7071                                 up_read(&obj->oo_ext_idx_sem);
7072                         }
7073
7074                         goto again;
7075                 }
7076
7077                 dirty = true;
7078                 if (unlikely(fid_is_sane(fid))) {
7079                         /*
7080                          * FID-in-dirent exists, but FID-in-LMA is lost.
7081                          * Trust the FID-in-dirent, and add FID-in-LMA.
7082                          */
7083                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7084                         if (rc == 0)
7085                                 *attr |= LUDA_REPAIR;
7086                         else
7087                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7088                                        "update dirent, dir = %lu/%u, "
7089                                        "name = %.*s, ino = %llu, "
7090                                        DFID": rc = %d\n",
7091                                        devname, dir->i_ino, dir->i_generation,
7092                                        ent->oied_namelen, ent->oied_name,
7093                                        ent->oied_ino, PFID(fid), rc);
7094                 } else if (dev->od_index == 0) {
7095                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7096                         /*
7097                          * It is probably IGIF object. Only aappend the
7098                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7099                          */
7100                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7101                                                  bh, de, hlock, dotdot);
7102                         if (rc == 0)
7103                                 *attr |= LUDA_UPGRADE;
7104                         else
7105                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7106                                        "after the dirent, dir = %lu/%u, "
7107                                        "name = %.*s, ino = %llu, "
7108                                        DFID": rc = %d\n",
7109                                        devname, dir->i_ino, dir->i_generation,
7110                                        ent->oied_namelen, ent->oied_name,
7111                                        ent->oied_ino, PFID(fid), rc);
7112                 }
7113         }
7114
7115         GOTO(out, rc);
7116
7117 out:
7118         if (!IS_ERR(bh))
7119                 brelse(bh);
7120         if (hlock != NULL) {
7121                 ldiskfs_htree_unlock(hlock);
7122         } else {
7123                 if (dev->od_dirent_journal != 0)
7124                         up_write(&obj->oo_ext_idx_sem);
7125                 else
7126                         up_read(&obj->oo_ext_idx_sem);
7127         }
7128
7129         if (jh != NULL)
7130                 ldiskfs_journal_stop(jh);
7131
7132 out_inode:
7133         iput(inode);
7134         if (rc >= 0 && !dirty)
7135                 dev->od_dirent_journal = 0;
7136
7137         return rc;
7138 }
7139
7140 /**
7141  * Returns the value at current position from iterator's in memory structure.
7142  *
7143  * \param di struct osd_it_ea, iterator's in memory structure
7144  * \param attr attr requested for dirent.
7145  * \param lde lustre dirent
7146  *
7147  * \retval   0 no error and \param lde has correct lustre dirent.
7148  * \retval -ve on error
7149  */
7150 static inline int osd_it_ea_rec(const struct lu_env *env,
7151                                 const struct dt_it *di,
7152                                 struct dt_rec *dtrec, __u32 attr)
7153 {
7154         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7155         struct osd_object      *obj   = it->oie_obj;
7156         struct osd_device      *dev   = osd_obj2dev(obj);
7157         struct osd_thread_info *oti   = osd_oti_get(env);
7158         struct osd_inode_id    *id    = &oti->oti_id;
7159         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7160         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7161         __u32 ino = it->oie_dirent->oied_ino;
7162         int rc = 0;
7163
7164         ENTRY;
7165
7166         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7167
7168         if (attr & LUDA_VERIFY) {
7169                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7170                         attr |= LUDA_IGNORE;
7171                         /*
7172                          * If the parent is on remote MDT, and there
7173                          * is no FID-in-dirent, then we have to get
7174                          * the parent FID from the linkEA.
7175                          */
7176                         if (!fid_is_sane(fid) &&
7177                             it->oie_dirent->oied_namelen == 2 &&
7178                             it->oie_dirent->oied_name[0] == '.' &&
7179                             it->oie_dirent->oied_name[1] == '.')
7180                                 osd_get_pfid_from_linkea(env, obj, fid);
7181                 } else {
7182                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7183                                                      &attr);
7184                 }
7185
7186                 if (!fid_is_sane(fid))
7187                         attr |= LUDA_UNKNOWN;
7188         } else {
7189                 attr &= ~LU_DIRENT_ATTRS_MASK;
7190                 if (!fid_is_sane(fid)) {
7191                         bool is_dotdot = false;
7192
7193                         if (it->oie_dirent->oied_namelen == 2 &&
7194                             it->oie_dirent->oied_name[0] == '.' &&
7195                             it->oie_dirent->oied_name[1] == '.')
7196                                 is_dotdot = true;
7197                         /*
7198                          * If the parent is on remote MDT, and there
7199                          * is no FID-in-dirent, then we have to get
7200                          * the parent FID from the linkEA.
