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 int ldiskfs_pdo = 1;
  81 module_param(ldiskfs_pdo, int, 0644);
  82 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  83
  84 int ldiskfs_track_declares_assert;
  85 module_param(ldiskfs_track_declares_assert, int, 0644);
  86 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  87
  88 /* Slab to allocate dynlocks */
  89 struct kmem_cache *dynlock_cachep;
  90
  91 /* Slab to allocate osd_it_ea */
  92 struct kmem_cache *osd_itea_cachep;
  93
  94 static struct lu_kmem_descr ldiskfs_caches[] = {
  95         {
  96                 .ckd_cache = &dynlock_cachep,
  97                 .ckd_name  = "dynlock_cache",
  98                 .ckd_size  = sizeof(struct dynlock_handle)
  99         },
 100         {
 101                 .ckd_cache = &osd_itea_cachep,
 102                 .ckd_name  = "osd_itea_cache",
 103                 .ckd_size  = sizeof(struct osd_it_ea)
 104         },
 105         {
 106                 .ckd_cache = NULL
 107         }
 108 };
 109
 110 static const char dot[] = ".";
 111 static const char dotdot[] = "..";
 112
 113 static const struct lu_object_operations      osd_lu_obj_ops;
 114 static const struct dt_object_operations      osd_obj_ops;
 115 static const struct dt_object_operations      osd_obj_otable_it_ops;
 116 static const struct dt_index_operations       osd_index_iam_ops;
 117 static const struct dt_index_operations       osd_index_ea_ops;
 118
 119 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 120                           const struct lu_fid *fid);
 121 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 122                                                 struct osd_device *osd);
 123
 124 int osd_trans_declare_op2rb[] = {
 125         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 126         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 127         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 128         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 129         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 130         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 131         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 132         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 133         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 134         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 135         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 136 };
 137
 138 static int osd_has_index(const struct osd_object *obj)
 139 {
 140         return obj->oo_dt.do_index_ops != NULL;
 141 }
 142
 143 static int osd_object_invariant(const struct lu_object *l)
 144 {
 145         return osd_invariant(osd_obj(l));
 146 }
 147
 148 /*
 149  * Concurrency: doesn't matter
 150  */
 151 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 152 {
 153         struct osd_thread_info *oti = osd_oti_get(env);
 154         return oti->oti_w_locks > 0 && o->oo_owner == env;
 155 }
 156
 157 /*
 158  * Concurrency: doesn't access mutable data
 159  */
 160 static int osd_root_get(const struct lu_env *env,
 161                         struct dt_device *dev, struct lu_fid *f)
 162 {
 163         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 164         return 0;
 165 }
 166
 167 /*
 168  * the following set of functions are used to maintain per-thread
 169  * cache of FID->ino mapping. this mechanism is needed to resolve
 170  * FID to inode at dt_insert() which in turn stores ino in the
 171  * directory entries to keep ldiskfs compatible with ext[34].
 172  * due to locking-originated restrictions we can't lookup ino
 173  * using LU cache (deadlock is possible). lookup using OI is quite
 174  * expensive. so instead we maintain this cache and methods like
 175  * dt_create() fill it. so in the majority of cases dt_insert() is
 176  * able to find needed mapping in lockless manner.
 177  */
 178 static struct osd_idmap_cache *
 179 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 180              const struct lu_fid *fid)
 181 {
 182         struct osd_thread_info  *oti   = osd_oti_get(env);
 183         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
 184         int i;
 185         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 186                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 187                         continue;
 188                 if (idc[i].oic_dev != osd)
 189                         continue;
 190
 191                 return idc + i;
 192         }
 193
 194         return NULL;
 195 }
 196
 197 static struct osd_idmap_cache *
 198 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 199             const struct lu_fid *fid)
 200 {
 201         struct osd_thread_info  *oti   = osd_oti_get(env);
 202         struct osd_idmap_cache  *idc;
 203         int i;
 204
 205         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 206                 i = oti->oti_ins_cache_size * 2;
 207                 if (i == 0)
 208                         i = OSD_INS_CACHE_SIZE;
 209                 OBD_ALLOC(idc, sizeof(*idc) * i);
 210                 if (idc == NULL)
 211                         return ERR_PTR(-ENOMEM);
 212                 if (oti->oti_ins_cache != NULL) {
 213                         memcpy(idc, oti->oti_ins_cache,
 214                                oti->oti_ins_cache_used * sizeof(*idc));
 215                         OBD_FREE(oti->oti_ins_cache,
 216                                  oti->oti_ins_cache_used * sizeof(*idc));
 217                 }
 218                 oti->oti_ins_cache = idc;
 219                 oti->oti_ins_cache_size = i;
 220         }
 221
 222         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 223         idc->oic_fid = *fid;
 224         idc->oic_dev = osd;
 225         idc->oic_lid.oii_ino = 0;
 226         idc->oic_lid.oii_gen = 0;
 227         idc->oic_remote = 0;
 228
 229         return idc;
 230 }
 231
 232 /*
 233  * lookup mapping for the given fid in the cache, initialize a
 234  * new one if not found. the initialization checks whether the
 235  * object is local or remote. for local objects, OI is used to
 236  * learn ino/generation. the function is used when the caller
 237  * has no information about the object, e.g. at dt_insert().
 238  */
 239 static struct osd_idmap_cache *
 240 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 241                      const struct lu_fid *fid)
 242 {
 243         struct osd_idmap_cache *idc;
 244         int rc;
 245
 246         idc = osd_idc_find(env, osd, fid);
 247         LASSERT(!IS_ERR(idc));
 248         if (idc != NULL)
 249                 return idc;
 250
 251         /* new mapping is needed */
 252         idc = osd_idc_add(env, osd, fid);
 253         if (IS_ERR(idc))
 254                 return idc;
 255
 256         /* initialize it */
 257         rc = osd_remote_fid(env, osd, fid);
 258         if (unlikely(rc < 0))
 259                 return ERR_PTR(rc);
 260
 261         if (rc == 0) {
 262                 /* the object is local, lookup in OI */
 263                 /* XXX: probably cheaper to lookup in LU first? */
 264                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 265                                    &idc->oic_lid, 0);
 266                 if (unlikely(rc < 0)) {
 267                         CERROR("can't lookup: rc = %d\n", rc);
 268                         return ERR_PTR(rc);
 269                 }
 270         } else {
 271                 /* the object is remote */
 272                 idc->oic_remote = 1;
 273         }
 274
 275         return idc;
 276 }
 277
 278 /*
 279  * lookup mapping for given FID and fill it from the given object.
 280  * the object is lolcal by definition.
 281  */
 282 static int osd_idc_find_and_init(const struct lu_env *env,
 283                                  struct osd_device *osd,
 284                                  struct osd_object *obj)
 285 {
 286         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
 287         struct osd_idmap_cache  *idc;
 288
 289         idc = osd_idc_find(env, osd, fid);
 290         LASSERT(!IS_ERR(idc));
 291         if (idc != NULL) {
 292                 if (obj->oo_inode == NULL)
 293                         return 0;
 294                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 295                         LASSERT(idc->oic_lid.oii_ino == 0);
 296                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 297                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 298                 }
 299                 return 0;
 300         }
 301
 302         /* new mapping is needed */
 303         idc = osd_idc_add(env, osd, fid);
 304         if (IS_ERR(idc))
 305                 return PTR_ERR(idc);
 306
 307         if (obj->oo_inode != NULL) {
 308                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 309                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 310         }
 311         return 0;
 312 }
 313
 314 /*
 315  * OSD object methods.
 316  */
 317
 318 /*
 319  * Concurrency: no concurrent access is possible that early in object
 320  * life-cycle.
 321  */
 322 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 323                                           const struct lu_object_header *hdr,
 324                                           struct lu_device *d)
 325 {
 326         struct osd_object *mo;
 327
 328         OBD_ALLOC_PTR(mo);
 329         if (mo != NULL) {
 330                 struct lu_object *l;
 331                 struct lu_object_header *h;
 332
 333                 l = &mo->oo_dt.do_lu;
 334                 h = &osd_dev(d)->od_scrub.os_scrub.os_obj_header;
 335                 if (unlikely(fid_is_zero(&h->loh_fid))) {
 336                         /* For the OI_scrub object during OSD device init. */
 337                         lu_object_header_init(h);
 338                         lu_object_init(l, h, d);
 339                         lu_object_add_top(h, l);
 340                 } else {
 341                         dt_object_init(&mo->oo_dt, NULL, d);
 342                 }
 343                 mo->oo_dt.do_ops = &osd_obj_ops;
 344                 l->lo_ops = &osd_lu_obj_ops;
 345                 init_rwsem(&mo->oo_sem);
 346                 init_rwsem(&mo->oo_ext_idx_sem);
 347                 spin_lock_init(&mo->oo_guard);
 348                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 349                 return l;
 350         } else {
 351                 return NULL;
 352         }
 353 }
 354
 355 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 356                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 357 {
 358         int rc;
 359
 360         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 361                              (void *)loa, sizeof(*loa));
 362         if (rc > 0) {
 363                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 364
 365                 if (rc < sizeof(*lma))
 366                         return -EINVAL;
 367
 368                 rc = 0;
 369                 lustre_loa_swab(loa, true);
 370                 /* Check LMA compatibility */
 371                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 372                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 373                               "for fid = "DFID", ino = %lu\n",
 374                               osd_ino2name(inode),
 375                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 376                               PFID(&lma->lma_self_fid), inode->i_ino);
 377                         rc = -EOPNOTSUPP;
 378                 }
 379         } else if (rc == 0) {
 380                 rc = -ENODATA;
 381         }
 382
 383         return rc;
 384 }
 385
 386 /*
 387  * retrieve object from backend ext fs.
 388  **/
 389 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 390                        struct osd_inode_id *id)
 391 {
 392         int rc;
 393         struct inode *inode = NULL;
 394
 395         /* if we look for an inode withing a running
 396          * transaction, then we risk to deadlock */
 397         /* osd_dirent_check_repair() breaks this */
 398         /*LASSERT(current->journal_info == NULL);*/
 399
 400         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 401         if (IS_ERR(inode)) {
 402                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 403                        id->oii_ino, PTR_ERR(inode));
 404         } else if (id->oii_gen != OSD_OII_NOGEN &&
 405                    inode->i_generation != id->oii_gen) {
 406                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 407                        "i_generation = %u\n",
 408                        id->oii_ino, id->oii_gen, inode->i_generation);
 409                 iput(inode);
 410                 inode = ERR_PTR(-ESTALE);
 411         } else if (inode->i_nlink == 0) {
 412                 /* due to parallel readdir and unlink,
 413                 * we can have dead inode here. */
 414                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 415                 iput(inode);
 416                 inode = ERR_PTR(-ESTALE);
 417         } else if (is_bad_inode(inode)) {
 418                 CWARN("%s: bad inode: ino = %u\n",
 419                 osd_dev2name(dev), id->oii_ino);
 420                 iput(inode);
 421                 inode = ERR_PTR(-ENOENT);
 422         } else if ((rc = osd_attach_jinode(inode))) {
 423                 iput(inode);
 424                 inode = ERR_PTR(rc);
 425         } else {
 426                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 427                 if (id->oii_gen == OSD_OII_NOGEN)
 428                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 429
 430                 /* Do not update file c/mtime in ldiskfs.
 431                  * NB: we don't have any lock to protect this because we don't
 432                  * have reference on osd_object now, but contention with
 433                  * another lookup + attr_set can't happen in the tiny window
 434                  * between if (...) and set S_NOCMTIME. */
 435                 if (!(inode->i_flags & S_NOCMTIME))
 436                         inode->i_flags |= S_NOCMTIME;
 437         }
 438         return inode;
 439 }
 440
 441 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 442                           handle_t *handle, struct dentry *child,
 443                           struct inode *inode, struct htree_lock *hlock)
 444 {
 445         int rc, rc2;
 446
 447         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 448         if (rc == -ENOBUFS || rc == -ENOSPC) {
 449                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 450                 struct inode *parent = child->d_parent->d_inode;
 451                 struct lu_fid *fid = NULL;
 452
 453                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 454                 if (!rc2) {
 455                         fid = &loa->loa_lma.lma_self_fid;
 456                 } else if (rc2 == -ENODATA) {
 457                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 458                                 fid = &info->oti_fid3;
 459                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 460                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 461                                 fid = &info->oti_fid3;
 462                                 lu_igif_build(fid, parent->i_ino,
 463                                               parent->i_generation);
 464                         }
 465                 }
 466
 467                 if (fid != NULL)
 468                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 469                               "maximum entry limit\n",
 470                               osd_name(osd), parent->i_ino, PFID(fid),
 471                               rc == -ENOSPC ? "has reached" : "is approaching");
 472                 else
 473                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 474                               "maximum entry limit\n",
 475                               osd_name(osd), parent->i_ino,
 476                               rc == -ENOSPC ? "has reached" : "is approaching");
 477
 478                 /* ignore such error now */
 479                 if (rc == -ENOBUFS)
 480                         rc = 0;
 481         }
 482
 483         return rc;
 484 }
 485
 486
 487 static struct inode *
 488 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 489              struct osd_inode_id *id, struct lu_fid *fid)
 490 {
 491         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 492         struct inode *inode;
 493         int rc;
 494
 495         inode = osd_iget(info, dev, id);
 496         if (IS_ERR(inode))
 497                 return inode;
 498
 499         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 500         if (!rc) {
 501                 *fid = loa->loa_lma.lma_self_fid;
 502         } else if (rc == -ENODATA) {
 503                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 504                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 505                 else
 506                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 507         } else {
 508                 iput(inode);
 509                 inode = ERR_PTR(rc);
 510         }
 511         return inode;
 512 }
 513
 514 static struct inode *osd_iget_check(struct osd_thread_info *info,
 515                                     struct osd_device *dev,
 516                                     const struct lu_fid *fid,
 517                                     struct osd_inode_id *id,
 518                                     bool trusted)
 519 {
 520         struct inode *inode;
 521         int rc = 0;
 522         ENTRY;
 523
 524         /* The cached OI mapping is trustable. If we cannot locate the inode
 525          * via the cached OI mapping, then return the failure to the caller
 526          * directly without further OI checking. */
 527
 528 again:
 529         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 530         if (IS_ERR(inode)) {
 531                 rc = PTR_ERR(inode);
 532                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 533                         goto check_oi;
 534
 535                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 536                        PFID(fid), id->oii_ino, rc);
 537                 GOTO(put, rc);
 538         }
 539
 540         if (is_bad_inode(inode)) {
 541                 rc = -ENOENT;
 542                 if (!trusted)
 543                         goto check_oi;
 544
 545                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 546                        PFID(fid), id->oii_ino);
 547                 GOTO(put, rc);
 548         }
 549
 550         if (id->oii_gen != OSD_OII_NOGEN &&
 551             inode->i_generation != id->oii_gen) {
 552                 rc = -ESTALE;
 553                 if (!trusted)
 554                         goto check_oi;
 555
 556                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 557                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 558                        id->oii_ino, id->oii_gen, inode->i_generation);
 559                 GOTO(put, rc);
 560         }
 561
 562         if (inode->i_nlink == 0) {
 563                 rc = -ENOENT;
 564                 if (!trusted)
 565                         goto check_oi;
 566
 567                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 568                        PFID(fid), id->oii_ino);
 569                 GOTO(put, rc);
 570         }
 571
 572         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 573
 574 check_oi:
 575         if (rc != 0) {
 576                 __u32 saved_ino = id->oii_ino;
 577                 __u32 saved_gen = id->oii_gen;
 578
 579                 LASSERT(!trusted);
 580                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 581
 582                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 583                 /* XXX: There are four possible cases:
 584                  *      1. rc = 0.
 585                  *         Backup/restore caused the OI invalid.
 586                  *      2. rc = 0.
 587                  *         Someone unlinked the object but NOT removed
 588                  *         the OI mapping, such as mount target device
 589                  *         as ldiskfs, and modify something directly.
 590                  *      3. rc = -ENOENT.
 591                  *         Someone just removed the object between the
 592                  *         former oi_lookup and the iget. It is normal.
 593                  *      4. Other failure cases.
 594                  *
 595                  *      Generally, when the device is mounted, it will
 596                  *      auto check whether the system is restored from
 597                  *      file-level backup or not. We trust such detect
 598                  *      to distinguish the 1st case from the 2nd case:
 599                  *      if the OI files are consistent but may contain
 600                  *      stale OI mappings because of case 2, if iget()
 601                  *      returns -ENOENT or -ESTALE, then it should be
 602                  *      the case 2. */
 603                 if (rc != 0)
 604                         /* If the OI mapping was in OI file before the
 605                          * osd_iget_check(), but now, it is disappear,
 606                          * then it must be removed by race. That is a
 607                          * normal race case. */
 608                         GOTO(put, rc);
 609
 610                 /* It is the OI scrub updated the OI mapping by race.
