lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 499                               "maximum entry limit\n",
 500                               osd_name(osd), parent->i_ino, PFID(fid),
 501                               rc == -ENOSPC ? "has reached" : "is approaching");
 502                 else
 503                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 504                               "maximum entry limit\n",
 505                               osd_name(osd), parent->i_ino,
 506                               rc == -ENOSPC ? "has reached" : "is approaching");
 507
 508                 /* ignore such error now */
 509                 if (rc == -ENOBUFS)
 510                         rc = 0;
 511         }
 512
 513         return rc;
 514 }
 515
 516
 517 struct inode *
 518 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 519              struct osd_inode_id *id, struct lu_fid *fid)
 520 {
 521         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 522         struct inode *inode;
 523         int rc;
 524
 525         inode = osd_iget(info, dev, id);
 526         if (IS_ERR(inode))
 527                 return inode;
 528
 529         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 530         if (!rc) {
 531                 *fid = loa->loa_lma.lma_self_fid;
 532         } else if (rc == -ENODATA) {
 533                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 534                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 535                 else
 536                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 537         } else {
 538                 iput(inode);
 539                 inode = ERR_PTR(rc);
 540         }
 541         return inode;
 542 }
 543
 544 static struct inode *osd_iget_check(struct osd_thread_info *info,
 545                                     struct osd_device *dev,
 546                                     const struct lu_fid *fid,
 547                                     struct osd_inode_id *id,
 548                                     bool trusted)
 549 {
 550         struct inode *inode;
 551         int rc = 0;
 552
 553         ENTRY;
 554
 555         /*
 556          * The cached OI mapping is trustable. If we cannot locate the inode
 557          * via the cached OI mapping, then return the failure to the caller
 558          * directly without further OI checking.
 559          */
 560
 561 again:
 562         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid, does not contain self-fid
 710  * \retval 0:  filter_fid_old, 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_old *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(*ff)) {
 722                 rc = 0;
 723                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 724                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 725                 /*
 726                  * XXX: use 0 as the index for compatibility, the caller will
 727                  * handle index related issues when necessary.
 728                  */
 729                 if (!rc)
 730                         ostid_to_fid(fid, ostid, 0);
 731         } else if (rc == sizeof(struct filter_fid)) {
 732                 rc = 1;
 733         } else if (rc >= 0) {
 734                 rc = -EINVAL;
 735         }
 736
 737         return rc;
 738 }
 739
 740 static int osd_lma_self_repair(struct osd_thread_info *info,
 741                                struct osd_device *osd, struct inode *inode,
 742                                const struct lu_fid *fid, __u32 compat)
 743 {
 744         handle_t *jh;
 745         int rc;
 746
 747         LASSERT(current->journal_info == NULL);
 748
 749         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 750                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 751         if (IS_ERR(jh)) {
 752                 rc = PTR_ERR(jh);
 753                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 754                       osd_name(osd), rc);
 755                 return rc;
 756         }
 757
 758         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 759         if (rc != 0)
 760                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 761                       osd_name(osd), rc);
 762         ldiskfs_journal_stop(jh);
 763         return rc;
 764 }
 765
 766 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 767 {
 768         struct osd_thread_info *info = osd_oti_get(env);
 769         struct osd_device *osd = osd_obj2dev(obj);
 770         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 771         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 772         struct inode *inode = obj->oo_inode;
 773         struct dentry *dentry = &info->oti_obj_dentry;
 774         struct lu_fid *fid = NULL;
 775         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 776         int rc;
 777
 778         ENTRY;
 779
 780         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 781                              (void *)loa, sizeof(*loa));
 782         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 783                 fid = &lma->lma_self_fid;
 784                 rc = osd_get_idif(info, inode, dentry, fid);
 785                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 786                         /*
 787                          * For the given OST-object, if it has neither LMA nor
 788                          * FID in XATTR_NAME_FID, then the given FID (which is
 789                          * contained in the @obj, from client RPC for locating
 790                          * the OST-object) is trusted. We use it to generate
 791                          * the LMA.
 792                          */
 793                         osd_lma_self_repair(info, osd, inode, rfid,
 794                                             LMAC_FID_ON_OST);
 795                         RETURN(0);
 796                 }
 797         }
 798
 799         if (rc < 0)
 800                 RETURN(rc);
 801
 802         if (rc > 0) {
 803                 rc = 0;
 804                 lustre_lma_swab(lma);
 805                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 806                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 807                               S_ISREG(inode->i_mode)))) {
 808                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 809                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 810                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 811                               PFID(rfid), inode->i_ino);
 812                         rc = -EOPNOTSUPP;
 813                 } else {
 814                         fid = &lma->lma_self_fid;
 815                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 816                             osd->od_is_ost)
 817                                 obj->oo_pfid_in_lma = 1;
 818                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 819                             !osd->od_is_ost)
 820                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 821                 }
 822         }
 823
 824         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 825                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 826                         struct ost_id   *oi   = &info->oti_ostid;
 827                         struct lu_fid   *fid1 = &info->oti_fid3;
 828                         __u32            idx  = fid_idif_ost_idx(rfid);
 829
 830                         /*
 831                          * For old IDIF, the OST index is not part of the IDIF,
 832                          * Means that different OSTs may have the same IDIFs.
 833                          * Under such case, we need to make some compatible
 834                          * check to make sure to trigger OI scrub properly.
 835                          */
 836                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 837                                 /* Given @rfid is new, LMA is old. */
 838                                 fid_to_ostid(fid, oi);
 839                                 ostid_to_fid(fid1, oi, idx);
 840                                 if (lu_fid_eq(fid1, rfid)) {
 841                                         if (osd->od_index_in_idif)
 842                                                 osd_lma_self_repair(info, osd,
 843                                                         inode, rfid,
 844                                                         LMAC_FID_ON_OST);
 845                                         RETURN(0);
 846                                 }
 847                         }
 848                 }
 849
 850                 rc = -EREMCHG;
 851         }
 852
 853         RETURN(rc);
 854 }
 855
 856 struct osd_check_lmv_buf {
 857 #ifdef HAVE_DIR_CONTEXT
 858         /* please keep it as first member */
 859         struct dir_context ctx;
 860 #endif
 861         struct osd_thread_info *oclb_info;
 862         struct osd_device *oclb_dev;
 863         struct osd_idmap_cache *oclb_oic;
 864         int oclb_items;
 865         bool oclb_found;
 866 };
 867
 868 /**
 869  * It is called internally by ->readdir() to filter out the
 870  * local slave object's FID of the striped directory.
 871  *
 872  * \retval      1 found the local slave's FID
 873  * \retval      0 continue to check next item
 874  * \retval      -ve for failure
 875  */
 876 #ifdef HAVE_FILLDIR_USE_CTX
 877 static int osd_stripe_dir_filldir(struct dir_context *buf,
 878 #else
 879 static int osd_stripe_dir_filldir(void *buf,
 880 #endif
 881                                   const char *name, int namelen,
 882                                   loff_t offset, __u64 ino, unsigned int d_type)
 883 {
 884         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 885         struct osd_thread_info *oti = oclb->oclb_info;
 886         struct lu_fid *fid = &oti->oti_fid3;
 887         struct osd_inode_id *id = &oti->oti_id3;
 888         struct osd_device *dev = oclb->oclb_dev;
 889         struct osd_idmap_cache *oic = oclb->oclb_oic;
 890         struct inode *inode;
 891
 892         oclb->oclb_items++;
 893
 894         if (name[0] == '.')
 895                 return 0;
 896
 897         fid_zero(fid);
 898         sscanf(name + 1, SFID, RFID(fid));
 899         if (!fid_is_sane(fid))
 900                 return 0;
 901
 902         if (osd_remote_fid(oti->oti_env, dev, fid))
 903                 return 0;
 904
 905         osd_id_gen(id, ino, OSD_OII_NOGEN);
 906         inode = osd_iget(oti, dev, id);
 907         if (IS_ERR(inode))
 908                 return PTR_ERR(inode);
 909
 910         iput(inode);
 911         osd_add_oi_cache(oti, dev, id, fid);
 912         oic->oic_fid = *fid;
 913         oic->oic_lid = *id;
 914         oic->oic_dev = dev;
 915         osd_oii_insert(dev, oic, true);
 916         oclb->oclb_found = true;
 917
 918         return 1;
 919 }
 920
 921 /*
 922  * When lookup item under striped directory, we need to locate the master
 923  * MDT-object of the striped directory firstly, then the client will send
 924  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 925  * and the item's name. If the system is restored from MDT file level backup,
 926  * then before the OI scrub completely built the OI files, the OI mappings of
 927  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 928  * not a problem for the master MDT-object. Because when locate the master
 929  * MDT-object, we will do name based lookup (for the striped directory itself)
 930  * firstly, during such process we can setup the correct OI mapping for the
 931  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 932  * the client will not trigger name based lookup on the MDT to locate the slave
 933  * MDT-object before locating item under the striped directory, then when
 934  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 935  * is invalid and does not know what the right OI mapping is, then the MDT has
 936  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 937  * the OI file, related OI mapping is unknown yet, please try again later. And
 938  * then client will re-try the RPC again and again until related OI mapping has
 939  * been updated. That is quite inefficient.
 940  *
 941  * To resolve above trouble, we will handle it as the following two cases:
 942  *
 943  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 944  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 945  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 946  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 947  *    directly. Please check osd_fid_lookup().
 948  *
 949  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 950  *    Under such case, during lookup the master MDT-object, we will lookup the
 951  *    slave MDT-object via readdir against the master MDT-object, because the
 952  *    slave MDT-objects information are stored as sub-directories with the name
 953  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 954  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 955  *    the correct OI mapping for the slave MDT-object.
 956  */
 957 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 958                          struct inode *inode, struct osd_idmap_cache *oic)
 959 {
 960         struct lu_buf *buf = &oti->oti_big_buf;
 961         struct dentry *dentry = &oti->oti_obj_dentry;
 962         struct file *filp = &oti->oti_file;
 963         const struct file_operations *fops;
 964         struct lmv_mds_md_v1 *lmv1;
 965         struct osd_check_lmv_buf oclb = {
 966 #ifdef HAVE_DIR_CONTEXT
 967                 .ctx.actor = osd_stripe_dir_filldir,
 968 #endif
 969                 .oclb_info = oti,
 970                 .oclb_dev = dev,
 971                 .oclb_oic = oic,
 972                 .oclb_found = false,
 973         };
 974         int rc = 0;
 975
 976         ENTRY;
 977
 978 again:
 979         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 980                              buf->lb_len);
 981         if (rc == -ERANGE) {
 982                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 983                 if (rc > 0) {
 984                         lu_buf_realloc(buf, rc);
 985                         if (buf->lb_buf == NULL)
 986                                 GOTO(out, rc = -ENOMEM);
 987
 988                         goto again;
 989                 }
 990         }
 991
 992         if (unlikely(rc == 0 || rc == -ENODATA))
 993                 GOTO(out, rc = 0);
 994
 995         if (rc < 0)
 996                 GOTO(out, rc);
 997
 998         if (unlikely(buf->lb_buf == NULL)) {
 999                 lu_buf_realloc(buf, rc);
1000                 if (buf->lb_buf == NULL)
1001                         GOTO(out, rc = -ENOMEM);
1002
1003                 goto again;
1004         }
1005
1006         lmv1 = buf->lb_buf;
1007         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1008                 GOTO(out, rc = 0);
1009
1010         fops = inode->i_fop;
1011         dentry->d_inode = inode;
1012         dentry->d_sb = inode->i_sb;
1013         filp->f_pos = 0;
1014         filp->f_path.dentry = dentry;
1015         filp->f_mode = FMODE_64BITHASH;
1016         filp->f_mapping = inode->i_mapping;
1017         filp->f_op = fops;
1018         filp->private_data = NULL;
1019         set_file_inode(filp, inode);
1020
1021         do {
1022                 oclb.oclb_items = 0;
1023 #ifdef HAVE_DIR_CONTEXT
1024                 oclb.ctx.pos = filp->f_pos;
1025 #ifdef HAVE_ITERATE_SHARED
1026                 rc = fops->iterate_shared(filp, &oclb.ctx);
1027 #else
1028                 rc = fops->iterate(filp, &oclb.ctx);
1029 #endif
1030                 filp->f_pos = oclb.ctx.pos;
1031 #else
1032                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1033 #endif
1034         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1035                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1036         fops->release(inode, filp);
1037
1038 out:
1039         if (rc < 0)
1040                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1041                        DFID": rc = %d\n", osd_ino2name(inode),
1042                        inode->i_ino, inode->i_generation,
1043                        PFID(&oic->oic_fid), rc);
1044         else
1045                 rc = 0;
1046
1047         RETURN(rc);
1048 }
1049
1050 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1051                           const struct lu_fid *fid,
1052                           const struct lu_object_conf *conf)
1053 {
1054         struct osd_thread_info *info;
1055         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1056         struct osd_device *dev;
1057         struct osd_idmap_cache *oic;
1058         struct osd_inode_id *id;
1059         struct inode *inode = NULL;
1060         struct lustre_scrub *scrub;
1061         struct scrub_file *sf;
1062         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1063         __u32 saved_ino;
1064         __u32 saved_gen;
1065         int result = 0;
1066         int rc1 = 0;
1067         bool remote = false;
1068         bool trusted = true;
1069         bool updated = false;
1070         bool checked = false;
1071
1072         ENTRY;
1073
1074         LINVRNT(osd_invariant(obj));
1075         LASSERT(obj->oo_inode == NULL);
1076         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1077
1078         dev = osd_dev(ldev);
1079         scrub = &dev->od_scrub.os_scrub;
1080         sf = &scrub->os_file;
1081         info = osd_oti_get(env);
1082         LASSERT(info);
1083         oic = &info->oti_cache;
1084
1085         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1086                 RETURN(-ENOENT);
1087
1088         /*
1089          * For the object is created as locking anchor, or for the object to
1090          * be created on disk. No need to osd_oi_lookup() at here because FID
1091          * shouldn't never be re-used, if it's really a duplicate FID from
1092          * unexpected reason, we should be able to detect it later by calling
1093          * do_create->osd_oi_insert().
1094          */
1095         if (conf && conf->loc_flags & LOC_F_NEW)
1096                 GOTO(out, result = 0);
1097
1098         /* Search order: 1. per-thread cache. */
1099         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1100                 id = &oic->oic_lid;
1101                 goto iget;
1102         }
1103
1104         id = &info->oti_id;
1105         if (!list_empty(&scrub->os_inconsistent_items)) {
1106                 /* Search order: 2. OI scrub pending list. */
1107                 result = osd_oii_lookup(dev, fid, id);
1108                 if (!result)
1109                         goto iget;
1110         }
1111
1112         /*
1113          * The OI mapping in the OI file can be updated by the OI scrub
1114          * when we locate the inode via FID. So it may be not trustable.
1115          */
1116         trusted = false;
1117
1118         /* Search order: 3. OI files. */
1119         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1120         if (result == -ENOENT) {
1121                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1122                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1123                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1124                                       sf->sf_oi_bitmap))
1125                         GOTO(out, result = 0);
1126
1127                 goto trigger;
1128         }
1129
1130         if (result)
1131                 GOTO(out, result);
1132
1133 iget:
1134         obj->oo_inode = NULL;
1135         /* for later passes through checks, not true on first pass */
1136         if (!IS_ERR_OR_NULL(inode))
1137                 iput(inode);
1138
1139         inode = osd_iget_check(info, dev, fid, id, trusted);
1140         if (!IS_ERR(inode)) {
1141                 obj->oo_inode = inode;
1142                 result = 0;
1143                 if (remote)
1144                         goto trigger;
1145
1146                 goto check_lma;
1147         }
1148
1149         result = PTR_ERR(inode);
1150         if (result == -ENOENT || result == -ESTALE)
1151                 GOTO(out, result = 0);
1152
1153         if (result != -EREMCHG)
1154                 GOTO(out, result);
1155
1156 trigger:
1157         /*
1158          * We still have chance to get the valid inode: for the
1159          * object which is referenced by remote name entry, the
1160          * object on the local MDT will be linked under the dir
1161          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1162          *
1163          * We do not know whether the object for the given FID
1164          * is referenced by some remote name entry or not, and
1165          * especially for DNE II, a multiple-linked object may
1166          * have many name entries reside on many MDTs.
1167          *
1168          * To simplify the operation, OSD will not distinguish
1169          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1170          * only happened for the RPC from other MDT during the
1171          * OI scrub, or for the client side RPC with FID only,
1172          * such as FID to path, or from old connected client.
1173          */
1174         if (!remote) {
1175                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1176                 if (!rc1) {
1177                         remote = true;
1178                         trusted = true;
1179                         flags |= SS_AUTO_PARTIAL;
1180                         flags &= ~SS_AUTO_FULL;
1181                         goto iget;
1182                 }
1183         }
1184
1185         if (thread_is_running(&scrub->os_thread)) {
1186                 if (scrub->os_partial_scan && !scrub->os_in_join)
1187                         goto join;
1188
1189                 if (IS_ERR_OR_NULL(inode) || result)
1190                         GOTO(out, result = -EINPROGRESS);
1191
1192                 LASSERT(remote);
1193                 LASSERT(obj->oo_inode == inode);
1194
1195                 osd_add_oi_cache(info, dev, id, fid);
1196                 osd_oii_insert(dev, oic, true);
1197                 goto found;
1198         }
1199
1200         if (dev->od_auto_scrub_interval == AS_NEVER) {
1201                 if (!remote)
1202                         GOTO(out, result = -EREMCHG);
1203
1204                 LASSERT(!result);
1205                 LASSERT(obj->oo_inode == inode);
1206
1207                 osd_add_oi_cache(info, dev, id, fid);
1208                 goto found;
1209         }
1210
1211 join:
1212         rc1 = osd_scrub_start(env, dev, flags);
1213         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1214                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1215         if (rc1 && rc1 != -EALREADY)
1216                 GOTO(out, result = -EREMCHG);
1217
1218         if (IS_ERR_OR_NULL(inode) || result)
1219                 GOTO(out, result = -EINPROGRESS);
1220
1221         LASSERT(remote);
1222         LASSERT(obj->oo_inode == inode);
1223
1224         osd_add_oi_cache(info, dev, id, fid);
1225         osd_oii_insert(dev, oic, true);
1226         goto found;
1227
1228 check_lma:
1229         checked = true;
1230         if (unlikely(obj->oo_header))
1231                 goto found;
1232
1233         result = osd_check_lma(env, obj);
1234         if (!result)
1235                 goto found;
1236
1237         LASSERTF(id->oii_ino == inode->i_ino &&
1238                  id->oii_gen == inode->i_generation,
1239                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1240                  PFID(fid), id->oii_ino, id->oii_gen,
1241                  inode->i_ino, inode->i_generation);
1242
1243         saved_ino = inode->i_ino;
1244         saved_gen = inode->i_generation;
1245
1246         if (unlikely(result == -ENODATA)) {
1247                 /*
1248                  * If the OI scrub updated the OI mapping by race, it
1249                  * must be valid. Trust the inode that has no LMA EA.
1250                  */
1251                 if (updated)
1252                         goto found;
1253
1254                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1255                 if (!result) {
1256                         /*
1257                          * The OI mapping is still there, the inode is still
1258                          * valid. It is just becaues the inode has no LMA EA.
1259                          */
1260                         if (saved_ino == id->oii_ino &&
1261                             saved_gen == id->oii_gen)
1262                                 goto found;
1263
1264                         /*
1265                          * It is the OI scrub updated the OI mapping by race.
1266                          * The new OI mapping must be valid.
1267                          */
1268                         trusted = true;
1269                         updated = true;
1270                         goto iget;
1271                 }
1272
1273                 /*
1274                  * "result == -ENOENT" means that the OI mappinghas been
1275                  * removed by race, so the inode belongs to other object.
1276                  *
1277                  * Others error can be returned  directly.
1278                  */
1279                 if (result == -ENOENT) {
1280                         LASSERT(trusted);
1281
1282                         obj->oo_inode = NULL;
1283                         result = 0;
1284                 }
1285         }
1286
1287         if (result != -EREMCHG)
1288                 GOTO(out, result);
1289
1290         LASSERT(!updated);
1291
1292         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1293         /*
1294          * "result == -ENOENT" means the cached OI mapping has been removed
1295          * from the OI file by race, above inode belongs to other object.
1296          */
1297         if (result == -ENOENT) {
1298                 LASSERT(trusted);
1299
1300                 obj->oo_inode = NULL;
1301                 GOTO(out, result = 0);
1302         }
1303
1304         if (result)
1305                 GOTO(out, result);
1306
1307         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1308                 result = -EREMCHG;
1309                 goto trigger;
1310         }
1311
1312         /*
1313          * It is the OI scrub updated the OI mapping by race.
1314          * The new OI mapping must be valid.
1315          */
1316         trusted = true;
1317         updated = true;
1318         goto iget;
1319
1320 found:
1321         if (!checked) {
1322                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1323                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1324
1325                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1326                 if (!result) {
1327                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1328                             dev->od_is_ost)
1329                                 obj->oo_pfid_in_lma = 1;
1330                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1331                             !dev->od_is_ost)
1332                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1333                 } else if (result != -ENODATA) {
1334                         GOTO(out, result);
1335                 }
1336         }
1337
1338         obj->oo_compat_dot_created = 1;
1339         obj->oo_compat_dotdot_created = 1;
1340
1341         if (S_ISDIR(inode->i_mode) &&
1342             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1343                 osd_check_lmv(info, dev, inode, oic);
1344
1345         result = osd_attach_jinode(inode);
1346         if (result)
1347                 GOTO(out, result);
1348
1349         if (!ldiskfs_pdo)
1350                 GOTO(out, result = 0);
1351
1352         LASSERT(!obj->oo_hl_head);
1353         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1354
1355         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1356
1357 out:
1358         if (result || !obj->oo_inode) {
1359                 if (!IS_ERR_OR_NULL(inode))
1360                         iput(inode);
1361
1362                 obj->oo_inode = NULL;
1363                 if (trusted)
1364                         fid_zero(&oic->oic_fid);
1365         }
1366
1367         LINVRNT(osd_invariant(obj));
1368         return result;
1369 }
1370
1371 /*
1372  * Concurrency: shouldn't matter.
1373  */
1374 static void osd_object_init0(struct osd_object *obj)
1375 {
1376         LASSERT(obj->oo_inode != NULL);
1377         obj->oo_dt.do_body_ops = &osd_body_ops;
1378         obj->oo_dt.do_lu.lo_header->loh_attr |=
1379                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1380 }
1381
1382 /*
1383  * Concurrency: no concurrent access is possible that early in object
1384  * life-cycle.
1385  */
1386 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1387                            const struct lu_object_conf *conf)
1388 {
1389         struct osd_object *obj = osd_obj(l);
1390         int result;
1391
1392         LINVRNT(osd_invariant(obj));
1393
1394         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1395                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1396                 l->lo_header->loh_attr |= LOHA_EXISTS;
1397                 return 0;
1398         }
1399
1400         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1401         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1402         if (result == 0 && obj->oo_inode != NULL) {
1403                 struct osd_thread_info *oti = osd_oti_get(env);
1404                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1405
1406                 osd_object_init0(obj);
1407                 if (unlikely(obj->oo_header))
1408                         return 0;
1409
1410                 result = osd_get_lma(oti, obj->oo_inode,
1411                                      &oti->oti_obj_dentry, loa);
1412                 if (!result) {
1413                         /*
1414                          * Convert LMAI flags to lustre LMA flags
1415                          * and cache it to oo_lma_flags
1416                          */
1417                         obj->oo_lma_flags =
1418                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1419                 } else if (result == -ENODATA) {
1420                         result = 0;
1421                 }
1422         }
1423
1424         LINVRNT(osd_invariant(obj));
1425         return result;
1426 }
1427
1428 /*
1429  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1430  * The left part is for value, binary mode.
1431  */
1432 struct osd_xattr_entry {
1433         struct list_head        oxe_list;
1434         size_t                  oxe_len;
1435         size_t                  oxe_namelen;
1436         bool                    oxe_exist;
1437         struct rcu_head         oxe_rcu;
1438         char                    oxe_buf[0];
1439 };
1440
1441 static int osd_oxc_get(struct osd_object *obj, const char *name,
1442                        struct lu_buf *buf)
1443 {
1444         struct osd_xattr_entry *tmp;
1445         struct osd_xattr_entry *oxe = NULL;
1446         size_t namelen = strlen(name);
1447         int rc;
1448
1449         ENTRY;
1450
1451         rcu_read_lock();
1452         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1453                 if (namelen == tmp->oxe_namelen &&
1454                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1455                         oxe = tmp;
1456                         break;
1457                 }
1458         }
1459
1460         if (oxe == NULL)
1461                 GOTO(out, rc = -ENOENT);
1462
1463         if (!oxe->oxe_exist)
1464                 GOTO(out, rc = -ENODATA);
1465
1466         /* vallen */
1467         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1468         LASSERT(rc > 0);
1469
1470         if (buf->lb_buf == NULL)
1471                 GOTO(out, rc);
1472
1473         if (buf->lb_len < rc)
1474                 GOTO(out, rc = -ERANGE);
1475
1476         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1477         EXIT;
1478 out:
1479         rcu_read_unlock();
1480
1481         return rc;
1482 }
1483
1484 static void osd_oxc_free(struct rcu_head *head)
1485 {
1486         struct osd_xattr_entry *oxe;
1487
1488         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1489         OBD_FREE(oxe, oxe->oxe_len);
1490 }
1491
1492 static void osd_oxc_add(struct osd_object *obj, const char *name,
1493                         const char *buf, int buflen)
1494 {
1495         struct osd_xattr_entry *oxe;
1496         struct osd_xattr_entry *old = NULL;
1497         struct osd_xattr_entry *tmp;
1498         size_t namelen = strlen(name);
1499         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1500
1501         OBD_ALLOC(oxe, len);
1502         if (oxe == NULL)
1503                 return;
1504
1505         INIT_LIST_HEAD(&oxe->oxe_list);
1506         oxe->oxe_len = len;
1507         oxe->oxe_namelen = namelen;
1508         memcpy(oxe->oxe_buf, name, namelen);
1509         if (buflen > 0) {
1510                 LASSERT(buf != NULL);
1511                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1512                 oxe->oxe_exist = true;
1513         } else {
1514                 oxe->oxe_exist = false;
1515         }
1516
1517         /* this should be rarely called, just remove old and add new */
1518         spin_lock(&obj->oo_guard);
1519         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1520                 if (namelen == tmp->oxe_namelen &&
1521                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1522                         old = tmp;
1523                         break;
1524                 }
1525         }
1526         if (old != NULL) {
1527                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1528                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1529         } else {
1530                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1531         }
1532         spin_unlock(&obj->oo_guard);
1533 }
1534
1535 static void osd_oxc_del(struct osd_object *obj, const char *name)
1536 {
1537         struct osd_xattr_entry *oxe;
1538         size_t namelen = strlen(name);
1539
1540         spin_lock(&obj->oo_guard);
1541         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1542                 if (namelen == oxe->oxe_namelen &&
1543                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1544                         list_del_rcu(&oxe->oxe_list);
1545                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1546                         break;
1547                 }
1548         }
1549         spin_unlock(&obj->oo_guard);
1550 }
1551
1552 static void osd_oxc_fini(struct osd_object *obj)
1553 {
1554         struct osd_xattr_entry *oxe, *next;
1555
1556         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1557                 list_del(&oxe->oxe_list);
1558                 OBD_FREE(oxe, oxe->oxe_len);
1559         }
1560 }
1561
1562 /*
1563  * Concurrency: no concurrent access is possible that late in object
1564  * life-cycle.
1565  */
1566 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1567 {
1568         struct osd_object *obj = osd_obj(l);
1569         struct lu_object_header *h = obj->oo_header;
1570
1571         LINVRNT(osd_invariant(obj));
1572
1573         osd_oxc_fini(obj);
1574         dt_object_fini(&obj->oo_dt);
1575         if (obj->oo_hl_head != NULL)
1576                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1577         OBD_FREE_PTR(obj);
1578         if (unlikely(h)) {
1579                 lu_object_header_fini(h);
1580                 OBD_FREE_PTR(h);
1581         }
1582 }
1583
1584 /*
1585  * Concurrency: no concurrent access is possible that late in object
1586  * life-cycle.