7201                          */
7202                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7203                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7204                         } else {
7205                                 if (is_dotdot == false &&
7206                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7207                                         RETURN(-ENOENT);
7208
7209                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7210                         }
7211                 } else {
7212                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7213                 }
7214         }
7215
7216         /* Pack the entry anyway, at least the offset is right. */
7217         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7218                            it->oie_dirent->oied_name,
7219                            it->oie_dirent->oied_namelen,
7220                            it->oie_dirent->oied_type, attr);
7221
7222         if (rc < 0)
7223                 RETURN(rc);
7224
7225         if (osd_remote_fid(env, dev, fid))
7226                 RETURN(0);
7227
7228         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7229                 osd_add_oi_cache(oti, dev, id, fid);
7230
7231         RETURN(rc > 0 ? 0 : rc);
7232 }
7233
7234 /**
7235  * Returns the record size size at current position.
7236  *
7237  * This function will return record(lu_dirent) size in bytes.
7238  *
7239  * \param[in] env       execution environment
7240  * \param[in] di        iterator's in memory structure
7241  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7242  *                      calculate the lu_dirent size.
7243  *
7244  * \retval      record size(in bytes & in memory) of the current lu_dirent
7245  *              entry.
7246  */
7247 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7248                               __u32 attr)
7249 {
7250         struct osd_it_ea *it = (struct osd_it_ea *)di;
7251
7252         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7253 }
7254
7255 /**
7256  * Returns a cookie for current position of the iterator head, so that
7257  * user can use this cookie to load/start the iterator next time.
7258  *
7259  * \param di iterator's in memory structure
7260  *
7261  * \retval cookie for current position, on success
7262  */
7263 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7264 {
7265         struct osd_it_ea *it = (struct osd_it_ea *)di;
7266
7267         return it->oie_dirent->oied_off;
7268 }
7269
7270 /**
7271  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7272  * to load a directory entry at a time and stored it i inn,
7273  * in iterator's in-memory data structure.
7274  *
7275  * \param di struct osd_it_ea, iterator's in memory structure
7276  *
7277  * \retval +ve on success
7278  * \retval -ve on error
7279  */
7280 static int osd_it_ea_load(const struct lu_env *env,
7281                           const struct dt_it *di, __u64 hash)
7282 {
7283         struct osd_it_ea *it = (struct osd_it_ea *)di;
7284         int rc;
7285
7286         ENTRY;
7287         it->oie_file.f_pos = hash;
7288
7289         rc =  osd_ldiskfs_it_fill(env, di);
7290         if (rc > 0)
7291                 rc = -ENODATA;
7292
7293         if (rc == 0)
7294                 rc = 1;
7295
7296         RETURN(rc);
7297 }
7298
7299 /**
7300  * Index lookup function for interoperability mode (b11826).
7301  *
7302  * \param key,  key i.e. file name to be searched
7303  *
7304  * \retval +ve, on success
7305  * \retval -ve, on error
7306  */
7307 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7308                                struct dt_rec *rec, const struct dt_key *key)
7309 {
7310         struct osd_object *obj = osd_dt_obj(dt);
7311         int rc = 0;
7312
7313         ENTRY;
7314
7315         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7316         LINVRNT(osd_invariant(obj));
7317
7318         rc = osd_ea_lookup_rec(env, obj, rec, key);
7319         if (rc == 0)
7320                 rc = 1;
7321         RETURN(rc);
7322 }
7323
7324 /**
7325  * Index and Iterator operations for interoperability
7326  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7327  */
7328 static const struct dt_index_operations osd_index_ea_ops = {
7329         .dio_lookup         = osd_index_ea_lookup,
7330         .dio_declare_insert = osd_index_declare_ea_insert,
7331         .dio_insert         = osd_index_ea_insert,
7332         .dio_declare_delete = osd_index_declare_ea_delete,
7333         .dio_delete         = osd_index_ea_delete,
7334         .dio_it     = {
7335                 .init     = osd_it_ea_init,
7336                 .fini     = osd_it_ea_fini,
7337                 .get      = osd_it_ea_get,
7338                 .put      = osd_it_ea_put,
7339                 .next     = osd_it_ea_next,
7340                 .key      = osd_it_ea_key,