 611                  * The new OI mapping must be valid. */
 612                 if (saved_ino != id->oii_ino ||
 613                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 614                         if (!IS_ERR(inode))
 615                                 iput(inode);
 616
 617                         trusted = true;
 618                         goto again;
 619                 }
 620
 621                 if (IS_ERR(inode)) {
 622                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 623                             SF_INCONSISTENT)
 624                                 /* It still can be the case 2, but we cannot
 625                                  * distinguish it from the case 1. So return
 626                                  * -EREMCHG to block current operation until
 627                                  *  OI scrub rebuilt the OI mappings. */
 628                                 rc = -EREMCHG;
 629                         else
 630                                 rc = -ENOENT;
 631
 632                         GOTO(put, rc);
 633                 }
 634
 635                 if (inode->i_generation == id->oii_gen)
 636                         rc = -ENOENT;
 637                 else
 638                         rc = -EREMCHG;
 639         } else {
 640                 if (id->oii_gen == OSD_OII_NOGEN)
 641                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 642
 643                 /* Do not update file c/mtime in ldiskfs.
 644                  * NB: we don't have any lock to protect this because we don't
 645                  * have reference on osd_object now, but contention with
 646                  * another lookup + attr_set can't happen in the tiny window
 647                  * between if (...) and set S_NOCMTIME. */
 648                 if (!(inode->i_flags & S_NOCMTIME))
 649                         inode->i_flags |= S_NOCMTIME;
 650         }
 651
 652         GOTO(put, rc);
 653
 654 put:
 655         if (rc != 0) {
 656                 if (!IS_ERR(inode))
 657                         iput(inode);
 658
 659                 inode = ERR_PTR(rc);
 660         }
 661
 662         return inode;
 663 }
 664
 665 /**
 666  * \retval +v: new filter_fid, does not contain self-fid
 667  * \retval 0:  filter_fid_old, contains self-fid
 668  * \retval -v: other failure cases
 669  */
 670 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 671                  struct dentry *dentry, struct lu_fid *fid)
 672 {
 673         struct filter_fid_old   *ff     = &info->oti_ff;
 674         struct ost_id           *ostid  = &info->oti_ostid;
 675         int                      rc;
 676
 677         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 678         if (rc == sizeof(*ff)) {
 679                 rc = 0;
 680                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 681                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 682                 /*
 683                  * XXX: use 0 as the index for compatibility, the caller will
 684                  *      handle index related issues when necessary.
 685                  */
 686                 if (!rc)
 687                         ostid_to_fid(fid, ostid, 0);
 688         } else if (rc == sizeof(struct filter_fid)) {
 689                 rc = 1;
 690         } else if (rc >= 0) {
 691                 rc = -EINVAL;
 692         }
 693
 694         return rc;
 695 }
 696
 697 static int osd_lma_self_repair(struct osd_thread_info *info,
 698                                struct osd_device *osd, struct inode *inode,
 699                                const struct lu_fid *fid, __u32 compat)
 700 {
 701         handle_t *jh;
 702         int       rc;
 703
 704         LASSERT(current->journal_info == NULL);
 705
 706         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 707                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 708         if (IS_ERR(jh)) {
 709                 rc = PTR_ERR(jh);
 710                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 711                       osd_name(osd), rc);
 712                 return rc;
 713         }
 714
 715         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 716         if (rc != 0)
 717                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 718                       osd_name(osd), rc);
 719         ldiskfs_journal_stop(jh);
 720         return rc;
 721 }
 722
 723 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 724 {
 725         struct osd_thread_info  *info   = osd_oti_get(env);
 726         struct osd_device       *osd    = osd_obj2dev(obj);
 727         struct lustre_ost_attrs *loa    = &info->oti_ost_attrs;
 728         struct lustre_mdt_attrs *lma    = &loa->loa_lma;
 729         struct inode            *inode  = obj->oo_inode;
 730         struct dentry           *dentry = &info->oti_obj_dentry;
 731         struct lu_fid           *fid    = NULL;
 732         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 733         int                      rc;
 734         ENTRY;
 735
 736         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 737                              (void *)loa, sizeof(*loa));
 738         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 739                 fid = &lma->lma_self_fid;
 740                 rc = osd_get_idif(info, inode, dentry, fid);
 741                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 742                         /* For the given OST-object, if it has neither LMA nor
 743                          * FID in XATTR_NAME_FID, then the given FID (which is
 744                          * contained in the @obj, from client RPC for locating
 745                          * the OST-object) is trusted. We use it to generate
 746                          * the LMA. */
 747                         osd_lma_self_repair(info, osd, inode, rfid,
 748                                             LMAC_FID_ON_OST);
 749                         RETURN(0);
 750                 }
 751         }
 752
 753         if (rc < 0)
 754                 RETURN(rc);
 755
 756         if (rc > 0) {
 757                 rc = 0;
 758                 lustre_lma_swab(lma);
 759                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 760                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 761                               S_ISREG(inode->i_mode)))) {
 762                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 763                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 764                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 765                               PFID(rfid), inode->i_ino);
 766                         rc = -EOPNOTSUPP;
 767                 } else {
 768                         fid = &lma->lma_self_fid;
 769                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 770                             osd->od_is_ost)
 771                                 obj->oo_pfid_in_lma = 1;
 772                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 773                             !osd->od_is_ost)
 774                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 775                 }
 776         }
 777
 778         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 779                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 780                         struct ost_id   *oi   = &info->oti_ostid;
 781                         struct lu_fid   *fid1 = &info->oti_fid3;
 782                         __u32            idx  = fid_idif_ost_idx(rfid);
 783
 784                         /* For old IDIF, the OST index is not part of the IDIF,
 785                          * Means that different OSTs may have the same IDIFs.
 786                          * Under such case, we need to make some compatible
 787                          * check to make sure to trigger OI scrub properly. */
 788                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 789                                 /* Given @rfid is new, LMA is old. */
 790                                 fid_to_ostid(fid, oi);
 791                                 ostid_to_fid(fid1, oi, idx);
 792                                 if (lu_fid_eq(fid1, rfid)) {
 793                                         if (osd->od_index_in_idif)
 794                                                 osd_lma_self_repair(info, osd,
 795                                                         inode, rfid,
 796                                                         LMAC_FID_ON_OST);
 797                                         RETURN(0);
 798                                 }
 799                         }
 800                 }
 801
 802                 rc = -EREMCHG;
 803         }
 804
 805         RETURN(rc);
 806 }
 807
 808 struct osd_check_lmv_buf {
 809 #ifdef HAVE_DIR_CONTEXT
 810         /* please keep it as first member */
 811         struct dir_context       ctx;
 812 #endif
 813         struct osd_thread_info  *oclb_info;
 814         struct osd_device       *oclb_dev;
 815         struct osd_idmap_cache  *oclb_oic;
 816 };
 817
 818 /**
 819  * It is called internally by ->readdir() to filter out the
 820  * local slave object's FID of the striped directory.
 821  *
 822  * \retval      1 found the local slave's FID
 823  * \retval      0 continue to check next item
 824  * \retval      -ve for failure
 825  */
 826 #ifdef HAVE_FILLDIR_USE_CTX
 827 static int osd_stripe_dir_filldir(struct dir_context *buf,
 828 #else
 829 static int osd_stripe_dir_filldir(void *buf,
 830 #endif
 831                                   const char *name, int namelen,
 832                                   loff_t offset, __u64 ino, unsigned d_type)
 833 {
 834         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 835         struct osd_thread_info *oti = oclb->oclb_info;
 836         struct lu_fid *fid = &oti->oti_fid3;
 837         struct osd_inode_id *id = &oti->oti_id3;
 838         struct osd_device *dev = oclb->oclb_dev;
 839         struct osd_idmap_cache *oic = oclb->oclb_oic;
 840         struct inode *inode;
 841         int rc;
 842
 843         if (name[0] == '.')
 844                 return 0;
 845
 846         fid_zero(fid);
 847         sscanf(name + 1, SFID, RFID(fid));
 848         if (!fid_is_sane(fid))
 849                 return 0;
 850
 851         if (osd_remote_fid(oti->oti_env, dev, fid))
 852                 return 0;
 853
 854         osd_id_gen(id, ino, OSD_OII_NOGEN);
 855         inode = osd_iget(oti, dev, id);
 856         if (IS_ERR(inode))
 857                 return PTR_ERR(inode);
 858
 859         iput(inode);
 860         osd_add_oi_cache(oti, dev, id, fid);
 861         oic->oic_fid = *fid;
 862         oic->oic_lid = *id;
 863         oic->oic_dev = dev;
 864         rc = osd_oii_insert(dev, oic, true);
 865
 866         return rc == 0 ? 1 : rc;
 867 }
 868
 869 /* When lookup item under striped directory, we need to locate the master
 870  * MDT-object of the striped directory firstly, then the client will send
 871  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 872  * and the item's name. If the system is restored from MDT file level backup,
 873  * then before the OI scrub completely built the OI files, the OI mappings of
 874  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 875  * not a problem for the master MDT-object. Because when locate the master
 876  * MDT-object, we will do name based lookup (for the striped directory itself)
 877  * firstly, during such process we can setup the correct OI mapping for the
 878  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 879  * the client will not trigger name based lookup on the MDT to locate the slave
 880  * MDT-object before locating item under the striped directory, then when
 881  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 882  * is invalid and does not know what the right OI mapping is, then the MDT has
 883  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 884  * the OI file, related OI mapping is unknown yet, please try again later. And
 885  * then client will re-try the RPC again and again until related OI mapping has
 886  * been updated. That is quite inefficient.
 887  *
 888  * To resolve above trouble, we will handle it as the following two cases:
 889  *
 890  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 891  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 892  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 893  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 894  *    directly. Please check osd_fid_lookup().
 895  *
 896  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 897  *    Under such case, during lookup the master MDT-object, we will lookup the
 898  *    slave MDT-object via readdir against the master MDT-object, because the
 899  *    slave MDT-objects information are stored as sub-directories with the name
 900  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 901  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 902  *    the correct OI mapping for the slave MDT-object. */
 903 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 904                          struct inode *inode, struct osd_idmap_cache *oic)
 905 {
 906         struct lu_buf *buf = &oti->oti_big_buf;
 907         struct dentry *dentry = &oti->oti_obj_dentry;
 908         struct file *filp = &oti->oti_file;
 909         const struct file_operations *fops;
 910         struct lmv_mds_md_v1 *lmv1;
 911         struct osd_check_lmv_buf oclb = {
 912 #ifdef HAVE_DIR_CONTEXT
 913                 .ctx.actor = osd_stripe_dir_filldir,
 914 #endif
 915                 .oclb_info = oti,
 916                 .oclb_dev = dev,
 917                 .oclb_oic = oic
 918         };
 919         int rc = 0;
 920         ENTRY;
 921
 922 again:
 923         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 924                              buf->lb_len);
 925         if (rc == -ERANGE) {
 926                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 927                 if (rc > 0) {
 928                         lu_buf_realloc(buf, rc);
 929                         if (buf->lb_buf == NULL)
 930                                 GOTO(out, rc = -ENOMEM);
 931
 932                         goto again;
 933                 }
 934         }
 935
 936         if (unlikely(rc == 0 || rc == -ENODATA))
 937                 GOTO(out, rc = 0);
 938
 939         if (rc < 0)
 940                 GOTO(out, rc);
 941
 942         if (unlikely(buf->lb_buf == NULL)) {
 943                 lu_buf_realloc(buf, rc);
 944                 if (buf->lb_buf == NULL)
 945                         GOTO(out, rc = -ENOMEM);
 946
 947                 goto again;
 948         }
 949
 950         lmv1 = buf->lb_buf;
 951         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
 952                 GOTO(out, rc = 0);
 953
 954         fops = inode->i_fop;
 955         dentry->d_inode = inode;
 956         dentry->d_sb = inode->i_sb;
 957         filp->f_pos = 0;
 958         filp->f_path.dentry = dentry;
 959         filp->f_mode = FMODE_64BITHASH;
 960         filp->f_mapping = inode->i_mapping;
 961         filp->f_op = fops;
 962         filp->private_data = NULL;
 963         set_file_inode(filp, inode);
 964
 965 #ifdef HAVE_DIR_CONTEXT
 966         oclb.ctx.pos = filp->f_pos;
 967         rc = fops->iterate(filp, &oclb.ctx);
 968         filp->f_pos = oclb.ctx.pos;
 969 #else
 970         rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
 971 #endif
 972         fops->release(inode, filp);
 973
 974 out:
 975         if (rc < 0)
 976                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
 977                        DFID": rc = %d\n", osd_ino2name(inode),
 978                        inode->i_ino, inode->i_generation,
 979                        PFID(&oic->oic_fid), rc);
 980         else
 981                 rc = 0;
 982
 983         RETURN(rc);
 984 }
 985
 986 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 987                           const struct lu_fid *fid,
 988                           const struct lu_object_conf *conf)
 989 {
 990         struct osd_thread_info *info;
 991         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
 992         struct osd_device *dev;
 993         struct osd_idmap_cache *oic;
 994         struct osd_inode_id *id;
 995         struct inode *inode = NULL;
 996         struct lustre_scrub *scrub;
 997         struct scrub_file *sf;
 998         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
 999         __u32 saved_ino;
1000         __u32 saved_gen;
1001         int result = 0;
1002         int rc1 = 0;
1003         bool remote = false;
1004         bool trusted = true;
1005         bool updated = false;
1006         bool checked = false;
1007         ENTRY;
1008
1009         LINVRNT(osd_invariant(obj));
1010         LASSERT(obj->oo_inode == NULL);
1011         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1012
1013         dev = osd_dev(ldev);
1014         scrub = &dev->od_scrub.os_scrub;
1015         sf = &scrub->os_file;
1016         info = osd_oti_get(env);
1017         LASSERT(info);
1018         oic = &info->oti_cache;
1019
1020         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1021                 RETURN(-ENOENT);
1022
1023         /* For the object is created as locking anchor, or for the object to
1024          * be created on disk. No need to osd_oi_lookup() at here because FID
1025          * shouldn't never be re-used, if it's really a duplicate FID from
1026          * unexpected reason, we should be able to detect it later by calling
1027          * do_create->osd_oi_insert(). */
1028         if (conf && conf->loc_flags & LOC_F_NEW)
1029                 GOTO(out, result = 0);
1030
1031         /* Search order: 1. per-thread cache. */
1032         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1033                 id = &oic->oic_lid;
1034                 goto iget;
1035         }
1036
1037         id = &info->oti_id;
1038         if (!list_empty(&scrub->os_inconsistent_items)) {
1039                 /* Search order: 2. OI scrub pending list. */
1040                 result = osd_oii_lookup(dev, fid, id);
1041                 if (!result)
1042                         goto iget;
1043         }
1044
1045         /* The OI mapping in the OI file can be updated by the OI scrub
1046          * when we locate the inode via FID. So it may be not trustable. */
1047         trusted = false;
1048
1049         /* Search order: 3. OI files. */
1050         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1051         if (result == -ENOENT) {
1052                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1053                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1054                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1055                                       sf->sf_oi_bitmap))
1056                         GOTO(out, result = 0);
1057
1058                 goto trigger;
1059         }
1060
1061         if (result)
1062                 GOTO(out, result);
1063
1064 iget:
1065         obj->oo_inode = NULL;
1066         /* for later passes through checks, not true on first pass */
1067         if (!IS_ERR_OR_NULL(inode))
1068                 iput(inode);
1069
1070         inode = osd_iget_check(info, dev, fid, id, trusted);
1071         if (!IS_ERR(inode)) {
1072                 obj->oo_inode = inode;
1073                 result = 0;
1074                 if (remote)
1075                         goto trigger;
1076
1077                 goto check_lma;
1078         }
1079
1080         result = PTR_ERR(inode);
1081         if (result == -ENOENT || result == -ESTALE)
1082                 GOTO(out, result = 0);
1083
1084         if (result != -EREMCHG)
1085                 GOTO(out, result);
1086
1087 trigger:
1088         /* We still have chance to get the valid inode: for the
1089          * object which is referenced by remote name entry, the
1090          * object on the local MDT will be linked under the dir
1091          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1092          *
1093          * We do not know whether the object for the given FID
1094          * is referenced by some remote name entry or not, and
1095          * especially for DNE II, a multiple-linked object may
1096          * have many name entries reside on many MDTs.