1587  */
1588 static void osd_index_fini(struct osd_object *o)
1589 {
1590         struct iam_container *bag;
1591
1592         if (o->oo_dir != NULL) {
1593                 bag = &o->oo_dir->od_container;
1594                 if (o->oo_inode != NULL) {
1595                         if (bag->ic_object == o->oo_inode)
1596                                 iam_container_fini(bag);
1597                 }
1598                 OBD_FREE_PTR(o->oo_dir);
1599                 o->oo_dir = NULL;
1600         }
1601 }
1602
1603 /*
1604  * Concurrency: no concurrent access is possible that late in object
1605  * life-cycle (for all existing callers, that is. New callers have to provide
1606  * their own locking.)
1607  */
1608 static int osd_inode_unlinked(const struct inode *inode)
1609 {
1610         return inode->i_nlink == 0;
1611 }
1612
1613 enum {
1614         OSD_TXN_OI_DELETE_CREDITS    = 20,
1615         OSD_TXN_INODE_DELETE_CREDITS = 20
1616 };
1617
1618 /*
1619  * Journal
1620  */
1621
1622 #if OSD_THANDLE_STATS
1623 /**
1624  * Set time when the handle is allocated
1625  */
1626 static void osd_th_alloced(struct osd_thandle *oth)
1627 {
1628         oth->oth_alloced = ktime_get();
1629 }
1630
1631 /**
1632  * Set time when the handle started
1633  */
1634 static void osd_th_started(struct osd_thandle *oth)
1635 {
1636         oth->oth_started = ktime_get();
1637 }
1638
1639 /**
1640  * Check whether the we deal with this handle for too long.
1641  */
1642 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1643                                 ktime_t alloced, ktime_t started,
1644                                 ktime_t closed)
1645 {
1646         ktime_t now = ktime_get();
1647
1648         LASSERT(dev != NULL);
1649
1650         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1651                             ktime_us_delta(started, alloced));
1652         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1653                             ktime_us_delta(closed, started));
1654         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1655                             ktime_us_delta(now, closed));
1656
1657         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1658                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1659                                 oth, now, alloced, started, closed);
1660                 libcfs_debug_dumpstack(NULL);
1661         }
1662 }
1663
1664 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1665 {                                                                       \
1666         ktime_t __closed = ktime_get();                                 \
1667         ktime_t __alloced = oth->oth_alloced;                           \
1668         ktime_t __started = oth->oth_started;                           \
1669                                                                         \
1670         expr;                                                           \
1671         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1672 }
1673
1674 #else /* OSD_THANDLE_STATS */
1675
1676 #define osd_th_alloced(h)                  do {} while(0)
1677 #define osd_th_started(h)                  do {} while(0)
1678 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1679
1680 #endif /* OSD_THANDLE_STATS */
1681
1682 /*
1683  * Concurrency: doesn't access mutable data.
1684  */
1685 static int osd_param_is_not_sane(const struct osd_device *dev,
1686                                  const struct thandle *th)
1687 {
1688         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1689
1690         return oh->ot_credits > osd_transaction_size(dev);
1691 }
1692
1693 /*
1694  * Concurrency: shouldn't matter.
1695  */
1696 static void osd_trans_commit_cb(struct super_block *sb,
1697                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1698 {
1699         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1700         struct thandle *th = &oh->ot_super;
1701         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1702         struct dt_txn_commit_cb *dcb, *tmp;
1703
1704         LASSERT(oh->ot_handle == NULL);
1705
1706         if (error)
1707                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1708
1709         dt_txn_hook_commit(th);
1710
1711         /* call per-transaction callbacks if any */
1712         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1713                                  dcb_linkage) {
1714                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1715                          "commit callback entry: magic=%x name='%s'\n",
1716                          dcb->dcb_magic, dcb->dcb_name);
1717                 list_del_init(&dcb->dcb_linkage);
1718                 dcb->dcb_func(NULL, th, dcb, error);
1719         }
1720
1721         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1722         lu_device_put(lud);
1723         th->th_dev = NULL;
1724
1725         OBD_FREE_PTR(oh);
1726 }
1727
1728 #ifndef HAVE_SB_START_WRITE
1729 # define sb_start_write(sb) do {} while (0)
1730 # define sb_end_write(sb) do {} while (0)
1731 #endif
1732
1733 static struct thandle *osd_trans_create(const struct lu_env *env,
1734                                         struct dt_device *d)
1735 {
1736         struct osd_thread_info *oti = osd_oti_get(env);
1737         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1738         struct osd_thandle *oh;
1739         struct thandle *th;
1740
1741         ENTRY;
1742
1743         if (d->dd_rdonly) {
1744                 CERROR("%s: someone try to start transaction under "
1745                        "readonly mode, should be disabled.\n",
1746                        osd_name(osd_dt_dev(d)));
1747                 dump_stack();
1748                 RETURN(ERR_PTR(-EROFS));
1749         }
1750
1751         /* on pending IO in this thread should left from prev. request */
1752         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1753
1754         sb_start_write(osd_sb(osd_dt_dev(d)));
1755
1756         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1757         if (!oh) {
1758                 sb_end_write(osd_sb(osd_dt_dev(d)));
1759                 RETURN(ERR_PTR(-ENOMEM));
1760         }
1761
1762         oh->ot_quota_trans = &oti->oti_quota_trans;
1763         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1764         th = &oh->ot_super;
1765         th->th_dev = d;
1766         th->th_result = 0;
1767         oh->ot_credits = 0;
1768         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1769         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1770         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1771         osd_th_alloced(oh);
1772
1773         memset(oti->oti_declare_ops, 0,
1774                sizeof(oti->oti_declare_ops));
1775         memset(oti->oti_declare_ops_cred, 0,
1776                sizeof(oti->oti_declare_ops_cred));
1777         memset(oti->oti_declare_ops_used, 0,
1778                sizeof(oti->oti_declare_ops_used));
1779
1780         oti->oti_ins_cache_depth++;
1781
1782         RETURN(th);
1783 }
1784
1785 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1786 {
1787         struct osd_thread_info *oti = osd_oti_get(env);
1788         struct osd_thandle *oh;
1789
1790         oh = container_of0(th, struct osd_thandle, ot_super);
1791         LASSERT(oh != NULL);
1792
1793         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1794               oti->oti_declare_ops[OSD_OT_CREATE],
1795               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1796               oti->oti_declare_ops_used[OSD_OT_CREATE],
1797               oti->oti_declare_ops[OSD_OT_DESTROY],
1798               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1799               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1800         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1801               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1802               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1803               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1805               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1806               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1807         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1808               oti->oti_declare_ops[OSD_OT_WRITE],
1809               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1810               oti->oti_declare_ops_used[OSD_OT_WRITE],
1811               oti->oti_declare_ops[OSD_OT_PUNCH],
1812               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1813               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1814               oti->oti_declare_ops[OSD_OT_QUOTA],
1815               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1816               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1817         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_INSERT],
1819               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1820               oti->oti_declare_ops_used[OSD_OT_INSERT],
1821               oti->oti_declare_ops[OSD_OT_DELETE],
1822               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1823               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1824         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1825               oti->oti_declare_ops[OSD_OT_REF_ADD],
1826               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1827               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops[OSD_OT_REF_DEL],
1829               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1830               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1831 }
1832
1833 /*
1834  * Concurrency: shouldn't matter.
1835  */
1836 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1837                            struct thandle *th)
1838 {
1839         struct osd_thread_info *oti = osd_oti_get(env);
1840         struct osd_device *dev = osd_dt_dev(d);
1841         handle_t *jh;
1842         struct osd_thandle *oh;
1843         int rc;
1844
1845         ENTRY;
1846
1847         LASSERT(current->journal_info == NULL);
1848
1849         oh = container_of0(th, struct osd_thandle, ot_super);
1850         LASSERT(oh != NULL);
1851         LASSERT(oh->ot_handle == NULL);
1852
1853         rc = dt_txn_hook_start(env, d, th);
1854         if (rc != 0)
1855                 GOTO(out, rc);
1856
1857         if (unlikely(osd_param_is_not_sane(dev, th))) {
1858                 static unsigned long last_printed;
1859                 static int last_credits;
1860
1861                 /*
1862                  * don't make noise on a tiny testing systems
1863                  * actual credits misuse will be caught anyway
1864                  */
1865                 if (last_credits != oh->ot_credits &&
1866                     time_after(jiffies, last_printed +
1867                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1868                     osd_transaction_size(dev) > 512) {
1869                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1870                               oh->ot_credits, osd_transaction_size(dev));
1871                         osd_trans_dump_creds(env, th);
1872                         libcfs_debug_dumpstack(NULL);
1873                         last_credits = oh->ot_credits;
1874                         last_printed = jiffies;
1875                 }
1876                 /*
1877                  * XXX Limit the credits to 'max_transaction_buffers', and
1878                  *     let the underlying filesystem to catch the error if
1879                  *     we really need so many credits.
1880                  *
1881                  *     This should be removed when we can calculate the
1882                  *     credits precisely.
1883                  */
1884                 oh->ot_credits = osd_transaction_size(dev);
1885         } else if (ldiskfs_track_declares_assert != 0) {
1886                 /*
1887                  * reserve few credits to prevent an assertion in JBD
1888                  * our debugging mechanism will be able to detected
1889                  * overuse. this can help to debug single-update
1890                  * transactions
1891                  */
1892                 oh->ot_credits += 10;
1893                 if (unlikely(osd_param_is_not_sane(dev, th)))
1894                         oh->ot_credits = osd_transaction_size(dev);
1895         }
1896
1897         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1898                 GOTO(out, rc = -EIO);
1899
1900         /*
1901          * XXX temporary stuff. Some abstraction layer should
1902          * be used.
1903          */
1904         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1905         osd_th_started(oh);
1906         if (!IS_ERR(jh)) {
1907                 oh->ot_handle = jh;
1908                 LASSERT(oti->oti_txns == 0);
1909
1910                 lu_device_get(&d->dd_lu_dev);
1911                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1912                               "osd-tx", th);
1913                 oti->oti_txns++;
1914                 rc = 0;
1915         } else {
1916                 rc = PTR_ERR(jh);
1917         }
1918 out:
1919         RETURN(rc);
1920 }
1921
1922 static int osd_seq_exists(const struct lu_env *env,
1923                           struct osd_device *osd, u64 seq)
1924 {
1925         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1926         struct seq_server_site *ss = osd_seq_site(osd);
1927         int rc;
1928
1929         ENTRY;
1930
1931         LASSERT(ss != NULL);
1932         LASSERT(ss->ss_server_fld != NULL);
1933
1934         rc = osd_fld_lookup(env, osd, seq, range);
1935         if (rc != 0) {
1936                 if (rc != -ENOENT)
1937                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1938                                osd_name(osd), seq, rc);
1939                 RETURN(0);
1940         }
1941
1942         RETURN(ss->ss_node_id == range->lsr_index);
1943 }
1944
1945 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1946 {
1947         struct dt_txn_commit_cb *dcb;
1948         struct dt_txn_commit_cb *tmp;
1949
1950         /* call per-transaction stop callbacks if any */
1951         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1952                                  dcb_linkage) {
1953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1954                          "commit callback entry: magic=%x name='%s'\n",
1955                          dcb->dcb_magic, dcb->dcb_name);
1956                 list_del_init(&dcb->dcb_linkage);
1957                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1958         }
1959 }
1960
1961 /*
1962  * Concurrency: shouldn't matter.
1963  */
1964 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1965                           struct thandle *th)
1966 {
1967         struct osd_thread_info *oti = osd_oti_get(env);
1968         struct osd_thandle *oh;
1969         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1970         struct osd_device *osd = osd_dt_dev(th->th_dev);
1971         struct qsd_instance *qsd = osd->od_quota_slave;
1972         struct lquota_trans *qtrans;
1973         struct list_head truncates = LIST_HEAD_INIT(truncates);
1974         int rc = 0, remove_agents = 0;
1975
1976         ENTRY;
1977
1978         oh = container_of0(th, struct osd_thandle, ot_super);
1979
1980         remove_agents = oh->ot_remove_agents;
1981
1982         qtrans = oh->ot_quota_trans;
1983         oh->ot_quota_trans = NULL;
1984
1985         /* move locks to local list, stop tx, execute truncates */
1986         list_splice(&oh->ot_trunc_locks, &truncates);
1987
1988         if (oh->ot_handle != NULL) {
1989                 int rc2;
1990
1991                 handle_t *hdl = oh->ot_handle;
1992
1993                 /*
1994                  * add commit callback
1995                  * notice we don't do this in osd_trans_start()
1996                  * as underlying transaction can change during truncate
1997                  */
1998                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1999                                              &oh->ot_jcb);
2000
2001                 LASSERT(oti->oti_txns == 1);
2002                 oti->oti_txns--;
2003
2004                 rc = dt_txn_hook_stop(env, th);
2005                 if (rc != 0)
2006                         CERROR("%s: failed in transaction hook: rc = %d\n",
2007                                osd_name(osd), rc);
2008
2009                 osd_trans_stop_cb(oh, rc);
2010                 /* hook functions might modify th_sync */
2011                 hdl->h_sync = th->th_sync;
2012
2013                 oh->ot_handle = NULL;
2014                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2015                 if (rc2 != 0)
2016                         CERROR("%s: failed to stop transaction: rc = %d\n",
2017                                osd_name(osd), rc2);
2018                 if (!rc)
2019                         rc = rc2;
2020
2021                 osd_process_truncates(&truncates);
2022         } else {
2023                 osd_trans_stop_cb(oh, th->th_result);
2024                 OBD_FREE_PTR(oh);
2025         }
2026
2027         osd_trunc_unlock_all(&truncates);
2028
2029         /* inform the quota slave device that the transaction is stopping */
2030         qsd_op_end(env, qsd, qtrans);
2031
2032         /*
2033          * as we want IO to journal and data IO be concurrent, we don't block
2034          * awaiting data IO completion in osd_do_bio(), instead we wait here
2035          * once transaction is submitted to the journal. all reqular requests
2036          * don't do direct IO (except read/write), thus this wait_event becomes
2037          * no-op for them.
2038          *
2039          * IMPORTANT: we have to wait till any IO submited by the thread is
2040          * completed otherwise iobuf may be corrupted by different request
2041          */
2042         wait_event(iobuf->dr_wait,
2043                        atomic_read(&iobuf->dr_numreqs) == 0);
2044         osd_fini_iobuf(osd, iobuf);
2045         if (!rc)
2046                 rc = iobuf->dr_error;
2047
2048         if (unlikely(remove_agents != 0))
2049                 osd_process_scheduled_agent_removals(env, osd);
2050
2051         oti->oti_ins_cache_depth--;
2052         /* reset OI cache for safety */
2053         if (oti->oti_ins_cache_depth == 0)
2054                 oti->oti_ins_cache_used = 0;
2055
2056         sb_end_write(osd_sb(osd));
2057
2058         RETURN(rc);
2059 }
2060
2061 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2062 {
2063         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2064                                                ot_super);
2065
2066         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2067         LASSERT(&dcb->dcb_func != NULL);
2068         if (dcb->dcb_flags & DCB_TRANS_STOP)
2069                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2070         else
2071                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2072
2073         return 0;
2074 }
2075
2076 /*
2077  * Called just before object is freed. Releases all resources except for
2078  * object itself (that is released by osd_object_free()).
2079  *
2080  * Concurrency: no concurrent access is possible that late in object
2081  * life-cycle.
2082  */
2083 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2084 {
2085         struct osd_object *obj = osd_obj(l);
2086         struct inode *inode = obj->oo_inode;
2087
2088         LINVRNT(osd_invariant(obj));
2089
2090         /*
2091          * If object is unlinked remove fid->ino mapping from object index.
2092          */
2093
2094         osd_index_fini(obj);
2095         if (inode != NULL) {
2096                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
2097                 qid_t                    uid = i_uid_read(inode);
2098                 qid_t                    gid = i_gid_read(inode);
2099
2100                 obj->oo_inode = NULL;
2101                 iput(inode);
2102                 if (!obj->oo_header && qsd) {
2103                         struct osd_thread_info *info = osd_oti_get(env);
2104                         struct lquota_id_info *qi = &info->oti_qi;
2105
2106                         /* Release granted quota to master if necessary */
2107                         qi->lqi_id.qid_uid = uid;
2108                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2109
2110                         qi->lqi_id.qid_uid = gid;
2111                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2112
2113                         qi->lqi_id.qid_uid = i_projid_read(inode);
2114                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2115                 }
2116         }
2117 }
2118
2119 /*
2120  * Concurrency: ->loo_object_release() is called under site spin-lock.
2121  */
2122 static void osd_object_release(const struct lu_env *env,
2123                                struct lu_object *l)
2124 {
2125         struct osd_object *o = osd_obj(l);
2126
2127         /*
2128          * nobody should be releasing a non-destroyed object with nlink=0
2129          * the API allows this, but ldiskfs doesn't like and then report
2130          * this inode as deleted
2131          */
2132         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2133                   o->oo_inode->i_nlink == 0));
2134 }
2135
2136 /*
2137  * Concurrency: shouldn't matter.
2138  */
2139 static int osd_object_print(const struct lu_env *env, void *cookie,
2140                             lu_printer_t p, const struct lu_object *l)
2141 {
2142         struct osd_object *o = osd_obj(l);
2143         struct iam_descr *d;
2144
2145         if (o->oo_dir != NULL)
2146                 d = o->oo_dir->od_container.ic_descr;
2147         else
2148                 d = NULL;
2149         return (*p)(env, cookie,
2150                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2151                     o, o->oo_inode,
2152                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2153                     o->oo_inode ? o->oo_inode->i_generation : 0,
2154                     d ? d->id_ops->id_name : "plain");
2155 }
2156
2157 /*
2158  * Concurrency: shouldn't matter.
2159  */
2160 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2161                 struct obd_statfs *sfs)
2162 {
2163         struct osd_device *osd = osd_dt_dev(d);
2164         struct super_block *sb = osd_sb(osd);
2165         struct kstatfs *ksfs;
2166         __u64 reserved;
2167         int result = 0;
2168
2169         if (unlikely(osd->od_mnt == NULL))
2170                 return -EINPROGRESS;
2171
2172         /* osd_lproc.c call this without env, allocate ksfs for that case */
2173         if (unlikely(env == NULL)) {
2174                 OBD_ALLOC_PTR(ksfs);
2175                 if (ksfs == NULL)
2176                         return -ENOMEM;
2177         } else {
2178                 ksfs = &osd_oti_get(env)->oti_ksfs;
2179         }
2180
2181         result = sb->s_op->statfs(sb->s_root, ksfs);
2182         if (result)
2183                 goto out;
2184
2185         statfs_pack(sfs, ksfs);
2186         if (unlikely(sb->s_flags & MS_RDONLY))
2187                 sfs->os_state |= OS_STATE_READONLY;
2188         if (ldiskfs_has_feature_extents(sb))
2189                 sfs->os_maxbytes = sb->s_maxbytes;
2190         else
2191                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2192
2193         /*
2194          * Reserve some space so to avoid fragmenting the filesystem too much.
2195          * Fragmentation not only impacts performance, but can also increase
2196          * metadata overhead significantly, causing grant calculation to be
2197          * wrong.
2198          *
2199          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2200          */
2201         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2202         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2203         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2204                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2205
2206         sfs->os_blocks -= reserved;
2207         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2208         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2209
2210 out:
2211         if (unlikely(env == NULL))
2212                 OBD_FREE_PTR(ksfs);
2213         return result;
2214 }
2215
2216 /**
2217  * Estimate space needed for file creations. We assume the largest filename
2218  * which is 2^64 - 1, hence a filename of 20 chars.
2219  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2220  */
2221 #ifdef __LDISKFS_DIR_REC_LEN
2222 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2223 #else
2224 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2225 #endif
2226
2227 /*
2228  * Concurrency: doesn't access mutable data.
2229  */
2230 static void osd_conf_get(const struct lu_env *env,
2231                          const struct dt_device *dev,
2232                          struct dt_device_param *param)
2233 {
2234         struct osd_device *d = osd_dt_dev(dev);
2235         struct super_block *sb = osd_sb(d);
2236         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2237         const char *name;
2238         int ea_overhead;
2239
2240         /*
2241          * XXX should be taken from not-yet-existing fs abstraction layer.
2242          */
2243         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2244         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2245         param->ddp_symlink_max  = sb->s_blocksize;
2246         param->ddp_mount_type   = LDD_MT_LDISKFS;
2247         if (ldiskfs_has_feature_extents(sb))
2248                 param->ddp_maxbytes = sb->s_maxbytes;
2249         else
2250                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2251         /*
2252          * inode are statically allocated, so per-inode space consumption
2253          * is the space consumed by the directory entry
2254          */
2255         param->ddp_inodespace     = PER_OBJ_USAGE;
2256         /*
2257          * EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2258          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2259          * maximum length to not under count the actual number of extents
2260          * needed for writing a file.
2261          */
2262         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2263         /* worst-case extent insertion metadata overhead */
2264         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2265         param->ddp_mntopts = 0;
2266         if (test_opt(sb, XATTR_USER))
2267                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2268         if (test_opt(sb, POSIX_ACL))
2269                 param->ddp_mntopts |= MNTOPT_ACL;
2270
2271         /*
2272          * LOD might calculate the max stripe count based on max_ea_size,
2273          * so we need take account in the overhead as well,
2274          * xattr_header + magic + xattr_entry_head
2275          */
2276         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2277                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2278
2279 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2280         if (ldiskfs_has_feature_ea_inode(sb))
2281                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2282                                                                 ea_overhead;
2283         else
2284 #endif
2285                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2286
2287         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2288                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2289
2290         /*
2291          * Preferred RPC size for efficient disk IO.  4MB shows good
2292          * all-around performance for ldiskfs, but use bigalloc chunk size
2293          * by default if larger.
2294          */
2295 #if defined(LDISKFS_CLUSTER_SIZE)
2296         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2297                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2298         else
2299 #endif
2300                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2301
2302         param->ddp_t10_cksum_type = 0;
2303         if (bi) {
2304                 unsigned short interval = blk_integrity_interval(bi);
2305                 name = blk_integrity_name(bi);
2306                 /*
2307                  * Expected values:
2308                  * T10-DIF-TYPE1-CRC
2309                  * T10-DIF-TYPE3-CRC
2310                  * T10-DIF-TYPE1-IP
2311                  * T10-DIF-TYPE3-IP
2312                  */
2313                 if (strncmp(name, "T10-DIF-TYPE",
2314                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2315                         /* also skip "1/3-" at end */
2316                         const int type_off = sizeof("T10-DIF-TYPE.");
2317
2318                         if (interval != 512 && interval != 4096)
2319                                 CERROR("%s: unsupported T10PI sector size %u\n",
2320                                        d->od_svname, interval);
2321                         else if (strcmp(name + type_off, "CRC") == 0)
2322                                 param->ddp_t10_cksum_type = interval == 512 ?
2323                                         OBD_CKSUM_T10CRC512 :
2324                                         OBD_CKSUM_T10CRC4K;
2325                         else if (strcmp(name + type_off, "IP") == 0)
2326                                 param->ddp_t10_cksum_type = interval == 512 ?
2327                                         OBD_CKSUM_T10IP512 :
2328                                         OBD_CKSUM_T10IP4K;
2329                         else
2330                                 CERROR("%s: unsupported checksum type of "
2331                                        "T10PI type '%s'",
2332                                        d->od_svname, name);
2333                 } else {
2334                         CERROR("%s: unsupported T10PI type '%s'",
2335                                d->od_svname, name);
2336                 }
2337         }
2338 }
2339
2340 /*
2341  * Concurrency: shouldn't matter.
2342  */
2343 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2344 {
2345         int rc;
2346         struct super_block *s = osd_sb(osd_dt_dev(d));
2347         ENTRY;
2348
2349         down_read(&s->s_umount);
2350         rc = s->s_op->sync_fs(s, 1);
2351         up_read(&s->s_umount);
2352
2353         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2354                rc);
2355
2356         return rc;
2357 }
2358
2359 /**
2360  * Start commit for OSD device.
2361  *
2362  * An implementation of dt_commit_async method for OSD device.
2363  * Asychronously starts underlayng fs sync and thereby a transaction
2364  * commit.
2365  *
2366  * \param env environment
2367  * \param d dt device
2368  *
2369  * \see dt_device_operations
2370  */
2371 static int osd_commit_async(const struct lu_env *env,
2372                             struct dt_device *d)
2373 {
2374         struct super_block *s = osd_sb(osd_dt_dev(d));
2375         int rc;
2376
2377         ENTRY;
2378
2379         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2380         down_read(&s->s_umount);
2381         rc = s->s_op->sync_fs(s, 0);
2382         up_read(&s->s_umount);
2383
2384         RETURN(rc);
2385 }
2386
2387 /* Our own copy of the set readonly functions if present, or NU if not. */
2388 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2389 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2390 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2391
2392 /*
2393  * Concurrency: shouldn't matter.
2394  */
2395 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2396 {
2397         struct super_block *sb = osd_sb(osd_dt_dev(d));
2398         struct block_device *dev = sb->s_bdev;
2399         int rc = -EOPNOTSUPP;
2400
2401         ENTRY;
2402
2403         if (priv_dev_set_rdonly) {
2404                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2405
2406                 rc = 0;
2407                 CERROR("*** setting %s read-only ***\n",
2408                        osd_dt_dev(d)->od_svname);
2409
2410                 if (sb->s_op->freeze_fs) {
2411                         rc = sb->s_op->freeze_fs(sb);
2412                         if (rc)
2413                                 goto out;
2414                 }
2415
2416                 if (jdev && (jdev != dev)) {
2417                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2418                                (long)jdev);
2419                         priv_dev_set_rdonly(jdev);
2420                 }
2421                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2422                 priv_dev_set_rdonly(dev);
2423
2424                 if (sb->s_op->unfreeze_fs)
2425                         sb->s_op->unfreeze_fs(sb);
2426         }
2427
2428 out:
2429         if (rc)
2430                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2431                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2432
2433         RETURN(rc);
2434 }
2435
2436 /**
2437  * Note: we do not count into QUOTA here.
2438  * If we mount with --data_journal we may need more.
2439  */
2440 const int osd_dto_credits_noquota[DTO_NR] = {
2441         /**
2442          * Insert.
2443          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2444          * SINGLEDATA_TRANS_BLOCKS(8)
2445          * XXX Note: maybe iam need more, since iam have more level than
2446          *           EXT3 htree.
2447          */
2448         [DTO_INDEX_INSERT]  = 16,
2449         /**
2450          * Delete
2451          * just modify a single entry, probably merge few within a block
2452          */
2453         [DTO_INDEX_DELETE]  = 1,
2454         /**
2455          * Used for OI scrub
2456          */
2457         [DTO_INDEX_UPDATE]  = 16,
2458         /**
2459          * 4(inode, inode bits, groups, GDT)
2460          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2461          */
2462         [DTO_OBJECT_CREATE] = 4,
2463         /**
2464          * 4(inode, inode bits, groups, GDT)
2465          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2466          */
2467         [DTO_OBJECT_DELETE] = 4,
2468         /**
2469          * Attr set credits (inode)
2470          */
2471         [DTO_ATTR_SET_BASE] = 1,
2472         /**
2473          * Xattr set. The same as xattr of EXT3.
2474          * DATA_TRANS_BLOCKS(14)
2475          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2476          * are also counted in. Do not know why?
2477          */
2478         [DTO_XATTR_SET]     = 14,
2479         /**
2480          * credits for inode change during write.
2481          */
2482         [DTO_WRITE_BASE]    = 3,
2483         /**
2484          * credits for single block write.
2485          */
2486         [DTO_WRITE_BLOCK]   = 14,
2487         /**
2488          * Attr set credits for chown.