7341                 .key_size = osd_it_ea_key_size,
7342                 .rec      = osd_it_ea_rec,
7343                 .rec_size = osd_it_ea_rec_size,
7344                 .store    = osd_it_ea_store,
7345                 .load     = osd_it_ea_load
7346         }
7347 };
7348
7349 static void *osd_key_init(const struct lu_context *ctx,
7350                           struct lu_context_key *key)
7351 {
7352         struct osd_thread_info *info;
7353
7354         OBD_ALLOC_PTR(info);
7355         if (info == NULL)
7356                 return ERR_PTR(-ENOMEM);
7357
7358         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7359         if (info->oti_it_ea_buf == NULL)
7360                 goto out_free_info;
7361
7362         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7363
7364         info->oti_hlock = ldiskfs_htree_lock_alloc();
7365         if (info->oti_hlock == NULL)
7366                 goto out_free_ea;
7367
7368         return info;
7369
7370 out_free_ea:
7371         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7372 out_free_info:
7373         OBD_FREE_PTR(info);
7374         return ERR_PTR(-ENOMEM);
7375 }
7376
7377 static void osd_key_fini(const struct lu_context *ctx,
7378                          struct lu_context_key *key, void *data)
7379 {
7380         struct osd_thread_info *info = data;
7381         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7382         struct osd_idmap_cache *idc = info->oti_ins_cache;
7383
7384         if (info->oti_dio_pages) {
7385                 int i;
7386                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7387                         if (info->oti_dio_pages[i])
7388                                 __free_page(info->oti_dio_pages[i]);
7389                 }
7390                 OBD_FREE(info->oti_dio_pages,
7391                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7392         }
7393
7394         if (info->oti_inode != NULL)
7395                 OBD_FREE_PTR(lli);
7396         if (info->oti_hlock != NULL)
7397                 ldiskfs_htree_lock_free(info->oti_hlock);
7398         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7399         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7400         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7401         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7402         lu_buf_free(&info->oti_big_buf);
7403         if (idc != NULL) {
7404                 LASSERT(info->oti_ins_cache_size > 0);
7405                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7406                 info->oti_ins_cache = NULL;
7407                 info->oti_ins_cache_size = 0;
7408         }
7409         OBD_FREE_PTR(info);
7410 }
7411
7412 static void osd_key_exit(const struct lu_context *ctx,
7413                          struct lu_context_key *key, void *data)
7414 {
7415         struct osd_thread_info *info = data;
7416
7417         LASSERT(info->oti_r_locks == 0);
7418         LASSERT(info->oti_w_locks == 0);
7419         LASSERT(info->oti_txns    == 0);
7420 }
7421
7422 /* type constructor/destructor: osd_type_init, osd_type_fini */
7423 LU_TYPE_INIT_FINI(osd, &osd_key);
7424
7425 struct lu_context_key osd_key = {
7426         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7427         .lct_init = osd_key_init,
7428         .lct_fini = osd_key_fini,
7429         .lct_exit = osd_key_exit
7430 };
7431
7432
7433 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7434                            const char *name, struct lu_device *next)
7435 {
7436         struct osd_device *osd = osd_dev(d);
7437
7438         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7439             sizeof(osd->od_svname))
7440                 return -E2BIG;
7441         return osd_procfs_init(osd, name);
7442 }
7443
7444 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7445 {
7446         struct seq_server_site *ss = osd_seq_site(osd);
7447         int rc;
7448
7449         ENTRY;
7450
7451         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7452                 RETURN(0);
7453
7454         if (unlikely(ss == NULL))
7455                 RETURN(-ENODEV);
7456
7457         OBD_ALLOC_PTR(osd->od_cl_seq);
7458         if (osd->od_cl_seq == NULL)
7459                 RETURN(-ENOMEM);
7460
7461         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7462                              osd->od_svname, ss->ss_server_seq);
7463         if (rc != 0) {
7464                 OBD_FREE_PTR(osd->od_cl_seq);
7465                 osd->od_cl_seq = NULL;
7466                 RETURN(rc);
7467         }
7468
7469         if (ss->ss_node_id == 0) {
7470                 /*
7471                  * If the OSD on the sequence controller(MDT0), then allocate
7472                  * sequence here, otherwise allocate sequence after connected
7473                  * to MDT0 (see mdt_register_lwp_callback()).