1097          *
1098          * To simplify the operation, OSD will not distinguish
1099          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1100          * only happened for the RPC from other MDT during the
1101          * OI scrub, or for the client side RPC with FID only,
1102          * such as FID to path, or from old connected client. */
1103         if (!remote && !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1104                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1105                 if (!rc1) {
1106                         remote = true;
1107                         trusted = true;
1108                         flags |= SS_AUTO_PARTIAL;
1109                         flags &= ~SS_AUTO_FULL;
1110                         goto iget;
1111                 }
1112         }
1113
1114         if (thread_is_running(&scrub->os_thread)) {
1115                 if (scrub->os_partial_scan && !scrub->os_in_join)
1116                         goto join;
1117
1118                 if (IS_ERR_OR_NULL(inode) || result)
1119                         GOTO(out, result = -EINPROGRESS);
1120
1121                 LASSERT(remote);
1122                 LASSERT(obj->oo_inode == inode);
1123
1124                 osd_add_oi_cache(info, dev, id, fid);
1125                 osd_oii_insert(dev, oic, true);
1126                 goto found;
1127         }
1128
1129         if (dev->od_auto_scrub_interval == AS_NEVER) {
1130                 if (!remote)
1131                         GOTO(out, result = -EREMCHG);
1132
1133                 LASSERT(!result);
1134                 LASSERT(obj->oo_inode == inode);
1135
1136                 osd_add_oi_cache(info, dev, id, fid);
1137                 goto found;
1138         }
1139
1140 join:
1141         rc1 = osd_scrub_start(env, dev, flags);
1142         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1143                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1144         if (rc1 && rc1 != -EALREADY)
1145                 GOTO(out, result = -EREMCHG);
1146
1147         if (IS_ERR_OR_NULL(inode) || result)
1148                 GOTO(out, result = -EINPROGRESS);
1149
1150         LASSERT(remote);
1151         LASSERT(obj->oo_inode == inode);
1152
1153         osd_add_oi_cache(info, dev, id, fid);
1154         osd_oii_insert(dev, oic, true);
1155         goto found;
1156
1157 check_lma:
1158         result = osd_check_lma(env, obj);
1159         checked = true;
1160         if (!result)
1161                 goto found;
1162
1163         LASSERTF(id->oii_ino == inode->i_ino &&
1164                  id->oii_gen == inode->i_generation,
1165                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1166                  PFID(fid), id->oii_ino, id->oii_gen,
1167                  inode->i_ino, inode->i_generation);
1168
1169         saved_ino = inode->i_ino;
1170         saved_gen = inode->i_generation;
1171
1172         if (unlikely(result == -ENODATA)) {
1173                 /* If the OI scrub updated the OI mapping by race, it
1174                  * must be valid. Trust the inode that has no LMA EA. */
1175                 if (updated)
1176                         goto found;
1177
1178                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1179                 if (!result) {
1180                         /* The OI mapping is still there, the inode is still
1181                          * valid. It is just becaues the inode has no LMA EA. */
1182                         if (saved_ino == id->oii_ino &&
1183                             saved_gen == id->oii_gen)
1184                                 goto found;
1185
1186                         /* It is the OI scrub updated the OI mapping by race.
1187                          * The new OI mapping must be valid. */
1188                         trusted = true;
1189                         updated = true;
1190                         goto iget;
1191                 }
1192
1193                 /* "result == -ENOENT" means that the OI mappinghas been
1194                  * removed by race, so the inode belongs to other object.
1195                  *
1196                  * Others error can be returned  directly. */
1197                 if (result == -ENOENT) {
1198                         LASSERT(trusted);
1199
1200                         obj->oo_inode = NULL;
1201                         result = 0;
1202                 }
1203         }
1204
1205         if (result != -EREMCHG)
1206                 GOTO(out, result);
1207
1208         LASSERT(!updated);
1209
1210         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1211         /* "result == -ENOENT" means the cached OI mapping has been removed
1212          * from the OI file by race, above inode belongs to other object. */
1213         if (result == -ENOENT) {
1214                 LASSERT(trusted);
1215
1216                 obj->oo_inode = NULL;
1217                 GOTO(out, result = 0);
1218         }
1219
1220         if (result)
1221                 GOTO(out, result);
1222
1223         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1224                 result = -EREMCHG;
1225                 goto trigger;
1226         }
1227
1228         /* It is the OI scrub updated the OI mapping by race.
1229          * The new OI mapping must be valid. */
1230         trusted = true;
1231         updated = true;
1232         goto iget;
1233
1234 found:
1235         if (!checked) {
1236                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1237                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1238
1239                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1240                 if (!result) {
1241                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1242                             dev->od_is_ost)
1243                                 obj->oo_pfid_in_lma = 1;
1244                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1245                             !dev->od_is_ost)
1246                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1247                 } else if (result != -ENODATA) {
1248                         GOTO(out, result);
1249                 }
1250         }
1251
1252         obj->oo_compat_dot_created = 1;
1253         obj->oo_compat_dotdot_created = 1;
1254
1255         if (S_ISDIR(inode->i_mode) &&
1256             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1257                 osd_check_lmv(info, dev, inode, oic);
1258
1259         result = osd_attach_jinode(inode);
1260         if (result)
1261                 GOTO(out, result);
1262
1263         if (!ldiskfs_pdo)
1264                 GOTO(out, result = 0);
1265
1266         LASSERT(!obj->oo_hl_head);
1267         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1268
1269         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1270
1271 out:
1272         if (result || !obj->oo_inode) {
1273                 if (!IS_ERR_OR_NULL(inode))
1274                         iput(inode);
1275
1276                 obj->oo_inode = NULL;
1277                 if (trusted)
1278                         fid_zero(&oic->oic_fid);
1279         }
1280
1281         LINVRNT(osd_invariant(obj));
1282         return result;
1283 }
1284
1285 /*
1286  * Concurrency: shouldn't matter.
1287  */
1288 static void osd_object_init0(struct osd_object *obj)
1289 {
1290         LASSERT(obj->oo_inode != NULL);
1291         obj->oo_dt.do_body_ops = &osd_body_ops;
1292         obj->oo_dt.do_lu.lo_header->loh_attr |=
1293                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1294 }
1295
1296 /*
1297  * Concurrency: no concurrent access is possible that early in object
1298  * life-cycle.
1299  */
1300 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1301                            const struct lu_object_conf *conf)
1302 {
1303         struct osd_object *obj = osd_obj(l);
1304         int result;
1305
1306         LINVRNT(osd_invariant(obj));
1307
1308         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1309                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1310                 l->lo_header->loh_attr |= LOHA_EXISTS;
1311                 return 0;
1312         }
1313
1314         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1315         if (unlikely(l->lo_header ==
1316                      &osd_obj2dev(obj)->od_scrub.os_scrub.os_obj_header)) {
1317                 /* For the OI_scrub object during OSD device init. */
1318                 l->lo_header->loh_attr |= LOHA_EXISTS;
1319                 obj->oo_dt.do_body_ops = &osd_body_ops;
1320         } else {
1321                 obj->oo_dt.do_body_ops = &osd_body_ops_new;
1322         }
1323         if (result == 0 && obj->oo_inode != NULL) {
1324                 struct osd_thread_info *oti = osd_oti_get(env);
1325                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1326
1327                 osd_object_init0(obj);
1328                 result = osd_get_lma(oti, obj->oo_inode,
1329                                      &oti->oti_obj_dentry, loa);
1330                 if (!result) {
1331                         /* Convert LMAI flags to lustre LMA flags
1332                          * and cache it to oo_lma_flags */
1333                         obj->oo_lma_flags =
1334                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1335                 } else if (result == -ENODATA) {
1336                         result = 0;
1337                 }
1338         }
1339
1340         LINVRNT(osd_invariant(obj));
1341         return result;
1342 }
1343
1344 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1345  * The left part is for value, binary mode. */
1346 struct osd_xattr_entry {
1347         struct list_head        oxe_list;
1348         size_t                  oxe_len;
1349         size_t                  oxe_namelen;
1350         bool                    oxe_exist;
1351         struct rcu_head         oxe_rcu;
1352         char                    oxe_buf[0];
1353 };
1354
1355 static int osd_oxc_get(struct osd_object *obj, const char *name,
1356                        struct lu_buf *buf)
1357 {
1358         struct osd_xattr_entry *tmp;
1359         struct osd_xattr_entry *oxe = NULL;
1360         size_t namelen = strlen(name);
1361         int rc;
1362         ENTRY;
1363
1364         rcu_read_lock();
1365         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1366                 if (namelen == tmp->oxe_namelen &&
1367                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1368                         oxe = tmp;
1369                         break;
1370                 }
1371         }
1372
1373         if (oxe == NULL)
1374                 GOTO(out, rc = -ENOENT);
1375
1376         if (!oxe->oxe_exist)
1377                 GOTO(out, rc = -ENODATA);
1378
1379         /* vallen */
1380         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1381         LASSERT(rc > 0);
1382
1383         if (buf->lb_buf == NULL)
1384                 GOTO(out, rc);
1385
1386         if (buf->lb_len < rc)
1387                 GOTO(out, rc = -ERANGE);
1388
1389         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1390         EXIT;
1391 out:
1392         rcu_read_unlock();
1393
1394         return rc;
1395 }
1396
1397 static void osd_oxc_free(struct rcu_head *head)
1398 {
1399         struct osd_xattr_entry *oxe;
1400
1401         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1402         OBD_FREE(oxe, oxe->oxe_len);
1403 }
1404
1405 static void osd_oxc_add(struct osd_object *obj, const char *name,
1406                         const char *buf, int buflen)
1407 {
1408         struct osd_xattr_entry *oxe;
1409         struct osd_xattr_entry *old = NULL;
1410         struct osd_xattr_entry *tmp;
1411         size_t namelen = strlen(name);
1412         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1413
1414         OBD_ALLOC(oxe, len);
1415         if (oxe == NULL)
1416                 return;
1417
1418         INIT_LIST_HEAD(&oxe->oxe_list);
1419         oxe->oxe_len = len;
1420         oxe->oxe_namelen = namelen;
1421         memcpy(oxe->oxe_buf, name, namelen);
1422         if (buflen > 0) {
1423                 LASSERT(buf != NULL);
1424                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1425                 oxe->oxe_exist = true;
1426         } else {
1427                 oxe->oxe_exist = false;
1428         }
1429
1430         /* this should be rarely called, just remove old and add new */
1431         spin_lock(&obj->oo_guard);
1432         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1433                 if (namelen == tmp->oxe_namelen &&
1434                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1435                         old = tmp;
1436                         break;
1437                 }
1438         }
1439         if (old != NULL) {
1440                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1441                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1442         } else {
1443                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1444         }
1445         spin_unlock(&obj->oo_guard);
1446 }
1447
1448 static void osd_oxc_del(struct osd_object *obj, const char *name)
1449 {
1450         struct osd_xattr_entry *oxe;
1451         size_t namelen = strlen(name);
1452
1453         spin_lock(&obj->oo_guard);
1454         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1455                 if (namelen == oxe->oxe_namelen &&
1456                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1457                         list_del_rcu(&oxe->oxe_list);
1458                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1459                         break;
1460                 }
1461         }
1462         spin_unlock(&obj->oo_guard);
1463 }
1464
1465 static void osd_oxc_fini(struct osd_object *obj)
1466 {
1467         struct osd_xattr_entry *oxe, *next;
1468
1469         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1470                 list_del(&oxe->oxe_list);
1471                 OBD_FREE(oxe, oxe->oxe_len);
1472         }
1473 }
1474
1475 /*
1476  * Concurrency: no concurrent access is possible that late in object
1477  * life-cycle.
1478  */
1479 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1480 {
1481         struct osd_object *obj = osd_obj(l);
1482         struct lu_object_header *h = NULL;
1483
1484         LINVRNT(osd_invariant(obj));
1485
1486         if (unlikely(l->lo_header ==
1487                      &osd_obj2dev(obj)->od_scrub.os_scrub.os_obj_header))
1488                 h = l->lo_header;
1489
1490         osd_oxc_fini(obj);
1491         dt_object_fini(&obj->oo_dt);
1492         if (obj->oo_hl_head != NULL)
1493                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1494         OBD_FREE_PTR(obj);
1495         if (unlikely(h))
1496                 lu_object_header_fini(h);
1497 }
1498
1499 /*
1500  * Concurrency: no concurrent access is possible that late in object
1501  * life-cycle.
1502  */
1503 static void osd_index_fini(struct osd_object *o)
1504 {
1505         struct iam_container *bag;
1506
1507         if (o->oo_dir != NULL) {
1508                 bag = &o->oo_dir->od_container;
1509                 if (o->oo_inode != NULL) {
1510                         if (bag->ic_object == o->oo_inode)
1511                                 iam_container_fini(bag);
1512                 }
1513                 OBD_FREE_PTR(o->oo_dir);
1514                 o->oo_dir = NULL;
1515         }
1516 }
1517
1518 /*
1519  * Concurrency: no concurrent access is possible that late in object
1520  * life-cycle (for all existing callers, that is. New callers have to provide
1521  * their own locking.)
1522  */
1523 static int osd_inode_unlinked(const struct inode *inode)
1524 {
1525         return inode->i_nlink == 0;
1526 }
1527
1528 enum {
1529         OSD_TXN_OI_DELETE_CREDITS    = 20,
1530         OSD_TXN_INODE_DELETE_CREDITS = 20
1531 };
1532
1533 /*
1534  * Journal
1535  */
1536
1537 #if OSD_THANDLE_STATS
1538 /**
1539  * Set time when the handle is allocated
1540  */
1541 static void osd_th_alloced(struct osd_thandle *oth)
1542 {
1543         oth->oth_alloced = ktime_get();
1544 }
1545
1546 /**
1547  * Set time when the handle started
1548  */
1549 static void osd_th_started(struct osd_thandle *oth)
1550 {
1551         oth->oth_started = ktime_get();
1552 }
1553
1554 /**
1555  * Check whether the we deal with this handle for too long.
1556  */
1557 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1558                                 ktime_t alloced, ktime_t started,
1559                                 ktime_t closed)
1560 {
1561         ktime_t now = ktime_get();
1562
1563         LASSERT(dev != NULL);
1564
1565         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1566                             ktime_us_delta(started, alloced));
1567         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1568                             ktime_us_delta(closed, started));
1569         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1570                             ktime_us_delta(now, closed));
1571
1572         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1573                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1574                       oth, now, alloced, started, closed);
1575                 libcfs_debug_dumpstack(NULL);
1576         }
1577 }
1578
1579 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1580 {                                                                       \
1581         ktime_t __closed = ktime_get();                                 \
1582         ktime_t __alloced = oth->oth_alloced;                           \
1583         ktime_t __started = oth->oth_started;                           \
1584                                                                         \
1585         expr;                                                           \
1586         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1587 }
1588
1589 #else /* OSD_THANDLE_STATS */
1590
1591 #define osd_th_alloced(h)                  do {} while(0)
1592 #define osd_th_started(h)                  do {} while(0)
1593 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1594
1595 #endif /* OSD_THANDLE_STATS */
1596
1597 /*
1598  * Concurrency: doesn't access mutable data.
1599  */
1600 static int osd_param_is_not_sane(const struct osd_device *dev,
1601                                  const struct thandle *th)
1602 {
1603         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1604
1605         return oh->ot_credits > osd_transaction_size(dev);
1606 }
1607
1608 /*
1609  * Concurrency: shouldn't matter.