2489          * This is extra credits for setattr, and it is null without quota
2490          */
2491         [DTO_ATTR_SET_CHOWN] = 0
2492 };
2493
2494 static const struct dt_device_operations osd_dt_ops = {
2495         .dt_root_get       = osd_root_get,
2496         .dt_statfs         = osd_statfs,
2497         .dt_trans_create   = osd_trans_create,
2498         .dt_trans_start    = osd_trans_start,
2499         .dt_trans_stop     = osd_trans_stop,
2500         .dt_trans_cb_add   = osd_trans_cb_add,
2501         .dt_conf_get       = osd_conf_get,
2502         .dt_sync           = osd_sync,
2503         .dt_ro             = osd_ro,
2504         .dt_commit_async   = osd_commit_async,
2505 };
2506
2507 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2508                           unsigned int role)
2509 {
2510         struct osd_object *obj = osd_dt_obj(dt);
2511         struct osd_thread_info *oti = osd_oti_get(env);
2512
2513         LINVRNT(osd_invariant(obj));
2514
2515         LASSERT(obj->oo_owner != env);
2516         down_read_nested(&obj->oo_sem, role);
2517
2518         LASSERT(obj->oo_owner == NULL);
2519         oti->oti_r_locks++;
2520 }
2521
2522 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2523                            unsigned int role)
2524 {
2525         struct osd_object *obj = osd_dt_obj(dt);
2526         struct osd_thread_info *oti = osd_oti_get(env);
2527
2528         LINVRNT(osd_invariant(obj));
2529
2530         LASSERT(obj->oo_owner != env);
2531         down_write_nested(&obj->oo_sem, role);
2532
2533         LASSERT(obj->oo_owner == NULL);
2534         obj->oo_owner = env;
2535         oti->oti_w_locks++;
2536 }
2537
2538 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2539 {
2540         struct osd_object *obj = osd_dt_obj(dt);
2541         struct osd_thread_info *oti = osd_oti_get(env);
2542
2543         LINVRNT(osd_invariant(obj));
2544
2545         LASSERT(oti->oti_r_locks > 0);
2546         oti->oti_r_locks--;
2547         up_read(&obj->oo_sem);
2548 }
2549
2550 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2551 {
2552         struct osd_object *obj = osd_dt_obj(dt);
2553         struct osd_thread_info *oti = osd_oti_get(env);
2554
2555         LINVRNT(osd_invariant(obj));
2556
2557         LASSERT(obj->oo_owner == env);
2558         LASSERT(oti->oti_w_locks > 0);
2559         oti->oti_w_locks--;
2560         obj->oo_owner = NULL;
2561         up_write(&obj->oo_sem);
2562 }
2563
2564 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2565 {
2566         struct osd_object *obj = osd_dt_obj(dt);
2567
2568         LINVRNT(osd_invariant(obj));
2569
2570         return obj->oo_owner == env;
2571 }
2572
2573 static struct timespec *osd_inode_time(const struct lu_env *env,
2574                                        struct inode *inode, __u64 seconds)
2575 {
2576         struct osd_thread_info *oti = osd_oti_get(env);
2577         struct timespec *t = &oti->oti_time;
2578
2579         t->tv_sec = seconds;
2580         t->tv_nsec = 0;
2581         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2582         return t;
2583 }
2584
2585 static void osd_inode_getattr(const struct lu_env *env,
2586                               struct inode *inode, struct lu_attr *attr)
2587 {
2588         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2589                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2590                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2591                            LA_BLKSIZE | LA_TYPE;
2592
2593         attr->la_atime   = LTIME_S(inode->i_atime);
2594         attr->la_mtime   = LTIME_S(inode->i_mtime);
2595         attr->la_ctime   = LTIME_S(inode->i_ctime);
2596         attr->la_mode    = inode->i_mode;
2597         attr->la_size    = i_size_read(inode);
2598         attr->la_blocks  = inode->i_blocks;
2599         attr->la_uid     = i_uid_read(inode);
2600         attr->la_gid     = i_gid_read(inode);
2601         attr->la_projid  = i_projid_read(inode);
2602         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2603         attr->la_nlink   = inode->i_nlink;
2604         attr->la_rdev    = inode->i_rdev;
2605         attr->la_blksize = 1 << inode->i_blkbits;
2606         attr->la_blkbits = inode->i_blkbits;
2607         /*
2608          * Ext4 did not transfer inherit flags from raw inode
2609          * to inode flags, and ext4 internally test raw inode
2610          * @i_flags directly. Instead of patching ext4, we do it here.
2611          */
2612         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2613                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2614 }
2615
2616 static int osd_attr_get(const struct lu_env *env,
2617                         struct dt_object *dt,
2618                         struct lu_attr *attr)
2619 {
2620         struct osd_object *obj = osd_dt_obj(dt);
2621
2622         if (unlikely(!dt_object_exists(dt)))
2623                 return -ENOENT;
2624         if (unlikely(obj->oo_destroyed))
2625                 return -ENOENT;
2626
2627         LASSERT(!dt_object_remote(dt));
2628         LINVRNT(osd_invariant(obj));
2629
2630         spin_lock(&obj->oo_guard);
2631         osd_inode_getattr(env, obj->oo_inode, attr);
2632         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2633                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2634         spin_unlock(&obj->oo_guard);
2635
2636         return 0;
2637 }
2638
2639 static int osd_declare_attr_qid(const struct lu_env *env,
2640                                 struct osd_object *obj,
2641                                 struct osd_thandle *oh, long long bspace,
2642                                 qid_t old_id, qid_t new_id, bool enforce,
2643                                 unsigned int type, bool ignore_edquot)
2644 {
2645         int rc;
2646         struct osd_thread_info *info = osd_oti_get(env);
2647         struct lquota_id_info  *qi = &info->oti_qi;
2648
2649         qi->lqi_type = type;
2650         /* inode accounting */
2651         qi->lqi_is_blk = false;
2652
2653         /* one more inode for the new id ... */
2654         qi->lqi_id.qid_uid = new_id;
2655         qi->lqi_space      = 1;
2656         /* Reserve credits for the new id */
2657         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2658         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2659                 rc = 0;
2660         if (rc)
2661                 RETURN(rc);
2662
2663         /* and one less inode for the current id */
2664         qi->lqi_id.qid_uid = old_id;
2665         qi->lqi_space = -1;
2666         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2667         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2668                 rc = 0;
2669         if (rc)
2670                 RETURN(rc);
2671
2672         /* block accounting */
2673         qi->lqi_is_blk = true;
2674
2675         /* more blocks for the new id ... */
2676         qi->lqi_id.qid_uid = new_id;
2677         qi->lqi_space      = bspace;
2678         /*
2679          * Credits for the new uid has been reserved, re-use "obj"
2680          * to save credit reservation.
2681          */
2682         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2683         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2684                 rc = 0;
2685         if (rc)
2686                 RETURN(rc);
2687
2688         /* and finally less blocks for the current uid */
2689         qi->lqi_id.qid_uid = old_id;
2690         qi->lqi_space      = -bspace;
2691         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2692         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2693                 rc = 0;
2694
2695         RETURN(rc);
2696 }
2697
2698 static int osd_declare_attr_set(const struct lu_env *env,
2699                                 struct dt_object *dt,
2700                                 const struct lu_attr *attr,
2701                                 struct thandle *handle)
2702 {
2703         struct osd_thandle *oh;
2704         struct osd_object *obj;
2705         qid_t uid;
2706         qid_t gid;
2707         long long bspace;
2708         int rc = 0;
2709         bool enforce;
2710
2711         ENTRY;
2712
2713         LASSERT(dt != NULL);
2714         LASSERT(handle != NULL);
2715
2716         obj = osd_dt_obj(dt);
2717         LASSERT(osd_invariant(obj));
2718
2719         oh = container_of0(handle, struct osd_thandle, ot_super);
2720         LASSERT(oh->ot_handle == NULL);
2721
2722         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2723                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2724
2725         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2726                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2727
2728         if (attr == NULL || obj->oo_inode == NULL)
2729                 RETURN(rc);
2730
2731         bspace   = obj->oo_inode->i_blocks << 9;
2732         bspace   = toqb(bspace);
2733
2734         /*
2735          * Changing ownership is always preformed by super user, it should not
2736          * fail with EDQUOT unless required explicitly.
2737          *
2738          * We still need to call the osd_declare_qid() to calculate the journal
2739          * credits for updating quota accounting files and to trigger quota
2740          * space adjustment once the operation is completed.
2741          */
2742         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2743                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2744
2745                 if (!ignore_edquot)
2746                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2747                                "(the quota space is %lld)\n",
2748                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2749                                attr->la_gid, bspace);
2750
2751                 /* USERQUOTA */
2752                 uid = i_uid_read(obj->oo_inode);
2753                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2754                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2755                                           attr->la_uid, enforce, USRQUOTA,
2756                                           true);
2757                 if (rc)
2758                         RETURN(rc);
2759
2760                 gid = i_gid_read(obj->oo_inode);
2761                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2762                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2763                                           i_gid_read(obj->oo_inode),
2764                                           attr->la_gid, enforce, GRPQUOTA,
2765                                           ignore_edquot);
2766                 if (rc)
2767                         RETURN(rc);
2768
2769         }
2770 #ifdef HAVE_PROJECT_QUOTA
2771         if (attr->la_valid & LA_PROJID) {
2772                 __u32 projid = i_projid_read(obj->oo_inode);
2773
2774                 enforce = (attr->la_valid & LA_PROJID) &&
2775                                         (attr->la_projid != projid);
2776                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2777                                           (qid_t)projid, (qid_t)attr->la_projid,
2778                                           enforce, PRJQUOTA, true);
2779                 if (rc)
2780                         RETURN(rc);
2781         }
2782 #endif
2783         RETURN(rc);
2784 }
2785
2786 static int osd_inode_setattr(const struct lu_env *env,
2787                              struct inode *inode, const struct lu_attr *attr)
2788 {
2789         __u64 bits = attr->la_valid;
2790
2791         /* Only allow set size for regular file */
2792         if (!S_ISREG(inode->i_mode))
2793                 bits &= ~(LA_SIZE | LA_BLOCKS);
2794
2795         if (bits == 0)
2796                 return 0;
2797
2798         if (bits & LA_ATIME)
2799                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2800         if (bits & LA_CTIME)
2801                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2802         if (bits & LA_MTIME)
2803                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2804         if (bits & LA_SIZE) {
2805                 spin_lock(&inode->i_lock);
2806                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2807                 i_size_write(inode, attr->la_size);
2808                 spin_unlock(&inode->i_lock);
2809         }
2810
2811         /*
2812          * OSD should not change "i_blocks" which is used by quota.
2813          * "i_blocks" should be changed by ldiskfs only.
2814          */
2815         if (bits & LA_MODE)
2816                 inode->i_mode = (inode->i_mode & S_IFMT) |
2817                                 (attr->la_mode & ~S_IFMT);
2818         if (bits & LA_UID)
2819                 i_uid_write(inode, attr->la_uid);
2820         if (bits & LA_GID)
2821                 i_gid_write(inode, attr->la_gid);
2822         if (bits & LA_PROJID)
2823                 i_projid_write(inode, attr->la_projid);
2824         if (bits & LA_NLINK)
2825                 set_nlink(inode, attr->la_nlink);
2826         if (bits & LA_RDEV)
2827                 inode->i_rdev = attr->la_rdev;
2828
2829         if (bits & LA_FLAGS) {
2830                 /* always keep S_NOCMTIME */
2831                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2832                                  S_NOCMTIME;
2833                 /*
2834                  * Ext4 did not transfer inherit flags from
2835                  * @inode->i_flags to raw inode i_flags when writing
2836                  * flags, we do it explictly here.
2837                  */
2838                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2839                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2840                 else
2841                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2842         }
2843         return 0;
2844 }
2845
2846 #ifdef HAVE_PROJECT_QUOTA
2847 static int osd_transfer_project(struct inode *inode, __u32 projid)
2848 {
2849         struct super_block *sb = inode->i_sb;
2850         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2851         int err;
2852         kprojid_t kprojid;
2853         struct ldiskfs_iloc iloc;
2854         struct ldiskfs_inode *raw_inode;
2855         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2856
2857         if (!ldiskfs_has_feature_project(sb)) {
2858                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2859                         == LDISKFS_DEF_PROJID);
2860                 if (projid != LDISKFS_DEF_PROJID)
2861                         return -EOPNOTSUPP;
2862                 else
2863                         return 0;
2864         }
2865
2866         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2867                 return -EOPNOTSUPP;
2868
2869         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2870         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2871                 return 0;
2872
2873         err = ldiskfs_get_inode_loc(inode, &iloc);
2874         if (err)
2875                 return err;
2876
2877         raw_inode = ldiskfs_raw_inode(&iloc);
2878         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2879                 err = -EOVERFLOW;
2880                 brelse(iloc.bh);
2881                 return err;
2882         }
2883         brelse(iloc.bh);
2884
2885         dquot_initialize(inode);
2886         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2887         if (transfer_to[PRJQUOTA]) {
2888                 err = __dquot_transfer(inode, transfer_to);
2889                 dqput(transfer_to[PRJQUOTA]);
2890                 if (err)
2891                         return err;
2892         }
2893
2894         return err;
2895 }
2896 #endif
2897
2898 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2899 {
2900         int rc;
2901
2902         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2903             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2904                 struct iattr iattr;
2905
2906                 ll_vfs_dq_init(inode);
2907                 iattr.ia_valid = 0;
2908                 if (attr->la_valid & LA_UID)
2909                         iattr.ia_valid |= ATTR_UID;
2910                 if (attr->la_valid & LA_GID)
2911                         iattr.ia_valid |= ATTR_GID;
2912                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2913                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2914
2915                 rc = ll_vfs_dq_transfer(inode, &iattr);
2916                 if (rc) {
2917                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2918                                "enforcement enabled on the ldiskfs "
2919                                "filesystem?\n", inode->i_sb->s_id, rc);
2920                         return rc;
2921                 }
2922         }
2923
2924         /* Handle project id transfer here properly */
2925         if (attr->la_valid & LA_PROJID &&
2926             attr->la_projid != i_projid_read(inode)) {
2927 #ifdef HAVE_PROJECT_QUOTA
2928                 rc = osd_transfer_project(inode, attr->la_projid);
2929 #else
2930                 rc = -ENOTSUPP;
2931 #endif
2932                 if (rc) {
2933                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2934                                "enforcement enabled on the ldiskfs "
2935                                "filesystem?\n", inode->i_sb->s_id, rc);
2936                         return rc;
2937                 }
2938         }
2939         return 0;
2940 }
2941
2942 static int osd_attr_set(const struct lu_env *env,
2943                         struct dt_object *dt,
2944                         const struct lu_attr *attr,
2945                         struct thandle *handle)
2946 {
2947         struct osd_object *obj = osd_dt_obj(dt);
2948         struct inode *inode;
2949         int rc;
2950
2951         if (!dt_object_exists(dt))
2952                 return -ENOENT;
2953
2954         LASSERT(handle != NULL);
2955         LASSERT(!dt_object_remote(dt));
2956         LASSERT(osd_invariant(obj));
2957
2958         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2959
2960         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2961             !osd_obj2dev(obj)->od_is_ost) {
2962                 struct osd_thread_info *oti = osd_oti_get(env);
2963                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2964                 struct lu_fid *fid1 = &oti->oti_fid;
2965                 struct osd_inode_id *id = &oti->oti_id;
2966                 struct iam_path_descr *ipd;
2967                 struct iam_container *bag;
2968                 struct osd_thandle *oh;
2969                 int rc;
2970
2971                 fid_cpu_to_be(fid1, fid0);
2972                 memset(id, 1, sizeof(*id));
2973                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2974                                   fid0)->oi_dir.od_container;
2975                 ipd = osd_idx_ipd_get(env, bag);
2976                 if (unlikely(ipd == NULL))
2977                         RETURN(-ENOMEM);
2978
2979                 oh = container_of0(handle, struct osd_thandle, ot_super);
2980                 rc = iam_update(oh->ot_handle, bag,
2981                                 (const struct iam_key *)fid1,
2982                                 (const struct iam_rec *)id, ipd);
2983                 osd_ipd_put(env, bag, ipd);
2984                 return(rc > 0 ? 0 : rc);
2985         }
2986
2987         inode = obj->oo_inode;
2988
2989         rc = osd_quota_transfer(inode, attr);
2990         if (rc)
2991                 return rc;
2992
2993         spin_lock(&obj->oo_guard);
2994         rc = osd_inode_setattr(env, inode, attr);
2995         spin_unlock(&obj->oo_guard);
2996         if (rc != 0)
2997                 GOTO(out, rc);
2998
2999         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3000
3001         if (!(attr->la_valid & LA_FLAGS))
3002                 GOTO(out, rc);
3003
3004         /* Let's check if there are extra flags need to be set into LMA */
3005         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3006                 struct osd_thread_info *info = osd_oti_get(env);
3007                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3008
3009                 LASSERT(!obj->oo_pfid_in_lma);
3010
3011                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3012                                  &info->oti_ost_attrs);
3013                 if (rc)
3014                         GOTO(out, rc);
3015
3016                 lma->lma_incompat |=
3017                         lustre_to_lma_flags(attr->la_flags);
3018                 lustre_lma_swab(lma);
3019                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3020                                      lma, sizeof(*lma), XATTR_REPLACE);
3021                 if (rc != 0) {
3022                         struct osd_device *osd = osd_obj2dev(obj);
3023
3024                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3025                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3026                               lma->lma_incompat, rc);
3027                 } else {
3028                         obj->oo_lma_flags =
3029                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3030                 }
3031                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3032         }
3033 out:
3034         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3035
3036         return rc;
3037 }
3038
3039 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3040                                            struct osd_object *obj,
3041                                            const char *name, const int namelen)
3042 {
3043         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3044 }
3045
3046 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3047                       umode_t mode, struct dt_allocation_hint *hint,
3048                       struct thandle *th)
3049 {
3050         int result;
3051         struct osd_device *osd = osd_obj2dev(obj);
3052         struct osd_thandle *oth;
3053         struct dt_object *parent = NULL;
3054         struct inode *inode;
3055
3056         LINVRNT(osd_invariant(obj));
3057         LASSERT(obj->oo_inode == NULL);
3058         LASSERT(obj->oo_hl_head == NULL);
3059
3060         if (S_ISDIR(mode) && ldiskfs_pdo) {
3061                 obj->oo_hl_head =
3062                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3063                 if (obj->oo_hl_head == NULL)
3064                         return -ENOMEM;
3065         }
3066
3067         oth = container_of(th, struct osd_thandle, ot_super);
3068         LASSERT(oth->ot_handle->h_transaction != NULL);
3069
3070         if (hint != NULL && hint->dah_parent != NULL &&
3071             !dt_object_remote(hint->dah_parent))
3072                 parent = hint->dah_parent;
3073
3074         inode = ldiskfs_create_inode(oth->ot_handle,
3075                                      parent ? osd_dt_obj(parent)->oo_inode :
3076                                               osd_sb(osd)->s_root->d_inode,
3077                                      mode);
3078         if (!IS_ERR(inode)) {
3079                 /* Do not update file c/mtime in ldiskfs. */
3080                 inode->i_flags |= S_NOCMTIME;
3081
3082                 /*
3083                  * For new created object, it must be consistent,
3084                  * and it is unnecessary to scrub against it.
3085                  */
3086                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3087
3088                 obj->oo_inode = inode;
3089                 result = 0;
3090         } else {
3091                 if (obj->oo_hl_head != NULL) {
3092                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3093                         obj->oo_hl_head = NULL;
3094                 }
3095                 result = PTR_ERR(inode);
3096         }
3097         LINVRNT(osd_invariant(obj));
3098         return result;
3099 }
3100
3101 enum {
3102         OSD_NAME_LEN = 255
3103 };
3104
3105 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3106                      struct lu_attr *attr,
3107                      struct dt_allocation_hint *hint,
3108                      struct dt_object_format *dof,
3109                      struct thandle *th)
3110 {
3111         int result;
3112         struct osd_thandle *oth;
3113         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3114
3115         LASSERT(S_ISDIR(attr->la_mode));
3116
3117         oth = container_of(th, struct osd_thandle, ot_super);
3118         LASSERT(oth->ot_handle->h_transaction != NULL);
3119         result = osd_mkfile(info, obj, mode, hint, th);
3120
3121         return result;
3122 }
3123
3124 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3125                         struct lu_attr *attr,
3126                         struct dt_allocation_hint *hint,
3127                         struct dt_object_format *dof,
3128                         struct thandle *th)
3129 {
3130         int result;
3131         struct osd_thandle *oth;
3132         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3133
3134         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3135
3136         LASSERT(S_ISREG(attr->la_mode));
3137
3138         oth = container_of(th, struct osd_thandle, ot_super);
3139         LASSERT(oth->ot_handle->h_transaction != NULL);
3140
3141         result = osd_mkfile(info, obj, mode, hint, th);
3142         if (result == 0) {
3143                 LASSERT(obj->oo_inode != NULL);
3144                 if (feat->dif_flags & DT_IND_VARKEY)
3145                         result = iam_lvar_create(obj->oo_inode,
3146                                                  feat->dif_keysize_max,
3147                                                  feat->dif_ptrsize,
3148                                                  feat->dif_recsize_max,
3149                                                  oth->ot_handle);
3150                 else
3151                         result = iam_lfix_create(obj->oo_inode,
3152                                                  feat->dif_keysize_max,
3153                                                  feat->dif_ptrsize,
3154                                                  feat->dif_recsize_max,
3155                                                  oth->ot_handle);
3156         }
3157         return result;
3158 }
3159
3160 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3161                      struct lu_attr *attr,
3162                      struct dt_allocation_hint *hint,
3163                      struct dt_object_format *dof,
3164                      struct thandle *th)
3165 {
3166         LASSERT(S_ISREG(attr->la_mode));
3167         return osd_mkfile(info, obj, (attr->la_mode &
3168                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3169 }
3170
3171 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3172                      struct lu_attr *attr,
3173                      struct dt_allocation_hint *hint,
3174                      struct dt_object_format *dof,
3175                      struct thandle *th)
3176 {
3177         LASSERT(S_ISLNK(attr->la_mode));
3178         return osd_mkfile(info, obj, (attr->la_mode &
3179                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
3180 }
3181
3182 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3183                      struct lu_attr *attr,
3184                      struct dt_allocation_hint *hint,
3185                      struct dt_object_format *dof,
3186                      struct thandle *th)
3187 {
3188         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3189         int result;
3190
3191         LINVRNT(osd_invariant(obj));
3192         LASSERT(obj->oo_inode == NULL);
3193         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3194                 S_ISFIFO(mode) || S_ISSOCK(mode));
3195
3196         result = osd_mkfile(info, obj, mode, hint, th);
3197         if (result == 0) {
3198                 LASSERT(obj->oo_inode != NULL);
3199                 /*
3200                  * This inode should be marked dirty for i_rdev.  Currently
3201                  * that is done in the osd_attr_init().
3202                  */
3203                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3204                                    attr->la_rdev);
3205         }
3206         LINVRNT(osd_invariant(obj));
3207         return result;
3208 }
3209
3210 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3211                               struct lu_attr *,
3212                               struct dt_allocation_hint *hint,
3213                               struct dt_object_format *dof,
3214                               struct thandle *);
3215
3216 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3217 {
3218         osd_obj_type_f result;
3219
3220         switch (type) {
3221         case DFT_DIR:
3222                 result = osd_mkdir;
3223                 break;
3224         case DFT_REGULAR:
3225                 result = osd_mkreg;
3226                 break;
3227         case DFT_SYM:
3228                 result = osd_mksym;
3229                 break;
3230         case DFT_NODE:
3231                 result = osd_mknod;
3232                 break;
3233         case DFT_INDEX:
3234                 result = osd_mk_index;
3235                 break;
3236
3237         default:
3238                 LBUG();
3239                 break;
3240         }
3241         return result;
3242 }
3243
3244
3245 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3246                         struct dt_object *parent, struct dt_object *child,
3247                         umode_t child_mode)
3248 {
3249         LASSERT(ah);
3250
3251         ah->dah_parent = parent;
3252         ah->dah_mode = child_mode;
3253
3254         if (parent != NULL && !dt_object_remote(parent)) {
3255                 /* will help to find FID->ino at dt_insert("..") */
3256                 struct osd_object *pobj = osd_dt_obj(parent);
3257
3258                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3259         }
3260 }
3261
3262 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3263                           struct lu_attr *attr, struct dt_object_format *dof)
3264 {
3265         struct inode *inode = obj->oo_inode;
3266         __u64 valid = attr->la_valid;
3267         int result;
3268
3269         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3270
3271         if (dof->dof_type != DFT_NODE)
3272                 attr->la_valid &= ~LA_RDEV;
3273         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3274                 attr->la_valid &= ~LA_ATIME;
3275         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3276                 attr->la_valid &= ~LA_CTIME;
3277         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3278                 attr->la_valid &= ~LA_MTIME;
3279
3280         result = osd_quota_transfer(inode, attr);
3281         if (result)
3282                 return;
3283
3284         if (attr->la_valid != 0) {
3285                 result = osd_inode_setattr(info->oti_env, inode, attr);
3286                 /*
3287                  * The osd_inode_setattr() should always succeed here.  The
3288                  * only error that could be returned is EDQUOT when we are
3289                  * trying to change the UID or GID of the inode. However, this
3290                  * should not happen since quota enforcement is no longer
3291                  * enabled on ldiskfs (lquota takes care of it).
3292                  */
3293                 LASSERTF(result == 0, "%d\n", result);
3294                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3295         }
3296
3297         attr->la_valid = valid;
3298 }
3299
3300 /**
3301  * Helper function for osd_create()
3302  *
3303  * \retval 0, on success
3304  */
3305 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3306                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3307                         struct dt_object_format *dof, struct thandle *th)
3308 {
3309         int result;
3310         __u32 umask;
3311
3312         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3313
3314         /* we drop umask so that permissions we pass are not affected */
3315         umask = current->fs->umask;
3316         current->fs->umask = 0;
3317
3318         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3319                                                   th);
3320         if (likely(obj->oo_inode != NULL)) {
3321                 LASSERT(obj->oo_inode->i_state & I_NEW);
3322
3323                 /*
3324                  * Unlock the inode before attr initialization to avoid
3325                  * unnecessary dqget operations. LU-6378
3326                  */
3327                 unlock_new_inode(obj->oo_inode);
3328         }
3329
3330         if (likely(result == 0)) {
3331                 osd_attr_init(info, obj, attr, dof);
3332                 osd_object_init0(obj);
3333         }
3334
3335         /* restore previous umask value */
3336         current->fs->umask = umask;
3337
3338         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3339
3340         return result;
3341 }
3342
3343 /**
3344  * Helper function for osd_create()
3345  *
3346  * \retval 0, on success
3347  */
3348 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3349                            const struct lu_fid *fid, struct thandle *th)
3350 {
3351         struct osd_thread_info *info = osd_oti_get(env);
3352         struct osd_inode_id    *id   = &info->oti_id;
3353         struct osd_device      *osd  = osd_obj2dev(obj);
3354         struct osd_thandle     *oh;
3355         int rc;
3356
3357         LASSERT(obj->oo_inode != NULL);
3358
3359         oh = container_of0(th, struct osd_thandle, ot_super);
3360         LASSERT(oh->ot_handle);
3361         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3362
3363         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3364         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3365                            OI_CHECK_FLD, NULL);
3366         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3367
3368         return rc;
3369 }
3370
3371 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3372                    u64 seq, struct lu_seq_range *range)
3373 {
3374         struct seq_server_site *ss = osd_seq_site(osd);
3375
3376         if (fid_seq_is_idif(seq)) {
3377                 fld_range_set_ost(range);
3378                 range->lsr_index = idif_ost_idx(seq);
3379                 return 0;
3380         }
3381
3382         if (!fid_seq_in_fldb(seq)) {
3383                 fld_range_set_mdt(range);
3384                 if (ss != NULL)
3385                         /*
3386                          * FIXME: If ss is NULL, it suppose not get lsr_index
3387                          * at all
3388                          */
3389                         range->lsr_index = ss->ss_node_id;
3390                 return 0;
3391         }
3392
3393         LASSERT(ss != NULL);
3394         fld_range_set_any(range);
3395         /* OSD will only do local fld lookup */
3396         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3397 }
3398
3399 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3400                               struct lu_attr *attr,
3401                               struct dt_allocation_hint *hint,
3402                               struct dt_object_format *dof,
3403                               struct thandle *handle)
3404 {
3405         struct osd_thandle *oh;
3406         int rc;
3407
3408         ENTRY;
3409
3410         LASSERT(handle != NULL);
3411
3412         oh = container_of0(handle, struct osd_thandle, ot_super);
3413         LASSERT(oh->ot_handle == NULL);
3414
3415         /*
3416          * EA object consumes more credits than regular object: osd_mk_index
3417          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3418          * leaf_node, could involves the block, block bitmap, groups, GDT
3419          * change for each block, so add 4 * 2 credits in that case.
3420          */
3421         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3422                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3423                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3424         /*
3425          * Reuse idle OI block may cause additional one OI block
3426          * to be changed.