7474                  */
7475                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7476                                    &osd->od_cl_seq->lcs_space, env);
7477         }
7478
7479         RETURN(rc);
7480 }
7481
7482 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7483 {
7484         if (osd->od_cl_seq == NULL)
7485                 return;
7486
7487         seq_client_fini(osd->od_cl_seq);
7488         OBD_FREE_PTR(osd->od_cl_seq);
7489         osd->od_cl_seq = NULL;
7490 }
7491
7492 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7493 {
7494         ENTRY;
7495
7496         /* shutdown quota slave instance associated with the device */
7497         if (o->od_quota_slave_md != NULL) {
7498                 struct qsd_instance *qsd = o->od_quota_slave_md;
7499
7500                 o->od_quota_slave_md = NULL;
7501                 qsd_fini(env, qsd);
7502         }
7503
7504         if (o->od_quota_slave_dt != NULL) {
7505                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7506
7507                 o->od_quota_slave_dt = NULL;
7508                 qsd_fini(env, qsd);
7509         }
7510
7511         osd_fid_fini(env, o);
7512         osd_scrub_cleanup(env, o);
7513
7514         RETURN(0);
7515 }
7516
7517 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7518 {
7519         ENTRY;
7520
7521         if (o->od_mnt != NULL) {
7522                 shrink_dcache_sb(osd_sb(o));
7523                 osd_sync(env, &o->od_dt_dev);
7524
7525                 mntput(o->od_mnt);
7526                 o->od_mnt = NULL;
7527         }
7528
7529         EXIT;
7530 }
7531
7532 static int osd_mount(const struct lu_env *env,
7533                      struct osd_device *o, struct lustre_cfg *cfg)
7534 {
7535         const char *name = lustre_cfg_string(cfg, 0);
7536         const char *dev = lustre_cfg_string(cfg, 1);
7537         const char *opts;
7538         unsigned long page, s_flags, lmd_flags = 0;
7539         struct page *__page;
7540         struct file_system_type *type;
7541         char *options = NULL;
7542         char *str;
7543         struct osd_thread_info *info = osd_oti_get(env);
7544         struct lu_fid *fid = &info->oti_fid;
7545         struct inode *inode;
7546         int rc = 0, force_over_512tb = 0;
7547
7548         ENTRY;
7549
7550         if (o->od_mnt != NULL)
7551                 RETURN(0);
7552
7553         if (strlen(dev) >= sizeof(o->od_mntdev))
7554                 RETURN(-E2BIG);
7555         strcpy(o->od_mntdev, dev);
7556
7557         str = lustre_cfg_string(cfg, 2);
7558         s_flags = simple_strtoul(str, NULL, 0);
7559         str = strstr(str, ":");
7560         if (str)
7561                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7562         opts = lustre_cfg_string(cfg, 3);
7563 #ifdef __BIG_ENDIAN
7564         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7565                 CERROR("%s: device %s extents feature is not guaranteed to "
7566                        "work on big-endian systems. Use \"bigendian_extents\" "
7567                        "mount option to override.\n", name, dev);
7568                 RETURN(-EINVAL);
7569         }
7570 #endif
7571 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7572         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7573                 CWARN("force_over_128tb option is deprecated. "
7574                       "Filesystems less than 512TB can be created without any "
7575                       "force options. Use force_over_512tb option for "
7576                       "filesystems greater than 512TB.\n");
7577         }
7578 #endif
7579 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7580         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7581                 CWARN("force_over_256tb option is deprecated. "
7582                       "Filesystems less than 512TB can be created without any "
7583                       "force options. Use force_over_512tb option for "
7584                       "filesystems greater than 512TB.\n");
7585         }
7586 #endif
7587
7588         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7589                 force_over_512tb = 1;
7590
7591         __page = alloc_page(GFP_KERNEL);
7592         if (__page == NULL)
7593                 GOTO(out, rc = -ENOMEM);
7594         page = (unsigned long)page_address(__page);
7595         options = (char *)page;
7596         *options = '\0';
7597         if (opts != NULL) {
7598                 /* strip out the options for back compatiblity */
7599                 static char *sout[] = {
7600                         "mballoc",
7601                         "iopen",
7602                         "noiopen",
7603                         "iopen_nopriv",
7604                         "extents",
7605                         "noextents",
7606                         /* strip out option we processed in osd */
7607                         "bigendian_extents",
7608                         "force_over_128tb",
7609                         "force_over_256tb",
7610                         "force_over_512tb",
7611                         NULL
7612                 };
7613                 strncat(options, opts, PAGE_SIZE);
7614                 for (rc = 0, str = options; sout[rc]; ) {
7615                         char *op = strstr(str, sout[rc]);
7616
7617                         if (op == NULL) {
7618                                 rc++;
7619                                 str = options;
7620                                 continue;
7621                         }
7622                         if (op == options || *(op - 1) == ',') {
7623                                 str = op + strlen(sout[rc]);
7624                                 if (*str == ',' || *str == '\0') {
7625                                         *str == ',' ? str++ : str;
7626                                         memmove(op, str, strlen(str) + 1);
7627                                 }
7628                         }