1610  */
1611 static void osd_trans_commit_cb(struct super_block *sb,
1612                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1613 {
1614         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1615         struct thandle     *th  = &oh->ot_super;
1616         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1617         struct dt_txn_commit_cb *dcb, *tmp;
1618
1619         LASSERT(oh->ot_handle == NULL);
1620
1621         if (error)
1622                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1623
1624         dt_txn_hook_commit(th);
1625
1626         /* call per-transaction callbacks if any */
1627         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1628                                  dcb_linkage) {
1629                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1630                          "commit callback entry: magic=%x name='%s'\n",
1631                          dcb->dcb_magic, dcb->dcb_name);
1632                 list_del_init(&dcb->dcb_linkage);
1633                 dcb->dcb_func(NULL, th, dcb, error);
1634         }
1635
1636         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1637         lu_device_put(lud);
1638         th->th_dev = NULL;
1639
1640         OBD_FREE_PTR(oh);
1641 }
1642
1643 #ifndef HAVE_SB_START_WRITE
1644 # define sb_start_write(sb) do {} while (0)
1645 # define sb_end_write(sb) do {} while (0)
1646 #endif
1647
1648 static struct thandle *osd_trans_create(const struct lu_env *env,
1649                                         struct dt_device *d)
1650 {
1651         struct osd_thread_info  *oti = osd_oti_get(env);
1652         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1653         struct osd_thandle      *oh;
1654         struct thandle          *th;
1655         ENTRY;
1656
1657         if (d->dd_rdonly) {
1658                 CERROR("%s: someone try to start transaction under "
1659                        "readonly mode, should be disabled.\n",
1660                        osd_name(osd_dt_dev(d)));
1661                 dump_stack();
1662                 RETURN(ERR_PTR(-EROFS));
1663         }
1664
1665         /* on pending IO in this thread should left from prev. request */
1666         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1667
1668         sb_start_write(osd_sb(osd_dt_dev(d)));
1669
1670         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1671         if (oh != NULL) {
1672                 oh->ot_quota_trans = &oti->oti_quota_trans;
1673                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1674                 th = &oh->ot_super;
1675                 th->th_dev = d;
1676                 th->th_result = 0;
1677                 oh->ot_credits = 0;
1678                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1679                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1680                 osd_th_alloced(oh);
1681
1682                 memset(oti->oti_declare_ops, 0,
1683                        sizeof(oti->oti_declare_ops));
1684                 memset(oti->oti_declare_ops_cred, 0,
1685                        sizeof(oti->oti_declare_ops_cred));
1686                 memset(oti->oti_declare_ops_used, 0,
1687                        sizeof(oti->oti_declare_ops_used));
1688         } else {
1689                 sb_end_write(osd_sb(osd_dt_dev(d)));
1690                 th = ERR_PTR(-ENOMEM);
1691         }
1692         RETURN(th);
1693 }
1694
1695 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1696 {
1697         struct osd_thread_info  *oti = osd_oti_get(env);
1698         struct osd_thandle      *oh;
1699
1700         oh = container_of0(th, struct osd_thandle, ot_super);
1701         LASSERT(oh != NULL);
1702
1703         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1704               oti->oti_declare_ops[OSD_OT_CREATE],
1705               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1706               oti->oti_declare_ops_used[OSD_OT_CREATE],
1707               oti->oti_declare_ops[OSD_OT_DESTROY],
1708               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1709               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1710         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1711               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1712               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1713               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1714               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1715               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1716               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1717         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1718               oti->oti_declare_ops[OSD_OT_WRITE],
1719               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1720               oti->oti_declare_ops_used[OSD_OT_WRITE],
1721               oti->oti_declare_ops[OSD_OT_PUNCH],
1722               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1723               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1724               oti->oti_declare_ops[OSD_OT_QUOTA],
1725               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1726               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1727         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1728               oti->oti_declare_ops[OSD_OT_INSERT],
1729               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1730               oti->oti_declare_ops_used[OSD_OT_INSERT],
1731               oti->oti_declare_ops[OSD_OT_DELETE],
1732               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1733               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1734         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1735               oti->oti_declare_ops[OSD_OT_REF_ADD],
1736               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1737               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1738               oti->oti_declare_ops[OSD_OT_REF_DEL],
1739               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1740               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1741 }
1742
1743 /*
1744  * Concurrency: shouldn't matter.
1745  */
1746 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1747                            struct thandle *th)
1748 {
1749         struct osd_thread_info *oti = osd_oti_get(env);
1750         struct osd_device  *dev = osd_dt_dev(d);
1751         handle_t           *jh;
1752         struct osd_thandle *oh;
1753         int rc;
1754
1755         ENTRY;
1756
1757         LASSERT(current->journal_info == NULL);
1758
1759         oh = container_of0(th, struct osd_thandle, ot_super);
1760         LASSERT(oh != NULL);
1761         LASSERT(oh->ot_handle == NULL);
1762
1763         rc = dt_txn_hook_start(env, d, th);
1764         if (rc != 0)
1765                 GOTO(out, rc);
1766
1767         if (unlikely(osd_param_is_not_sane(dev, th))) {
1768                 static unsigned long last_printed;
1769                 static int last_credits;
1770
1771                 /* don't make noise on a tiny testing systems
1772                  * actual credits misuse will be caught anyway */
1773                 if (last_credits != oh->ot_credits &&
1774                     time_after(jiffies, last_printed +
1775                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1776                     osd_transaction_size(dev) > 512) {
1777                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1778                               oh->ot_credits, osd_transaction_size(dev));
1779                         osd_trans_dump_creds(env, th);
1780                         libcfs_debug_dumpstack(NULL);
1781                         last_credits = oh->ot_credits;
1782                         last_printed = jiffies;
1783                 }
1784                 /* XXX Limit the credits to 'max_transaction_buffers', and
1785                  *     let the underlying filesystem to catch the error if
1786                  *     we really need so many credits.
1787                  *
1788                  *     This should be removed when we can calculate the
1789                  *     credits precisely. */
1790                 oh->ot_credits = osd_transaction_size(dev);
1791         } else if (ldiskfs_track_declares_assert != 0) {
1792                 /* reserve few credits to prevent an assertion in JBD
1793                  * our debugging mechanism will be able to detected
1794                  * overuse. this can help to debug single-update
1795                  * transactions */
1796                 oh->ot_credits += 10;
1797                 if (unlikely(osd_param_is_not_sane(dev, th)))
1798                         oh->ot_credits = osd_transaction_size(dev);
1799         }
1800
1801         /*
1802          * XXX temporary stuff. Some abstraction layer should
1803          * be used.
1804          */
1805         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1806         osd_th_started(oh);
1807         if (!IS_ERR(jh)) {
1808                 oh->ot_handle = jh;
1809                 LASSERT(oti->oti_txns == 0);
1810
1811                 lu_device_get(&d->dd_lu_dev);
1812                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1813                               "osd-tx", th);
1814                 oti->oti_txns++;
1815                 rc = 0;
1816         } else {
1817                 rc = PTR_ERR(jh);
1818         }
1819 out:
1820         RETURN(rc);
1821 }
1822
1823 static int osd_seq_exists(const struct lu_env *env,
1824                           struct osd_device *osd, u64 seq)
1825 {
1826         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1827         struct seq_server_site  *ss = osd_seq_site(osd);
1828         int                     rc;
1829         ENTRY;
1830
1831         LASSERT(ss != NULL);
1832         LASSERT(ss->ss_server_fld != NULL);
1833
1834         rc = osd_fld_lookup(env, osd, seq, range);
1835         if (rc != 0) {
1836                 if (rc != -ENOENT)
1837                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1838                                osd_name(osd), seq, rc);
1839                 RETURN(0);
1840         }
1841
1842         RETURN(ss->ss_node_id == range->lsr_index);
1843 }
1844
1845 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1846 {
1847         struct dt_txn_commit_cb *dcb;
1848         struct dt_txn_commit_cb *tmp;
1849
1850         /* call per-transaction stop callbacks if any */
1851         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1852                                  dcb_linkage) {
1853                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1854                          "commit callback entry: magic=%x name='%s'\n",
1855                          dcb->dcb_magic, dcb->dcb_name);
1856                 list_del_init(&dcb->dcb_linkage);
1857                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1858         }
1859 }
1860
1861 /*
1862  * Concurrency: shouldn't matter.
1863  */
1864 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1865                           struct thandle *th)
1866 {
1867         int                     rc = 0, remove_agents = 0;
1868         struct osd_thandle     *oh;
1869         struct osd_thread_info *oti = osd_oti_get(env);
1870         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1871         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1872         struct qsd_instance    *qsd = osd->od_quota_slave;
1873         struct lquota_trans    *qtrans;
1874         ENTRY;
1875
1876         oh = container_of0(th, struct osd_thandle, ot_super);
1877
1878         /* reset OI cache for safety */
1879         oti->oti_ins_cache_used = 0;
1880
1881         remove_agents = oh->ot_remove_agents;
1882
1883         qtrans = oh->ot_quota_trans;
1884         oh->ot_quota_trans = NULL;
1885
1886         if (oh->ot_handle != NULL) {
1887                 int rc2;
1888                 handle_t *hdl = oh->ot_handle;
1889
1890                 /*
1891                  * add commit callback
1892                  * notice we don't do this in osd_trans_start()
1893                  * as underlying transaction can change during truncate
1894                  */
1895                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1896                                          &oh->ot_jcb);
1897
1898                 LASSERT(oti->oti_txns == 1);
1899                 oti->oti_txns--;
1900
1901                 rc = dt_txn_hook_stop(env, th);
1902                 if (rc != 0)
1903                         CERROR("%s: failed in transaction hook: rc = %d\n",
1904                                osd_name(osd), rc);
1905
1906                 osd_trans_stop_cb(oh, rc);
1907                 /* hook functions might modify th_sync */
1908                 hdl->h_sync = th->th_sync;
1909
1910                 oh->ot_handle = NULL;
1911                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1912                 if (rc2 != 0)
1913                         CERROR("%s: failed to stop transaction: rc = %d\n",
1914                                osd_name(osd), rc2);
1915                 if (!rc)
1916                         rc = rc2;
1917         } else {
1918                 osd_trans_stop_cb(oh, th->th_result);
1919                 OBD_FREE_PTR(oh);
1920         }
1921
1922         /* inform the quota slave device that the transaction is stopping */
1923         qsd_op_end(env, qsd, qtrans);
1924
1925         /* as we want IO to journal and data IO be concurrent, we don't block
1926          * awaiting data IO completion in osd_do_bio(), instead we wait here
1927          * once transaction is submitted to the journal. all reqular requests
1928          * don't do direct IO (except read/write), thus this wait_event becomes
1929          * no-op for them.
1930          *
1931          * IMPORTANT: we have to wait till any IO submited by the thread is
1932          * completed otherwise iobuf may be corrupted by different request
1933          */
1934         wait_event(iobuf->dr_wait,
1935                        atomic_read(&iobuf->dr_numreqs) == 0);
1936         osd_fini_iobuf(osd, iobuf);
1937         if (!rc)
1938                 rc = iobuf->dr_error;
1939
1940         if (unlikely(remove_agents != 0))
1941                 osd_process_scheduled_agent_removals(env, osd);
1942
1943         sb_end_write(osd_sb(osd));
1944
1945         RETURN(rc);
1946 }
1947
1948 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1949 {
1950         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1951                                                ot_super);
1952
1953         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1954         LASSERT(&dcb->dcb_func != NULL);
1955         if (dcb->dcb_flags & DCB_TRANS_STOP)
1956                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1957         else
1958                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1959
1960         return 0;
1961 }
1962
1963 /*
1964  * Called just before object is freed. Releases all resources except for
1965  * object itself (that is released by osd_object_free()).
1966  *
1967  * Concurrency: no concurrent access is possible that late in object
1968  * life-cycle.
1969  */
1970 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1971 {
1972         struct osd_object *obj   = osd_obj(l);
1973         struct inode      *inode = obj->oo_inode;
1974
1975         LINVRNT(osd_invariant(obj));
1976
1977         /*
1978          * If object is unlinked remove fid->ino mapping from object index.
1979          */
1980
1981         osd_index_fini(obj);
1982         if (inode != NULL) {
1983                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1984                 qid_t                    uid = i_uid_read(inode);
1985                 qid_t                    gid = i_gid_read(inode);
1986
1987                 iput(inode);
1988                 obj->oo_inode = NULL;
1989
1990                 if (qsd != NULL) {
1991                         struct osd_thread_info  *info = osd_oti_get(env);
1992                         struct lquota_id_info   *qi = &info->oti_qi;
1993
1994                         /* Release granted quota to master if necessary */
1995                         qi->lqi_id.qid_uid = uid;
1996                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1997
1998                         qi->lqi_id.qid_uid = gid;
1999                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2000
2001                         qi->lqi_id.qid_uid = i_projid_read(inode);
2002                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2003                 }
2004         }
2005 }
2006
2007 /*
2008  * Concurrency: ->loo_object_release() is called under site spin-lock.
2009  */
2010 static void osd_object_release(const struct lu_env *env,
2011                                struct lu_object *l)
2012 {
2013         struct osd_object *o = osd_obj(l);
2014         /* nobody should be releasing a non-destroyed object with nlink=0
2015          * the API allows this, but ldiskfs doesn't like and then report
2016          * this inode as deleted */
2017         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
2018                 LBUG();
2019 }
2020
2021 /*
2022  * Concurrency: shouldn't matter.
2023  */
2024 static int osd_object_print(const struct lu_env *env, void *cookie,
2025                             lu_printer_t p, const struct lu_object *l)
2026 {
2027         struct osd_object *o = osd_obj(l);
2028         struct iam_descr  *d;
2029
2030         if (o->oo_dir != NULL)
2031                 d = o->oo_dir->od_container.ic_descr;
2032         else
2033                 d = NULL;
2034         return (*p)(env, cookie,
2035                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2036                     o, o->oo_inode,
2037                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2038                     o->oo_inode ? o->oo_inode->i_generation : 0,
2039                     d ? d->id_ops->id_name : "plain");
2040 }
2041
2042 /*
2043  * Concurrency: shouldn't matter.
2044  */
2045 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2046                struct obd_statfs *sfs)
2047 {
2048         struct osd_device       *osd = osd_dt_dev(d);
2049         struct super_block      *sb = osd_sb(osd);
2050         struct kstatfs          *ksfs;
2051         __u64                    reserved;
2052         int                      result = 0;
2053
2054         if (unlikely(osd->od_mnt == NULL))
2055                 return -EINPROGRESS;
2056
2057         /* osd_lproc.c call this without env, allocate ksfs for that case */
2058         if (unlikely(env == NULL)) {
2059                 OBD_ALLOC_PTR(ksfs);
2060                 if (ksfs == NULL)
2061                         return -ENOMEM;
2062         } else {
2063                 ksfs = &osd_oti_get(env)->oti_ksfs;
2064         }
2065
2066         result = sb->s_op->statfs(sb->s_root, ksfs);
2067         if (result)
2068                 goto out;
2069
2070         statfs_pack(sfs, ksfs);
2071         if (unlikely(sb->s_flags & MS_RDONLY))
2072                 sfs->os_state |= OS_STATE_READONLY;
2073         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2074                                          LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2075                 sfs->os_maxbytes = sb->s_maxbytes;
2076         else
2077                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2078
2079         /*
2080          * Reserve some space so to avoid fragmenting the filesystem too much.
2081          * Fragmentation not only impacts performance, but can also increase
2082          * metadata overhead significantly, causing grant calculation to be
2083          * wrong.
2084          *
2085          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2086          */
2087         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2088         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2089         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2090                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2091
2092         sfs->os_blocks -= reserved;
2093         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2094         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2095
2096 out:
2097         if (unlikely(env == NULL))
2098                 OBD_FREE_PTR(ksfs);
2099         return result;
2100 }
2101
2102 /**
2103  * Estimate space needed for file creations. We assume the largest filename
2104  * which is 2^64 - 1, hence a filename of 20 chars.
2105  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2106  */
2107 #ifdef __LDISKFS_DIR_REC_LEN
2108 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2109 #else
2110 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2111 #endif
2112
2113 /*
2114  * Concurrency: doesn't access mutable data.
2115  */
2116 static void osd_conf_get(const struct lu_env *env,
2117                          const struct dt_device *dev,
2118                          struct dt_device_param *param)
2119 {
2120         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2121         int                ea_overhead;
2122
2123         /*
2124          * XXX should be taken from not-yet-existing fs abstraction layer.
2125          */
2126         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2127         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2128         param->ddp_symlink_max  = sb->s_blocksize;
2129         param->ddp_mount_type     = LDD_MT_LDISKFS;
2130         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2131                 param->ddp_maxbytes = sb->s_maxbytes;
2132         else
2133                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2134         /* inode are statically allocated, so per-inode space consumption
2135          * is the space consumed by the directory entry */
2136         param->ddp_inodespace     = PER_OBJ_USAGE;
2137         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2138          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2139          * maximum length to not under count the actual number of extents
2140          * needed for writing a file. */
2141         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2142         /* worst-case extent insertion metadata overhead */
2143         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2144         param->ddp_mntopts      = 0;
2145         if (test_opt(sb, XATTR_USER))
2146                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2147         if (test_opt(sb, POSIX_ACL))
2148                 param->ddp_mntopts |= MNTOPT_ACL;
2149
2150         /* LOD might calculate the max stripe count based on max_ea_size,
2151          * so we need take account in the overhead as well,
2152          * xattr_header + magic + xattr_entry_head */
2153         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2154                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2155
2156 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2157         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2158                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2159                                                                 ea_overhead;
2160         else
2161 #endif
2162                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2163
2164         /* Preferred RPC size for efficient disk IO.  4MB shows good
2165          * all-around performance for ldiskfs, but use bigalloc chunk size
2166          * by default if larger. */
2167 #if defined(LDISKFS_CLUSTER_SIZE)
2168         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2169                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2170         else
2171 #endif
2172                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2173 }
2174
2175 /*
2176  * Concurrency: shouldn't matter.