3427          */
3428         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3429                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3430
3431         /* will help to find FID->ino mapping at dt_insert() */
3432         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3433                                    osd_dt_obj(dt));
3434         if (rc != 0)
3435                 RETURN(rc);
3436
3437         if (!attr)
3438                 RETURN(0);
3439
3440         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3441                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3442                                    NULL, OSD_QID_INODE);
3443         if (rc != 0)
3444                 RETURN(rc);
3445
3446         RETURN(rc);
3447 }
3448
3449 /**
3450  * Called to destroy on-disk representation of the object
3451  *
3452  * Concurrency: must be locked
3453  */
3454 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3455                                struct thandle *th)
3456 {
3457         struct osd_object *obj = osd_dt_obj(dt);
3458         struct inode *inode = obj->oo_inode;
3459         struct osd_thandle *oh;
3460         int rc;
3461
3462         ENTRY;
3463
3464         if (inode == NULL)
3465                 RETURN(-ENOENT);
3466
3467         oh = container_of0(th, struct osd_thandle, ot_super);
3468         LASSERT(oh->ot_handle == NULL);
3469
3470         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3471                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3472
3473         /* For removing agent entry */
3474         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3475                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3476
3477         /*
3478          * Recycle idle OI leaf may cause additional three OI blocks
3479          * to be changed.
3480          */
3481         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3482                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3483                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3484         /* one less inode */
3485         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3486                                    i_projid_read(inode), -1, oh, obj, NULL,
3487                                    OSD_QID_INODE);
3488         if (rc)
3489                 RETURN(rc);
3490         /* data to be truncated */
3491         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3492                                    i_projid_read(inode), 0, oh, obj, NULL,
3493                                    OSD_QID_BLK);
3494         if (rc)
3495                 RETURN(rc);
3496
3497         /*
3498          * will help to find FID->ino when this object is being
3499          * added to PENDING
3500          */
3501         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3502
3503         RETURN(rc);
3504 }
3505
3506 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3507                        struct thandle *th)
3508 {
3509         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3510         struct osd_object *obj = osd_dt_obj(dt);
3511         struct inode *inode = obj->oo_inode;
3512         struct osd_device *osd = osd_obj2dev(obj);
3513         struct osd_thandle *oh;
3514         int result;
3515
3516         ENTRY;
3517
3518         oh = container_of0(th, struct osd_thandle, ot_super);
3519         LASSERT(oh->ot_handle);
3520         LASSERT(inode);
3521         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3522
3523         if (unlikely(fid_is_acct(fid)))
3524                 RETURN(-EPERM);
3525
3526         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3527                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3528                 if (result != 0)
3529                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3530                                osd_name(osd), PFID(fid), result);
3531         }
3532
3533         if (S_ISDIR(inode->i_mode)) {
3534                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3535                         inode->i_nlink == 2);
3536
3537                 spin_lock(&obj->oo_guard);
3538                 clear_nlink(inode);
3539                 spin_unlock(&obj->oo_guard);
3540                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3541         }
3542
3543         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3544
3545         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3546
3547         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3548                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3549                                        oh->ot_handle, OI_CHECK_FLD);
3550
3551         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3552         /* XXX: add to ext3 orphan list */
3553         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3554
3555         /* not needed in the cache anymore */
3556         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3557         obj->oo_destroyed = 1;
3558
3559         RETURN(0);
3560 }
3561
3562 /**
3563  * Put the fid into lustre_mdt_attrs, and then place the structure
3564  * inode's ea. This fid should not be altered during the life time
3565  * of the inode.
3566  *
3567  * \retval +ve, on success
3568  * \retval -ve, on error
3569  *
3570  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3571  */
3572 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3573                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3574 {
3575         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3576         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3577         int rc;
3578
3579         ENTRY;
3580
3581         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3582                 RETURN(0);
3583
3584         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3585                 rc = -ENOMEM;
3586
3587         lustre_loa_init(loa, fid, compat, incompat);
3588         lustre_loa_swab(loa, false);
3589
3590         /*
3591          * For the OST device with 256 bytes inode size by default,
3592          * the PFID EA will be stored together with LMA EA to avoid
3593          * performance trouble. Otherwise the PFID EA can be stored
3594          * independently. LU-8998
3595          */
3596         if ((compat & LMAC_FID_ON_OST) &&
3597             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3598                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3599                                      sizeof(*loa), XATTR_CREATE);
3600         else
3601                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3602                                      sizeof(*lma), XATTR_CREATE);
3603         /*
3604          * LMA may already exist, but we need to check that all the
3605          * desired compat/incompat flags have been added.
3606          */
3607         if (unlikely(rc == -EEXIST)) {
3608                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3609                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3610                 if (rc < 0)
3611                         RETURN(rc);
3612
3613                 if (rc < sizeof(*lma))
3614                         RETURN(-EINVAL);
3615
3616                 lustre_loa_swab(loa, true);
3617                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3618                     ((compat == 0 && incompat == 0) ||
3619                      (!(~lma->lma_compat & compat) &&
3620                       !(~lma->lma_incompat & incompat))))
3621                         RETURN(0);
3622
3623                 lma->lma_self_fid = *fid;
3624                 lma->lma_compat |= compat;
3625                 lma->lma_incompat |= incompat;
3626                 if (rc == sizeof(*lma)) {
3627                         lustre_lma_swab(lma);
3628                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3629                                              sizeof(*lma), XATTR_REPLACE);
3630                 } else {
3631                         lustre_loa_swab(loa, false);
3632                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3633                                              sizeof(*loa), XATTR_REPLACE);
3634                 }
3635         }
3636
3637         RETURN(rc);
3638 }
3639
3640 /**
3641  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3642  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3643  * To have compatilibility with 1.8 ldiskfs driver we need to have
3644  * magic number at start of fid data.
3645  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3646  * its inmemory API.
3647  */
3648 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3649                                          const struct lu_fid *fid)
3650 {
3651         if (!fid_is_namespace_visible(fid) ||
3652             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3653                 param->edp_magic = 0;
3654                 return;
3655         }
3656
3657         param->edp_magic = LDISKFS_LUFID_MAGIC;
3658         param->edp_len =  sizeof(struct lu_fid) + 1;
3659         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3660 }
3661
3662 /**
3663  * Try to read the fid from inode ea into dt_rec.
3664  *
3665  * \param fid object fid.
3666  *
3667  * \retval 0 on success
3668  */
3669 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3670                           __u32 ino, struct lu_fid *fid,
3671                           struct osd_inode_id *id)
3672 {
3673         struct osd_thread_info *info  = osd_oti_get(env);
3674         struct inode *inode;
3675
3676         ENTRY;
3677
3678         osd_id_gen(id, ino, OSD_OII_NOGEN);
3679         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3680         if (IS_ERR(inode))
3681                 RETURN(PTR_ERR(inode));
3682
3683         iput(inode);
3684         RETURN(0);
3685 }
3686
3687 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3688                                         struct inode *dir,
3689                                         struct inode *parent_dir,
3690                                         const struct lu_fid *dot_fid,
3691                                         const struct lu_fid *dot_dot_fid,
3692                                         struct osd_thandle *oth)
3693 {
3694         struct ldiskfs_dentry_param *dot_ldp;
3695         struct ldiskfs_dentry_param *dot_dot_ldp;
3696         __u32 saved_nlink = dir->i_nlink;
3697         int rc;
3698
3699         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3700         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3701
3702         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3703         dot_ldp->edp_magic = 0;
3704
3705         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3706                                     dir, dot_ldp, dot_dot_ldp);
3707         /*
3708          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3709          * the subseqent ref_add() will increase the dir->i_nlink
3710          * as 3. That is incorrect for new created directory.
3711          *
3712          * It looks like hack, because we want to make the OSD API
3713          * to be order-independent for new created directory object
3714          * between dt_insert(..) and ref_add() operations.
3715          *
3716          * Here, we only restore the in-RAM dir-inode's nlink attr,
3717          * becuase if the nlink attr is not 2, then there will be
3718          * ref_add() called following the dt_insert(..), such call
3719          * will make both the in-RAM and on-disk dir-inode's nlink
3720          * attr to be set as 2. LU-7447
3721          */
3722         set_nlink(dir, saved_nlink);
3723         return rc;
3724 }
3725
3726 /**
3727  * Create an local agent inode for remote entry
3728  */
3729 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3730                                                   struct osd_device *osd,
3731                                                   struct osd_object *pobj,
3732                                                   const struct lu_fid *fid,
3733                                                   __u32 type,
3734                                                   struct thandle *th)
3735 {
3736         struct osd_thread_info *info = osd_oti_get(env);
3737         struct inode *local;
3738         struct osd_thandle *oh;
3739         int rc;
3740
3741         ENTRY;
3742
3743         LASSERT(th);
3744         oh = container_of(th, struct osd_thandle, ot_super);
3745         LASSERT(oh->ot_handle->h_transaction != NULL);
3746
3747         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3748         if (IS_ERR(local)) {
3749                 CERROR("%s: create local error %d\n", osd_name(osd),
3750                        (int)PTR_ERR(local));
3751                 RETURN(local);
3752         }
3753
3754         /*
3755          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3756          * correct gid on remote file, not agent here
3757          */
3758         local->i_gid = current_fsgid();
3759         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3760         unlock_new_inode(local);
3761
3762         /* Agent inode should not have project ID */
3763 #ifdef  HAVE_PROJECT_QUOTA
3764         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3765             i_projid_read(pobj->oo_inode) != 0) {
3766                 rc = osd_transfer_project(local, 0);
3767                 if (rc) {
3768                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3769                                "quota enforcement enabled on the ldiskfs "
3770                                "filesystem?\n", local->i_sb->s_id, rc);
3771                         RETURN(ERR_PTR(rc));
3772                 }
3773         }
3774 #endif
3775         /* Set special LMA flag for local agent inode */
3776         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3777         if (rc != 0) {
3778                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3779                        osd_name(osd), PFID(fid), rc);
3780                 RETURN(ERR_PTR(rc));
3781         }
3782
3783         if (!S_ISDIR(type))
3784                 RETURN(local);
3785
3786         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3787                                          lu_object_fid(&pobj->oo_dt.do_lu),
3788                                          fid, oh);
3789         if (rc != 0) {
3790                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3791                         osd_name(osd), PFID(fid), rc);
3792                 RETURN(ERR_PTR(rc));
3793         }
3794
3795         RETURN(local);
3796 }
3797
3798 /**
3799  * when direntry is deleted, we have to take care of possible agent inode
3800  * referenced by that. unfortunately we can't do this at that point:
3801  * iget() within a running transaction leads to deadlock and we better do
3802  * not call that every delete declaration to save performance. so we put
3803  * a potention agent inode on a list and process that once the transaction
3804  * is over. Notice it's not any worse in terms of real orphans as regular
3805  * object destroy doesn't put inodes on the on-disk orphan list. this should
3806  * be addressed separately
3807  */
3808 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3809                                             struct osd_thandle *oh,
3810                                             __u32 ino)
3811 {
3812         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3813         struct osd_obj_orphan *oor;
3814
3815         OBD_ALLOC_PTR(oor);
3816         if (oor == NULL)
3817                 return -ENOMEM;
3818
3819         oor->oor_ino = ino;
3820         oor->oor_env = (struct lu_env *)env;
3821         spin_lock(&osd->od_osfs_lock);
3822         list_add(&oor->oor_list, &osd->od_orphan_list);
3823         spin_unlock(&osd->od_osfs_lock);
3824
3825         oh->ot_remove_agents = 1;
3826
3827         return 0;
3828
3829 }
3830
3831 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3832                                                 struct osd_device *osd)
3833 {
3834         struct osd_thread_info *info = osd_oti_get(env);
3835         struct osd_obj_orphan *oor, *tmp;
3836         struct osd_inode_id id;
3837         struct list_head list;
3838         struct inode *inode;
3839         struct lu_fid fid;
3840         handle_t *jh;
3841         __u32 ino;
3842
3843         INIT_LIST_HEAD(&list);
3844
3845         spin_lock(&osd->od_osfs_lock);
3846         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3847                 if (oor->oor_env == env) {
3848                         list_del(&oor->oor_list);
3849                         list_add(&oor->oor_list, &list);
3850                 }
3851         }
3852         spin_unlock(&osd->od_osfs_lock);
3853
3854         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3855
3856                 ino = oor->oor_ino;
3857
3858                 list_del(&oor->oor_list);
3859                 OBD_FREE_PTR(oor);
3860
3861                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3862                 inode = osd_iget_fid(info, osd, &id, &fid);
3863                 if (IS_ERR(inode))
3864                         continue;
3865
3866                 if (!osd_remote_fid(env, osd, &fid)) {
3867                         iput(inode);
3868                         continue;
3869                 }
3870
3871                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3872                 clear_nlink(inode);
3873                 mark_inode_dirty(inode);
3874                 ldiskfs_journal_stop(jh);
3875                 iput(inode);
3876         }
3877
3878         return 0;
3879 }
3880
3881 /**
3882  * OSD layer object create function for OST objects (b=11826).
3883  *
3884  * The FID is inserted into inode xattr here.
3885  *
3886  * \retval   0, on success
3887  * \retval -ve, on error
3888  */
3889 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3890                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3891                       struct dt_object_format *dof, struct thandle *th)
3892 {
3893         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3894         struct osd_object *obj = osd_dt_obj(dt);
3895         struct osd_thread_info *info = osd_oti_get(env);
3896         int result, on_ost = 0;
3897
3898         ENTRY;
3899
3900         if (dt_object_exists(dt))
3901                 RETURN(-EEXIST);
3902
3903         LINVRNT(osd_invariant(obj));
3904         LASSERT(!dt_object_remote(dt));
3905         LASSERT(osd_is_write_locked(env, obj));
3906         LASSERT(th != NULL);
3907
3908         if (unlikely(fid_is_acct(fid)))
3909                 /*
3910                  * Quota files can't be created from the kernel any more,
3911                  * 'tune2fs -O quota' will take care of creating them
3912                  */
3913                 RETURN(-EPERM);
3914
3915         result = __osd_create(info, obj, attr, hint, dof, th);
3916         if (result == 0) {
3917                 if (fid_is_idif(fid) &&
3918                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3919                         struct lu_fid *tfid = &info->oti_fid;
3920                         struct ost_id *oi   = &info->oti_ostid;
3921
3922                         fid_to_ostid(fid, oi);
3923                         ostid_to_fid(tfid, oi, 0);
3924                         on_ost = 1;
3925                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3926                                                 LMAC_FID_ON_OST, 0);
3927                 } else {
3928                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3929                                                fid, OI_CHECK_FLD);
3930                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3931                                                 on_ost ? LMAC_FID_ON_OST : 0,
3932                                                 0);
3933                 }
3934                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3935                         obj->oo_dt.do_body_ops = &osd_body_ops;
3936         }
3937
3938         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3939                 result = __osd_oi_insert(env, obj, fid, th);
3940
3941         /*
3942          * a small optimization - dt_insert() isn't usually applied
3943          * to OST objects, so we don't need to cache OI mapping for
3944          * OST objects
3945          */
3946         if (result == 0 && on_ost == 0) {
3947                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3948
3949                 result = osd_idc_find_and_init(env, osd, obj);
3950                 LASSERT(result == 0);
3951         }
3952
3953         LASSERT(ergo(result == 0,
3954                      dt_object_exists(dt) && !dt_object_remote(dt)));
3955         LINVRNT(osd_invariant(obj));
3956         RETURN(result);
3957 }
3958
3959 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3960                                struct thandle *handle)
3961 {
3962         struct osd_thandle *oh;
3963
3964         /* it's possible that object doesn't exist yet */
3965         LASSERT(handle != NULL);
3966
3967         oh = container_of0(handle, struct osd_thandle, ot_super);
3968         LASSERT(oh->ot_handle == NULL);
3969
3970         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3971                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3972
3973         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
3974
3975         return 0;
3976 }
3977
3978 /*
3979  * Concurrency: @dt is write locked.
3980  */
3981 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3982                        struct thandle *th)
3983 {
3984         struct osd_object *obj = osd_dt_obj(dt);
3985         struct inode *inode = obj->oo_inode;
3986         struct osd_thandle *oh;
3987         int rc = 0;
3988
3989         if (!dt_object_exists(dt) || obj->oo_destroyed)
3990                 return -ENOENT;
3991
3992         LINVRNT(osd_invariant(obj));
3993         LASSERT(!dt_object_remote(dt));
3994         LASSERT(osd_is_write_locked(env, obj));
3995         LASSERT(th != NULL);
3996
3997         oh = container_of0(th, struct osd_thandle, ot_super);
3998         LASSERT(oh->ot_handle != NULL);
3999
4000         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4001
4002         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4003                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4004         /*
4005          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4006          * (65000) subdirectories by storing "1" in i_nlink if the link count
4007          * would otherwise overflow. Directory tranversal tools understand
4008          * that (st_nlink == 1) indicates that the filesystem dose not track
4009          * hard links count on the directory, and will not abort subdirectory
4010          * scanning early once (st_nlink - 2) subdirs have been found.
4011          *
4012          * This also has to properly handle the case of inodes with nlink == 0
4013          * in case they are being linked into the PENDING directory
4014          */
4015         spin_lock(&obj->oo_guard);
4016         if (unlikely(inode->i_nlink == 0))
4017                 /* inc_nlink from 0 may cause WARN_ON */
4018                 set_nlink(inode, 1);
4019         else {
4020                 ldiskfs_inc_count(oh->ot_handle, inode);
4021                 if (!S_ISDIR(inode->i_mode))
4022                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4023         }
4024         spin_unlock(&obj->oo_guard);
4025
4026         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4027         LINVRNT(osd_invariant(obj));
4028
4029         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4030
4031         return rc;
4032 }
4033
4034 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4035                                struct thandle *handle)
4036 {
4037         struct osd_thandle *oh;
4038
4039         if (!dt_object_exists(dt))
4040                 return -ENOENT;
4041
4042         LASSERT(!dt_object_remote(dt));
4043         LASSERT(handle != NULL);
4044
4045         oh = container_of0(handle, struct osd_thandle, ot_super);
4046         LASSERT(oh->ot_handle == NULL);
4047
4048         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4049                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4050
4051         return 0;
4052 }
4053
4054 /*
4055  * Concurrency: @dt is write locked.
4056  */
4057 static int osd_ref_del(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_device *osd = osd_dev(dt->do_lu.lo_dev);
4063         struct osd_thandle *oh;
4064
4065         if (!dt_object_exists(dt))
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_DEL);
4077
4078         spin_lock(&obj->oo_guard);
4079         /*
4080          * That can be result of upgrade from old Lustre version and
4081          * applied only to local files.  Just skip this ref_del call.
4082          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4083          */
4084         if (inode->i_nlink == 0) {
4085                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4086                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4087                              ", maybe an upgraded file? (LU-3915)\n",
4088                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4089                 spin_unlock(&obj->oo_guard);
4090                 return 0;
4091         }
4092
4093         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4094                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4095
4096         ldiskfs_dec_count(oh->ot_handle, inode);
4097         spin_unlock(&obj->oo_guard);
4098
4099         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4100         LINVRNT(osd_invariant(obj));
4101
4102         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4103
4104         return 0;
4105 }
4106
4107 /*
4108  * Concurrency: @dt is read locked.
4109  */
4110 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4111                          struct lu_buf *buf, const char *name)
4112 {
4113         struct osd_object      *obj    = osd_dt_obj(dt);
4114         struct inode           *inode  = obj->oo_inode;
4115         struct osd_thread_info *info   = osd_oti_get(env);
4116         struct dentry          *dentry = &info->oti_obj_dentry;
4117         bool cache_xattr = false;
4118         int rc;
4119
4120         LASSERT(buf);
4121
4122         /* version get is not real XATTR but uses xattr API */
4123         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4124                 dt_obj_version_t *ver = buf->lb_buf;
4125
4126                 /*
4127                  * for version we are just using xattr API but change inode
4128                  * field instead
4129                  */
4130                 if (buf->lb_len == 0)
4131                         return sizeof(dt_obj_version_t);
4132
4133                 if (buf->lb_len < sizeof(dt_obj_version_t))
4134                         return -ERANGE;
4135
4136                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4137                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4138
4139                 *ver = LDISKFS_I(inode)->i_fs_version;
4140
4141                 return sizeof(dt_obj_version_t);
4142         }
4143
4144         if (!dt_object_exists(dt))
4145                 return -ENOENT;
4146
4147         LASSERT(!dt_object_remote(dt));
4148         LASSERT(inode->i_op != NULL);
4149 #ifdef HAVE_IOP_XATTR
4150         LASSERT(inode->i_op->getxattr != NULL);
4151 #endif
4152
4153         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4154             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4155                 cache_xattr = true;
4156
4157         if (cache_xattr) {
4158                 rc = osd_oxc_get(obj, name, buf);
4159                 if (rc != -ENOENT)
4160                         return rc;
4161         }
4162
4163         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4164                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4165                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4166                 struct filter_fid *ff;
4167                 struct ost_layout *ol;
4168
4169                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4170                 if (rc)
4171                         return rc;
4172
4173                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4174
4175                 rc = sizeof(*ff);
4176                 if (buf->lb_len == 0 || !buf->lb_buf)
4177                         return rc;
4178
4179                 if (buf->lb_len < rc)
4180                         return -ERANGE;
4181
4182                 ff = buf->lb_buf;
4183                 ol = &ff->ff_layout;
4184                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4185                                                   PFID_STRIPE_IDX_BITS);
4186                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4187                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4188                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4189                 if (lma->lma_compat & LMAC_COMP_INFO) {
4190                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4191                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4192                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4193                 } else {
4194                         ol->ol_comp_start = 0;
4195                         ol->ol_comp_end = 0;
4196                         ol->ol_comp_id = 0;
4197                 }
4198         } else {
4199                 rc = __osd_xattr_get(inode, dentry, name,
4200                                      buf->lb_buf, buf->lb_len);
4201         }
4202
4203         if (cache_xattr) {
4204                 if (rc == -ENOENT || rc == -ENODATA)
4205                         osd_oxc_add(obj, name, NULL, 0);
4206                 else if (rc > 0 && buf->lb_buf != NULL)
4207                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4208         }
4209
4210         return rc;
4211 }
4212
4213 static int osd_declare_xattr_set(const struct lu_env *env,
4214                                  struct dt_object *dt,
4215                                  const struct lu_buf *buf, const char *name,
4216                                  int fl, struct thandle *handle)
4217 {
4218         struct osd_thandle *oh;
4219         int credits = 0;
4220         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4221
4222         LASSERT(handle != NULL);
4223
4224         oh = container_of0(handle, struct osd_thandle, ot_super);
4225         LASSERT(oh->ot_handle == NULL);
4226
4227         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4228                 /*
4229                  * For non-upgrading case, the LMA is set first and
4230                  * usually fit inode. But for upgrade case, the LMA
4231                  * may be in another separated EA block.
4232                  */
4233                 if (dt_object_exists(dt)) {
4234                         if (fl == LU_XATTR_REPLACE)
4235                                 credits = 1;
4236                         else
4237                                 goto upgrade;
4238                 }
4239         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4240                 credits = 1;
4241         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4242                 /* We may need to delete the old PFID EA. */
4243                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4244                 if (fl == LU_XATTR_REPLACE)
4245                         credits += 1;
4246                 else
4247                         goto upgrade;
4248         } else {
4249                 /*
4250                  * If some name entry resides on remote MDT, then will create
4251                  * agent entry under remote parent. On the other hand, if the
4252                  * remote entry will be removed, then related agent entry may
4253                  * need to be removed from the remote parent. So there may be
4254                  * kinds of cases, let's declare enough credits. The credits
4255                  * for create agent entry is enough for remove case.
4256                  */
4257                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4258                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4259                         if (dt_object_exists(dt))
4260                                 credits += 1; /* For updating LMA */
4261                 }
4262
4263 upgrade:
4264                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4265
4266                 if (buf != NULL) {
4267                         ssize_t buflen;
4268
4269                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4270                                 /*
4271                                  * learn xattr size from osd_xattr_get if
4272                                  * attribute has not been read yet
4273                                  */
4274                                 buflen = __osd_xattr_get(
4275                                     osd_dt_obj(dt)->oo_inode,
4276                                     &osd_oti_get(env)->oti_obj_dentry,
4277                                     name, NULL, 0);
4278                                 if (buflen < 0)
4279                                         buflen = 0;
4280                         } else {
4281                                 buflen = buf->lb_len;
4282                         }
4283
4284                         if (buflen > sb->s_blocksize) {
4285                                 credits += osd_calc_bkmap_credits(
4286                                     sb, NULL, 0, -1,
4287                                     (buflen + sb->s_blocksize - 1) >>
4288                                     sb->s_blocksize_bits);
4289                         }
4290                 }
4291                 /*
4292                  * xattr set may involve inode quota change, reserve credits for
4293                  * dquot_initialize()
4294                  */
4295                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4296         }
4297
4298         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4299
4300         return 0;
4301 }
4302
4303 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4304                               const struct lu_buf *buf, int fl,
4305                               struct thandle *handle)
4306 {
4307         struct osd_thread_info *info = osd_oti_get(env);
4308         struct dentry *dentry = &info->oti_obj_dentry;
4309         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4310         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4311         struct inode *inode = obj->oo_inode;
4312         struct filter_fid *ff = buf->lb_buf;
4313         struct ost_layout *ol = &ff->ff_layout;
4314         int flags = XATTR_REPLACE;
4315         int rc;
4316
4317         ENTRY;
4318
4319         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4320                 RETURN(-EINVAL);
4321
4322         rc = osd_get_lma(info, inode, dentry, loa);
4323         if (rc == -ENODATA) {
4324                 /* Usually for upgarding from old device */
4325                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4326                                 LMAC_FID_ON_OST, 0);
4327                 flags = XATTR_CREATE;
4328         } else if (rc) {
4329                 RETURN(rc);
4330         }
4331
4332         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4333                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4334                         RETURN(-EEXIST);
4335
4336                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4337                         /* Separate PFID EA from LMA */
4338                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4339                         lustre_lma_swab(lma);
4340                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4341                                              sizeof(*lma), XATTR_REPLACE);
4342                         if (!rc) {
4343                                 obj->oo_pfid_in_lma = 0;
4344                                 rc = LU_XATTR_CREATE;
4345                         }
4346
4347                         RETURN(rc);
4348                 }
4349         } else {
4350                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4351                         RETURN(fl);
4352
4353                 /*
4354                  * Old client does not send stripe information,
4355                  * then store the PFID EA on disk separatedly.
4356                  */
4357                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4358                              ol->ol_stripe_size == 0))
4359                         RETURN(fl);
4360
4361                 /* Remove old PFID EA entry firstly. */
4362                 ll_vfs_dq_init(inode);
4363                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4364                 if (rc == -ENODATA) {
4365                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4366                                 RETURN(rc);
4367                 } else if (rc) {
4368                         RETURN(rc);
4369                 }
4370         }
4371
4372         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4373         if (likely(ol->ol_stripe_size != 0)) {
4374                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4375                                              PFID_STRIPE_IDX_BITS;
4376                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4377                 lma->lma_compat |= LMAC_STRIPE_INFO;
4378                 if (ol->ol_comp_id != 0) {
4379                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4380                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4381                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4382                         lma->lma_compat |= LMAC_COMP_INFO;
4383                 }
4384         }
4385
4386         lustre_loa_swab(loa, false);
4387
4388         /* Store the PFID EA inside LMA. */
4389         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4390                              flags);
4391         if (!rc)
4392                 obj->oo_pfid_in_lma = 1;
4393
4394         RETURN(rc);
4395 }
4396
4397 /*
4398  * In DNE environment, the object (in spite of regular file or directory)
4399  * and its name entry may reside on different MDTs. Under such case, we will
4400  * create an agent entry on the MDT where the object resides. The agent entry
4401  * references the object locally, that makes the object to be visible to the
4402  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4403  * such as 'tar' can handle the object properly.
4404  *
4405  * We handle the agent entry during set linkEA that is the common interface
4406  * for both regular file and directroy, can handle kinds of cases, such as
4407  * create/link/unlink/rename, and so on.
4408  *
4409  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4410  *
4411  * XXX: There are two known issues:
4412  * 1. For one object, we will create at most one agent entry even if there
4413  *    may be more than one cross-MDTs hard links on the object. So the local
4414  *    e2fsck may claim that the object's nlink is larger than the name entries
4415  *    that reference such inode. And in further, the e2fsck will fix the nlink
4416  *    attribute to match the local references. Then it will cause the object's
4417  *    nlink attribute to be inconsistent with the global references. it is bad
4418  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4419  *    other hand, the global namespace LFSCK can repair the object's attribute
4420  *    according to the linkEA.