7629                         for (str = op; *str != ',' && *str != '\0'; str++)
7630                                 ;
7631                 }
7632         } else {
7633                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7634         }
7635
7636         /* Glom up mount options */
7637         if (*options != '\0')
7638                 strncat(options, ",", PAGE_SIZE);
7639         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7640
7641         type = get_fs_type("ldiskfs");
7642         if (!type) {
7643                 CERROR("%s: cannot find ldiskfs module\n", name);
7644                 GOTO(out, rc = -ENODEV);
7645         }
7646
7647         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7648         module_put(type->owner);
7649
7650         if (IS_ERR(o->od_mnt)) {
7651                 rc = PTR_ERR(o->od_mnt);
7652                 o->od_mnt = NULL;
7653                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7654                 GOTO(out, rc);
7655         }
7656
7657         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7658                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7659                                  force_over_512tb == 0) {
7660                 CERROR("%s: device %s LDISKFS does not support filesystems "
7661                        "greater than 512TB and can cause data corruption. "
7662                        "Use \"force_over_512tb\" mount option to override.\n",
7663                        name, dev);
7664                 GOTO(out_mnt, rc = -EINVAL);
7665         }
7666
7667         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7668                 if (priv_dev_set_rdonly) {
7669                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7670                         o->od_dt_dev.dd_rdonly = 1;
7671                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7672                                       name);
7673                 } else {
7674                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7675                                       name);
7676
7677                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7678                 }
7679         } else if (priv_dev_check_rdonly &&
7680                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7681                 CERROR("%s: underlying device %s is marked as "
7682                        "read-only. Setup failed\n", name, dev);
7683
7684                 GOTO(out_mnt, rc = -EROFS);
7685         }
7686
7687         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7688                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7689                 GOTO(out_mnt, rc = -EINVAL);
7690         }
7691
7692 #ifdef LDISKFS_MOUNT_DIRDATA
7693         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7694                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7695         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7696                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7697                       "enabling the dirdata feature. If you do not want to "
7698                       "downgrade to Lustre-1.x again, you can enable it via "
7699                       "'tune2fs -O dirdata device'\n", name, dev);
7700 #endif
7701         inode = osd_sb(o)->s_root->d_inode;
7702         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7703         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7704         if (rc != 0) {
7705                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7706                 GOTO(out_mnt, rc);
7707         }
7708
7709         if (lmd_flags & LMD_FLG_NOSCRUB)
7710                 o->od_auto_scrub_interval = AS_NEVER;
7711
7712         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7713                 /* do not use pagecache with flash-backed storage */
7714                 o->od_writethrough_cache = 0;
7715                 o->od_read_cache = 0;
7716         }
7717
7718         GOTO(out, rc = 0);
7719
7720 out_mnt:
7721         mntput(o->od_mnt);
7722         o->od_mnt = NULL;
7723
7724 out:
7725         if (__page)
7726                 __free_page(__page);
7727
7728         return rc;
7729 }
7730
7731 static struct lu_device *osd_device_fini(const struct lu_env *env,
7732                                          struct lu_device *d)
7733 {
7734         struct osd_device *o = osd_dev(d);
7735
7736         ENTRY;
7737
7738         osd_index_backup(env, o, false);
7739         osd_shutdown(env, o);
7740         osd_procfs_fini(o);
7741         osd_obj_map_fini(o);
7742         osd_umount(env, o);
7743
7744         RETURN(NULL);
7745 }
7746
7747 static int osd_device_init0(const struct lu_env *env,
7748                             struct osd_device *o,
7749                             struct lustre_cfg *cfg)
7750 {
7751         struct lu_device *l = osd2lu_dev(o);
7752         struct osd_thread_info *info;
7753         int rc;
7754         int cplen = 0;
7755
7756         /* if the module was re-loaded, env can loose its keys */
7757         rc = lu_env_refill((struct lu_env *)env);
7758         if (rc)
7759                 GOTO(out, rc);
7760         info = osd_oti_get(env);
7761         LASSERT(info);
7762
7763         l->ld_ops = &osd_lu_ops;
7764         o->od_dt_dev.dd_ops = &osd_dt_ops;
7765
7766         spin_lock_init(&o->od_osfs_lock);
7767         mutex_init(&o->od_otable_mutex);
7768         INIT_LIST_HEAD(&o->od_orphan_list);
7769         INIT_LIST_HEAD(&o->od_index_backup_list);
7770         INIT_LIST_HEAD(&o->od_index_restore_list);
7771         spin_lock_init(&o->od_lock);
7772         o->od_index_backup_policy = LIBP_NONE;
7773         o->od_t10_type = 0;
7774
7775         o->od_read_cache = 1;
7776         o->od_writethrough_cache = 1;
7777         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7778         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7779         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7780         o->od_auto_scrub_interval = AS_DEFAULT;