2177  */
2178 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2179 {
2180         int rc;
2181
2182         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2183
2184         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2185
2186         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2187                rc);
2188
2189         return rc;
2190 }
2191
2192 /**
2193  * Start commit for OSD device.
2194  *
2195  * An implementation of dt_commit_async method for OSD device.
2196  * Asychronously starts underlayng fs sync and thereby a transaction
2197  * commit.
2198  *
2199  * \param env environment
2200  * \param d dt device
2201  *
2202  * \see dt_device_operations
2203  */
2204 static int osd_commit_async(const struct lu_env *env,
2205                             struct dt_device *d)
2206 {
2207         struct super_block *s = osd_sb(osd_dt_dev(d));
2208         ENTRY;
2209
2210         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2211         RETURN(s->s_op->sync_fs(s, 0));
2212 }
2213
2214 /* Our own copy of the set readonly functions if present, or NU if not. */
2215 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2216 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2217 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2218
2219 /*
2220  * Concurrency: shouldn't matter.
2221  */
2222 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2223 {
2224         struct super_block *sb = osd_sb(osd_dt_dev(d));
2225         struct block_device *dev = sb->s_bdev;
2226         int rc = -EOPNOTSUPP;
2227         ENTRY;
2228
2229         if (priv_dev_set_rdonly) {
2230                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2231
2232                 rc = 0;
2233                 CERROR("*** setting %s read-only ***\n",
2234                        osd_dt_dev(d)->od_svname);
2235
2236                 if (sb->s_op->freeze_fs) {
2237                         rc = sb->s_op->freeze_fs(sb);
2238                         if (rc)
2239                                 goto out;
2240                 }
2241
2242                 if (jdev && (jdev != dev)) {
2243                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2244                                (long)jdev);
2245                         priv_dev_set_rdonly(jdev);
2246                 }
2247                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2248                 priv_dev_set_rdonly(dev);
2249
2250                 if (sb->s_op->unfreeze_fs)
2251                         sb->s_op->unfreeze_fs(sb);
2252         }
2253
2254 out:
2255         if (rc)
2256                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2257                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2258
2259         RETURN(rc);
2260 }
2261
2262 /**
2263  * Note: we do not count into QUOTA here.
2264  * If we mount with --data_journal we may need more.
2265  */
2266 const int osd_dto_credits_noquota[DTO_NR] = {
2267         /**
2268          * Insert.
2269          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2270          * SINGLEDATA_TRANS_BLOCKS(8)
2271          * XXX Note: maybe iam need more, since iam have more level than
2272          *           EXT3 htree.
2273          */
2274         [DTO_INDEX_INSERT]  = 16,
2275         /**
2276          * Delete
2277          * just modify a single entry, probably merge few within a block
2278          */
2279         [DTO_INDEX_DELETE]  = 1,
2280         /**
2281          * Used for OI scrub
2282          */
2283         [DTO_INDEX_UPDATE]  = 16,
2284         /**
2285          * 4(inode, inode bits, groups, GDT)
2286          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2287          */
2288         [DTO_OBJECT_CREATE] = 4,
2289         /**
2290          * 4(inode, inode bits, groups, GDT)
2291          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2292          */
2293         [DTO_OBJECT_DELETE] = 4,
2294         /**
2295          * Attr set credits (inode)
2296          */
2297         [DTO_ATTR_SET_BASE] = 1,
2298         /**
2299          * Xattr set. The same as xattr of EXT3.
2300          * DATA_TRANS_BLOCKS(14)
2301          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2302          * are also counted in. Do not know why?
2303          */
2304         [DTO_XATTR_SET]     = 14,
2305         /**
2306          * credits for inode change during write.
2307          */
2308         [DTO_WRITE_BASE]    = 3,
2309         /**
2310          * credits for single block write.
2311          */
2312         [DTO_WRITE_BLOCK]   = 14,
2313         /**
2314          * Attr set credits for chown.
2315          * This is extra credits for setattr, and it is null without quota
2316          */
2317         [DTO_ATTR_SET_CHOWN] = 0
2318 };
2319
2320 static const struct dt_device_operations osd_dt_ops = {
2321         .dt_root_get       = osd_root_get,
2322         .dt_statfs         = osd_statfs,
2323         .dt_trans_create   = osd_trans_create,
2324         .dt_trans_start    = osd_trans_start,
2325         .dt_trans_stop     = osd_trans_stop,
2326         .dt_trans_cb_add   = osd_trans_cb_add,
2327         .dt_conf_get       = osd_conf_get,
2328         .dt_sync           = osd_sync,
2329         .dt_ro             = osd_ro,
2330         .dt_commit_async   = osd_commit_async,
2331 };
2332
2333 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2334                           unsigned role)
2335 {
2336         struct osd_object *obj = osd_dt_obj(dt);
2337         struct osd_thread_info *oti = osd_oti_get(env);
2338
2339         LINVRNT(osd_invariant(obj));
2340
2341         LASSERT(obj->oo_owner != env);
2342         down_read_nested(&obj->oo_sem, role);
2343
2344         LASSERT(obj->oo_owner == NULL);
2345         oti->oti_r_locks++;
2346 }
2347
2348 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2349                            unsigned role)
2350 {
2351         struct osd_object *obj = osd_dt_obj(dt);
2352         struct osd_thread_info *oti = osd_oti_get(env);
2353
2354         LINVRNT(osd_invariant(obj));
2355
2356         LASSERT(obj->oo_owner != env);
2357         down_write_nested(&obj->oo_sem, role);
2358
2359         LASSERT(obj->oo_owner == NULL);
2360         obj->oo_owner = env;
2361         oti->oti_w_locks++;
2362 }
2363
2364 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2365 {
2366         struct osd_object *obj = osd_dt_obj(dt);
2367         struct osd_thread_info *oti = osd_oti_get(env);
2368
2369         LINVRNT(osd_invariant(obj));
2370
2371         LASSERT(oti->oti_r_locks > 0);
2372         oti->oti_r_locks--;
2373         up_read(&obj->oo_sem);
2374 }
2375
2376 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2377 {
2378         struct osd_object *obj = osd_dt_obj(dt);
2379         struct osd_thread_info *oti = osd_oti_get(env);
2380
2381         LINVRNT(osd_invariant(obj));
2382
2383         LASSERT(obj->oo_owner == env);
2384         LASSERT(oti->oti_w_locks > 0);
2385         oti->oti_w_locks--;
2386         obj->oo_owner = NULL;
2387         up_write(&obj->oo_sem);
2388 }
2389
2390 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2391 {
2392         struct osd_object *obj = osd_dt_obj(dt);
2393
2394         LINVRNT(osd_invariant(obj));
2395
2396         return obj->oo_owner == env;
2397 }
2398
2399 static struct timespec *osd_inode_time(const struct lu_env *env,
2400                                        struct inode *inode, __u64 seconds)
2401 {
2402         struct osd_thread_info  *oti = osd_oti_get(env);
2403         struct timespec         *t   = &oti->oti_time;
2404
2405         t->tv_sec = seconds;
2406         t->tv_nsec = 0;
2407         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2408         return t;
2409 }
2410
2411 static void osd_inode_getattr(const struct lu_env *env,
2412                               struct inode *inode, struct lu_attr *attr)
2413 {
2414         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2415                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2416                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2417                            LA_BLKSIZE | LA_TYPE;
2418
2419         attr->la_atime   = LTIME_S(inode->i_atime);
2420         attr->la_mtime   = LTIME_S(inode->i_mtime);
2421         attr->la_ctime   = LTIME_S(inode->i_ctime);
2422         attr->la_mode    = inode->i_mode;
2423         attr->la_size    = i_size_read(inode);
2424         attr->la_blocks  = inode->i_blocks;
2425         attr->la_uid     = i_uid_read(inode);
2426         attr->la_gid     = i_gid_read(inode);
2427         attr->la_projid  = i_projid_read(inode);
2428         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2429         attr->la_nlink   = inode->i_nlink;
2430         attr->la_rdev    = inode->i_rdev;
2431         attr->la_blksize = 1 << inode->i_blkbits;
2432         attr->la_blkbits = inode->i_blkbits;
2433         /*
2434          * Ext4 did not transfer inherit flags from raw inode
2435          * to inode flags, and ext4 internally test raw inode
2436          * @i_flags directly. Instead of patching ext4, we do it here.
2437          */
2438         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2439                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2440 }
2441
2442 static int osd_attr_get(const struct lu_env *env,
2443                         struct dt_object *dt,
2444                         struct lu_attr *attr)
2445 {
2446         struct osd_object *obj = osd_dt_obj(dt);
2447
2448         if (unlikely(!dt_object_exists(dt)))
2449                 return -ENOENT;
2450         if (unlikely(obj->oo_destroyed))
2451                 return -ENOENT;
2452
2453         LASSERT(!dt_object_remote(dt));
2454         LINVRNT(osd_invariant(obj));
2455
2456         spin_lock(&obj->oo_guard);
2457         osd_inode_getattr(env, obj->oo_inode, attr);
2458         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2459                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2460         spin_unlock(&obj->oo_guard);
2461
2462         return 0;
2463 }
2464
2465 static int osd_declare_attr_qid(const struct lu_env *env,
2466                                 struct osd_object *obj,
2467                                 struct osd_thandle *oh, long long bspace,
2468                                 qid_t old_id, qid_t new_id, bool enforce,
2469                                 unsigned type)
2470 {
2471         int rc;
2472         struct osd_thread_info *info = osd_oti_get(env);
2473         struct lquota_id_info  *qi = &info->oti_qi;
2474
2475         qi->lqi_type = type;
2476         /* inode accounting */
2477         qi->lqi_is_blk = false;
2478
2479         /* one more inode for the new id ... */
2480         qi->lqi_id.qid_uid = new_id;
2481         qi->lqi_space      = 1;
2482         /* Reserve credits for the new id */
2483         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2484         if (rc == -EDQUOT || rc == -EINPROGRESS)
2485                 rc = 0;
2486         if (rc)
2487                 RETURN(rc);
2488
2489         /* and one less inode for the current id */
2490         qi->lqi_id.qid_uid = old_id;
2491         qi->lqi_space      = -1;
2492         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2493         if (rc == -EDQUOT || rc == -EINPROGRESS)
2494                 rc = 0;
2495         if (rc)
2496                 RETURN(rc);
2497
2498         /* block accounting */
2499         qi->lqi_is_blk = true;
2500
2501         /* more blocks for the new id ... */
2502         qi->lqi_id.qid_uid = new_id;
2503         qi->lqi_space      = bspace;
2504         /*
2505          * Credits for the new uid has been reserved, re-use "obj"
2506          * to save credit reservation.
2507          */
2508         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2509         if (rc == -EDQUOT || rc == -EINPROGRESS)
2510                 rc = 0;
2511         if (rc)
2512                 RETURN(rc);
2513
2514         /* and finally less blocks for the current uid */
2515         qi->lqi_id.qid_uid = old_id;
2516         qi->lqi_space      = -bspace;
2517         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2518         if (rc == -EDQUOT || rc == -EINPROGRESS)
2519                 rc = 0;
2520
2521         RETURN(rc);
2522 }
2523
2524 static int osd_declare_attr_set(const struct lu_env *env,
2525                                 struct dt_object *dt,
2526                                 const struct lu_attr *attr,
2527                                 struct thandle *handle)
2528 {
2529         struct osd_thandle     *oh;
2530         struct osd_object      *obj;
2531         qid_t                   uid;
2532         qid_t                   gid;
2533         long long               bspace;
2534         int                     rc = 0;
2535         bool                    enforce;
2536         ENTRY;
2537
2538         LASSERT(dt != NULL);
2539         LASSERT(handle != NULL);
2540
2541         obj = osd_dt_obj(dt);
2542         LASSERT(osd_invariant(obj));
2543
2544         oh = container_of0(handle, struct osd_thandle, ot_super);
2545         LASSERT(oh->ot_handle == NULL);
2546
2547         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2548                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2549
2550         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2551                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2552
2553         if (attr == NULL || obj->oo_inode == NULL)
2554                 RETURN(rc);
2555
2556         bspace   = obj->oo_inode->i_blocks;
2557         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2558         bspace   = toqb(bspace);
2559
2560         /* Changing ownership is always preformed by super user, it should not
2561          * fail with EDQUOT.
2562          *
2563          * We still need to call the osd_declare_qid() to calculate the journal
2564          * credits for updating quota accounting files and to trigger quota
2565          * space adjustment once the operation is completed.*/
2566         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2567                 /* USERQUOTA */
2568                 uid = i_uid_read(obj->oo_inode);
2569                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2570                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2571                                           attr->la_uid, enforce, USRQUOTA);
2572                 if (rc)
2573                         RETURN(rc);
2574
2575                 gid = i_gid_read(obj->oo_inode);
2576                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2577                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2578                                           i_gid_read(obj->oo_inode),
2579                                           attr->la_gid, enforce, GRPQUOTA);
2580                 if (rc)
2581                         RETURN(rc);
2582
2583         }
2584 #ifdef HAVE_PROJECT_QUOTA
2585         if (attr->la_valid & LA_PROJID) {
2586                 __u32 projid = i_projid_read(obj->oo_inode);
2587                 enforce = (attr->la_valid & LA_PROJID) &&
2588                                         (attr->la_projid != projid);
2589                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2590                                           (qid_t)projid, (qid_t)attr->la_projid,
2591                                           enforce, PRJQUOTA);
2592                 if (rc)
2593                         RETURN(rc);
2594         }
2595 #endif
2596         RETURN(rc);
2597 }
2598
2599 static int osd_inode_setattr(const struct lu_env *env,
2600                              struct inode *inode, const struct lu_attr *attr)
2601 {
2602         __u64 bits = attr->la_valid;
2603
2604         /* Only allow set size for regular file */
2605         if (!S_ISREG(inode->i_mode))
2606                 bits &= ~(LA_SIZE | LA_BLOCKS);
2607
2608         if (bits == 0)
2609                 return 0;
2610
2611         if (bits & LA_ATIME)
2612                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2613         if (bits & LA_CTIME)
2614                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2615         if (bits & LA_MTIME)
2616                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2617         if (bits & LA_SIZE) {
2618                 spin_lock(&inode->i_lock);
2619                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2620                 i_size_write(inode, attr->la_size);
2621                 spin_unlock(&inode->i_lock);
2622         }
2623
2624         /* OSD should not change "i_blocks" which is used by quota.
2625          * "i_blocks" should be changed by ldiskfs only. */
2626         if (bits & LA_MODE)
2627                 inode->i_mode = (inode->i_mode & S_IFMT) |
2628                                 (attr->la_mode & ~S_IFMT);
2629         if (bits & LA_UID)
2630                 i_uid_write(inode, attr->la_uid);
2631         if (bits & LA_GID)
2632                 i_gid_write(inode, attr->la_gid);
2633         if (bits & LA_PROJID)
2634                 i_projid_write(inode, attr->la_projid);
2635         if (bits & LA_NLINK)
2636                 set_nlink(inode, attr->la_nlink);
2637         if (bits & LA_RDEV)
2638                 inode->i_rdev = attr->la_rdev;
2639
2640         if (bits & LA_FLAGS) {
2641                 /* always keep S_NOCMTIME */
2642                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2643                                  S_NOCMTIME;
2644                 /*
2645                  * Ext4 did not transfer inherit flags from
2646                  * @inode->i_flags to raw inode i_flags when writing
2647                  * flags, we do it explictly here.