4421  * 2. There may be too many hard links on the object as to its linkEA overflow,
4422  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4423  *    case happened, then at this point, we do not know whether there are some
4424  *    cross-MDTs reference. But there are local references, it guarantees that
4425  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4426  */
4427 static int osd_xattr_handle_linkea(const struct lu_env *env,
4428                                    struct osd_device *osd,
4429                                    struct osd_object *obj,
4430                                    const struct lu_buf *buf,
4431                                    struct thandle *handle)
4432 {
4433         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4434         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4435         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4436         struct lu_name tmpname;
4437         struct osd_thandle *oh;
4438         int rc;
4439         bool remote = false;
4440
4441         ENTRY;
4442
4443         oh = container_of0(handle, struct osd_thandle, ot_super);
4444         LASSERT(oh->ot_handle != NULL);
4445
4446         rc = linkea_init_with_rec(&ldata);
4447         if (!rc) {
4448                 linkea_first_entry(&ldata);
4449                 while (ldata.ld_lee != NULL && !remote) {
4450                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4451                                             &tmpname, tfid);
4452                         if (osd_remote_fid(env, osd, tfid) > 0)
4453                                 remote = true;
4454                         else
4455                                 linkea_next_entry(&ldata);
4456                 }
4457         } else if (rc == -ENODATA) {
4458                 rc = 0;
4459         } else {
4460                 RETURN(rc);
4461         }
4462
4463         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4464                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4465                 if (rc)
4466                         CERROR("%s: failed to remove agent entry for "DFID
4467                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4468         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4469                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4470                 if (rc)
4471                         CERROR("%s: failed to create agent entry for "DFID
4472                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4473         }
4474
4475         RETURN(rc);
4476 }
4477
4478 /*
4479  * Concurrency: @dt is write locked.
4480  */
4481 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4482                          const struct lu_buf *buf, const char *name, int fl,
4483                          struct thandle *handle)
4484 {
4485         struct osd_object *obj = osd_dt_obj(dt);
4486         struct osd_device *osd = osd_obj2dev(obj);
4487         struct inode *inode = obj->oo_inode;
4488         struct osd_thread_info *info = osd_oti_get(env);
4489         int fs_flags = 0;
4490         int len;
4491         int rc;
4492
4493         ENTRY;
4494
4495         LASSERT(handle);
4496         LASSERT(buf);
4497
4498         /* version set is not real XATTR */
4499         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4500                 dt_obj_version_t *version = buf->lb_buf;
4501
4502                 /*
4503                  * for version we are just using xattr API but change inode
4504                  * field instead
4505                  */
4506                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4507
4508                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4509                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4510
4511                 LDISKFS_I(inode)->i_fs_version = *version;
4512                 /*
4513                  * Version is set after all inode operations are finished,
4514                  * so we should mark it dirty here
4515                  */
4516                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4517
4518                 RETURN(0);
4519         }
4520
4521         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4522                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4523
4524         len = buf->lb_len;
4525         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4526
4527         /*
4528          * For the OST device with 256 bytes inode size by default,
4529          * the PFID EA will be stored together with LMA EA to avoid
4530          * performance trouble. Otherwise the PFID EA can be stored
4531          * independently. LU-8998
4532          */
4533         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4534             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4535                 LASSERT(buf->lb_buf);
4536
4537                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4538                 if (fl <= 0)
4539                         RETURN(fl);
4540         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4541                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4542                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4543
4544                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4545                 if (rc)
4546                         RETURN(rc);
4547
4548                 lma->lma_incompat |= LMAI_STRIPED;
4549                 lustre_lma_swab(lma);
4550                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4551                                      sizeof(*lma), XATTR_REPLACE);
4552                 if (rc != 0)
4553                         RETURN(rc);
4554         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4555                 LASSERT(!osd->od_is_ost);
4556
4557                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4558                 if (rc)
4559                         RETURN(rc);
4560         }
4561
4562         if (fl & LU_XATTR_REPLACE)
4563                 fs_flags |= XATTR_REPLACE;
4564
4565         if (fl & LU_XATTR_CREATE)
4566                 fs_flags |= XATTR_CREATE;
4567
4568         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4569         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4570
4571         if (rc == 0 &&
4572             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4573              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4574                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4575
4576         return rc;
4577 }
4578
4579 /*
4580  * Concurrency: @dt is read locked.
4581  */
4582 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4583                           const struct lu_buf *buf)
4584 {
4585         struct osd_object *obj = osd_dt_obj(dt);
4586         struct inode *inode = obj->oo_inode;
4587         struct osd_thread_info *info = osd_oti_get(env);
4588         struct dentry *dentry = &info->oti_obj_dentry;
4589
4590         if (!dt_object_exists(dt))
4591                 return -ENOENT;
4592
4593         LASSERT(!dt_object_remote(dt));
4594         LASSERT(inode->i_op != NULL);
4595         LASSERT(inode->i_op->listxattr != NULL);
4596
4597         dentry->d_inode = inode;
4598         dentry->d_sb = inode->i_sb;
4599         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4600 }
4601
4602 static int osd_declare_xattr_del(const struct lu_env *env,
4603                                  struct dt_object *dt, const char *name,
4604                                  struct thandle *handle)
4605 {
4606         struct osd_thandle *oh;
4607         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4608
4609         LASSERT(!dt_object_remote(dt));
4610         LASSERT(handle != NULL);
4611
4612         oh = container_of0(handle, struct osd_thandle, ot_super);
4613         LASSERT(oh->ot_handle == NULL);
4614
4615         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4616                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4617         /*
4618          * xattr del may involve inode quota change, reserve credits for
4619          * dquot_initialize()
4620          */
4621         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4622
4623         return 0;
4624 }
4625
4626 /*
4627  * Concurrency: @dt is write locked.
4628  */
4629 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4630                          const char *name, struct thandle *handle)
4631 {
4632         struct osd_object *obj = osd_dt_obj(dt);
4633         struct inode *inode = obj->oo_inode;
4634         struct osd_thread_info *info = osd_oti_get(env);
4635         struct dentry *dentry = &info->oti_obj_dentry;
4636         int rc;
4637
4638         if (!dt_object_exists(dt))
4639                 return -ENOENT;
4640
4641         LASSERT(!dt_object_remote(dt));
4642         LASSERT(inode->i_op != NULL);
4643         LASSERT(handle != NULL);
4644 #ifdef HAVE_IOP_XATTR
4645         LASSERT(inode->i_op->removexattr != NULL);
4646 #endif
4647
4648         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4649
4650         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4651                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4652
4653                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4654                                  &info->oti_ost_attrs);
4655                 if (!rc) {
4656                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4657
4658                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4659                         lustre_lma_swab(lma);
4660                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4661                                              sizeof(*lma), XATTR_REPLACE);
4662                         if (!rc)
4663                                 obj->oo_pfid_in_lma = 0;
4664                 }
4665         } else {
4666                 ll_vfs_dq_init(inode);
4667                 dentry->d_inode = inode;
4668                 dentry->d_sb = inode->i_sb;
4669                 rc = osd_removexattr(dentry, inode, name);
4670         }
4671
4672         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4673
4674         if (rc == 0 &&
4675             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4676              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4677                 osd_oxc_del(obj, name);
4678
4679         return rc;
4680 }
4681
4682 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4683                            __u64 start, __u64 end)
4684 {
4685         struct osd_object *obj = osd_dt_obj(dt);
4686         struct inode *inode = obj->oo_inode;
4687         struct osd_thread_info *info = osd_oti_get(env);
4688         struct dentry *dentry = &info->oti_obj_dentry;
4689         struct file *file = &info->oti_file;
4690         int rc;
4691
4692         ENTRY;
4693
4694         dentry->d_inode = inode;
4695         dentry->d_sb = inode->i_sb;
4696         file->f_path.dentry = dentry;
4697         file->f_mapping = inode->i_mapping;
4698         file->f_op = inode->i_fop;
4699         set_file_inode(file, inode);
4700
4701         rc = ll_vfs_fsync_range(file, start, end, 0);
4702
4703         RETURN(rc);
4704 }
4705
4706 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4707 {
4708         return 0;
4709 }
4710
4711 /*
4712  * Index operations.
4713  */
4714
4715 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4716                                const struct dt_index_features *feat)
4717 {
4718         struct iam_descr *descr;
4719
4720         if (osd_object_is_root(o))
4721                 return feat == &dt_directory_features;
4722
4723         LASSERT(o->oo_dir != NULL);
4724
4725         descr = o->oo_dir->od_container.ic_descr;
4726         if (feat == &dt_directory_features) {
4727                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4728                         return 1;
4729                 else
4730                         return 0;
4731         } else {
4732                 return feat->dif_keysize_min <= descr->id_key_size &&
4733                        descr->id_key_size <= feat->dif_keysize_max &&
4734                        feat->dif_recsize_min <= descr->id_rec_size &&
4735                        descr->id_rec_size <= feat->dif_recsize_max &&
4736                        !(feat->dif_flags & (DT_IND_VARKEY |
4737                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4738                        ergo(feat->dif_flags & DT_IND_UPDATE,
4739                             1 /* XXX check that object (and fs) is writable */);
4740         }
4741 }
4742
4743 static int osd_iam_container_init(const struct lu_env *env,
4744                                   struct osd_object *obj,
4745                                   struct osd_directory *dir)
4746 {
4747         struct iam_container *bag = &dir->od_container;
4748         int result;
4749
4750         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4751         if (result != 0)
4752                 return result;
4753
4754         result = iam_container_setup(bag);
4755         if (result == 0)
4756                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4757         else
4758                 iam_container_fini(bag);
4759
4760         return result;
4761 }
4762
4763
4764 /*
4765  * Concurrency: no external locking is necessary.
4766  */
4767 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4768                          const struct dt_index_features *feat)
4769 {
4770         int result;
4771         int skip_iam = 0;
4772         struct osd_object *obj = osd_dt_obj(dt);
4773
4774         LINVRNT(osd_invariant(obj));
4775
4776         if (osd_object_is_root(obj)) {
4777                 dt->do_index_ops = &osd_index_ea_ops;
4778                 result = 0;
4779         } else if (feat == &dt_directory_features) {
4780                 dt->do_index_ops = &osd_index_ea_ops;
4781                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4782                         result = 0;
4783                 else
4784                         result = -ENOTDIR;
4785                 skip_iam = 1;
4786         } else if (unlikely(feat == &dt_otable_features)) {
4787                 dt->do_index_ops = &osd_otable_ops;
4788                 return 0;
4789         } else if (unlikely(feat == &dt_acct_features)) {
4790                 dt->do_index_ops = &osd_acct_index_ops;
4791                 result = 0;
4792                 skip_iam = 1;
4793         } else if (!osd_has_index(obj)) {
4794                 struct osd_directory *dir;
4795                 struct osd_device *osd = osd_obj2dev(obj);
4796                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4797
4798                 OBD_ALLOC_PTR(dir);
4799                 if (dir) {
4800
4801                         spin_lock(&obj->oo_guard);
4802                         if (obj->oo_dir == NULL)
4803                                 obj->oo_dir = dir;
4804                         else
4805                                 /*
4806                                  * Concurrent thread allocated container data.
4807                                  */
4808                                 OBD_FREE_PTR(dir);
4809                         spin_unlock(&obj->oo_guard);
4810                         /*
4811                          * Now, that we have container data, serialize its
4812                          * initialization.
4813                          */
4814                         down_write(&obj->oo_ext_idx_sem);
4815                         /*
4816                          * recheck under lock.
4817                          */
4818
4819                         if (osd_has_index(obj)) {
4820                                 result = 0;
4821                                 goto unlock;
4822                         }
4823
4824                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4825                         if (result || feat == &dt_lfsck_namespace_features ||
4826                             feat == &dt_lfsck_layout_orphan_features ||
4827                             feat == &dt_lfsck_layout_dangling_features)
4828                                 goto unlock;
4829
4830                         result = osd_index_register(osd, fid,
4831                                                     feat->dif_keysize_max,
4832                                                     feat->dif_recsize_max);
4833                         if (result < 0)
4834                                 CWARN("%s: failed to register index "
4835                                       DFID": rc = %d\n",
4836                                       osd_name(osd), PFID(fid), result);
4837                         else if (result > 0)
4838                                 result = 0;
4839                         else
4840                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4841                                        " (%d/%d) registered\n",
4842                                        osd_name(osd), PFID(fid),
4843                                        (int)feat->dif_keysize_max,
4844                                        (int)feat->dif_recsize_max);
4845
4846 unlock:
4847                         up_write(&obj->oo_ext_idx_sem);
4848                 } else {
4849                         result = -ENOMEM;
4850                 }
4851         } else {
4852                 result = 0;
4853         }
4854
4855         if (result == 0 && skip_iam == 0) {
4856                 if (!osd_iam_index_probe(env, obj, feat))
4857                         result = -ENOTDIR;
4858         }
4859         LINVRNT(osd_invariant(obj));
4860
4861         return result;
4862 }
4863
4864 static int osd_otable_it_attr_get(const struct lu_env *env,
4865                                  struct dt_object *dt,
4866                                  struct lu_attr *attr)
4867 {
4868         attr->la_valid = 0;
4869         return 0;
4870 }
4871
4872 static const struct dt_object_operations osd_obj_ops = {
4873         .do_read_lock           = osd_read_lock,
4874         .do_write_lock          = osd_write_lock,
4875         .do_read_unlock         = osd_read_unlock,
4876         .do_write_unlock        = osd_write_unlock,
4877         .do_write_locked        = osd_write_locked,
4878         .do_attr_get            = osd_attr_get,
4879         .do_declare_attr_set    = osd_declare_attr_set,
4880         .do_attr_set            = osd_attr_set,
4881         .do_ah_init             = osd_ah_init,
4882         .do_declare_create      = osd_declare_create,
4883         .do_create              = osd_create,
4884         .do_declare_destroy     = osd_declare_destroy,
4885         .do_destroy             = osd_destroy,
4886         .do_index_try           = osd_index_try,
4887         .do_declare_ref_add     = osd_declare_ref_add,
4888         .do_ref_add             = osd_ref_add,
4889         .do_declare_ref_del     = osd_declare_ref_del,
4890         .do_ref_del             = osd_ref_del,
4891         .do_xattr_get           = osd_xattr_get,
4892         .do_declare_xattr_set   = osd_declare_xattr_set,
4893         .do_xattr_set           = osd_xattr_set,
4894         .do_declare_xattr_del   = osd_declare_xattr_del,
4895         .do_xattr_del           = osd_xattr_del,
4896         .do_xattr_list          = osd_xattr_list,
4897         .do_object_sync         = osd_object_sync,
4898         .do_invalidate          = osd_invalidate,
4899 };
4900
4901 static const struct dt_object_operations osd_obj_otable_it_ops = {
4902         .do_attr_get    = osd_otable_it_attr_get,
4903         .do_index_try   = osd_index_try,
4904 };
4905
4906 static int osd_index_declare_iam_delete(const struct lu_env *env,
4907                                         struct dt_object *dt,
4908                                         const struct dt_key *key,
4909                                         struct thandle *handle)
4910 {
4911         struct osd_thandle *oh;
4912
4913         oh = container_of0(handle, struct osd_thandle, ot_super);
4914         LASSERT(oh->ot_handle == NULL);
4915
4916         /* Recycle  may cause additional three blocks to be changed. */
4917         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4918                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4919
4920         return 0;
4921 }
4922
4923 /**
4924  *      delete a (key, value) pair from index \a dt specified by \a key
4925  *
4926  *      \param  dt      osd index object
4927  *      \param  key     key for index
4928  *      \param  rec     record reference
4929  *      \param  handle  transaction handler
4930  *
4931  *      \retval  0  success
4932  *      \retval -ve   failure
4933  */
4934 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4935                                 const struct dt_key *key,
4936                                 struct thandle *handle)
4937 {
4938         struct osd_thread_info *oti = osd_oti_get(env);
4939         struct osd_object *obj = osd_dt_obj(dt);
4940         struct osd_thandle *oh;
4941         struct iam_path_descr *ipd;
4942         struct iam_container *bag = &obj->oo_dir->od_container;
4943         int rc;
4944
4945         ENTRY;
4946
4947         if (!dt_object_exists(dt))
4948                 RETURN(-ENOENT);
4949
4950         LINVRNT(osd_invariant(obj));
4951         LASSERT(!dt_object_remote(dt));
4952         LASSERT(bag->ic_object == obj->oo_inode);
4953         LASSERT(handle != NULL);
4954
4955         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4956
4957         ipd = osd_idx_ipd_get(env, bag);
4958         if (unlikely(ipd == NULL))
4959                 RETURN(-ENOMEM);
4960
4961         oh = container_of0(handle, struct osd_thandle, ot_super);
4962         LASSERT(oh->ot_handle != NULL);
4963         LASSERT(oh->ot_handle->h_transaction != NULL);
4964
4965         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4966                 /* swab quota uid/gid provided by caller */
4967                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4968                 key = (const struct dt_key *)&oti->oti_quota_id;
4969         }
4970
4971         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4972         osd_ipd_put(env, bag, ipd);
4973         LINVRNT(osd_invariant(obj));
4974         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4975         RETURN(rc);
4976 }
4977
4978 static int osd_index_declare_ea_delete(const struct lu_env *env,
4979                                        struct dt_object *dt,
4980                                        const struct dt_key *key,
4981                                        struct thandle *handle)
4982 {
4983         struct osd_thandle *oh;
4984         struct inode *inode;
4985         int rc, credits;
4986
4987         ENTRY;
4988
4989         LASSERT(!dt_object_remote(dt));
4990         LASSERT(handle != NULL);
4991
4992         oh = container_of0(handle, struct osd_thandle, ot_super);
4993         LASSERT(oh->ot_handle == NULL);
4994
4995         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4996         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4997
4998         inode = osd_dt_obj(dt)->oo_inode;
4999         if (inode == NULL)
5000                 RETURN(-ENOENT);
5001
5002         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5003                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5004                                    NULL, OSD_QID_BLK);
5005         RETURN(rc);
5006 }
5007
5008 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5009                                           struct dt_rec *fid)
5010 {
5011         struct osd_fid_pack *rec;
5012         int rc = -ENODATA;
5013
5014         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5015                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5016                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5017                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5018                         rc = -EINVAL;
5019         }
5020         return rc;
5021 }
5022
5023 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5024                           const struct lu_fid *fid)
5025 {
5026         struct seq_server_site *ss = osd_seq_site(osd);
5027
5028         ENTRY;
5029
5030         /* FID seqs not in FLDB, must be local seq */
5031         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5032                 RETURN(0);
5033
5034         /*
5035          * If FLD is not being initialized yet, it only happens during the
5036          * initialization, likely during mgs initialization, and we assume
5037          * this is local FID.
5038          */
5039         if (ss == NULL || ss->ss_server_fld == NULL)
5040                 RETURN(0);
5041
5042         /* Only check the local FLDB here */
5043         if (osd_seq_exists(env, osd, fid_seq(fid)))
5044                 RETURN(0);
5045
5046         RETURN(1);
5047 }
5048
5049 static void osd_take_care_of_agent(const struct lu_env *env,
5050                                    struct osd_device *osd,
5051                                    struct osd_thandle *oh,
5052                                    struct ldiskfs_dir_entry_2 *de)
5053 {
5054         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5055         struct osd_idmap_cache *idc;
5056         int rc, schedule = 0;
5057
5058         LASSERT(de != NULL);
5059
5060         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5061         if (likely(rc == 0)) {
5062                 idc = osd_idc_find_or_init(env, osd, fid);
5063                 if (IS_ERR(idc) || idc->oic_remote)
5064                         schedule = 1;
5065         } else if (rc == -ENODATA) {
5066                 /*
5067                  * can't get FID, postpone to the end of the
5068                  * transaction when iget() is safe
5069                  */
5070                 schedule = 1;
5071         } else {
5072                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5073         }
5074         if (schedule)
5075                 osd_schedule_agent_inode_removal(env, oh,
5076                                                  le32_to_cpu(de->inode));
5077 }
5078
5079 /**
5080  * Index delete function for interoperability mode (b11826).
5081  * It will remove the directory entry added by osd_index_ea_insert().
5082  * This entry is needed to maintain name->fid mapping.
5083  *
5084  * \param key,  key i.e. file entry to be deleted
5085  *
5086  * \retval   0, on success
5087  * \retval -ve, on error
5088  */
5089 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5090                                const struct dt_key *key, struct thandle *handle)
5091 {
5092         struct osd_object *obj = osd_dt_obj(dt);
5093         struct inode *dir = obj->oo_inode;
5094         struct dentry *dentry;
5095         struct osd_thandle *oh;
5096         struct ldiskfs_dir_entry_2 *de = NULL;
5097         struct buffer_head *bh;
5098         struct htree_lock *hlock = NULL;
5099         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5100         int rc;
5101
5102         ENTRY;
5103
5104         if (!dt_object_exists(dt))
5105                 RETURN(-ENOENT);
5106
5107         LINVRNT(osd_invariant(obj));
5108         LASSERT(!dt_object_remote(dt));
5109         LASSERT(handle != NULL);
5110
5111         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5112
5113         oh = container_of(handle, struct osd_thandle, ot_super);
5114         LASSERT(oh->ot_handle != NULL);
5115         LASSERT(oh->ot_handle->h_transaction != NULL);
5116
5117         ll_vfs_dq_init(dir);
5118         dentry = osd_child_dentry_get(env, obj,
5119                                       (char *)key, strlen((char *)key));
5120
5121         if (obj->oo_hl_head != NULL) {
5122                 hlock = osd_oti_get(env)->oti_hlock;
5123                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5124                                    dir, LDISKFS_HLOCK_DEL);
5125         } else {
5126                 down_write(&obj->oo_ext_idx_sem);
5127         }
5128
5129         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5130         if (!IS_ERR(bh)) {
5131                 /*
5132                  * If this is not the ".." entry, it might be a remote DNE
5133                  * entry and  we need to check if the FID is for a remote
5134                  * MDT.  If the FID is  not in the directory entry (e.g.
5135                  * upgraded 1.8 filesystem without dirdata enabled) then
5136                  * we need to get the FID from the LMA. For a remote directory
5137                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5138                  * case.
5139                  *
5140                  * Delete the entry before the agent inode in order to
5141                  * simplify error handling.  At worst an error after deleting
5142                  * the entry first might leak the agent inode afterward. The
5143                  * reverse would need filesystem abort in case of error deleting
5144                  * the entry after the agent had been removed, or leave a
5145                  * dangling entry pointing at a random inode.
5146                  */
5147                 if (strcmp((char *)key, dotdot) != 0)
5148                         osd_take_care_of_agent(env, osd, oh, de);
5149                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5150                 brelse(bh);
5151         } else {
5152                 rc = PTR_ERR(bh);
5153         }
5154         if (hlock != NULL)
5155                 ldiskfs_htree_unlock(hlock);
5156         else
5157                 up_write(&obj->oo_ext_idx_sem);
5158
5159         GOTO(out, rc);
5160 out:
5161         LASSERT(osd_invariant(obj));
5162         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5163         RETURN(rc);
5164 }
5165
5166 /**
5167  *      Lookup index for \a key and copy record to \a rec.
5168  *
5169  *      \param  dt      osd index object
5170  *      \param  key     key for index
5171  *      \param  rec     record reference
5172  *
5173  *      \retval  +ve  success : exact mach
5174  *      \retval  0    return record with key not greater than \a key
5175  *      \retval -ve   failure
5176  */
5177 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5178                                 struct dt_rec *rec, const struct dt_key *key)
5179 {
5180         struct osd_object *obj = osd_dt_obj(dt);
5181         struct iam_path_descr *ipd;
5182         struct iam_container *bag = &obj->oo_dir->od_container;
5183         struct osd_thread_info *oti = osd_oti_get(env);
5184         struct iam_iterator *it = &oti->oti_idx_it;
5185         struct iam_rec *iam_rec;
5186         int rc;
5187
5188         ENTRY;
5189
5190         if (!dt_object_exists(dt))
5191                 RETURN(-ENOENT);
5192
5193         LASSERT(osd_invariant(obj));
5194         LASSERT(!dt_object_remote(dt));
5195         LASSERT(bag->ic_object == obj->oo_inode);
5196
5197         ipd = osd_idx_ipd_get(env, bag);
5198         if (IS_ERR(ipd))
5199                 RETURN(-ENOMEM);
5200
5201         /* got ipd now we can start iterator. */
5202         iam_it_init(it, bag, 0, ipd);
5203
5204         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5205                 /* swab quota uid/gid provided by caller */
5206                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5207                 key = (const struct dt_key *)&oti->oti_quota_id;
5208         }
5209
5210         rc = iam_it_get(it, (struct iam_key *)key);
5211         if (rc >= 0) {
5212                 if (S_ISDIR(obj->oo_inode->i_mode))
5213                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5214                 else
5215                         iam_rec = (struct iam_rec *)rec;
5216
5217                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5218
5219                 if (S_ISDIR(obj->oo_inode->i_mode))
5220                         osd_fid_unpack((struct lu_fid *)rec,
5221                                        (struct osd_fid_pack *)iam_rec);
5222                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5223                         osd_quota_unpack(obj, rec);
5224         }
5225
5226         iam_it_put(it);
5227         iam_it_fini(it);
5228         osd_ipd_put(env, bag, ipd);
5229
5230         LINVRNT(osd_invariant(obj));
5231
5232         RETURN(rc);
5233 }
5234
5235 static int osd_index_declare_iam_insert(const struct lu_env *env,
5236                                         struct dt_object *dt,
5237                                         const struct dt_rec *rec,
5238                                         const struct dt_key *key,
5239                                         struct thandle *handle)
5240 {
5241         struct osd_thandle *oh;
5242
5243         LASSERT(handle != NULL);
5244
5245         oh = container_of0(handle, struct osd_thandle, ot_super);
5246         LASSERT(oh->ot_handle == NULL);
5247
5248         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5249                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5250
5251         return 0;
5252 }
5253
5254 /**
5255  *      Inserts (key, value) pair in \a dt index object.
5256  *
5257  *      \param  dt      osd index object
5258  *      \param  key     key for index
5259  *      \param  rec     record reference
5260  *      \param  th      transaction handler
5261  *
5262  *      \retval  0  success
5263  *      \retval -ve failure
5264  */
5265 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5266                                 const struct dt_rec *rec,
5267                                 const struct dt_key *key, struct thandle *th)
5268 {
5269         struct osd_object *obj = osd_dt_obj(dt);
5270         struct iam_path_descr *ipd;
5271         struct osd_thandle *oh;
5272         struct iam_container *bag;
5273         struct osd_thread_info *oti = osd_oti_get(env);
5274         struct iam_rec *iam_rec;
5275         int rc;
5276
5277         ENTRY;
5278
5279         if (!dt_object_exists(dt))
5280                 RETURN(-ENOENT);
5281
5282         LINVRNT(osd_invariant(obj));
5283         LASSERT(!dt_object_remote(dt));
5284
5285         bag = &obj->oo_dir->od_container;
5286         LASSERT(bag->ic_object == obj->oo_inode);
5287         LASSERT(th != NULL);
5288
5289         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5290
5291         ipd = osd_idx_ipd_get(env, bag);
5292         if (unlikely(ipd == NULL))
5293                 RETURN(-ENOMEM);
5294
5295         oh = container_of0(th, struct osd_thandle, ot_super);
5296         LASSERT(oh->ot_handle != NULL);
5297         LASSERT(oh->ot_handle->h_transaction != NULL);
5298         if (S_ISDIR(obj->oo_inode->i_mode)) {
5299                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5300                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5301                              &oti->oti_fid);
5302         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5303                 /* pack quota uid/gid */
5304                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5305                 key = (const struct dt_key *)&oti->oti_quota_id;
5306                 /* pack quota record */
5307                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5308                 iam_rec = (struct iam_rec *)rec;
5309         } else {
5310                 iam_rec = (struct iam_rec *)rec;
5311         }
5312
5313         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5314                         iam_rec, ipd);
5315         osd_ipd_put(env, bag, ipd);
5316         LINVRNT(osd_invariant(obj));
5317         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5318         RETURN(rc);
5319 }
5320
5321 /**
5322  * Calls ldiskfs_add_entry() to add directory entry
5323  * into the directory. This is required for
5324  * interoperability mode (b11826)
5325  *
5326  * \retval   0, on success
5327  * \retval -ve, on error
5328  */
5329 static int __osd_ea_add_rec(struct osd_thread_info *info,
5330                             struct osd_object *pobj, struct inode  *cinode,
5331                             const char *name, const struct lu_fid *fid,
5332                             struct htree_lock *hlock, struct thandle *th)
5333 {
5334         struct ldiskfs_dentry_param *ldp;
5335         struct dentry *child;
5336         struct osd_thandle *oth;
5337         int rc;
5338
5339         oth = container_of(th, struct osd_thandle, ot_super);
5340         LASSERT(oth->ot_handle != NULL);
5341         LASSERT(oth->ot_handle->h_transaction != NULL);
5342         LASSERT(pobj->oo_inode);
5343
5344         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5345         if (unlikely(pobj->oo_inode ==
5346                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5347                 ldp->edp_magic = 0;
5348         else
5349                 osd_get_ldiskfs_dirent_param(ldp, fid);
5350         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5351         child->d_fsdata = (void *)ldp;
5352         ll_vfs_dq_init(pobj->oo_inode);
5353         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5354                                    child, cinode, hlock);
5355         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5356                 struct ldiskfs_dir_entry_2      *de;
5357                 struct buffer_head              *bh;
5358                 int                              rc1;
5359
5360                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5361                                             NULL, hlock);
5362                 if (!IS_ERR(bh)) {
5363                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5364                                                                bh);
5365                         if (rc1 == 0) {
5366                                 if (S_ISDIR(cinode->i_mode))
5367                                         de->file_type = LDISKFS_DIRENT_LUFID |
5368                                                         LDISKFS_FT_REG_FILE;
5369                                 else
5370                                         de->file_type = LDISKFS_DIRENT_LUFID |
5371                                                         LDISKFS_FT_DIR;
5372                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5373                                                               NULL, bh);
5374                         }
5375                         brelse(bh);
5376                 }
5377         }
5378
5379         RETURN(rc);
5380 }
5381
5382 /**
5383  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5384  * into the directory.Also sets flags into osd object to
5385  * indicate dot and dotdot are created. This is required for
5386  * interoperability mode (b11826)
5387  *
5388  * \param dir   directory for dot and dotdot fixup.