7781
7782         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7783                         sizeof(o->od_svname));
7784         if (cplen >= sizeof(o->od_svname)) {
7785                 rc = -E2BIG;
7786                 GOTO(out, rc);
7787         }
7788
7789         o->od_index_backup_stop = 0;
7790         o->od_index = -1; /* -1 means index is invalid */
7791         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7792         if (rc == LDD_F_SV_TYPE_OST)
7793                 o->od_is_ost = 1;
7794
7795         o->od_full_scrub_ratio = OFSR_DEFAULT;
7796         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7797         rc = osd_mount(env, o, cfg);
7798         if (rc != 0)
7799                 GOTO(out, rc);
7800
7801         /* Can only check block device after mount */
7802         o->od_nonrotational =
7803                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7804
7805         rc = osd_obj_map_init(env, o);
7806         if (rc != 0)
7807                 GOTO(out_mnt, rc);
7808
7809         rc = lu_site_init(&o->od_site, l);
7810         if (rc != 0)
7811                 GOTO(out_compat, rc);
7812         o->od_site.ls_bottom_dev = l;
7813
7814         rc = lu_site_init_finish(&o->od_site);
7815         if (rc != 0)
7816                 GOTO(out_site, rc);
7817
7818         INIT_LIST_HEAD(&o->od_ios_list);
7819         /* setup scrub, including OI files initialization */
7820         o->od_in_init = 1;
7821         rc = osd_scrub_setup(env, o);
7822         o->od_in_init = 0;
7823         if (rc < 0)
7824                 GOTO(out_site, rc);
7825
7826         rc = osd_procfs_init(o, o->od_svname);
7827         if (rc != 0) {
7828                 CERROR("%s: can't initialize procfs: rc = %d\n",
7829                        o->od_svname, rc);
7830                 GOTO(out_scrub, rc);
7831         }
7832
7833         LASSERT(l->ld_site->ls_linkage.next != NULL);
7834         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7835
7836         /* initialize quota slave instance */
7837         /* currently it's no need to prepare qsd_instance_md for OST */
7838         if (!o->od_is_ost) {
7839                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7840                                                 &o->od_dt_dev,
7841                                                 o->od_proc_entry, true);
7842                 if (IS_ERR(o->od_quota_slave_md)) {
7843                         rc = PTR_ERR(o->od_quota_slave_md);
7844                         o->od_quota_slave_md = NULL;
7845                         GOTO(out_procfs, rc);
7846                 }
7847         }
7848
7849         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7850                                         o->od_proc_entry, false);
7851
7852         if (IS_ERR(o->od_quota_slave_dt)) {
7853                 if (o->od_quota_slave_md != NULL) {
7854                         qsd_fini(env, o->od_quota_slave_md);
7855                         o->od_quota_slave_md = NULL;
7856                 }
7857
7858                 rc = PTR_ERR(o->od_quota_slave_dt);
7859                 o->od_quota_slave_dt = NULL;
7860                 GOTO(out_procfs, rc);
7861         }
7862
7863         RETURN(0);
7864
7865 out_procfs:
7866         osd_procfs_fini(o);
7867 out_scrub:
7868         osd_scrub_cleanup(env, o);
7869 out_site:
7870         lu_site_fini(&o->od_site);
7871 out_compat:
7872         osd_obj_map_fini(o);
7873 out_mnt:
7874         osd_umount(env, o);
7875 out:
7876         return rc;
7877 }
7878
7879 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7880                                           struct lu_device_type *t,
7881                                           struct lustre_cfg *cfg)
7882 {
7883         struct osd_device *o;
7884         int rc;
7885
7886         OBD_ALLOC_PTR(o);
7887         if (o == NULL)
7888                 return ERR_PTR(-ENOMEM);
7889
7890         rc = dt_device_init(&o->od_dt_dev, t);
7891         if (rc == 0) {
7892                 /*
7893                  * Because the ctx might be revived in dt_device_init,
7894                  * refill the env here
7895                  */
7896                 lu_env_refill((struct lu_env *)env);
7897                 rc = osd_device_init0(env, o, cfg);
7898                 if (rc)
7899                         dt_device_fini(&o->od_dt_dev);
7900         }
7901
7902         if (unlikely(rc != 0))
7903                 OBD_FREE_PTR(o);
7904
7905         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7906 }
7907
7908 static struct lu_device *osd_device_free(const struct lu_env *env,
7909                                          struct lu_device *d)
7910 {
7911         struct osd_device *o = osd_dev(d);
7912
7913         ENTRY;
7914
7915         /* XXX: make osd top device in order to release reference */
7916         d->ld_site->ls_top_dev = d;
7917         lu_site_purge(env, d->ld_site, -1);
7918         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7919                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7920                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7921         }
7922         lu_site_fini(&o->od_site);
7923         dt_device_fini(&o->od_dt_dev);
7924         OBD_FREE_PTR(o);
7925         RETURN(NULL);
7926 }
7927
7928 static int osd_process_config(const struct lu_env *env,
7929                               struct lu_device *d, struct lustre_cfg *cfg)
7930 {
7931         struct osd_device *o = osd_dev(d);
7932         ssize_t count;
7933         int rc;
7934
7935         ENTRY;
7936
7937         switch (cfg->lcfg_command) {
7938         case LCFG_SETUP:
7939                 rc = osd_mount(env, o, cfg);
7940                 break;
7941         case LCFG_CLEANUP:
7942                 /*
7943                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7944                  * that may happend if hit failure during mount process.