2648                  */
2649                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2650                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2651                 else
2652                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2653         }
2654         return 0;
2655 }
2656
2657 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2658 {
2659         int rc;
2660
2661         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2662             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2663                 struct iattr    iattr;
2664
2665                 ll_vfs_dq_init(inode);
2666                 iattr.ia_valid = 0;
2667                 if (attr->la_valid & LA_UID)
2668                         iattr.ia_valid |= ATTR_UID;
2669                 if (attr->la_valid & LA_GID)
2670                         iattr.ia_valid |= ATTR_GID;
2671                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2672                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2673
2674                 rc = ll_vfs_dq_transfer(inode, &iattr);
2675                 if (rc) {
2676                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2677                                "enforcement enabled on the ldiskfs "
2678                                "filesystem?\n", inode->i_sb->s_id, rc);
2679                         return rc;
2680                 }
2681         }
2682
2683         /* Handle project id transfer here properly */
2684         if (attr->la_valid & LA_PROJID &&
2685             attr->la_projid != i_projid_read(inode)) {
2686 #ifdef HAVE_PROJECT_QUOTA
2687                 rc = ldiskfs_transfer_project(inode, attr->la_projid);
2688 #else
2689                 rc = -ENOTSUPP;
2690 #endif
2691                 if (rc) {
2692                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2693                                "enforcement enabled on the ldiskfs "
2694                                "filesystem?\n", inode->i_sb->s_id, rc);
2695                         return rc;
2696                 }
2697         }
2698         return 0;
2699 }
2700
2701 static int osd_attr_set(const struct lu_env *env,
2702                         struct dt_object *dt,
2703                         const struct lu_attr *attr,
2704                         struct thandle *handle)
2705 {
2706         struct osd_object *obj = osd_dt_obj(dt);
2707         struct inode      *inode;
2708         int rc;
2709
2710         if (!dt_object_exists(dt))
2711                 return -ENOENT;
2712
2713         LASSERT(handle != NULL);
2714         LASSERT(!dt_object_remote(dt));
2715         LASSERT(osd_invariant(obj));
2716
2717         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2718
2719         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2720             !osd_obj2dev(obj)->od_is_ost) {
2721                 struct osd_thread_info  *oti  = osd_oti_get(env);
2722                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2723                 struct lu_fid           *fid1 = &oti->oti_fid;
2724                 struct osd_inode_id     *id   = &oti->oti_id;
2725                 struct iam_path_descr   *ipd;
2726                 struct iam_container    *bag;
2727                 struct osd_thandle      *oh;
2728                 int                      rc;
2729
2730                 fid_cpu_to_be(fid1, fid0);
2731                 memset(id, 1, sizeof(*id));
2732                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2733                                   fid0)->oi_dir.od_container;
2734                 ipd = osd_idx_ipd_get(env, bag);
2735                 if (unlikely(ipd == NULL))
2736                         RETURN(-ENOMEM);
2737
2738                 oh = container_of0(handle, struct osd_thandle, ot_super);
2739                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2740                                 (const struct iam_rec *)id, ipd);
2741                 osd_ipd_put(env, bag, ipd);
2742                 return(rc > 0 ? 0 : rc);
2743         }
2744
2745         inode = obj->oo_inode;
2746
2747         rc = osd_quota_transfer(inode, attr);
2748         if (rc)
2749                 return rc;
2750
2751         spin_lock(&obj->oo_guard);
2752         rc = osd_inode_setattr(env, inode, attr);
2753         spin_unlock(&obj->oo_guard);
2754         if (rc != 0)
2755                 GOTO(out, rc);
2756
2757         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2758
2759         if (!(attr->la_valid & LA_FLAGS))
2760                 GOTO(out, rc);
2761
2762         /* Let's check if there are extra flags need to be set into LMA */
2763         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2764                 struct osd_thread_info *info = osd_oti_get(env);
2765                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2766
2767                 LASSERT(!obj->oo_pfid_in_lma);
2768
2769                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2770                                  &info->oti_ost_attrs);
2771                 if (rc)
2772                         GOTO(out, rc);
2773
2774                 lma->lma_incompat |=
2775                         lustre_to_lma_flags(attr->la_flags);
2776                 lustre_lma_swab(lma);
2777                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2778                                      lma, sizeof(*lma), XATTR_REPLACE);
2779                 if (rc != 0) {
2780                         struct osd_device *osd = osd_obj2dev(obj);
2781
2782                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2783                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2784                               lma->lma_incompat, rc);
2785                 } else {
2786                         obj->oo_lma_flags =
2787                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2788                 }
2789                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2790         }
2791 out:
2792         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2793
2794         return rc;
2795 }
2796
2797 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2798                                            struct osd_object *obj,
2799                                            const char *name, const int namelen)
2800 {
2801         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2802 }
2803
2804 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2805                       umode_t mode, struct dt_allocation_hint *hint,
2806                       struct thandle *th)
2807 {
2808         int result;
2809         struct osd_device  *osd = osd_obj2dev(obj);
2810         struct osd_thandle *oth;
2811         struct dt_object   *parent = NULL;
2812         struct inode       *inode;
2813
2814         LINVRNT(osd_invariant(obj));
2815         LASSERT(obj->oo_inode == NULL);
2816         LASSERT(obj->oo_hl_head == NULL);
2817
2818         if (S_ISDIR(mode) && ldiskfs_pdo) {
2819                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2820                 if (obj->oo_hl_head == NULL)
2821                         return -ENOMEM;
2822         }
2823
2824         oth = container_of(th, struct osd_thandle, ot_super);
2825         LASSERT(oth->ot_handle->h_transaction != NULL);
2826
2827         if (hint != NULL && hint->dah_parent != NULL &&
2828             !dt_object_remote(hint->dah_parent))
2829                 parent = hint->dah_parent;
2830
2831         inode = ldiskfs_create_inode(oth->ot_handle,
2832                                      parent ? osd_dt_obj(parent)->oo_inode :
2833                                               osd_sb(osd)->s_root->d_inode,
2834                                      mode);
2835         if (!IS_ERR(inode)) {
2836                 /* Do not update file c/mtime in ldiskfs. */
2837                 inode->i_flags |= S_NOCMTIME;
2838
2839                 /* For new created object, it must be consistent,
2840                  * and it is unnecessary to scrub against it. */
2841                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2842
2843                 obj->oo_inode = inode;
2844                 result = 0;
2845         } else {
2846                 if (obj->oo_hl_head != NULL) {
2847                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2848                         obj->oo_hl_head = NULL;
2849                 }
2850                 result = PTR_ERR(inode);
2851         }
2852         LINVRNT(osd_invariant(obj));
2853         return result;
2854 }
2855
2856 enum {
2857         OSD_NAME_LEN = 255
2858 };
2859
2860 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2861                      struct lu_attr *attr,
2862                      struct dt_allocation_hint *hint,
2863                      struct dt_object_format *dof,
2864                      struct thandle *th)
2865 {
2866         int result;
2867         struct osd_thandle *oth;
2868         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2869
2870         LASSERT(S_ISDIR(attr->la_mode));
2871
2872         oth = container_of(th, struct osd_thandle, ot_super);
2873         LASSERT(oth->ot_handle->h_transaction != NULL);
2874         result = osd_mkfile(info, obj, mode, hint, th);
2875
2876         return result;
2877 }
2878
2879 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2880                         struct lu_attr *attr,
2881                         struct dt_allocation_hint *hint,
2882                         struct dt_object_format *dof,
2883                         struct thandle *th)
2884 {
2885         int result;
2886         struct osd_thandle *oth;
2887         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2888
2889         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2890
2891         LASSERT(S_ISREG(attr->la_mode));
2892
2893         oth = container_of(th, struct osd_thandle, ot_super);
2894         LASSERT(oth->ot_handle->h_transaction != NULL);
2895
2896         result = osd_mkfile(info, obj, mode, hint, th);
2897         if (result == 0) {
2898                 LASSERT(obj->oo_inode != NULL);
2899                 if (feat->dif_flags & DT_IND_VARKEY)
2900                         result = iam_lvar_create(obj->oo_inode,
2901                                                  feat->dif_keysize_max,
2902                                                  feat->dif_ptrsize,
2903                                                  feat->dif_recsize_max,
2904                                                  oth->ot_handle);
2905                 else
2906                         result = iam_lfix_create(obj->oo_inode,
2907                                                  feat->dif_keysize_max,
2908                                                  feat->dif_ptrsize,
2909                                                  feat->dif_recsize_max,
2910                                                  oth->ot_handle);
2911
2912         }
2913         return result;
2914 }
2915
2916 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2917                      struct lu_attr *attr,
2918                      struct dt_allocation_hint *hint,
2919                      struct dt_object_format *dof,
2920                      struct thandle *th)
2921 {
2922         LASSERT(S_ISREG(attr->la_mode));
2923         return osd_mkfile(info, obj, (attr->la_mode &
2924                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2925 }
2926
2927 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2928                      struct lu_attr *attr,
2929                      struct dt_allocation_hint *hint,
2930                      struct dt_object_format *dof,
2931                      struct thandle *th)
2932 {
2933         LASSERT(S_ISLNK(attr->la_mode));
2934         return osd_mkfile(info, obj, (attr->la_mode &
2935                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2936 }
2937
2938 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2939                      struct lu_attr *attr,
2940                      struct dt_allocation_hint *hint,
2941                      struct dt_object_format *dof,
2942                      struct thandle *th)
2943 {
2944         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2945         int result;
2946
2947         LINVRNT(osd_invariant(obj));
2948         LASSERT(obj->oo_inode == NULL);
2949         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2950                 S_ISFIFO(mode) || S_ISSOCK(mode));
2951
2952         result = osd_mkfile(info, obj, mode, hint, th);
2953         if (result == 0) {
2954                 LASSERT(obj->oo_inode != NULL);
2955                 /*
2956                  * This inode should be marked dirty for i_rdev.  Currently
2957                  * that is done in the osd_attr_init().
2958                  */
2959                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2960                                    attr->la_rdev);
2961         }
2962         LINVRNT(osd_invariant(obj));
2963         return result;
2964 }
2965
2966 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2967                               struct lu_attr *,
2968                               struct dt_allocation_hint *hint,
2969                               struct dt_object_format *dof,
2970                               struct thandle *);
2971
2972 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2973 {
2974         osd_obj_type_f result;
2975
2976         switch (type) {
2977         case DFT_DIR:
2978                 result = osd_mkdir;
2979                 break;
2980         case DFT_REGULAR:
2981                 result = osd_mkreg;
2982                 break;
2983         case DFT_SYM:
2984                 result = osd_mksym;
2985                 break;
2986         case DFT_NODE:
2987                 result = osd_mknod;
2988                 break;
2989         case DFT_INDEX:
2990                 result = osd_mk_index;
2991                 break;
2992
2993         default:
2994                 LBUG();
2995                 break;
2996         }
2997         return result;
2998 }
2999
3000
3001 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3002                         struct dt_object *parent, struct dt_object *child,
3003                         umode_t child_mode)
3004 {
3005         LASSERT(ah);
3006
3007         ah->dah_parent = parent;
3008         ah->dah_mode = child_mode;
3009
3010         if (parent != NULL && !dt_object_remote(parent)) {
3011                 /* will help to find FID->ino at dt_insert("..") */
3012                 struct osd_object *pobj = osd_dt_obj(parent);
3013                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3014         }
3015 }
3016
3017 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3018                           struct lu_attr *attr, struct dt_object_format *dof)
3019 {
3020         struct inode   *inode = obj->oo_inode;
3021         __u64           valid = attr->la_valid;
3022         int             result;
3023
3024         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3025
3026         if (dof->dof_type != DFT_NODE)
3027                 attr->la_valid &= ~LA_RDEV;
3028         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3029                 attr->la_valid &= ~LA_ATIME;
3030         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3031                 attr->la_valid &= ~LA_CTIME;
3032         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3033                 attr->la_valid &= ~LA_MTIME;
3034
3035         result = osd_quota_transfer(inode, attr);
3036         if (result)
3037                 return;
3038
3039         if (attr->la_valid != 0) {
3040                 result = osd_inode_setattr(info->oti_env, inode, attr);
3041                 /*
3042                  * The osd_inode_setattr() should always succeed here.  The
3043                  * only error that could be returned is EDQUOT when we are
3044                  * trying to change the UID or GID of the inode. However, this
3045                  * should not happen since quota enforcement is no longer
3046                  * enabled on ldiskfs (lquota takes care of it).
3047                  */
3048                 LASSERTF(result == 0, "%d\n", result);
3049                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3050         }
3051
3052         attr->la_valid = valid;
3053 }
3054
3055 /**
3056  * Helper function for osd_create()
3057  *
3058  * \retval 0, on success
3059  */
3060 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3061                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3062                         struct dt_object_format *dof, struct thandle *th)
3063 {
3064         int     result;
3065         __u32   umask;
3066
3067         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3068
3069         /* we drop umask so that permissions we pass are not affected */
3070         umask = current->fs->umask;
3071         current->fs->umask = 0;
3072
3073         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3074                                                   th);
3075         if (likely(obj->oo_inode != NULL)) {
3076                 LASSERT(obj->oo_inode->i_state & I_NEW);
3077
3078                 /* Unlock the inode before attr initialization to avoid
3079                  * unnecessary dqget operations. LU-6378 */
3080                 unlock_new_inode(obj->oo_inode);
3081         }
3082
3083         if (likely(result == 0)) {
3084                 osd_attr_init(info, obj, attr, dof);
3085                 osd_object_init0(obj);
3086         }
3087
3088         /* restore previous umask value */
3089         current->fs->umask = umask;
3090
3091         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3092
3093         return result;
3094 }
3095
3096 /**
3097  * Helper function for osd_create()
3098  *
3099  * \retval 0, on success
3100  */
3101 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3102                            const struct lu_fid *fid, struct thandle *th)
3103 {
3104         struct osd_thread_info *info = osd_oti_get(env);
3105         struct osd_inode_id    *id   = &info->oti_id;
3106         struct osd_device      *osd  = osd_obj2dev(obj);
3107         struct osd_thandle     *oh;
3108         int                     rc;
3109
3110         LASSERT(obj->oo_inode != NULL);
3111
3112         oh = container_of0(th, struct osd_thandle, ot_super);
3113         LASSERT(oh->ot_handle);
3114         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3115
3116         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3117         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3118                            OI_CHECK_FLD, NULL);
3119         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3120
3121         return rc;
3122 }
3123
3124 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3125                    u64 seq, struct lu_seq_range *range)
3126 {
3127         struct seq_server_site  *ss = osd_seq_site(osd);
3128
3129         if (fid_seq_is_idif(seq)) {
3130                 fld_range_set_ost(range);
3131                 range->lsr_index = idif_ost_idx(seq);
3132                 return 0;
3133         }
3134
3135         if (!fid_seq_in_fldb(seq)) {
3136                 fld_range_set_mdt(range);
3137                 if (ss != NULL)
3138                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3139                          * at all */
3140                         range->lsr_index = ss->ss_node_id;
3141                 return 0;
3142         }
3143
3144         LASSERT(ss != NULL);
3145         fld_range_set_any(range);
3146         /* OSD will only do local fld lookup */
3147         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3148 }
3149
3150 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3151                               struct lu_attr *attr,
3152                               struct dt_allocation_hint *hint,
3153                               struct dt_object_format *dof,
3154                               struct thandle *handle)
3155 {
3156         struct osd_thandle *oh;
3157         int rc;
3158         ENTRY;
3159
3160         LASSERT(handle != NULL);
3161
3162         oh = container_of0(handle, struct osd_thandle, ot_super);
3163         LASSERT(oh->ot_handle == NULL);
3164
3165         /* EA object consumes more credits than regular object: osd_mk_index
3166          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3167          * leaf_node, could involves the block, block bitmap, groups, GDT
3168          * change for each block, so add 4 * 2 credits in that case. */
3169         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3170                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3171                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3172         /* Reuse idle OI block may cause additional one OI block
3173          * to be changed. */
3174         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3175                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3176
3177         if (!attr)
3178                 RETURN(0);
3179
3180         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3181                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3182                                    NULL, OSD_QID_INODE);
3183         if (rc != 0)
3184                 RETURN(rc);
3185
3186         /* will help to find FID->ino mapping at dt_insert() */
3187         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3188                                    osd_dt_obj(dt));
3189
3190         RETURN(rc);
3191 }
3192
3193 /**
3194  * Called to destroy on-disk representation of the object
3195  *
3196  * Concurrency: must be locked
3197  */
3198 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3199                                struct thandle *th)
3200 {
3201         struct osd_object  *obj = osd_dt_obj(dt);
3202         struct inode       *inode = obj->oo_inode;
3203         struct osd_thandle *oh;
3204         int                 rc;
3205         ENTRY;
3206
3207         if (inode == NULL)
3208                 RETURN(-ENOENT);
3209
3210         oh = container_of0(th, struct osd_thandle, ot_super);
3211         LASSERT(oh->ot_handle == NULL);
3212
3213         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3214                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3215
3216         /* For removing agent entry */
3217         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3218                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3219
3220         /* Recycle idle OI leaf may cause additional three OI blocks
3221          * to be changed. */
3222         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3223                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3224                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3225         /* one less inode */
3226         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3227                                    i_projid_read(inode), -1, oh, obj, NULL,
3228                                    OSD_QID_INODE);
3229         if (rc)
3230                 RETURN(rc);
3231         /* data to be truncated */
3232         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3233                                    i_projid_read(inode), 0, oh, obj, NULL,
3234                                    OSD_QID_BLK);
3235         if (rc)
3236                 RETURN(rc);
3237
3238         /* will help to find FID->ino when this object is being
3239          * added to PENDING/ */
3240         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3241
3242         RETURN(rc);
3243 }
3244
3245 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3246                        struct thandle *th)
3247 {
3248         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3249         struct osd_object *obj = osd_dt_obj(dt);
3250         struct inode *inode = obj->oo_inode;
3251         struct osd_device *osd = osd_obj2dev(obj);
3252         struct osd_thandle *oh;
3253         int result;
3254         ENTRY;
3255
3256         oh = container_of0(th, struct osd_thandle, ot_super);
3257         LASSERT(oh->ot_handle);
3258         LASSERT(inode);
3259         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3260
3261         if (unlikely(fid_is_acct(fid)))
3262                 RETURN(-EPERM);
3263
3264         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3265                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3266                 if (result != 0)
3267                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3268                                osd_name(osd), PFID(fid), result);
3269         }
3270
3271         if (S_ISDIR(inode->i_mode)) {
3272                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3273                         inode->i_nlink == 2);
3274
3275                 spin_lock(&obj->oo_guard);
3276                 clear_nlink(inode);
3277                 spin_unlock(&obj->oo_guard);
3278                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3279         }
3280
3281         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3282
3283         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3284
3285         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3286                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3287                                        oh->ot_handle, OI_CHECK_FLD);
3288
3289         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3290         /* XXX: add to ext3 orphan list */
3291         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3292
3293         /* not needed in the cache anymore */
3294         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3295         obj->oo_destroyed = 1;
3296
3297         RETURN(0);
3298 }
3299
3300 /**
3301  * Put the fid into lustre_mdt_attrs, and then place the structure
3302  * inode's ea. This fid should not be altered during the life time
3303  * of the inode.