5389  * \param obj   child object for linking
5390  *
5391  * \retval   0, on success
5392  * \retval -ve, on error
5393  */
5394 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5395                               struct osd_object *dir,
5396                               struct inode *parent_dir, const char *name,
5397                               const struct lu_fid *dot_fid,
5398                               const struct lu_fid *dot_dot_fid,
5399                               struct thandle *th)
5400 {
5401         struct inode *inode = dir->oo_inode;
5402         struct osd_thandle *oth;
5403         int result = 0;
5404
5405         oth = container_of(th, struct osd_thandle, ot_super);
5406         LASSERT(oth->ot_handle->h_transaction != NULL);
5407         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5408
5409         if (strcmp(name, dot) == 0) {
5410                 if (dir->oo_compat_dot_created) {
5411                         result = -EEXIST;
5412                 } else {
5413                         LASSERT(inode->i_ino == parent_dir->i_ino);
5414                         dir->oo_compat_dot_created = 1;
5415                         result = 0;
5416                 }
5417         } else if (strcmp(name, dotdot) == 0) {
5418                 if (!dir->oo_compat_dot_created)
5419                         return -EINVAL;
5420                 /* in case of rename, dotdot is already created */
5421                 if (dir->oo_compat_dotdot_created) {
5422                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5423                                                 dot_dot_fid, NULL, th);
5424                 }
5425
5426                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5427                         struct lu_fid tfid = *dot_dot_fid;
5428
5429                         tfid.f_oid--;
5430                         result = osd_add_dot_dotdot_internal(info,
5431                                         dir->oo_inode, parent_dir, dot_fid,
5432                                         &tfid, oth);
5433                 } else {
5434                         result = osd_add_dot_dotdot_internal(info,
5435                                         dir->oo_inode, parent_dir, dot_fid,
5436                                         dot_dot_fid, oth);
5437                 }
5438
5439                 if (result == 0)
5440                         dir->oo_compat_dotdot_created = 1;
5441         }
5442
5443         return result;
5444 }
5445
5446
5447 /**
5448  * It will call the appropriate osd_add* function and return the
5449  * value, return by respective functions.
5450  */
5451 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5452                           struct inode *cinode, const char *name,
5453                           const struct lu_fid *fid, struct thandle *th)
5454 {
5455         struct osd_thread_info *info = osd_oti_get(env);
5456         struct htree_lock *hlock;
5457         int rc;
5458
5459         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5460
5461         if (name[0] == '.' && (name[1] == '\0' ||
5462                                (name[1] == '.' && name[2] == '\0'))) {
5463                 if (hlock != NULL) {
5464                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5465                                            pobj->oo_inode, 0);
5466                 } else {
5467                         down_write(&pobj->oo_ext_idx_sem);
5468                 }
5469
5470                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5471                                         lu_object_fid(&pobj->oo_dt.do_lu),
5472                                         fid, th);
5473         } else {
5474                 if (hlock != NULL) {
5475                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5476                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5477                 } else {
5478                         down_write(&pobj->oo_ext_idx_sem);
5479                 }
5480
5481                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5482                         struct lu_fid *tfid = &info->oti_fid;
5483
5484                         *tfid = *fid;
5485                         tfid->f_ver = ~0;
5486                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5487                                               tfid, hlock, th);
5488                 } else {
5489                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5490                                               hlock, th);
5491                 }
5492         }
5493         if (hlock != NULL)
5494                 ldiskfs_htree_unlock(hlock);
5495         else
5496                 up_write(&pobj->oo_ext_idx_sem);
5497
5498         return rc;
5499 }
5500
5501 static int
5502 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5503                       struct osd_idmap_cache *oic)
5504 {
5505         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5506         struct lu_fid *fid = &oic->oic_fid;
5507         struct osd_inode_id *id = &oic->oic_lid;
5508         struct inode *inode = NULL;
5509         int once = 0;
5510         bool insert;
5511         int rc;
5512
5513         ENTRY;
5514
5515         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5516                 RETURN(0);
5517
5518         if (thread_is_running(&scrub->os_thread) &&
5519             scrub->os_pos_current > id->oii_ino)
5520                 RETURN(0);
5521
5522         if (dev->od_auto_scrub_interval == AS_NEVER ||
5523             ktime_get_real_seconds() <
5524             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5525                 RETURN(0);
5526
5527 again:
5528         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5529         if (rc == -ENOENT) {
5530                 __u32 gen = id->oii_gen;
5531
5532                 insert = true;
5533                 if (inode != NULL)
5534                         goto trigger;
5535
5536                 inode = osd_iget(oti, dev, id);
5537                 /* The inode has been removed (by race maybe). */
5538                 if (IS_ERR(inode)) {
5539                         rc = PTR_ERR(inode);
5540
5541                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5542                 }
5543
5544                 /* The OI mapping is lost. */
5545                 if (gen != OSD_OII_NOGEN)
5546                         goto trigger;
5547
5548                 /*
5549                  * The inode may has been reused by others, we do not know,
5550                  * leave it to be handled by subsequent osd_fid_lookup().
5551                  */
5552                 GOTO(out, rc = 0);
5553         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5554                 GOTO(out, rc);
5555         }
5556
5557         insert = false;
5558
5559 trigger:
5560         if (thread_is_running(&scrub->os_thread)) {
5561                 if (inode == NULL) {
5562                         inode = osd_iget(oti, dev, id);
5563                         /* The inode has been removed (by race maybe). */
5564                         if (IS_ERR(inode)) {
5565                                 rc = PTR_ERR(inode);
5566
5567                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5568                         }
5569                 }
5570
5571                 rc = osd_oii_insert(dev, oic, insert);
5572                 /*
5573                  * There is race condition between osd_oi_lookup and OI scrub.
5574                  * The OI scrub finished just after osd_oi_lookup() failure.
5575                  * Under such case, it is unnecessary to trigger OI scrub again,
5576                  * but try to call osd_oi_lookup() again.
5577                  */
5578                 if (unlikely(rc == -EAGAIN))
5579                         goto again;
5580
5581                 if (!S_ISDIR(inode->i_mode))
5582                         rc = 0;
5583                 else
5584                         rc = osd_check_lmv(oti, dev, inode, oic);
5585
5586                 GOTO(out, rc);
5587         }
5588
5589         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5590                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5591                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5592                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5593                        "%s: trigger partial OI scrub for RPC inconsistency "
5594                        "checking FID "DFID": rc = %d\n",
5595                        osd_dev2name(dev), PFID(fid), rc);
5596                 if (rc == 0 || rc == -EALREADY)
5597                         goto again;
5598         }
5599
5600         GOTO(out, rc);
5601
5602 out:
5603         if (inode)
5604                 iput(inode);
5605
5606         RETURN(rc);
5607 }
5608
5609 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5610                                struct osd_device *dev,
5611                                struct osd_idmap_cache *oic,
5612                                struct lu_fid *fid, __u32 ino)
5613 {
5614         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5615         struct inode *inode;
5616         int rc;
5617
5618         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5619         inode = osd_iget(oti, dev, &oic->oic_lid);
5620         if (IS_ERR(inode)) {
5621                 fid_zero(&oic->oic_fid);
5622                 return PTR_ERR(inode);
5623         }
5624
5625         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5626         iput(inode);
5627         if (rc != 0)
5628                 fid_zero(&oic->oic_fid);
5629         else
5630                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5631         return rc;
5632 }
5633
5634 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5635                       struct osd_inode_id *id, const struct lu_fid *fid)
5636 {
5637         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5638                id->oii_ino, id->oii_gen, info);
5639         info->oti_cache.oic_lid = *id;
5640         info->oti_cache.oic_fid = *fid;
5641         info->oti_cache.oic_dev = osd;
5642 }
5643
5644 /**
5645  * Get parent FID from the linkEA.
5646  *
5647  * For a directory which parent resides on remote MDT, to satisfy the
5648  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5649  * the other hand, to make the lookup(..) on the directory can return
5650  * the real parent FID, we append the real parent FID after its ".."
5651  * name entry in the /REMOTE_PARENT_DIR.
5652  *
5653  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5654  * backup. So after the restore, we cannot get the right parent FID from
5655  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5656  * we have stored the real parent FID in the directory object's linkEA,
5657  * we can parse the linkEA for the real parent FID.
5658  *
5659  * \param[in] env       pointer to the thread context
5660  * \param[in] obj       pointer to the object to be handled
5661  * \param[out]fid       pointer to the buffer to hold the parent FID
5662  *
5663  * \retval              0 for getting the real parent FID successfully
5664  * \retval              negative error number on failure
5665  */
5666 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5667                                     struct osd_object *obj,
5668                                     struct lu_fid *fid)
5669 {
5670         struct osd_thread_info *oti = osd_oti_get(env);
5671         struct lu_buf *buf = &oti->oti_big_buf;
5672         struct dentry *dentry = &oti->oti_obj_dentry;
5673         struct inode *inode = obj->oo_inode;
5674         struct linkea_data ldata = { NULL };
5675         int rc;
5676
5677         ENTRY;
5678
5679         fid_zero(fid);
5680         if (!S_ISDIR(inode->i_mode))
5681                 RETURN(-EIO);
5682
5683 again:
5684         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5685                              buf->lb_buf, buf->lb_len);
5686         if (rc == -ERANGE) {
5687                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5688                                      NULL, 0);
5689                 if (rc > 0) {
5690                         lu_buf_realloc(buf, rc);
5691                         if (buf->lb_buf == NULL)
5692                                 RETURN(-ENOMEM);
5693
5694                         goto again;
5695                 }
5696         }
5697
5698         if (unlikely(rc == 0))
5699                 RETURN(-ENODATA);
5700
5701         if (rc < 0)
5702                 RETURN(rc);
5703
5704         if (unlikely(buf->lb_buf == NULL)) {
5705                 lu_buf_realloc(buf, rc);
5706                 if (buf->lb_buf == NULL)
5707                         RETURN(-ENOMEM);
5708
5709                 goto again;
5710         }
5711
5712         ldata.ld_buf = buf;
5713         rc = linkea_init_with_rec(&ldata);
5714         if (!rc) {
5715                 linkea_first_entry(&ldata);
5716                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5717         }
5718
5719         RETURN(rc);
5720 }
5721
5722 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5723                                     struct inode *inode,
5724                                     const struct lu_fid *pfid,
5725                                     const char *name, const int namelen)
5726 {
5727         struct osd_thread_info *oti = osd_oti_get(env);
5728         struct lu_buf *buf = &oti->oti_big_buf;
5729         struct dentry *dentry = &oti->oti_obj_dentry;
5730         struct linkea_data ldata = { NULL };
5731         struct lu_name cname = { .ln_name = name,
5732                                  .ln_namelen = namelen };
5733         int rc;
5734
5735         ENTRY;
5736
5737 again:
5738         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5739                              buf->lb_buf, buf->lb_len);
5740         if (rc == -ERANGE)
5741                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5742
5743         if (rc < 0)
5744                 RETURN(rc);
5745
5746         if (unlikely(rc == 0))
5747                 RETURN(-ENODATA);
5748
5749         if (buf->lb_len < rc) {
5750                 lu_buf_realloc(buf, rc);
5751                 if (buf->lb_buf == NULL)
5752                         RETURN(-ENOMEM);
5753
5754                 goto again;
5755         }
5756
5757         ldata.ld_buf = buf;
5758         rc = linkea_init_with_rec(&ldata);
5759         if (!rc)
5760                 rc = linkea_links_find(&ldata, &cname, pfid);
5761
5762         RETURN(rc);
5763 }
5764
5765 /**
5766  * Calls ->lookup() to find dentry. From dentry get inode and
5767  * read inode's ea to get fid. This is required for  interoperability
5768  * mode (b11826)
5769  *
5770  * \retval   0, on success
5771  * \retval -ve, on error
5772  */
5773 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5774                              struct dt_rec *rec, const struct dt_key *key)
5775 {
5776         struct inode *dir = obj->oo_inode;
5777         struct dentry *dentry;
5778         struct ldiskfs_dir_entry_2 *de;
5779         struct buffer_head *bh;
5780         struct lu_fid *fid = (struct lu_fid *)rec;
5781         struct htree_lock *hlock = NULL;
5782         int ino;
5783         int rc;
5784
5785         ENTRY;
5786
5787         LASSERT(dir->i_op != NULL);
5788         LASSERT(dir->i_op->lookup != NULL);
5789
5790         dentry = osd_child_dentry_get(env, obj,
5791                                       (char *)key, strlen((char *)key));
5792
5793         if (obj->oo_hl_head != NULL) {
5794                 hlock = osd_oti_get(env)->oti_hlock;
5795                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5796                                    dir, LDISKFS_HLOCK_LOOKUP);
5797         } else {
5798                 down_read(&obj->oo_ext_idx_sem);
5799         }
5800
5801         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5802         if (!IS_ERR(bh)) {
5803                 struct osd_thread_info *oti = osd_oti_get(env);
5804                 struct osd_inode_id *id = &oti->oti_id;
5805                 struct osd_idmap_cache *oic = &oti->oti_cache;
5806                 struct osd_device *dev = osd_obj2dev(obj);
5807
5808                 ino = le32_to_cpu(de->inode);
5809                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5810                         brelse(bh);
5811                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5812                         GOTO(out, rc);
5813                 }
5814
5815                 rc = osd_get_fid_from_dentry(de, rec);
5816
5817                 /* done with de, release bh */
5818                 brelse(bh);
5819                 if (rc != 0) {
5820                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5821                                 const char *name = (const char *)key;
5822
5823                                 /*
5824                                  * If the parent is on remote MDT, and there
5825                                  * is no FID-in-dirent, then we have to get
5826                                  * the parent FID from the linkEA.
5827                                  */
5828                                 if (likely(strlen(name) == 2 &&
5829                                            name[0] == '.' && name[1] == '.'))
5830                                         rc = osd_get_pfid_from_linkea(env, obj,
5831                                                                       fid);
5832                         } else {
5833                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5834                         }
5835                 } else {
5836                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5837                 }
5838
5839                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5840                         fid_zero(&oic->oic_fid);
5841
5842                         GOTO(out, rc);
5843                 }
5844
5845                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5846                 rc = osd_consistency_check(oti, dev, oic);
5847                 if (rc == -ENOENT)
5848                         fid_zero(&oic->oic_fid);
5849                 else
5850                         /* Other error should not affect lookup result. */
5851                         rc = 0;
5852         } else {
5853                 rc = PTR_ERR(bh);
5854         }
5855
5856         GOTO(out, rc);
5857
5858 out:
5859         if (hlock != NULL)
5860                 ldiskfs_htree_unlock(hlock);
5861         else
5862                 up_read(&obj->oo_ext_idx_sem);
5863         return rc;
5864 }
5865
5866 static int osd_index_declare_ea_insert(const struct lu_env *env,
5867                                        struct dt_object *dt,
5868                                        const struct dt_rec *rec,
5869                                        const struct dt_key *key,
5870                                        struct thandle *handle)
5871 {
5872         struct osd_thandle *oh;
5873         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5874         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5875         const struct lu_fid *fid = rec1->rec_fid;
5876         int credits, rc = 0;
5877         struct osd_idmap_cache *idc;
5878
5879         ENTRY;
5880
5881         LASSERT(!dt_object_remote(dt));
5882         LASSERT(handle != NULL);
5883         LASSERT(fid != NULL);
5884         LASSERT(rec1->rec_type != 0);
5885
5886         oh = container_of0(handle, struct osd_thandle, ot_super);
5887         LASSERT(oh->ot_handle == NULL);
5888
5889         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5890
5891         /*
5892          * we can't call iget() while a transactions is running
5893          * (this can lead to a deadlock), but we need to know
5894          * inum and object type. so we find this information at
5895          * declaration and cache in per-thread info
5896          */
5897         idc = osd_idc_find_or_init(env, osd, fid);
5898         if (IS_ERR(idc))
5899                 RETURN(PTR_ERR(idc));
5900         if (idc->oic_remote) {
5901                 /*
5902                  * a reference to remote inode is represented by an
5903                  * agent inode which we have to create
5904                  */
5905                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5906                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5907         }
5908
5909         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5910
5911         if (osd_dt_obj(dt)->oo_inode != NULL) {
5912                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5913
5914                 /*
5915                  * We ignore block quota on meta pool (MDTs), so needn't
5916                  * calculate how many blocks will be consumed by this index
5917                  * insert
5918                  */
5919                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5920                                            i_gid_read(inode),
5921                                            i_projid_read(inode), 0,
5922                                            oh, osd_dt_obj(dt), NULL,
5923                                            OSD_QID_BLK);
5924                 if (rc)
5925                         RETURN(rc);
5926
5927 #ifdef HAVE_PROJECT_QUOTA
5928                 /*
5929                  * Reserve credits for local agent inode to transfer
5930                  * to 0, quota enforcement is ignored in this case.
5931                  */
5932                 if (idc->oic_remote &&
5933                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5934                     i_projid_read(inode) != 0)
5935                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5936                                                   0, i_projid_read(inode),
5937                                                   0, false, PRJQUOTA, true);
5938 #endif
5939         }
5940
5941         RETURN(rc);
5942 }
5943
5944 /**
5945  * Index add function for interoperability mode (b11826).
5946  * It will add the directory entry.This entry is needed to
5947  * maintain name->fid mapping.
5948  *
5949  * \param key it is key i.e. file entry to be inserted
5950  * \param rec it is value of given key i.e. fid
5951  *
5952  * \retval   0, on success
5953  * \retval -ve, on error
5954  */
5955 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5956                                const struct dt_rec *rec,
5957                                const struct dt_key *key, struct thandle *th)
5958 {
5959         struct osd_object *obj = osd_dt_obj(dt);
5960         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5961         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5962         const struct lu_fid *fid = rec1->rec_fid;
5963         const char *name = (const char *)key;
5964         struct osd_thread_info *oti = osd_oti_get(env);
5965         struct inode *child_inode = NULL;
5966         struct osd_idmap_cache *idc;
5967         int rc;
5968
5969         ENTRY;
5970
5971         if (!dt_object_exists(dt))
5972                 RETURN(-ENOENT);
5973
5974         LASSERT(osd_invariant(obj));
5975         LASSERT(!dt_object_remote(dt));
5976         LASSERT(th != NULL);
5977
5978         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5979
5980         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5981
5982         idc = osd_idc_find(env, osd, fid);
5983         if (unlikely(idc == NULL)) {
5984                 idc = osd_idc_find_or_init(env, osd, fid);
5985                 if (IS_ERR(idc)) {
5986                         /*
5987                          * this dt_insert() wasn't declared properly, so
5988                          * FID is missing in OI cache. we better do not
5989                          * lookup FID in FLDB/OI and don't risk to deadlock,
5990                          * but in some special cases (lfsck testing, etc)
5991                          * it's much simpler than fixing a caller.
5992                          *
5993                          * normally this error should be placed after the first
5994                          * find, but migrate may attach source stripes to
5995                          * target, which doesn't create stripes.
5996                          */
5997                         CERROR("%s: "DFID" wasn't declared for insert\n",
5998                                osd_name(osd), PFID(fid));
5999                         dump_stack();
6000                         RETURN(PTR_ERR(idc));
6001                 }
6002         }
6003
6004         if (idc->oic_remote) {
6005                 /* Insert remote entry */
6006                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6007                         child_inode =
6008                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6009                 } else {
6010                         child_inode = osd_create_local_agent_inode(env, osd,
6011                                         obj, fid, rec1->rec_type & S_IFMT, th);
6012                         if (IS_ERR(child_inode))
6013                                 RETURN(PTR_ERR(child_inode));
6014                 }
6015         } else {
6016                 /* Insert local entry */
6017                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6018                         /* for a reason OI cache wasn't filled properly */
6019                         CERROR("%s: OIC for "DFID" isn't filled\n",
6020                                osd_name(osd), PFID(fid));
6021                         RETURN(-EINVAL);
6022                 }
6023                 child_inode = oti->oti_inode;
6024                 if (unlikely(child_inode == NULL)) {
6025                         struct ldiskfs_inode_info *lii;
6026
6027                         OBD_ALLOC_PTR(lii);
6028                         if (lii == NULL)
6029                                 RETURN(-ENOMEM);
6030                         child_inode = oti->oti_inode = &lii->vfs_inode;
6031                 }
6032                 child_inode->i_sb = osd_sb(osd);
6033                 child_inode->i_ino = idc->oic_lid.oii_ino;
6034                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6035         }
6036
6037         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6038
6039         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6040                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6041
6042         if (child_inode && child_inode != oti->oti_inode)
6043                 iput(child_inode);
6044         LASSERT(osd_invariant(obj));
6045         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6046         RETURN(rc);
6047 }
6048
6049 /**
6050  *  Initialize osd Iterator for given osd index object.
6051  *
6052  *  \param  dt      osd index object
6053  */
6054
6055 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6056                                      struct dt_object *dt,
6057                                      __u32 unused)
6058 {
6059         struct osd_it_iam *it;
6060         struct osd_object *obj = osd_dt_obj(dt);
6061         struct lu_object *lo = &dt->do_lu;
6062         struct iam_path_descr *ipd;
6063         struct iam_container *bag = &obj->oo_dir->od_container;
6064
6065         if (!dt_object_exists(dt))
6066                 return ERR_PTR(-ENOENT);
6067
6068         OBD_ALLOC_PTR(it);
6069         if (it == NULL)
6070                 return ERR_PTR(-ENOMEM);
6071
6072         ipd = osd_it_ipd_get(env, bag);
6073         if (likely(ipd != NULL)) {
6074                 it->oi_obj = obj;
6075                 it->oi_ipd = ipd;
6076                 lu_object_get(lo);
6077                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6078                 return (struct dt_it *)it;
6079         } else {
6080                 OBD_FREE_PTR(it);
6081                 return ERR_PTR(-ENOMEM);
6082         }
6083 }
6084
6085 /**
6086  * free given Iterator.
6087  */
6088 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6089 {
6090         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6091         struct osd_object *obj = it->oi_obj;
6092
6093         iam_it_fini(&it->oi_it);
6094         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6095         osd_object_put(env, obj);
6096         OBD_FREE_PTR(it);
6097 }
6098
6099 /**
6100  *  Move Iterator to record specified by \a key
6101  *
6102  *  \param  di      osd iterator
6103  *  \param  key     key for index
6104  *
6105  *  \retval +ve  di points to record with least key not larger than key
6106  *  \retval  0   di points to exact matched key
6107  *  \retval -ve  failure
6108  */
6109
6110 static int osd_it_iam_get(const struct lu_env *env,
6111                           struct dt_it *di, const struct dt_key *key)
6112 {
6113         struct osd_thread_info *oti = osd_oti_get(env);
6114         struct osd_it_iam *it = (struct osd_it_iam *)di;
6115
6116         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6117                 /* swab quota uid/gid */
6118                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6119                 key = (struct dt_key *)&oti->oti_quota_id;
6120         }
6121
6122         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6123 }
6124
6125 /**
6126  *  Release Iterator
6127  *
6128  *  \param  di      osd iterator
6129  */
6130 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6131 {
6132         struct osd_it_iam *it = (struct osd_it_iam *)di;
6133
6134         iam_it_put(&it->oi_it);
6135 }
6136
6137 /**
6138  *  Move iterator by one record
6139  *
6140  *  \param  di      osd iterator
6141  *
6142  *  \retval +1   end of container reached
6143  *  \retval  0   success
6144  *  \retval -ve  failure
6145  */
6146
6147 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6148 {
6149         struct osd_it_iam *it = (struct osd_it_iam *)di;
6150
6151         return iam_it_next(&it->oi_it);
6152 }
6153
6154 /**
6155  * Return pointer to the key under iterator.
6156  */
6157
6158 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6159                                      const struct dt_it *di)
6160 {
6161         struct osd_thread_info *oti = osd_oti_get(env);
6162         struct osd_it_iam *it = (struct osd_it_iam *)di;
6163         struct osd_object *obj = it->oi_obj;
6164         struct dt_key *key;
6165
6166         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6167
6168         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6169                 /* swab quota uid/gid */
6170                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6171                 key = (struct dt_key *)&oti->oti_quota_id;
6172         }
6173
6174         return key;
6175 }
6176
6177 /**
6178  * Return size of key under iterator (in bytes)
6179  */
6180
6181 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6182 {
6183         struct osd_it_iam *it = (struct osd_it_iam *)di;
6184
6185         return iam_it_key_size(&it->oi_it);
6186 }
6187
6188 static inline void
6189 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6190 {
6191         /* check if file type is required */
6192         if (ent->lde_attrs & LUDA_TYPE) {
6193                 struct luda_type *lt;
6194                 int align = sizeof(*lt) - 1;
6195
6196                 len = (len + align) & ~align;
6197                 lt = (struct luda_type *)(ent->lde_name + len);
6198                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6199         }
6200
6201         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6202 }
6203
6204 /**
6205  * build lu direct from backend fs dirent.
6206  */
6207
6208 static inline void
6209 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6210                    char *name, __u16 namelen, __u16 type, __u32 attr)
6211 {
6212         ent->lde_attrs = attr | LUDA_FID;
6213         fid_cpu_to_le(&ent->lde_fid, fid);
6214
6215         ent->lde_hash = cpu_to_le64(offset);
6216         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6217
6218         strncpy(ent->lde_name, name, namelen);
6219         ent->lde_name[namelen] = '\0';
6220         ent->lde_namelen = cpu_to_le16(namelen);
6221
6222         /* append lustre attributes */
6223         osd_it_append_attrs(ent, namelen, type);
6224 }
6225
6226 /**
6227  * Return pointer to the record under iterator.
6228  */
6229 static int osd_it_iam_rec(const struct lu_env *env,
6230                           const struct dt_it *di,
6231                           struct dt_rec *dtrec, __u32 attr)
6232 {
6233         struct osd_it_iam *it = (struct osd_it_iam *)di;
6234         struct osd_thread_info *info = osd_oti_get(env);
6235
6236         ENTRY;
6237
6238         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6239                 const struct osd_fid_pack *rec;
6240                 struct lu_fid *fid = &info->oti_fid;
6241                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6242                 char *name;
6243                 int namelen;
6244                 __u64 hash;
6245                 int rc;
6246
6247                 name = (char *)iam_it_key_get(&it->oi_it);
6248                 if (IS_ERR(name))
6249                         RETURN(PTR_ERR(name));
6250
6251                 namelen = iam_it_key_size(&it->oi_it);
6252
6253                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6254                 if (IS_ERR(rec))
6255                         RETURN(PTR_ERR(rec));
6256
6257                 rc = osd_fid_unpack(fid, rec);
6258                 if (rc)
6259                         RETURN(rc);
6260
6261                 hash = iam_it_store(&it->oi_it);
6262
6263                 /* IAM does not store object type in IAM index (dir) */
6264                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6265                                    0, LUDA_FID);
6266         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6267                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6268                            (struct iam_rec *)dtrec);
6269                 osd_quota_unpack(it->oi_obj, dtrec);
6270         } else {
6271                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6272                            (struct iam_rec *)dtrec);
6273         }
6274
6275         RETURN(0);
6276 }
6277
6278 /**
6279  * Returns cookie for current Iterator position.