7945                  */
7946                 osd_index_backup(env, o, false);
7947                 lu_dev_del_linkage(d->ld_site, d);
7948                 rc = osd_shutdown(env, o);
7949                 break;
7950         case LCFG_PARAM:
7951                 LASSERT(&o->od_dt_dev);
7952                 count  = class_modify_config(cfg, PARAM_OSD,
7953                                              &o->od_dt_dev.dd_kobj);
7954                 if (count < 0)
7955                         count = class_modify_config(cfg, PARAM_OST,
7956                                                     &o->od_dt_dev.dd_kobj);
7957                 rc = count > 0 ? 0 : count;
7958                 break;
7959         case LCFG_PRE_CLEANUP:
7960                 osd_scrub_stop(o);
7961                 osd_index_backup(env, o,
7962                                  o->od_index_backup_policy != LIBP_NONE);
7963                 rc = 0;
7964                 break;
7965         default:
7966                 rc = -ENOSYS;
7967         }
7968
7969         RETURN(rc);
7970 }
7971
7972 static int osd_recovery_complete(const struct lu_env *env,
7973                                  struct lu_device *d)
7974 {
7975         struct osd_device *osd = osd_dev(d);
7976         int rc = 0;
7977
7978         ENTRY;
7979
7980         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7981                 RETURN(0);
7982
7983         /*
7984          * start qsd instance on recovery completion, this notifies the quota
7985          * slave code that we are about to process new requests now
7986          */
7987         rc = qsd_start(env, osd->od_quota_slave_dt);
7988         if (rc == 0 && osd->od_quota_slave_md != NULL)
7989                 rc = qsd_start(env, osd->od_quota_slave_md);
7990
7991         RETURN(rc);
7992 }
7993
7994 /*
7995  * we use exports to track all osd users
7996  */
7997 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7998                            struct obd_device *obd, struct obd_uuid *cluuid,
7999                            struct obd_connect_data *data, void *localdata)
8000 {
8001         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8002         struct lustre_handle conn;
8003         int rc;
8004
8005         ENTRY;
8006
8007         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8008
8009         rc = class_connect(&conn, obd, cluuid);
8010         if (rc)
8011                 RETURN(rc);
8012
8013         *exp = class_conn2export(&conn);
8014
8015         spin_lock(&osd->od_osfs_lock);
8016         osd->od_connects++;
8017         spin_unlock(&osd->od_osfs_lock);
8018
8019         RETURN(0);
8020 }
8021
8022 /*
8023  * once last export (we don't count self-export) disappeared
8024  * osd can be released
8025  */
8026 static int osd_obd_disconnect(struct obd_export *exp)
8027 {
8028         struct obd_device *obd = exp->exp_obd;
8029         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8030         int rc, release = 0;
8031
8032         ENTRY;
8033
8034         /* Only disconnect the underlying layers on the final disconnect. */
8035         spin_lock(&osd->od_osfs_lock);
8036         osd->od_connects--;
8037         if (osd->od_connects == 0)
8038                 release = 1;
8039         spin_unlock(&osd->od_osfs_lock);
8040
8041         rc = class_disconnect(exp); /* bz 9811 */
8042
8043         if (rc == 0 && release)
8044                 class_manual_cleanup(obd);
8045         RETURN(rc);
8046 }
8047
8048 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8049                        struct lu_device *dev)
8050 {
8051         struct osd_device *osd = osd_dev(dev);
8052         struct lr_server_data *lsd =
8053                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8054         int result = 0;
8055
8056         ENTRY;
8057
8058         if (osd->od_quota_slave_md != NULL) {
8059                 /* set up quota slave objects for inode */
8060                 result = qsd_prepare(env, osd->od_quota_slave_md);
8061                 if (result != 0)
8062                         RETURN(result);
8063         }
8064
8065         if (osd->od_quota_slave_dt != NULL) {
8066                 /* set up quota slave objects for block */
8067                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8068                 if (result != 0)
8069                         RETURN(result);
8070         }
8071
8072
8073         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8074 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8075                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8076                         osd->od_index_in_idif = 1;
8077                 } else {
8078                         osd->od_index_in_idif = 0;
8079                         result = osd_register_proc_index_in_idif(osd);
8080                         if (result != 0)
8081                                 RETURN(result);
8082                 }
8083 #else
8084                 osd->od_index_in_idif = 1;
8085 #endif
8086         }
8087
8088         result = osd_fid_init(env, osd);
8089
8090         RETURN(result);
8091 }
8092
8093 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8094                          struct lu_fid *fid, struct md_op_data *op_data)
8095 {
8096         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8097