3304  *
3305  * \retval +ve, on success
3306  * \retval -ve, on error
3307  *
3308  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3309  */
3310 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3311                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3312 {
3313         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3314         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3315         int rc;
3316         ENTRY;
3317
3318         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3319                 RETURN(0);
3320
3321         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3322                 rc = -ENOMEM;
3323
3324         lustre_loa_init(loa, fid, compat, incompat);
3325         lustre_loa_swab(loa, false);
3326
3327         /* For the OST device with 256 bytes inode size by default,
3328          * the PFID EA will be stored together with LMA EA to avoid
3329          * performance trouble. Otherwise the PFID EA can be stored
3330          * independently. LU-8998 */
3331         if ((compat & LMAC_FID_ON_OST) &&
3332             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3333                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3334                                      sizeof(*loa), XATTR_CREATE);
3335         else
3336                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3337                                      sizeof(*lma), XATTR_CREATE);
3338         /* LMA may already exist, but we need to check that all the
3339          * desired compat/incompat flags have been added. */
3340         if (unlikely(rc == -EEXIST)) {
3341                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3342                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3343                 if (rc < 0)
3344                         RETURN(rc);
3345
3346                 if (rc < sizeof(*lma))
3347                         RETURN(-EINVAL);
3348
3349                 lustre_loa_swab(loa, true);
3350                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3351                     ((compat == 0 && incompat == 0) ||
3352                      (!(~lma->lma_compat & compat) &&
3353                       !(~lma->lma_incompat & incompat))))
3354                         RETURN(0);
3355
3356                 lma->lma_self_fid = *fid;
3357                 lma->lma_compat |= compat;
3358                 lma->lma_incompat |= incompat;
3359                 if (rc == sizeof(*lma)) {
3360                         lustre_lma_swab(lma);
3361                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3362                                              sizeof(*lma), XATTR_REPLACE);
3363                 } else {
3364                         lustre_loa_swab(loa, false);
3365                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3366                                              sizeof(*loa), XATTR_REPLACE);
3367                 }
3368         }
3369
3370         RETURN(rc);
3371 }
3372
3373 /**
3374  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3375  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3376  * To have compatilibility with 1.8 ldiskfs driver we need to have
3377  * magic number at start of fid data.
3378  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3379  * its inmemory API.
3380  */
3381 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3382                                          const struct lu_fid *fid)
3383 {
3384         if (!fid_is_namespace_visible(fid) ||
3385             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3386                 param->edp_magic = 0;
3387                 return;
3388         }
3389
3390         param->edp_magic = LDISKFS_LUFID_MAGIC;
3391         param->edp_len =  sizeof(struct lu_fid) + 1;
3392         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3393 }
3394
3395 /**
3396  * Try to read the fid from inode ea into dt_rec.
3397  *
3398  * \param fid object fid.
3399  *
3400  * \retval 0 on success
3401  */
3402 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3403                           __u32 ino, struct lu_fid *fid,
3404                           struct osd_inode_id *id)
3405 {
3406         struct osd_thread_info *info  = osd_oti_get(env);
3407         struct inode           *inode;
3408         ENTRY;
3409
3410         osd_id_gen(id, ino, OSD_OII_NOGEN);
3411         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3412         if (IS_ERR(inode))
3413                 RETURN(PTR_ERR(inode));
3414
3415         iput(inode);
3416         RETURN(0);
3417 }
3418
3419 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3420                                         struct inode *dir,
3421                                         struct inode *parent_dir,
3422                                         const struct lu_fid *dot_fid,
3423                                         const struct lu_fid *dot_dot_fid,
3424                                         struct osd_thandle *oth)
3425 {
3426         struct ldiskfs_dentry_param *dot_ldp;
3427         struct ldiskfs_dentry_param *dot_dot_ldp;
3428         __u32 saved_nlink = dir->i_nlink;
3429         int rc;
3430
3431         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3432         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3433
3434         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3435         dot_ldp->edp_magic = 0;
3436
3437         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3438                                     dir, dot_ldp, dot_dot_ldp);
3439         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3440          * the subseqent ref_add() will increase the dir->i_nlink
3441          * as 3. That is incorrect for new created directory.
3442          *
3443          * It looks like hack, because we want to make the OSD API
3444          * to be order-independent for new created directory object
3445          * between dt_insert(..) and ref_add() operations.
3446          *
3447          * Here, we only restore the in-RAM dir-inode's nlink attr,
3448          * becuase if the nlink attr is not 2, then there will be
3449          * ref_add() called following the dt_insert(..), such call
3450          * will make both the in-RAM and on-disk dir-inode's nlink
3451          * attr to be set as 2. LU-7447 */
3452         set_nlink(dir, saved_nlink);
3453         return rc;
3454 }
3455
3456 /**
3457  * Create an local agent inode for remote entry
3458  */
3459 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3460                                                   struct osd_device *osd,
3461                                                   struct osd_object *pobj,
3462                                                   const struct lu_fid *fid,
3463                                                   __u32 type,
3464                                                   struct thandle *th)
3465 {
3466         struct osd_thread_info  *info = osd_oti_get(env);
3467         struct inode            *local;
3468         struct osd_thandle      *oh;
3469         int                     rc;
3470         ENTRY;
3471
3472         LASSERT(th);
3473         oh = container_of(th, struct osd_thandle, ot_super);
3474         LASSERT(oh->ot_handle->h_transaction != NULL);
3475
3476         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3477         if (IS_ERR(local)) {
3478                 CERROR("%s: create local error %d\n", osd_name(osd),
3479                        (int)PTR_ERR(local));
3480                 RETURN(local);
3481         }
3482
3483         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3484          * correct gid on remote file, not agent here */
3485         local->i_gid = current_fsgid();
3486         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3487         unlock_new_inode(local);
3488
3489         /* Agent inode should not have project ID */
3490 #ifdef  HAVE_PROJECT_QUOTA
3491         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3492             i_projid_read(pobj->oo_inode) != 0) {
3493                 rc = ldiskfs_transfer_project(local, 0);
3494                 if (rc) {
3495                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3496                                "quota enforcement enabled on the ldiskfs "
3497                                "filesystem?\n", local->i_sb->s_id, rc);
3498                         RETURN(ERR_PTR(rc));
3499                 }
3500         }
3501 #endif
3502         /* Set special LMA flag for local agent inode */
3503         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3504         if (rc != 0) {
3505                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3506                        osd_name(osd), PFID(fid), rc);
3507                 RETURN(ERR_PTR(rc));
3508         }
3509
3510         if (!S_ISDIR(type))
3511                 RETURN(local);
3512
3513         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3514                                          lu_object_fid(&pobj->oo_dt.do_lu),
3515                                          fid, oh);
3516         if (rc != 0) {
3517                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3518                         osd_name(osd), PFID(fid), rc);
3519                 RETURN(ERR_PTR(rc));
3520         }
3521
3522         RETURN(local);
3523 }
3524
3525 /**
3526  * when direntry is deleted, we have to take care of possible agent inode
3527  * referenced by that. unfortunately we can't do this at that point:
3528  * iget() within a running transaction leads to deadlock and we better do
3529  * not call that every delete declaration to save performance. so we put
3530  * a potention agent inode on a list and process that once the transaction
3531  * is over. Notice it's not any worse in terms of real orphans as regular
3532  * object destroy doesn't put inodes on the on-disk orphan list. this should
3533  * be addressed separately
3534  */
3535 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3536                                             struct osd_thandle *oh,
3537                                             __u32 ino)
3538 {
3539         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3540         struct osd_obj_orphan *oor;
3541
3542         OBD_ALLOC_PTR(oor);
3543         if (oor == NULL)
3544                 return -ENOMEM;
3545
3546         oor->oor_ino = ino;
3547         oor->oor_env = (struct lu_env *)env;
3548         spin_lock(&osd->od_osfs_lock);
3549         list_add(&oor->oor_list, &osd->od_orphan_list);
3550         spin_unlock(&osd->od_osfs_lock);
3551
3552         oh->ot_remove_agents = 1;
3553
3554         return 0;
3555
3556 }
3557
3558 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3559                                                 struct osd_device *osd)
3560 {
3561         struct osd_thread_info *info = osd_oti_get(env);
3562         struct osd_obj_orphan *oor, *tmp;
3563         struct osd_inode_id id;
3564         struct list_head list;
3565         struct inode *inode;
3566         struct lu_fid fid;
3567         handle_t *jh;
3568         __u32 ino;
3569
3570         INIT_LIST_HEAD(&list);
3571
3572         spin_lock(&osd->od_osfs_lock);
3573         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3574                 if (oor->oor_env == env) {
3575                         list_del(&oor->oor_list);
3576                         list_add(&oor->oor_list, &list);
3577                 }
3578         }
3579         spin_unlock(&osd->od_osfs_lock);
3580
3581         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3582
3583                 ino = oor->oor_ino;
3584
3585                 list_del(&oor->oor_list);
3586                 OBD_FREE_PTR(oor);
3587
3588                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3589                 inode = osd_iget_fid(info, osd, &id, &fid);
3590                 if (IS_ERR(inode))
3591                         continue;
3592
3593                 if (!osd_remote_fid(env, osd, &fid)) {
3594                         iput(inode);
3595                         continue;
3596                 }
3597
3598                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3599                 clear_nlink(inode);
3600                 mark_inode_dirty(inode);
3601                 ldiskfs_journal_stop(jh);
3602                 iput(inode);
3603         }
3604
3605         return 0;
3606 }
3607
3608 /**
3609  * OSD layer object create function for OST objects (b=11826).
3610  *
3611  * The FID is inserted into inode xattr here.
3612  *
3613  * \retval   0, on success
3614  * \retval -ve, on error
3615  */
3616 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3617                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3618                       struct dt_object_format *dof, struct thandle *th)
3619 {
3620         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3621         struct osd_object       *obj    = osd_dt_obj(dt);
3622         struct osd_thread_info  *info   = osd_oti_get(env);
3623         int                      result, on_ost = 0;
3624
3625         ENTRY;
3626
3627         if (dt_object_exists(dt))
3628                 RETURN(-EEXIST);
3629
3630         LINVRNT(osd_invariant(obj));
3631         LASSERT(!dt_object_remote(dt));
3632         LASSERT(osd_is_write_locked(env, obj));
3633         LASSERT(th != NULL);
3634
3635         if (unlikely(fid_is_acct(fid)))
3636                 /* Quota files can't be created from the kernel any more,
3637                  * 'tune2fs -O quota' will take care of creating them */
3638                 RETURN(-EPERM);
3639
3640         result = __osd_create(info, obj, attr, hint, dof, th);
3641         if (result == 0) {
3642                 if (fid_is_idif(fid) &&
3643                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3644                         struct lu_fid *tfid = &info->oti_fid;
3645                         struct ost_id *oi   = &info->oti_ostid;
3646
3647                         fid_to_ostid(fid, oi);
3648                         ostid_to_fid(tfid, oi, 0);
3649                         on_ost = 1;
3650                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3651                                                 LMAC_FID_ON_OST, 0);
3652                 } else {
3653                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3654                                                fid, OI_CHECK_FLD);
3655                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3656                                                 on_ost ? LMAC_FID_ON_OST : 0,
3657                                                 0);
3658                 }
3659                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3660                         obj->oo_dt.do_body_ops = &osd_body_ops;
3661         }
3662
3663         if (result == 0)
3664                 result = __osd_oi_insert(env, obj, fid, th);
3665
3666         /* a small optimization - dt_insert() isn't usually applied
3667          * to OST objects, so we don't need to cache OI mapping for
3668          * OST objects */
3669         if (result == 0 && on_ost == 0) {
3670                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3671                 result = osd_idc_find_and_init(env, osd, obj);
3672                 LASSERT(result == 0);
3673         }
3674
3675         LASSERT(ergo(result == 0,
3676                      dt_object_exists(dt) && !dt_object_remote(dt)));
3677         LINVRNT(osd_invariant(obj));
3678         RETURN(result);
3679 }
3680
3681 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3682                                struct thandle *handle)
3683 {
3684         struct osd_thandle *oh;
3685
3686         /* it's possible that object doesn't exist yet */
3687         LASSERT(handle != NULL);
3688
3689         oh = container_of0(handle, struct osd_thandle, ot_super);
3690         LASSERT(oh->ot_handle == NULL);
3691
3692         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3693                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3694
3695         return 0;
3696 }
3697
3698 /*
3699  * Concurrency: @dt is write locked.
3700  */
3701 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3702                        struct thandle *th)
3703 {
3704         struct osd_object  *obj = osd_dt_obj(dt);
3705         struct inode       *inode = obj->oo_inode;
3706         struct osd_thandle *oh;
3707         int                 rc = 0;
3708
3709         if (!dt_object_exists(dt) || obj->oo_destroyed)
3710                 return -ENOENT;
3711
3712         LINVRNT(osd_invariant(obj));
3713         LASSERT(!dt_object_remote(dt));
3714         LASSERT(osd_is_write_locked(env, obj));
3715         LASSERT(th != NULL);
3716
3717         oh = container_of0(th, struct osd_thandle, ot_super);
3718         LASSERT(oh->ot_handle != NULL);
3719
3720         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3721
3722         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3723                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3724         /*
3725          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3726          * (65000) subdirectories by storing "1" in i_nlink if the link count
3727          * would otherwise overflow. Directory tranversal tools understand
3728          * that (st_nlink == 1) indicates that the filesystem dose not track
3729          * hard links count on the directory, and will not abort subdirectory
3730          * scanning early once (st_nlink - 2) subdirs have been found.
3731          *
3732          * This also has to properly handle the case of inodes with nlink == 0
3733          * in case they are being linked into the PENDING directory
3734          */
3735         spin_lock(&obj->oo_guard);
3736         if (unlikely(inode->i_nlink == 0))
3737                 /* inc_nlink from 0 may cause WARN_ON */
3738                 set_nlink(inode, 1);
3739         else {
3740                 ldiskfs_inc_count(oh->ot_handle, inode);
3741                 if (!S_ISDIR(inode->i_mode))
3742                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3743         }
3744         spin_unlock(&obj->oo_guard);
3745
3746         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3747         LINVRNT(osd_invariant(obj));
3748
3749         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3750
3751         return rc;
3752 }
3753
3754 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3755                                struct thandle *handle)
3756 {
3757         struct osd_thandle *oh;
3758
3759         if (!dt_object_exists(dt))
3760                 return -ENOENT;
3761
3762         LASSERT(!dt_object_remote(dt));
3763         LASSERT(handle != NULL);
3764
3765         oh = container_of0(handle, struct osd_thandle, ot_super);
3766         LASSERT(oh->ot_handle == NULL);
3767
3768         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3769                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3770
3771         return 0;
3772 }
3773
3774 /*
3775  * Concurrency: @dt is write locked.