6280  */
6281 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6282 {
6283         struct osd_it_iam *it = (struct osd_it_iam *)di;
6284
6285         return iam_it_store(&it->oi_it);
6286 }
6287
6288 /**
6289  * Restore iterator from cookie.
6290  *
6291  * \param  di      osd iterator
6292  * \param  hash    Iterator location cookie
6293  *
6294  * \retval +ve  di points to record with least key not larger than key.
6295  * \retval  0   di points to exact matched key
6296  * \retval -ve  failure
6297  */
6298
6299 static int osd_it_iam_load(const struct lu_env *env,
6300                            const struct dt_it *di, __u64 hash)
6301 {
6302         struct osd_it_iam *it = (struct osd_it_iam *)di;
6303
6304         return iam_it_load(&it->oi_it, hash);
6305 }
6306
6307 static const struct dt_index_operations osd_index_iam_ops = {
6308         .dio_lookup         = osd_index_iam_lookup,
6309         .dio_declare_insert = osd_index_declare_iam_insert,
6310         .dio_insert         = osd_index_iam_insert,
6311         .dio_declare_delete = osd_index_declare_iam_delete,
6312         .dio_delete         = osd_index_iam_delete,
6313         .dio_it     = {
6314                 .init     = osd_it_iam_init,
6315                 .fini     = osd_it_iam_fini,
6316                 .get      = osd_it_iam_get,
6317                 .put      = osd_it_iam_put,
6318                 .next     = osd_it_iam_next,
6319                 .key      = osd_it_iam_key,
6320                 .key_size = osd_it_iam_key_size,
6321                 .rec      = osd_it_iam_rec,
6322                 .store    = osd_it_iam_store,
6323                 .load     = osd_it_iam_load
6324         }
6325 };
6326
6327
6328 /**
6329  * Creates or initializes iterator context.
6330  *
6331  * \retval struct osd_it_ea, iterator structure on success
6332  *
6333  */
6334 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6335                                     struct dt_object *dt,
6336                                     __u32 attr)
6337 {
6338         struct osd_object *obj = osd_dt_obj(dt);
6339         struct osd_thread_info *info = osd_oti_get(env);
6340         struct osd_it_ea *oie;
6341         struct file *file;
6342         struct lu_object *lo = &dt->do_lu;
6343         struct dentry *obj_dentry;
6344
6345         ENTRY;
6346
6347         if (!dt_object_exists(dt) || obj->oo_destroyed)
6348                 RETURN(ERR_PTR(-ENOENT));
6349
6350         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6351         if (oie == NULL)
6352                 RETURN(ERR_PTR(-ENOMEM));
6353         obj_dentry = &oie->oie_dentry;
6354
6355         obj_dentry->d_inode = obj->oo_inode;
6356         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6357         obj_dentry->d_name.hash = 0;
6358
6359         oie->oie_rd_dirent       = 0;
6360         oie->oie_it_dirent       = 0;
6361         oie->oie_dirent          = NULL;
6362         if (unlikely(!info->oti_it_ea_buf_used)) {
6363                 oie->oie_buf = info->oti_it_ea_buf;
6364                 info->oti_it_ea_buf_used = 1;
6365         } else {
6366                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6367                 if (oie->oie_buf == NULL)
6368                         RETURN(ERR_PTR(-ENOMEM));
6369         }
6370         oie->oie_obj = obj;
6371
6372         file = &oie->oie_file;
6373
6374         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6375         if (attr & LUDA_64BITHASH)
6376                 file->f_mode    = FMODE_64BITHASH;
6377         else
6378                 file->f_mode    = FMODE_32BITHASH;
6379         file->f_path.dentry     = obj_dentry;
6380         file->f_mapping         = obj->oo_inode->i_mapping;
6381         file->f_op              = obj->oo_inode->i_fop;
6382         set_file_inode(file, obj->oo_inode);
6383
6384         lu_object_get(lo);
6385         RETURN((struct dt_it *)oie);
6386 }
6387
6388 /**
6389  * Destroy or finishes iterator context.
6390  *
6391  * \param di iterator structure to be destroyed
6392  */
6393 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6394 {
6395         struct osd_thread_info *info = osd_oti_get(env);
6396         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6397         struct osd_object *obj = oie->oie_obj;
6398         struct inode *inode = obj->oo_inode;
6399
6400         ENTRY;
6401         oie->oie_file.f_op->release(inode, &oie->oie_file);
6402         osd_object_put(env, obj);
6403         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6404                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6405         else
6406                 info->oti_it_ea_buf_used = 0;
6407         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6408         EXIT;
6409 }
6410
6411 /**
6412  * It position the iterator at given key, so that next lookup continues from
6413  * that key Or it is similar to dio_it->load() but based on a key,
6414  * rather than file position.
6415  *
6416  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6417  * to the beginning.
6418  *
6419  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6420  */
6421 static int osd_it_ea_get(const struct lu_env *env,
6422                          struct dt_it *di, const struct dt_key *key)
6423 {
6424         struct osd_it_ea *it = (struct osd_it_ea *)di;
6425
6426         ENTRY;
6427         LASSERT(((const char *)key)[0] == '\0');
6428         it->oie_file.f_pos = 0;
6429         it->oie_rd_dirent = 0;
6430         it->oie_it_dirent = 0;
6431         it->oie_dirent = NULL;
6432
6433         RETURN(+1);
6434 }
6435
6436 /**
6437  * Does nothing
6438  */
6439 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6440 {
6441 }
6442
6443 struct osd_filldir_cbs {
6444 #ifdef HAVE_DIR_CONTEXT
6445         struct dir_context ctx;
6446 #endif
6447         struct osd_it_ea  *it;
6448 };
6449 /**
6450  * It is called internally by ->readdir(). It fills the
6451  * iterator's in-memory data structure with required
6452  * information i.e. name, namelen, rec_size etc.
6453  *
6454  * \param buf in which information to be filled in.
6455  * \param name name of the file in given dir
6456  *
6457  * \retval 0 on success
6458  * \retval 1 on buffer full
6459  */
6460 #ifdef HAVE_FILLDIR_USE_CTX
6461 static int osd_ldiskfs_filldir(struct dir_context *buf,
6462 #else
6463 static int osd_ldiskfs_filldir(void *buf,
6464 #endif
6465                                const char *name, int namelen,
6466                                loff_t offset, __u64 ino, unsigned int d_type)
6467 {
6468         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6469         struct osd_object *obj = it->oie_obj;
6470         struct osd_it_ea_dirent *ent = it->oie_dirent;
6471         struct lu_fid *fid = &ent->oied_fid;
6472         struct osd_fid_pack *rec;
6473
6474         ENTRY;
6475
6476 /* this should never happen */
6477         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6478                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6479                 RETURN(-EIO);
6480         }
6481
6482         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6483             OSD_IT_EA_BUFSIZE)
6484                 RETURN(1);
6485
6486         /* "." is just the object itself. */
6487         if (namelen == 1 && name[0] == '.') {
6488                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6489         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6490                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6491                 if (osd_fid_unpack(fid, rec) != 0)
6492                         fid_zero(fid);
6493         } else {
6494                 fid_zero(fid);
6495         }
6496         d_type &= ~LDISKFS_DIRENT_LUFID;
6497
6498         /* NOT export local root. */
6499         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6500                 ino = obj->oo_inode->i_ino;
6501                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6502         }
6503
6504         ent->oied_ino     = ino;
6505         ent->oied_off     = offset;
6506         ent->oied_namelen = namelen;
6507         ent->oied_type    = d_type;
6508
6509         memcpy(ent->oied_name, name, namelen);
6510
6511         it->oie_rd_dirent++;
6512         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6513         RETURN(0);
6514 }
6515
6516 /**
6517  * Calls ->readdir() to load a directory entry at a time
6518  * and stored it in iterator's in-memory data structure.
6519  *
6520  * \param di iterator's in memory structure
6521  *
6522  * \retval   0 on success
6523  * \retval -ve on error
6524  * \retval +1 reach the end of entry
6525  */
6526 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6527                                const struct dt_it *di)
6528 {
6529         struct osd_it_ea *it = (struct osd_it_ea *)di;
6530         struct osd_object *obj = it->oie_obj;
6531         struct inode *inode = obj->oo_inode;
6532         struct htree_lock *hlock = NULL;
6533         struct file *filp = &it->oie_file;
6534         int rc = 0;
6535         struct osd_filldir_cbs buf = {
6536 #ifdef HAVE_DIR_CONTEXT
6537                 .ctx.actor = osd_ldiskfs_filldir,
6538 #endif
6539                 .it = it
6540         };
6541
6542         ENTRY;
6543         it->oie_dirent = it->oie_buf;
6544         it->oie_rd_dirent = 0;
6545
6546         if (obj->oo_hl_head != NULL) {
6547                 hlock = osd_oti_get(env)->oti_hlock;
6548                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6549                                    inode, LDISKFS_HLOCK_READDIR);
6550         } else {
6551                 down_read(&obj->oo_ext_idx_sem);
6552         }
6553
6554 #ifdef HAVE_DIR_CONTEXT
6555         buf.ctx.pos = filp->f_pos;
6556 #ifdef HAVE_ITERATE_SHARED
6557         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6558 #else
6559         rc = inode->i_fop->iterate(filp, &buf.ctx);
6560 #endif
6561         filp->f_pos = buf.ctx.pos;
6562 #else
6563         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6564 #endif
6565
6566         if (hlock != NULL)
6567                 ldiskfs_htree_unlock(hlock);
6568         else
6569                 up_read(&obj->oo_ext_idx_sem);
6570
6571         if (it->oie_rd_dirent == 0) {
6572                 /*
6573                  * If it does not get any dirent, it means it has been reached
6574                  * to the end of the dir
6575                  */
6576                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6577                 if (rc == 0)
6578                         rc = 1;
6579         } else {
6580                 it->oie_dirent = it->oie_buf;
6581                 it->oie_it_dirent = 1;
6582         }
6583
6584         RETURN(rc);
6585 }
6586
6587 /**
6588  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6589  * to load a directory entry at a time and stored it in
6590  * iterator's in-memory data structure.
6591  *
6592  * \param di iterator's in memory structure
6593  *
6594  * \retval +ve iterator reached to end
6595  * \retval   0 iterator not reached to end
6596  * \retval -ve on error
6597  */
6598 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6599 {
6600         struct osd_it_ea *it = (struct osd_it_ea *)di;
6601         int rc;
6602
6603         ENTRY;
6604
6605         if (it->oie_it_dirent < it->oie_rd_dirent) {
6606                 it->oie_dirent =
6607                         (void *)it->oie_dirent +
6608                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6609                                        it->oie_dirent->oied_namelen);
6610                 it->oie_it_dirent++;
6611                 rc = 0;
6612         } else {
6613                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6614                         rc = 1;
6615                 else
6616                         rc = osd_ldiskfs_it_fill(env, di);
6617         }
6618
6619         RETURN(rc);
6620 }
6621
6622 /**
6623  * Returns the key at current position from iterator's in memory structure.
6624  *
6625  * \param di iterator's in memory structure
6626  *
6627  * \retval key i.e. struct dt_key on success
6628  */
6629 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6630                                     const struct dt_it *di)
6631 {
6632         struct osd_it_ea *it = (struct osd_it_ea *)di;
6633
6634         return (struct dt_key *)it->oie_dirent->oied_name;
6635 }
6636
6637 /**
6638  * Returns key's size at current position from iterator's in memory structure.
6639  *
6640  * \param di iterator's in memory structure
6641  *
6642  * \retval key_size i.e. struct dt_key on success
6643  */
6644 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6645 {
6646         struct osd_it_ea *it = (struct osd_it_ea *)di;
6647
6648         return it->oie_dirent->oied_namelen;
6649 }
6650
6651 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6652 {
6653         if (LDISKFS_DIR_REC_LEN(de) >=
6654             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6655                 return true;
6656
6657         return false;
6658 }
6659
6660 static inline bool
6661 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6662                      unsigned int blocksize, bool dotdot)
6663 {
6664         if (dotdot)
6665                 return osd_dotdot_has_space(de);
6666
6667         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6668             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6669                 return true;
6670
6671         return false;
6672 }
6673
6674 static int
6675 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6676                     handle_t *jh, struct dentry *dentry,
6677                     const struct lu_fid *fid, struct buffer_head *bh,
6678                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6679                     bool dotdot)
6680 {
6681         struct inode *dir = dentry->d_parent->d_inode;
6682         struct inode *inode = dentry->d_inode;
6683         struct osd_fid_pack *rec;
6684         struct ldiskfs_dentry_param *ldp;
6685         int namelen = dentry->d_name.len;
6686         int rc;
6687         struct osd_thread_info *info = osd_oti_get(env);
6688
6689         ENTRY;
6690
6691         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6692                 RETURN(0);
6693
6694         /* There is enough space to hold the FID-in-dirent. */
6695         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6696                 rc = ldiskfs_journal_get_write_access(jh, bh);
6697                 if (rc != 0)
6698                         RETURN(rc);
6699
6700                 de->name[namelen] = 0;
6701                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6702                 rec->fp_len = sizeof(struct lu_fid) + 1;
6703                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6704                 de->file_type |= LDISKFS_DIRENT_LUFID;
6705                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6706
6707                 RETURN(rc);
6708         }
6709
6710         LASSERT(!dotdot);
6711
6712         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6713         if (rc != 0)
6714                 RETURN(rc);
6715
6716         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6717         osd_get_ldiskfs_dirent_param(ldp, fid);
6718         dentry->d_fsdata = (void *)ldp;
6719         ll_vfs_dq_init(dir);
6720         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6721         /*
6722          * It is too bad, we cannot reinsert the name entry back.
6723          * That means we lose it!
6724          */
6725         if (rc != 0)
6726                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6727                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6728                        osd_ino2name(inode),
6729                        dir->i_ino, dir->i_generation, namelen,
6730                        dentry->d_name.name, PFID(fid), rc);
6731
6732         RETURN(rc);
6733 }
6734
6735 static int
6736 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6737                         struct osd_it_ea *it, struct lu_fid *fid,
6738                         struct osd_inode_id *id, __u32 *attr)
6739 {
6740         struct osd_thread_info *info = osd_oti_get(env);
6741         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6742         struct osd_device *dev = osd_obj2dev(obj);
6743         struct super_block *sb = osd_sb(dev);
6744         const char *devname = osd_name(dev);
6745         struct osd_it_ea_dirent *ent = it->oie_dirent;
6746         struct inode *dir = obj->oo_inode;
6747         struct htree_lock *hlock = NULL;
6748         struct buffer_head *bh = NULL;
6749         handle_t *jh = NULL;
6750         struct ldiskfs_dir_entry_2 *de;
6751         struct dentry *dentry;
6752         struct inode *inode;
6753         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6754         int credits;
6755         int rc;
6756         bool dotdot = false;
6757         bool dirty = false;
6758
6759         ENTRY;
6760
6761         if (ent->oied_name[0] == '.') {
6762                 if (ent->oied_namelen == 1)
6763                         RETURN(0);
6764
6765                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6766                         dotdot = true;
6767         }
6768
6769         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6770         inode = osd_iget(info, dev, id);
6771         if (IS_ERR(inode)) {
6772                 rc = PTR_ERR(inode);
6773                 if (rc == -ENOENT || rc == -ESTALE) {
6774                         /*
6775                          * Maybe dangling name entry, or
6776                          * corrupted directory entry.
6777                          */
6778                         *attr |= LUDA_UNKNOWN;
6779                         rc = 0;
6780                 } else {
6781                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6782                                "check_repair, dir = %lu/%u, name = %.*s, "
6783                                "ino = %llu, rc = %d\n",
6784                                devname, dir->i_ino, dir->i_generation,
6785                                ent->oied_namelen, ent->oied_name,
6786                                ent->oied_ino, rc);
6787                 }
6788
6789                 RETURN(rc);
6790         }
6791
6792         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6793                                            ent->oied_namelen);
6794         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6795         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6796                 lma = NULL;
6797         else if (rc != 0)
6798                 GOTO(out, rc);
6799
6800         /*
6801          * We need to ensure that the name entry is still valid.
6802          * Because it may be removed or renamed by other already.
6803          *
6804          * The unlink or rename operation will start journal before PDO lock,
6805          * so to avoid deadlock, here we need to start journal handle before
6806          * related PDO lock also. But because we do not know whether there
6807          * will be something to be repaired before PDO lock, we just start
6808          * journal without conditions.
6809          *
6810          * We may need to remove the name entry firstly, then insert back.
6811          * One credit is for user quota file update.
6812          * One credit is for group quota file update.
6813          * Two credits are for dirty inode.
6814          */
6815         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6816                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6817
6818         if (dev->od_dirent_journal != 0) {
6819
6820 again:
6821                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6822                 if (IS_ERR(jh)) {
6823                         rc = PTR_ERR(jh);
6824                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6825                                "check_repair, dir = %lu/%u, credits = %d, "
6826                                "name = %.*s, ino = %llu: rc = %d\n",
6827                                devname, dir->i_ino, dir->i_generation, credits,
6828                                ent->oied_namelen, ent->oied_name,
6829                                ent->oied_ino, rc);
6830
6831                         GOTO(out_inode, rc);
6832                 }
6833
6834                 if (obj->oo_hl_head != NULL) {
6835                         hlock = osd_oti_get(env)->oti_hlock;
6836                         /*
6837                          * "0" means exclusive lock for the whole directory.
6838                          * We need to prevent others access such name entry
6839                          * during the delete + insert. Neither HLOCK_ADD nor
6840                          * HLOCK_DEL cannot guarantee the atomicity.
6841                          */
6842                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6843                 } else {
6844                         down_write(&obj->oo_ext_idx_sem);
6845                 }
6846         } else {
6847                 if (obj->oo_hl_head != NULL) {
6848                         hlock = osd_oti_get(env)->oti_hlock;
6849                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6850                                            LDISKFS_HLOCK_LOOKUP);
6851                 } else {
6852                         down_read(&obj->oo_ext_idx_sem);
6853                 }
6854         }
6855
6856         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6857         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6858                 *attr |= LUDA_IGNORE;
6859
6860                 GOTO(out, rc = 0);
6861         }
6862
6863         /*
6864          * For dotdot entry, if there is not enough space to hold the
6865          * FID-in-dirent, just keep them there. It only happens when the
6866          * device upgraded from 1.8 or restored from MDT file-level backup.
6867          * For the whole directory, only dotdot entry have no FID-in-dirent
6868          * and needs to get FID from LMA when readdir, it will not affect the
6869          * performance much.
6870          */
6871         if (dotdot && !osd_dotdot_has_space(de)) {
6872                 *attr |= LUDA_UNKNOWN;
6873
6874                 GOTO(out, rc = 0);
6875         }
6876
6877         if (lma != NULL) {
6878                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6879                         GOTO(out, rc = 0);
6880
6881                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6882                         struct lu_fid *tfid = &lma->lma_self_fid;
6883
6884                         if (likely(dotdot &&
6885                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6886                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6887                                 /*
6888                                  * It must be REMOTE_PARENT_DIR and as the
6889                                  * 'dotdot' entry of remote directory
6890                                  */
6891                                 *attr |= LUDA_IGNORE;
6892                         } else {
6893                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6894                                        "parent directory, but got %.*s under "
6895                                        "dir = %lu/%u with the FID "DFID"\n",
6896                                        devname, ent->oied_namelen,
6897                                        ent->oied_name, dir->i_ino,
6898                                        dir->i_generation, PFID(tfid));
6899
6900                                 *attr |= LUDA_UNKNOWN;
6901                         }
6902
6903                         GOTO(out, rc = 0);
6904                 }
6905         }
6906
6907         if (!fid_is_zero(fid)) {
6908                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6909                                               ent->oied_namelen);
6910                 if (rc == -ENOENT ||
6911                     (rc == -ENODATA &&
6912                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6913                         /*
6914                          * linkEA does not recognize the dirent entry,
6915                          * it may because the dirent entry corruption
6916                          * and points to other's inode.
6917                          */
6918                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6919                                "recognize the dirent, dir = %lu/%u, "
6920                                " name = %.*s, ino = %llu, "
6921                                DFID": rc = %d\n", devname, dir->i_ino,
6922                                dir->i_generation, ent->oied_namelen,
6923                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6924                         *attr |= LUDA_UNKNOWN;
6925
6926                         GOTO(out, rc = 0);
6927                 }
6928
6929                 if (rc && rc != -ENODATA) {
6930                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6931                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6932                                DFID": rc = %d\n", devname, dir->i_ino,
6933                                dir->i_generation, ent->oied_namelen,
6934                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6935                         *attr |= LUDA_UNKNOWN;
6936
6937                         GOTO(out, rc = 0);
6938                 }
6939         }
6940
6941         if (lma != NULL) {
6942                 /*
6943                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6944                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6945                  */
6946                 if (*attr & LUDA_VERIFY_DRYRUN) {
6947                         *fid = lma->lma_self_fid;
6948                         *attr |= LUDA_REPAIR;
6949
6950                         GOTO(out, rc = 0);
6951                 }
6952
6953                 if (jh == NULL) {
6954                         brelse(bh);
6955                         dev->od_dirent_journal = 1;
6956                         if (hlock != NULL) {
6957                                 ldiskfs_htree_unlock(hlock);
6958                                 hlock = NULL;
6959                         } else {
6960                                 up_read(&obj->oo_ext_idx_sem);
6961                         }
6962
6963                         goto again;
6964                 }
6965
6966                 *fid = lma->lma_self_fid;
6967                 dirty = true;
6968                 /* Update or append the FID-in-dirent. */
6969                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6970                                          bh, de, hlock, dotdot);
6971                 if (rc == 0)
6972                         *attr |= LUDA_REPAIR;
6973                 else
6974                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6975                                "the dirent, dir = %lu/%u, name = %.*s, "
6976                                "ino = %llu, "DFID": rc = %d\n",
6977                                devname, dir->i_ino, dir->i_generation,
6978                                ent->oied_namelen, ent->oied_name,
6979                                ent->oied_ino, PFID(fid), rc);
6980         } else {
6981                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6982                 if (*attr & LUDA_VERIFY_DRYRUN) {
6983                         if (fid_is_sane(fid)) {
6984                                 *attr |= LUDA_REPAIR;
6985                         } else if (dev->od_index == 0) {
6986                                 lu_igif_build(fid, inode->i_ino,
6987                                               inode->i_generation);
6988                                 *attr |= LUDA_UPGRADE;
6989                         }
6990
6991                         GOTO(out, rc = 0);
6992                 }
6993
6994                 if (jh == NULL) {
6995                         brelse(bh);
6996                         dev->od_dirent_journal = 1;
6997                         if (hlock != NULL) {
6998                                 ldiskfs_htree_unlock(hlock);
6999                                 hlock = NULL;
7000                         } else {
7001                                 up_read(&obj->oo_ext_idx_sem);
7002                         }
7003
7004                         goto again;
7005                 }
7006
7007                 dirty = true;
7008                 if (unlikely(fid_is_sane(fid))) {
7009                         /*
7010                          * FID-in-dirent exists, but FID-in-LMA is lost.
7011                          * Trust the FID-in-dirent, and add FID-in-LMA.
7012                          */
7013                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7014                         if (rc == 0)
7015                                 *attr |= LUDA_REPAIR;
7016                         else
7017                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7018                                        "update dirent, dir = %lu/%u, "
7019                                        "name = %.*s, ino = %llu, "
7020                                        DFID": rc = %d\n",
7021                                        devname, dir->i_ino, dir->i_generation,
7022                                        ent->oied_namelen, ent->oied_name,
7023                                        ent->oied_ino, PFID(fid), rc);
7024                 } else if (dev->od_index == 0) {
7025                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7026                         /*
7027                          * It is probably IGIF object. Only aappend the
7028                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7029                          */
7030                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7031                                                  bh, de, hlock, dotdot);
7032                         if (rc == 0)
7033                                 *attr |= LUDA_UPGRADE;
7034                         else
7035                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7036                                        "after the dirent, dir = %lu/%u, "
7037                                        "name = %.*s, ino = %llu, "
7038                                        DFID": rc = %d\n",
7039                                        devname, dir->i_ino, dir->i_generation,
7040                                        ent->oied_namelen, ent->oied_name,
7041                                        ent->oied_ino, PFID(fid), rc);
7042                 }
7043         }
7044
7045         GOTO(out, rc);
7046
7047 out:
7048         if (!IS_ERR(bh))
7049                 brelse(bh);
7050         if (hlock != NULL) {
7051                 ldiskfs_htree_unlock(hlock);
7052         } else {
7053                 if (dev->od_dirent_journal != 0)
7054                         up_write(&obj->oo_ext_idx_sem);
7055                 else
7056                         up_read(&obj->oo_ext_idx_sem);
7057         }
7058
7059         if (jh != NULL)
7060                 ldiskfs_journal_stop(jh);
7061
7062 out_inode:
7063         iput(inode);
7064         if (rc >= 0 && !dirty)
7065                 dev->od_dirent_journal = 0;
7066
7067         return rc;
7068 }
7069
7070 /**
7071  * Returns the value at current position from iterator's in memory structure.
7072  *
7073  * \param di struct osd_it_ea, iterator's in memory structure
7074  * \param attr attr requested for dirent.
7075  * \param lde lustre dirent
7076  *
7077  * \retval   0 no error and \param lde has correct lustre dirent.
7078  * \retval -ve on error
7079  */
7080 static inline int osd_it_ea_rec(const struct lu_env *env,
7081                                 const struct dt_it *di,
7082                                 struct dt_rec *dtrec, __u32 attr)
7083 {
7084         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7085         struct osd_object      *obj   = it->oie_obj;
7086         struct osd_device      *dev   = osd_obj2dev(obj);
7087         struct osd_thread_info *oti   = osd_oti_get(env);
7088         struct osd_inode_id    *id    = &oti->oti_id;
7089         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7090         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7091         __u32 ino = it->oie_dirent->oied_ino;
7092         int rc = 0;
7093
7094         ENTRY;
7095
7096         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7097
7098         if (attr & LUDA_VERIFY) {
7099                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7100                         attr |= LUDA_IGNORE;
7101                         /*
7102                          * If the parent is on remote MDT, and there
7103                          * is no FID-in-dirent, then we have to get
7104                          * the parent FID from the linkEA.
7105                          */
7106                         if (!fid_is_sane(fid) &&
7107                             it->oie_dirent->oied_namelen == 2 &&
7108                             it->oie_dirent->oied_name[0] == '.' &&
7109                             it->oie_dirent->oied_name[1] == '.')
7110                                 osd_get_pfid_from_linkea(env, obj, fid);
7111                 } else {
7112                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7113                                                      &attr);
7114                 }
7115
7116                 if (!fid_is_sane(fid))
7117                         attr |= LUDA_UNKNOWN;
7118         } else {
7119                 attr &= ~LU_DIRENT_ATTRS_MASK;
7120                 if (!fid_is_sane(fid)) {
7121                         bool is_dotdot = false;
7122
7123                         if (it->oie_dirent->oied_namelen == 2 &&
7124                             it->oie_dirent->oied_name[0] == '.' &&
7125                             it->oie_dirent->oied_name[1] == '.')
7126                                 is_dotdot = true;
7127                         /*
7128                          * If the parent is on remote MDT, and there
7129                          * is no FID-in-dirent, then we have to get
7130                          * the parent FID from the linkEA.
7131                          */
7132                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7133                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7134                         } else {
7135                                 if (is_dotdot == false &&
7136                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7137                                         RETURN(-ENOENT);
7138
7139                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7140                         }
7141                 } else {
7142                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7143                 }
7144         }
7145
7146         /* Pack the entry anyway, at least the offset is right. */
7147         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7148                            it->oie_dirent->oied_name,
7149                            it->oie_dirent->oied_namelen,
7150                            it->oie_dirent->oied_type, attr);
7151
7152         if (rc < 0)
7153                 RETURN(rc);
7154
7155         if (osd_remote_fid(env, dev, fid))
7156                 RETURN(0);
7157
7158         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7159                 osd_add_oi_cache(oti, dev, id, fid);
7160
7161         RETURN(rc > 0 ? 0 : rc);
7162 }
7163
7164 /**
7165  * Returns the record size size at current position.