8098         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8099 }
8100
8101 static const struct lu_object_operations osd_lu_obj_ops = {
8102         .loo_object_init      = osd_object_init,
8103         .loo_object_delete    = osd_object_delete,
8104         .loo_object_release   = osd_object_release,
8105         .loo_object_free      = osd_object_free,
8106         .loo_object_print     = osd_object_print,
8107         .loo_object_invariant = osd_object_invariant
8108 };
8109
8110 const struct lu_device_operations osd_lu_ops = {
8111         .ldo_object_alloc      = osd_object_alloc,
8112         .ldo_process_config    = osd_process_config,
8113         .ldo_recovery_complete = osd_recovery_complete,
8114         .ldo_prepare           = osd_prepare,
8115 };
8116
8117 static const struct lu_device_type_operations osd_device_type_ops = {
8118         .ldto_init = osd_type_init,
8119         .ldto_fini = osd_type_fini,
8120
8121         .ldto_start = osd_type_start,
8122         .ldto_stop  = osd_type_stop,
8123
8124         .ldto_device_alloc = osd_device_alloc,
8125         .ldto_device_free  = osd_device_free,
8126
8127         .ldto_device_init = osd_device_init,
8128         .ldto_device_fini = osd_device_fini
8129 };
8130
8131 static struct lu_device_type osd_device_type = {
8132         .ldt_tags     = LU_DEVICE_DT,
8133         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8134         .ldt_ops      = &osd_device_type_ops,
8135         .ldt_ctx_tags = LCT_LOCAL,
8136 };
8137
8138 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8139 {
8140         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8141         struct super_block *sb = osd_sb(osd);
8142
8143         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8144 }
8145
8146 /*
8147  * lprocfs legacy support.
8148  */
8149 static const struct obd_ops osd_obd_device_ops = {
8150         .o_owner = THIS_MODULE,
8151         .o_connect      = osd_obd_connect,
8152         .o_disconnect   = osd_obd_disconnect,
8153         .o_fid_alloc    = osd_fid_alloc,
8154         .o_health_check = osd_health_check,
8155 };
8156
8157 static ssize_t track_declares_assert_show(struct kobject *kobj,
8158                                    struct attribute *attr,
8159                                    char *buf)
8160 {
8161         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8162 }
8163
8164 static ssize_t track_declares_assert_store(struct kobject *kobj,
8165                                            struct attribute *attr,
8166                                            const char *buffer, size_t count)
8167 {
8168         bool track_declares_assert;
8169         int rc;
8170
8171         rc = kstrtobool(buffer, &track_declares_assert);
8172         if (rc)
8173                 return rc;
8174
8175         ldiskfs_track_declares_assert = track_declares_assert;
8176
8177         return count;
8178 }
8179 LUSTRE_RW_ATTR(track_declares_assert);
8180
8181 static int __init osd_init(void)
8182 {
8183         struct kobject *kobj;
8184         int rc;
8185
8186         BUILD_BUG_ON(BH_DXLock >=
8187                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8188 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8189         /* please, try to keep osd_thread_info smaller than a page */
8190         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8191 #endif
8192
8193         osd_oi_mod_init();
8194
8195         rc = lu_kmem_init(ldiskfs_caches);
8196         if (rc)
8197                 return rc;
8198
8199 #ifdef CONFIG_KALLSYMS
8200         priv_security_file_alloc =
8201                 (void *)kallsyms_lookup_name("security_file_alloc");
8202         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8203         priv_dev_check_rdonly =
8204                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8205 #endif
8206
8207         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8208                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8209         if (rc) {
8210                 lu_kmem_fini(ldiskfs_caches);
8211                 return rc;
8212         }
8213
8214         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8215         if (kobj) {
8216                 rc = sysfs_create_file(kobj,
8217                                        &lustre_attr_track_declares_assert.attr);
8218                 kobject_put(kobj);
8219                 if (rc) {
8220                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8221                         rc = 0;
8222                 }
8223         }
8224         return rc;
8225 }
8226
8227 static void __exit osd_exit(void)
8228 {
8229         struct kobject *kobj;
8230
8231         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8232         if (kobj) {
8233                 sysfs_remove_file(kobj,
8234                                   &lustre_attr_track_declares_assert.attr);
8235                 kobject_put(kobj);
8236         }
8237         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8238         lu_kmem_fini(ldiskfs_caches);
8239 }
8240
8241 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8242 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8243 MODULE_VERSION(LUSTRE_VERSION_STRING);
8244 MODULE_LICENSE("GPL");
8245
8246 module_init(osd_init);
8247 module_exit(osd_exit);