7166  *
7167  * This function will return record(lu_dirent) size in bytes.
7168  *
7169  * \param[in] env       execution environment
7170  * \param[in] di        iterator's in memory structure
7171  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7172  *                      calculate the lu_dirent size.
7173  *
7174  * \retval      record size(in bytes & in memory) of the current lu_dirent
7175  *              entry.
7176  */
7177 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7178                               __u32 attr)
7179 {
7180         struct osd_it_ea *it = (struct osd_it_ea *)di;
7181
7182         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7183 }
7184
7185 /**
7186  * Returns a cookie for current position of the iterator head, so that
7187  * user can use this cookie to load/start the iterator next time.
7188  *
7189  * \param di iterator's in memory structure
7190  *
7191  * \retval cookie for current position, on success
7192  */
7193 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7194 {
7195         struct osd_it_ea *it = (struct osd_it_ea *)di;
7196
7197         return it->oie_dirent->oied_off;
7198 }
7199
7200 /**
7201  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7202  * to load a directory entry at a time and stored it i inn,
7203  * in iterator's in-memory data structure.
7204  *
7205  * \param di struct osd_it_ea, iterator's in memory structure
7206  *
7207  * \retval +ve on success
7208  * \retval -ve on error
7209  */
7210 static int osd_it_ea_load(const struct lu_env *env,
7211                           const struct dt_it *di, __u64 hash)
7212 {
7213         struct osd_it_ea *it = (struct osd_it_ea *)di;
7214         int rc;
7215
7216         ENTRY;
7217         it->oie_file.f_pos = hash;
7218
7219         rc =  osd_ldiskfs_it_fill(env, di);
7220         if (rc > 0)
7221                 rc = -ENODATA;
7222
7223         if (rc == 0)
7224                 rc = 1;
7225
7226         RETURN(rc);
7227 }
7228
7229 /**
7230  * Index lookup function for interoperability mode (b11826).
7231  *
7232  * \param key,  key i.e. file name to be searched
7233  *
7234  * \retval +ve, on success
7235  * \retval -ve, on error
7236  */
7237 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7238                                struct dt_rec *rec, const struct dt_key *key)
7239 {
7240         struct osd_object *obj = osd_dt_obj(dt);
7241         int rc = 0;
7242
7243         ENTRY;
7244
7245         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7246         LINVRNT(osd_invariant(obj));
7247
7248         rc = osd_ea_lookup_rec(env, obj, rec, key);
7249         if (rc == 0)
7250                 rc = 1;
7251         RETURN(rc);
7252 }
7253
7254 /**
7255  * Index and Iterator operations for interoperability
7256  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7257  */
7258 static const struct dt_index_operations osd_index_ea_ops = {
7259         .dio_lookup         = osd_index_ea_lookup,
7260         .dio_declare_insert = osd_index_declare_ea_insert,
7261         .dio_insert         = osd_index_ea_insert,
7262         .dio_declare_delete = osd_index_declare_ea_delete,
7263         .dio_delete         = osd_index_ea_delete,
7264         .dio_it     = {
7265                 .init     = osd_it_ea_init,
7266                 .fini     = osd_it_ea_fini,
7267                 .get      = osd_it_ea_get,
7268                 .put      = osd_it_ea_put,
7269                 .next     = osd_it_ea_next,
7270                 .key      = osd_it_ea_key,
7271                 .key_size = osd_it_ea_key_size,
7272                 .rec      = osd_it_ea_rec,
7273                 .rec_size = osd_it_ea_rec_size,
7274                 .store    = osd_it_ea_store,
7275                 .load     = osd_it_ea_load
7276         }
7277 };
7278
7279 static void *osd_key_init(const struct lu_context *ctx,
7280                           struct lu_context_key *key)
7281 {
7282         struct osd_thread_info *info;
7283
7284         OBD_ALLOC_PTR(info);
7285         if (info == NULL)
7286                 return ERR_PTR(-ENOMEM);
7287
7288         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7289         if (info->oti_it_ea_buf == NULL)
7290                 goto out_free_info;
7291
7292         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7293
7294         info->oti_hlock = ldiskfs_htree_lock_alloc();
7295         if (info->oti_hlock == NULL)
7296                 goto out_free_ea;
7297
7298         return info;
7299
7300 out_free_ea:
7301         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7302 out_free_info:
7303         OBD_FREE_PTR(info);
7304         return ERR_PTR(-ENOMEM);
7305 }
7306
7307 static void osd_key_fini(const struct lu_context *ctx,
7308                          struct lu_context_key *key, void *data)
7309 {
7310         struct osd_thread_info *info = data;
7311         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7312         struct osd_idmap_cache *idc = info->oti_ins_cache;
7313
7314         if (info->oti_dio_pages) {
7315                 int i;
7316                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7317                         if (info->oti_dio_pages[i])
7318                                 __free_page(info->oti_dio_pages[i]);
7319                 }
7320                 OBD_FREE(info->oti_dio_pages,
7321                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7322         }
7323
7324         if (info->oti_inode != NULL)
7325                 OBD_FREE_PTR(lli);
7326         if (info->oti_hlock != NULL)
7327                 ldiskfs_htree_lock_free(info->oti_hlock);
7328         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7329         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7330         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7331         lu_buf_free(&info->oti_big_buf);
7332         if (idc != NULL) {
7333                 LASSERT(info->oti_ins_cache_size > 0);
7334                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7335                 info->oti_ins_cache = NULL;
7336                 info->oti_ins_cache_size = 0;
7337         }
7338         OBD_FREE_PTR(info);
7339 }
7340
7341 static void osd_key_exit(const struct lu_context *ctx,
7342                          struct lu_context_key *key, void *data)
7343 {
7344         struct osd_thread_info *info = data;
7345
7346         LASSERT(info->oti_r_locks == 0);
7347         LASSERT(info->oti_w_locks == 0);
7348         LASSERT(info->oti_txns    == 0);
7349 }
7350
7351 /* type constructor/destructor: osd_type_init, osd_type_fini */
7352 LU_TYPE_INIT_FINI(osd, &osd_key);
7353
7354 struct lu_context_key osd_key = {
7355         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7356         .lct_init = osd_key_init,
7357         .lct_fini = osd_key_fini,
7358         .lct_exit = osd_key_exit
7359 };
7360
7361
7362 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7363                            const char *name, struct lu_device *next)
7364 {
7365         struct osd_device *osd = osd_dev(d);
7366
7367         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7368             sizeof(osd->od_svname))
7369                 return -E2BIG;
7370         return osd_procfs_init(osd, name);
7371 }
7372
7373 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7374 {
7375         struct seq_server_site *ss = osd_seq_site(osd);
7376         int rc;
7377
7378         ENTRY;
7379
7380         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7381                 RETURN(0);
7382
7383         if (unlikely(ss == NULL))
7384                 RETURN(-ENODEV);
7385
7386         OBD_ALLOC_PTR(osd->od_cl_seq);
7387         if (osd->od_cl_seq == NULL)
7388                 RETURN(-ENOMEM);
7389
7390         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7391                              osd->od_svname, ss->ss_server_seq);
7392         if (rc != 0) {
7393                 OBD_FREE_PTR(osd->od_cl_seq);
7394                 osd->od_cl_seq = NULL;
7395                 RETURN(rc);
7396         }
7397
7398         if (ss->ss_node_id == 0) {
7399                 /*
7400                  * If the OSD on the sequence controller(MDT0), then allocate
7401                  * sequence here, otherwise allocate sequence after connected
7402                  * to MDT0 (see mdt_register_lwp_callback()).
7403                  */
7404                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7405                                    &osd->od_cl_seq->lcs_space, env);
7406         }
7407
7408         RETURN(rc);
7409 }
7410
7411 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7412 {
7413         if (osd->od_cl_seq == NULL)
7414                 return;
7415
7416         seq_client_fini(osd->od_cl_seq);
7417         OBD_FREE_PTR(osd->od_cl_seq);
7418         osd->od_cl_seq = NULL;
7419 }
7420
7421 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7422 {
7423         ENTRY;
7424
7425         /* shutdown quota slave instance associated with the device */
7426         if (o->od_quota_slave != NULL) {
7427                 struct qsd_instance *qsd = o->od_quota_slave;
7428
7429                 o->od_quota_slave = NULL;
7430                 qsd_fini(env, qsd);
7431         }
7432
7433         osd_fid_fini(env, o);
7434         osd_scrub_cleanup(env, o);
7435
7436         RETURN(0);
7437 }
7438
7439 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7440 {
7441         ENTRY;
7442
7443         if (o->od_mnt != NULL) {
7444                 shrink_dcache_sb(osd_sb(o));
7445                 osd_sync(env, &o->od_dt_dev);
7446
7447                 mntput(o->od_mnt);
7448                 o->od_mnt = NULL;
7449         }
7450
7451         EXIT;
7452 }
7453
7454 static int osd_mount(const struct lu_env *env,
7455                      struct osd_device *o, struct lustre_cfg *cfg)
7456 {
7457         const char *name = lustre_cfg_string(cfg, 0);
7458         const char *dev = lustre_cfg_string(cfg, 1);
7459         const char *opts;
7460         unsigned long page, s_flags, lmd_flags = 0;
7461         struct page *__page;
7462         struct file_system_type *type;
7463         char *options = NULL;
7464         char *str;
7465         struct osd_thread_info *info = osd_oti_get(env);
7466         struct lu_fid *fid = &info->oti_fid;
7467         struct inode *inode;
7468         int rc = 0, force_over_512tb = 0;
7469
7470         ENTRY;
7471
7472         if (o->od_mnt != NULL)
7473                 RETURN(0);
7474
7475         if (strlen(dev) >= sizeof(o->od_mntdev))
7476                 RETURN(-E2BIG);
7477         strcpy(o->od_mntdev, dev);
7478
7479         str = lustre_cfg_string(cfg, 2);
7480         s_flags = simple_strtoul(str, NULL, 0);
7481         str = strstr(str, ":");
7482         if (str)
7483                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7484         opts = lustre_cfg_string(cfg, 3);
7485 #ifdef __BIG_ENDIAN
7486         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7487                 CERROR("%s: device %s extents feature is not guaranteed to "
7488                        "work on big-endian systems. Use \"bigendian_extents\" "
7489                        "mount option to override.\n", name, dev);
7490                 RETURN(-EINVAL);
7491         }
7492 #endif
7493 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7494         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7495                 CWARN("force_over_128tb option is deprecated. "
7496                       "Filesystems less than 512TB can be created without any "
7497                       "force options. Use force_over_512tb option for "
7498                       "filesystems greater than 512TB.\n");
7499         }
7500 #endif
7501 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7502         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7503                 CWARN("force_over_256tb option is deprecated. "
7504                       "Filesystems less than 512TB can be created without any "
7505                       "force options. Use force_over_512tb option for "
7506                       "filesystems greater than 512TB.\n");
7507         }
7508 #endif
7509
7510         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7511                 force_over_512tb = 1;
7512
7513         __page = alloc_page(GFP_KERNEL);
7514         if (__page == NULL)
7515                 GOTO(out, rc = -ENOMEM);
7516         page = (unsigned long)page_address(__page);
7517         options = (char *)page;
7518         *options = '\0';
7519         if (opts != NULL) {
7520                 /* strip out the options for back compatiblity */
7521                 static char *sout[] = {
7522                         "mballoc",
7523                         "iopen",
7524                         "noiopen",
7525                         "iopen_nopriv",
7526                         "extents",
7527                         "noextents",
7528                         /* strip out option we processed in osd */
7529                         "bigendian_extents",
7530                         "force_over_128tb",
7531                         "force_over_256tb",
7532                         "force_over_512tb",
7533                         NULL
7534                 };
7535                 strcat(options, opts);
7536                 for (rc = 0, str = options; sout[rc]; ) {
7537                         char *op = strstr(str, sout[rc]);
7538
7539                         if (op == NULL) {
7540                                 rc++;
7541                                 str = options;
7542                                 continue;
7543                         }
7544                         if (op == options || *(op - 1) == ',') {
7545                                 str = op + strlen(sout[rc]);
7546                                 if (*str == ',' || *str == '\0') {
7547                                         *str == ',' ? str++ : str;
7548                                         memmove(op, str, strlen(str) + 1);
7549                                 }
7550                         }
7551                         for (str = op; *str != ',' && *str != '\0'; str++)
7552                                 ;
7553                 }
7554         } else {
7555                 strncat(options, "user_xattr,acl", 14);
7556         }
7557
7558         /* Glom up mount options */
7559         if (*options != '\0')
7560                 strcat(options, ",");
7561         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7562
7563         type = get_fs_type("ldiskfs");
7564         if (!type) {
7565                 CERROR("%s: cannot find ldiskfs module\n", name);
7566                 GOTO(out, rc = -ENODEV);
7567         }
7568
7569         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7570         module_put(type->owner);
7571
7572         if (IS_ERR(o->od_mnt)) {
7573                 rc = PTR_ERR(o->od_mnt);
7574                 o->od_mnt = NULL;
7575                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7576                 GOTO(out, rc);
7577         }
7578
7579         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7580                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7581                                  force_over_512tb == 0) {
7582                 CERROR("%s: device %s LDISKFS does not support filesystems "
7583                        "greater than 512TB and can cause data corruption. "
7584                        "Use \"force_over_512tb\" mount option to override.\n",
7585                        name, dev);
7586                 GOTO(out_mnt, rc = -EINVAL);
7587         }
7588
7589         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7590                 if (priv_dev_set_rdonly) {
7591                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7592                         o->od_dt_dev.dd_rdonly = 1;
7593                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7594                                       name);
7595                 } else {
7596                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7597                                       name);
7598
7599                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7600                 }
7601         } else if (priv_dev_check_rdonly &&
7602                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7603                 CERROR("%s: underlying device %s is marked as "
7604                        "read-only. Setup failed\n", name, dev);
7605
7606                 GOTO(out_mnt, rc = -EROFS);
7607         }
7608
7609         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7610                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7611                 GOTO(out_mnt, rc = -EINVAL);
7612         }
7613
7614 #ifdef LDISKFS_MOUNT_DIRDATA
7615         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7616                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7617         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7618                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7619                       "enabling the dirdata feature. If you do not want to "
7620                       "downgrade to Lustre-1.x again, you can enable it via "
7621                       "'tune2fs -O dirdata device'\n", name, dev);
7622 #endif
7623         inode = osd_sb(o)->s_root->d_inode;
7624         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7625         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7626         if (rc != 0) {
7627                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7628                 GOTO(out_mnt, rc);
7629         }
7630
7631         if (lmd_flags & LMD_FLG_NOSCRUB)
7632                 o->od_auto_scrub_interval = AS_NEVER;
7633
7634         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7635                 /* do not use pagecache with flash-backed storage */
7636                 o->od_writethrough_cache = 0;
7637                 o->od_read_cache = 0;
7638         }
7639
7640         GOTO(out, rc = 0);
7641
7642 out_mnt:
7643         mntput(o->od_mnt);
7644         o->od_mnt = NULL;
7645
7646 out:
7647         if (__page)
7648                 __free_page(__page);
7649
7650         return rc;
7651 }
7652
7653 static struct lu_device *osd_device_fini(const struct lu_env *env,
7654                                          struct lu_device *d)
7655 {
7656         struct osd_device *o = osd_dev(d);
7657
7658         ENTRY;
7659
7660         osd_index_backup(env, o, false);
7661         osd_shutdown(env, o);
7662         osd_procfs_fini(o);
7663         osd_obj_map_fini(o);
7664         osd_umount(env, o);
7665
7666         RETURN(NULL);
7667 }
7668
7669 static int osd_device_init0(const struct lu_env *env,
7670                             struct osd_device *o,
7671                             struct lustre_cfg *cfg)
7672 {
7673         struct lu_device *l = osd2lu_dev(o);
7674         struct osd_thread_info *info;
7675         int rc;
7676         int cplen = 0;
7677
7678         /* if the module was re-loaded, env can loose its keys */
7679         rc = lu_env_refill((struct lu_env *)env);
7680         if (rc)
7681                 GOTO(out, rc);
7682         info = osd_oti_get(env);
7683         LASSERT(info);
7684
7685         l->ld_ops = &osd_lu_ops;
7686         o->od_dt_dev.dd_ops = &osd_dt_ops;
7687
7688         spin_lock_init(&o->od_osfs_lock);
7689         mutex_init(&o->od_otable_mutex);
7690         INIT_LIST_HEAD(&o->od_orphan_list);
7691         INIT_LIST_HEAD(&o->od_index_backup_list);
7692         INIT_LIST_HEAD(&o->od_index_restore_list);
7693         spin_lock_init(&o->od_lock);
7694         o->od_index_backup_policy = LIBP_NONE;
7695
7696         o->od_read_cache = 1;
7697         o->od_writethrough_cache = 1;
7698         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7699         o->od_auto_scrub_interval = AS_DEFAULT;
7700
7701         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7702                         sizeof(o->od_svname));
7703         if (cplen >= sizeof(o->od_svname)) {
7704                 rc = -E2BIG;
7705                 GOTO(out, rc);
7706         }
7707
7708         o->od_index_backup_stop = 0;
7709         o->od_index = -1; /* -1 means index is invalid */
7710         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7711         if (rc == LDD_F_SV_TYPE_OST)
7712                 o->od_is_ost = 1;
7713
7714         o->od_full_scrub_ratio = OFSR_DEFAULT;
7715         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7716         rc = osd_mount(env, o, cfg);
7717         if (rc != 0)
7718                 GOTO(out, rc);
7719
7720         rc = osd_obj_map_init(env, o);
7721         if (rc != 0)
7722                 GOTO(out_mnt, rc);
7723
7724         rc = lu_site_init(&o->od_site, l);
7725         if (rc != 0)
7726                 GOTO(out_compat, rc);
7727         o->od_site.ls_bottom_dev = l;
7728
7729         rc = lu_site_init_finish(&o->od_site);
7730         if (rc != 0)
7731                 GOTO(out_site, rc);
7732
7733         INIT_LIST_HEAD(&o->od_ios_list);
7734         /* setup scrub, including OI files initialization */
7735         o->od_in_init = 1;
7736         rc = osd_scrub_setup(env, o);
7737         o->od_in_init = 0;
7738         if (rc < 0)
7739                 GOTO(out_site, rc);
7740
7741         rc = osd_procfs_init(o, o->od_svname);
7742         if (rc != 0) {
7743                 CERROR("%s: can't initialize procfs: rc = %d\n",
7744                        o->od_svname, rc);
7745                 GOTO(out_scrub, rc);
7746         }
7747
7748         LASSERT(l->ld_site->ls_linkage.next != NULL);
7749         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7750
7751         /* initialize quota slave instance */
7752         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7753                                      o->od_proc_entry);
7754         if (IS_ERR(o->od_quota_slave)) {
7755                 rc = PTR_ERR(o->od_quota_slave);
7756                 o->od_quota_slave = NULL;
7757                 GOTO(out_procfs, rc);
7758         }
7759
7760         RETURN(0);
7761
7762 out_procfs:
7763         osd_procfs_fini(o);
7764 out_scrub:
7765         osd_scrub_cleanup(env, o);
7766 out_site:
7767         lu_site_fini(&o->od_site);
7768 out_compat:
7769         osd_obj_map_fini(o);
7770 out_mnt:
7771         osd_umount(env, o);
7772 out:
7773         return rc;
7774 }
7775
7776 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7777                                           struct lu_device_type *t,
7778                                           struct lustre_cfg *cfg)
7779 {
7780         struct osd_device *o;
7781         int rc;
7782
7783         OBD_ALLOC_PTR(o);
7784         if (o == NULL)
7785                 return ERR_PTR(-ENOMEM);
7786
7787         rc = dt_device_init(&o->od_dt_dev, t);
7788         if (rc == 0) {
7789                 /*
7790                  * Because the ctx might be revived in dt_device_init,
7791                  * refill the env here
7792                  */
7793                 lu_env_refill((struct lu_env *)env);
7794                 rc = osd_device_init0(env, o, cfg);
7795                 if (rc)
7796                         dt_device_fini(&o->od_dt_dev);
7797         }
7798
7799         if (unlikely(rc != 0))
7800                 OBD_FREE_PTR(o);
7801
7802         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7803 }
7804
7805 static struct lu_device *osd_device_free(const struct lu_env *env,
7806                                          struct lu_device *d)
7807 {
7808         struct osd_device *o = osd_dev(d);
7809
7810         ENTRY;
7811
7812         /* XXX: make osd top device in order to release reference */
7813         d->ld_site->ls_top_dev = d;
7814         lu_site_purge(env, d->ld_site, -1);
7815         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7816                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7817                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7818         }
7819         lu_site_fini(&o->od_site);
7820         dt_device_fini(&o->od_dt_dev);
7821         OBD_FREE_PTR(o);
7822         RETURN(NULL);
7823 }
7824
7825 static int osd_process_config(const struct lu_env *env,
7826                               struct lu_device *d, struct lustre_cfg *cfg)
7827 {
7828         struct osd_device *o = osd_dev(d);
7829         int rc;
7830
7831         ENTRY;
7832
7833         switch (cfg->lcfg_command) {
7834         case LCFG_SETUP:
7835                 rc = osd_mount(env, o, cfg);
7836                 break;
7837         case LCFG_CLEANUP:
7838                 /*
7839                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7840                  * that may happend if hit failure during mount process.
7841                  */
7842                 osd_index_backup(env, o, false);
7843                 lu_dev_del_linkage(d->ld_site, d);
7844                 rc = osd_shutdown(env, o);
7845                 break;
7846         case LCFG_PARAM:
7847                 LASSERT(&o->od_dt_dev);
7848                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7849                                               cfg, &o->od_dt_dev);
7850                 if (rc > 0 || rc == -ENOSYS) {
7851                         rc = class_process_proc_param(PARAM_OST,
7852                                                       lprocfs_osd_obd_vars,
7853                                                       cfg, &o->od_dt_dev);
7854                         if (rc > 0)
7855                                 rc = 0;
7856                 }
7857                 break;
7858         case LCFG_PRE_CLEANUP:
7859                 osd_scrub_stop(o);
7860                 osd_index_backup(env, o,
7861                                  o->od_index_backup_policy != LIBP_NONE);
7862                 rc = 0;
7863                 break;
7864         default:
7865                 rc = -ENOSYS;
7866         }
7867
7868         RETURN(rc);
7869 }
7870
7871 static int osd_recovery_complete(const struct lu_env *env,
7872                                  struct lu_device *d)
7873 {
7874         struct osd_device *osd = osd_dev(d);
7875         int rc = 0;
7876
7877         ENTRY;
7878
7879         if (osd->od_quota_slave == NULL)
7880                 RETURN(0);
7881
7882         /*
7883          * start qsd instance on recovery completion, this notifies the quota
7884          * slave code that we are about to process new requests now
7885          */
7886         rc = qsd_start(env, osd->od_quota_slave);
7887         RETURN(rc);
7888 }
7889
7890 /*
7891  * we use exports to track all osd users
7892  */
7893 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7894                            struct obd_device *obd, struct obd_uuid *cluuid,
7895                            struct obd_connect_data *data, void *localdata)
7896 {
7897         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7898         struct lustre_handle conn;
7899         int rc;
7900
7901         ENTRY;
7902
7903         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7904
7905         rc = class_connect(&conn, obd, cluuid);
7906         if (rc)
7907                 RETURN(rc);
7908
7909         *exp = class_conn2export(&conn);
7910
7911         spin_lock(&osd->od_osfs_lock);
7912         osd->od_connects++;
7913         spin_unlock(&osd->od_osfs_lock);
7914
7915         RETURN(0);
7916 }
7917
7918 /*
7919  * once last export (we don't count self-export) disappeared
7920  * osd can be released
7921  */
7922 static int osd_obd_disconnect(struct obd_export *exp)
7923 {
7924         struct obd_device *obd = exp->exp_obd;
7925         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7926         int rc, release = 0;
7927
7928         ENTRY;
7929
7930         /* Only disconnect the underlying layers on the final disconnect. */
7931         spin_lock(&osd->od_osfs_lock);
7932         osd->od_connects--;
7933         if (osd->od_connects == 0)
7934                 release = 1;
7935         spin_unlock(&osd->od_osfs_lock);
7936
7937         rc = class_disconnect(exp); /* bz 9811 */
7938
7939         if (rc == 0 && release)
7940                 class_manual_cleanup(obd);
7941         RETURN(rc);
7942 }
7943
7944 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7945                        struct lu_device *dev)
7946 {
7947         struct osd_device *osd = osd_dev(dev);
7948         struct lr_server_data *lsd =
7949                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7950         int result = 0;
7951
7952         ENTRY;
7953
7954         if (osd->od_quota_slave != NULL) {
7955                 /* set up quota slave objects */
7956                 result = qsd_prepare(env, osd->od_quota_slave);
7957                 if (result != 0)
7958                         RETURN(result);
7959         }
7960
7961         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7962 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7963                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7964                         osd->od_index_in_idif = 1;
7965                 } else {
7966                         osd->od_index_in_idif = 0;
7967                         result = osd_register_proc_index_in_idif(osd);
7968                         if (result != 0)
7969                                 RETURN(result);
7970                 }
7971 #else
7972                 osd->od_index_in_idif = 1;
7973 #endif
7974         }
7975
7976         result = osd_fid_init(env, osd);
7977
7978         RETURN(result);
7979 }
7980
7981 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7982                          struct lu_fid *fid, struct md_op_data *op_data)
7983 {
7984         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7985
7986         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7987 }
7988
7989 static const struct lu_object_operations osd_lu_obj_ops = {
7990         .loo_object_init      = osd_object_init,
7991         .loo_object_delete    = osd_object_delete,
7992         .loo_object_release   = osd_object_release,
7993         .loo_object_free      = osd_object_free,
7994         .loo_object_print     = osd_object_print,
7995         .loo_object_invariant = osd_object_invariant
7996 };
7997
7998 const struct lu_device_operations osd_lu_ops = {
7999         .ldo_object_alloc      = osd_object_alloc,
8000         .ldo_process_config    = osd_process_config,
8001         .ldo_recovery_complete = osd_recovery_complete,
8002         .ldo_prepare           = osd_prepare,
8003 };
8004
8005 static const struct lu_device_type_operations osd_device_type_ops = {
8006         .ldto_init = osd_type_init,
8007         .ldto_fini = osd_type_fini,
8008
8009         .ldto_start = osd_type_start,
8010         .ldto_stop  = osd_type_stop,
8011
8012         .ldto_device_alloc = osd_device_alloc,
8013         .ldto_device_free  = osd_device_free,
8014
8015         .ldto_device_init = osd_device_init,
8016         .ldto_device_fini = osd_device_fini
8017 };
8018
8019 static struct lu_device_type osd_device_type = {
8020         .ldt_tags     = LU_DEVICE_DT,
8021         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8022         .ldt_ops      = &osd_device_type_ops,
8023         .ldt_ctx_tags = LCT_LOCAL,
8024 };
8025
8026 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8027 {
8028         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8029         struct super_block *sb = osd_sb(osd);
8030
8031         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8032 }
8033
8034 /*
8035  * lprocfs legacy support.
8036  */
8037 static struct obd_ops osd_obd_device_ops = {
8038         .o_owner = THIS_MODULE,
8039         .o_connect      = osd_obd_connect,
8040         .o_disconnect   = osd_obd_disconnect,
8041         .o_fid_alloc    = osd_fid_alloc,
8042         .o_health_check = osd_health_check,
8043 };
8044
8045 static int __init osd_init(void)
8046 {
8047         int rc;
8048
8049         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8050 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8051         /* please, try to keep osd_thread_info smaller than a page */
8052         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8053 #endif
8054
8055         osd_oi_mod_init();
8056
8057         rc = lu_kmem_init(ldiskfs_caches);
8058         if (rc)
8059                 return rc;
8060
8061 #ifdef CONFIG_KALLSYMS
8062         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8063         priv_dev_check_rdonly =
8064                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8065 #endif
8066
8067         rc = class_register_type(&osd_obd_device_ops, NULL, true,
8068                                  lprocfs_osd_module_vars,
8069                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8070         if (rc)
8071                 lu_kmem_fini(ldiskfs_caches);
8072         return rc;
8073 }
8074
8075 static void __exit osd_exit(void)
8076 {
8077         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8078         lu_kmem_fini(ldiskfs_caches);
8079 }
8080
8081 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8082 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8083 MODULE_VERSION(LUSTRE_VERSION_STRING);
8084 MODULE_LICENSE("GPL");
8085
8086 module_init(osd_init);
8087 module_exit(osd_exit);