3776  */
3777 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3778                        struct thandle *th)
3779 {
3780         struct osd_object       *obj = osd_dt_obj(dt);
3781         struct inode            *inode = obj->oo_inode;
3782         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3783         struct osd_thandle      *oh;
3784
3785         if (!dt_object_exists(dt))
3786                 return -ENOENT;
3787
3788         LINVRNT(osd_invariant(obj));
3789         LASSERT(!dt_object_remote(dt));
3790         LASSERT(osd_is_write_locked(env, obj));
3791         LASSERT(th != NULL);
3792
3793         oh = container_of0(th, struct osd_thandle, ot_super);
3794         LASSERT(oh->ot_handle != NULL);
3795
3796         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3797
3798         spin_lock(&obj->oo_guard);
3799         /* That can be result of upgrade from old Lustre version and
3800          * applied only to local files.  Just skip this ref_del call.
3801          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3802         if (inode->i_nlink == 0) {
3803                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3804                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3805                              ", maybe an upgraded file? (LU-3915)\n",
3806                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3807                 spin_unlock(&obj->oo_guard);
3808                 return 0;
3809         }
3810
3811         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3812                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3813
3814         ldiskfs_dec_count(oh->ot_handle, inode);
3815         spin_unlock(&obj->oo_guard);
3816
3817         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3818         LINVRNT(osd_invariant(obj));
3819
3820         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3821
3822         return 0;
3823 }
3824
3825 /*
3826  * Concurrency: @dt is read locked.
3827  */
3828 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3829                          struct lu_buf *buf, const char *name)
3830 {
3831         struct osd_object      *obj    = osd_dt_obj(dt);
3832         struct inode           *inode  = obj->oo_inode;
3833         struct osd_thread_info *info   = osd_oti_get(env);
3834         struct dentry          *dentry = &info->oti_obj_dentry;
3835         bool                    cache_xattr = false;
3836         int                     rc;
3837
3838         LASSERT(buf);
3839
3840         /* version get is not real XATTR but uses xattr API */
3841         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3842                 dt_obj_version_t *ver = buf->lb_buf;
3843
3844                 /* for version we are just using xattr API but change inode
3845                  * field instead */
3846                 if (buf->lb_len == 0)
3847                         return sizeof(dt_obj_version_t);
3848
3849                 if (buf->lb_len < sizeof(dt_obj_version_t))
3850                         return -ERANGE;
3851
3852                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3853                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3854
3855                 *ver = LDISKFS_I(inode)->i_fs_version;
3856
3857                 return sizeof(dt_obj_version_t);
3858         }
3859
3860         if (!dt_object_exists(dt))
3861                 return -ENOENT;
3862
3863         LASSERT(!dt_object_remote(dt));
3864         LASSERT(inode->i_op != NULL);
3865         LASSERT(inode->i_op->getxattr != NULL);
3866
3867         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3868             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3869                 cache_xattr = true;
3870
3871         if (cache_xattr) {
3872                 rc = osd_oxc_get(obj, name, buf);
3873                 if (rc != -ENOENT)
3874                         return rc;
3875         }
3876
3877         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
3878                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3879                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3880                 struct filter_fid *ff;
3881                 struct ost_layout *ol;
3882
3883                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3884                 if (rc)
3885                         return rc;
3886
3887                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
3888
3889                 rc = sizeof(*ff);
3890                 if (buf->lb_len == 0 || !buf->lb_buf)
3891                         return rc;
3892
3893                 if (buf->lb_len < rc)
3894                         return -ERANGE;
3895
3896                 ff = buf->lb_buf;
3897                 ol = &ff->ff_layout;
3898                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3899                                                   PFID_STRIPE_IDX_BITS);
3900                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3901                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3902                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3903                 if (lma->lma_compat & LMAC_COMP_INFO) {
3904                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3905                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3906                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3907                 } else {
3908                         ol->ol_comp_start = 0;
3909                         ol->ol_comp_end = 0;
3910                         ol->ol_comp_id = 0;
3911                 }
3912         } else {
3913                 rc = __osd_xattr_get(inode, dentry, name,
3914                                      buf->lb_buf, buf->lb_len);
3915         }
3916
3917         if (cache_xattr) {
3918                 if (rc == -ENOENT || rc == -ENODATA)
3919                         osd_oxc_add(obj, name, NULL, 0);
3920                 else if (rc > 0 && buf->lb_buf != NULL)
3921                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3922         }
3923
3924         return rc;
3925 }
3926
3927 static int osd_declare_xattr_set(const struct lu_env *env,
3928                                  struct dt_object *dt,
3929                                  const struct lu_buf *buf, const char *name,
3930                                  int fl, struct thandle *handle)
3931 {
3932         struct osd_thandle *oh;
3933         int credits = 0;
3934         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3935
3936         LASSERT(handle != NULL);
3937
3938         oh = container_of0(handle, struct osd_thandle, ot_super);
3939         LASSERT(oh->ot_handle == NULL);
3940
3941         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3942                 /* For non-upgrading case, the LMA is set first and
3943                  * usually fit inode. But for upgrade case, the LMA
3944                  * may be in another separated EA block. */
3945                 if (dt_object_exists(dt)) {
3946                         if (fl == LU_XATTR_REPLACE)
3947                                 credits = 1;
3948                         else
3949                                 goto upgrade;
3950                 }
3951         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3952                 credits = 1;
3953         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3954                 /* We may need to delete the old PFID EA. */
3955                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3956                 if (fl == LU_XATTR_REPLACE)
3957                         credits += 1;
3958                 else
3959                         goto upgrade;
3960         } else {
3961                 /* If some name entry resides on remote MDT, then will create
3962                  * agent entry under remote parent. On the other hand, if the
3963                  * remote entry will be removed, then related agent entry may
3964                  * need to be removed from the remote parent. So there may be
3965                  * kinds of cases, let's declare enough credits. The credits
3966                  * for create agent entry is enough for remove case. */
3967                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
3968                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3969                         if (dt_object_exists(dt))
3970                                 credits += 1; /* For updating LMA */
3971                 }
3972
3973 upgrade:
3974                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3975
3976                 if (buf != NULL) {
3977                         ssize_t buflen;
3978
3979                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3980                                 /* learn xattr size from osd_xattr_get if
3981                                    attribute has not been read yet */
3982                                 buflen = __osd_xattr_get(
3983                                     osd_dt_obj(dt)->oo_inode,
3984                                     &osd_oti_get(env)->oti_obj_dentry,
3985                                     name, NULL, 0);
3986                                 if (buflen < 0)
3987                                         buflen = 0;
3988                         } else {
3989                                 buflen = buf->lb_len;
3990                         }
3991
3992                         if (buflen > sb->s_blocksize) {
3993                                 credits += osd_calc_bkmap_credits(
3994                                     sb, NULL, 0, -1,
3995                                     (buflen + sb->s_blocksize - 1) >>
3996                                     sb->s_blocksize_bits);
3997                         }
3998                 }
3999                 /*
4000                  * xattr set may involve inode quota change, reserve credits for
4001                  * dquot_initialize()
4002                  */
4003                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4004         }
4005
4006         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4007
4008         return 0;
4009 }
4010
4011 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4012                               const struct lu_buf *buf, int fl,
4013                               struct thandle *handle)
4014 {
4015         struct osd_thread_info *info = osd_oti_get(env);
4016         struct dentry *dentry = &info->oti_obj_dentry;
4017         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4018         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4019         struct inode *inode = obj->oo_inode;
4020         struct filter_fid *ff = buf->lb_buf;
4021         struct ost_layout *ol = &ff->ff_layout;
4022         int flags = XATTR_REPLACE;
4023         int rc;
4024         ENTRY;
4025
4026         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4027                 RETURN(-EINVAL);
4028
4029         rc = osd_get_lma(info, inode, dentry, loa);
4030         if (rc == -ENODATA) {
4031                 /* Usually for upgarding from old device */
4032                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4033                                 LMAC_FID_ON_OST, 0);
4034                 flags = XATTR_CREATE;
4035         } else if (rc) {
4036                 RETURN(rc);
4037         }
4038
4039         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4040                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4041                         RETURN(-EEXIST);
4042
4043                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4044                         /* Separate PFID EA from LMA */
4045                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4046                         lustre_lma_swab(lma);
4047                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4048                                              sizeof(*lma), XATTR_REPLACE);
4049                         if (!rc) {
4050                                 obj->oo_pfid_in_lma = 0;
4051                                 rc = LU_XATTR_CREATE;
4052                         }
4053
4054                         RETURN(rc);
4055                 }
4056         } else {
4057                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4058                         RETURN(fl);
4059
4060                 /* Old client does not send stripe information,
4061                  * then store the PFID EA on disk separatedly. */
4062                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4063                              ol->ol_stripe_size == 0))
4064                         RETURN(fl);
4065
4066                 /* Remove old PFID EA entry firstly. */
4067                 ll_vfs_dq_init(inode);
4068                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
4069                 if (rc == -ENODATA) {
4070                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4071                                 RETURN(rc);
4072                 } else if (rc) {
4073                         RETURN(rc);
4074                 }
4075         }
4076
4077         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4078         if (likely(ol->ol_stripe_size != 0)) {
4079                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4080                                              PFID_STRIPE_IDX_BITS;
4081                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4082                 lma->lma_compat |= LMAC_STRIPE_INFO;
4083                 if (ol->ol_comp_id != 0) {
4084                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4085                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4086                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4087                         lma->lma_compat |= LMAC_COMP_INFO;
4088                 }
4089         }
4090
4091         lustre_loa_swab(loa, false);
4092
4093         /* Store the PFID EA inside LMA. */
4094         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4095                              flags);
4096         if (!rc)
4097                 obj->oo_pfid_in_lma = 1;
4098
4099         RETURN(rc);
4100 }
4101
4102 /*
4103  * In DNE environment, the object (in spite of regular file or directory)
4104  * and its name entry may reside on different MDTs. Under such case, we will
4105  * create an agent entry on the MDT where the object resides. The agent entry
4106  * references the object locally, that makes the object to be visible to the
4107  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4108  * such as 'tar' can handle the object properly.
4109  *
4110  * We handle the agent entry during set linkEA that is the common interface
4111  * for both regular file and directroy, can handle kinds of cases, such as
4112  * create/link/unlink/rename, and so on.
4113  *
4114  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4115  *
4116  * XXX: There are two known issues:
4117  * 1. For one object, we will create at most one agent entry even if there
4118  *    may be more than one cross-MDTs hard links on the object. So the local
4119  *    e2fsck may claim that the object's nlink is larger than the name entries
4120  *    that reference such inode. And in further, the e2fsck will fix the nlink
4121  *    attribute to match the local references. Then it will cause the object's
4122  *    nlink attribute to be inconsistent with the global references. it is bad
4123  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4124  *    other hand, the global namespace LFSCK can repair the object's attribute
4125  *    according to the linkEA.
4126  * 2. There may be too many hard links on the object as to its linkEA overflow,
4127  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4128  *    case happened, then at this point, we do not know whether there are some
4129  *    cross-MDTs reference. But there are local references, it guarantees that
4130  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4131  */
4132 static int osd_xattr_handle_linkea(const struct lu_env *env,
4133                                    struct osd_device *osd,
4134                                    struct osd_object *obj,
4135                                    const struct lu_buf *buf,
4136                                    struct thandle *handle)
4137 {
4138         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4139         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4140         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4141         struct lu_name tmpname;
4142         struct osd_thandle *oh;
4143         int rc;
4144         bool remote = false;
4145         ENTRY;
4146
4147         oh = container_of0(handle, struct osd_thandle, ot_super);
4148         LASSERT(oh->ot_handle != NULL);
4149
4150         rc = linkea_init_with_rec(&ldata);
4151         if (!rc) {
4152                 linkea_first_entry(&ldata);
4153                 while (ldata.ld_lee != NULL && !remote) {
4154                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4155                                             &tmpname, tfid);
4156                         if (osd_remote_fid(env, osd, tfid) > 0)
4157                                 remote = true;
4158                         else
4159                                 linkea_next_entry(&ldata);
4160                 }
4161         } else if (rc == -ENODATA) {
4162                 rc = 0;
4163         } else {
4164                 RETURN(rc);
4165         }
4166
4167         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4168                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4169                 if (rc)
4170                         CERROR("%s: failed to remove agent entry for "DFID
4171                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4172         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4173                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4174                 if (rc)
4175                         CERROR("%s: failed to create agent entry for "DFID
4176                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4177         }
4178
4179         RETURN(rc);
4180 }
4181
4182 /*
4183  * Concurrency: @dt is write locked.
4184  */
4185 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4186                          const struct lu_buf *buf, const char *name, int fl,
4187                          struct thandle *handle)
4188 {
4189         struct osd_object *obj = osd_dt_obj(dt);
4190         struct osd_device *osd = osd_obj2dev(obj);
4191         struct inode *inode = obj->oo_inode;
4192         struct osd_thread_info *info = osd_oti_get(env);
4193         int fs_flags = 0;
4194         int len;
4195         int rc;
4196         ENTRY;
4197
4198         LASSERT(handle);
4199         LASSERT(buf);
4200
4201         /* version set is not real XATTR */
4202         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4203                 dt_obj_version_t *version = buf->lb_buf;
4204
4205                 /* for version we are just using xattr API but change inode
4206                  * field instead */
4207                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4208
4209                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4210                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4211
4212                 LDISKFS_I(inode)->i_fs_version = *version;
4213                 /* Version is set after all inode operations are finished,
4214                  * so we should mark it dirty here */
4215                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4216
4217                 RETURN(0);
4218         }
4219
4220         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4221                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4222
4223         len = buf->lb_len;
4224         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4225
4226         /* For the OST device with 256 bytes inode size by default,
4227          * the PFID EA will be stored together with LMA EA to avoid
4228          * performance trouble. Otherwise the PFID EA can be stored
4229          * independently. LU-8998 */
4230         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4231             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4232                 LASSERT(buf->lb_buf);
4233
4234                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4235                 if (fl <= 0)
4236                         RETURN(fl);
4237         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4238                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4239                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4240
4241                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4242                 if (rc)
4243                         RETURN(rc);
4244
4245                 lma->lma_incompat |= LMAI_STRIPED;
4246                 lustre_lma_swab(lma);
4247                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4248                                      sizeof(*lma), XATTR_REPLACE);
4249                 if (rc != 0)
4250                         RETURN(rc);
4251         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4252                 LASSERT(!osd->od_is_ost);
4253
4254                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4255                 if (rc)
4256                         RETURN(rc);
4257         }
4258
4259         if (fl & LU_XATTR_REPLACE)
4260                 fs_flags |= XATTR_REPLACE;
4261
4262         if (fl & LU_XATTR_CREATE)
4263                 fs_flags |= XATTR_CREATE;
4264
4265         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4266         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4267
4268         if (rc == 0 &&
4269             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4270              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4271                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4272
4273         return rc;
4274 }
4275
4276 /*
4277  * Concurrency: @dt is read locked.
4278  */
4279 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4280                           const struct lu_buf *buf)
4281 {
4282         struct osd_object      *obj    = osd_dt_obj(dt);
4283         struct inode           *inode  = obj->oo_inode;
4284         struct osd_thread_info *info   = osd_oti_get(env);
4285         struct dentry          *dentry = &info->oti_obj_dentry;
4286
4287         if (!dt_object_exists(dt))
4288                 return -ENOENT;
4289
4290         LASSERT(!dt_object_remote(dt));
4291         LASSERT(inode->i_op != NULL);
4292         LASSERT(inode->i_op->listxattr != NULL);
4293
4294         dentry->d_inode = inode;
4295         dentry->d_sb = inode->i_sb;
4296         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4297 }
4298
4299 static int osd_declare_xattr_del(const struct lu_env *env,
4300                                  struct dt_object *dt, const char *name,
4301                                  struct thandle *handle)
4302 {
4303         struct osd_thandle *oh;
4304         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4305
4306         LASSERT(!dt_object_remote(dt));
4307         LASSERT(handle != NULL);
4308
4309         oh = container_of0(handle, struct osd_thandle, ot_super);
4310         LASSERT(oh->ot_handle == NULL);
4311
4312         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4313                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4314         /*
4315          * xattr del may involve inode quota change, reserve credits for
4316          * dquot_initialize()
4317          */
4318         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4319
4320         return 0;
4321 }
4322
4323 /*
4324  * Concurrency: @dt is write locked.
4325  */
4326 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4327                          const char *name, struct thandle *handle)
4328 {
4329         struct osd_object      *obj    = osd_dt_obj(dt);
4330         struct inode           *inode  = obj->oo_inode;
4331         struct osd_thread_info *info   = osd_oti_get(env);
4332         struct dentry          *dentry = &info->oti_obj_dentry;
4333         int                     rc;
4334
4335         if (!dt_object_exists(dt))
4336                 return -ENOENT;
4337
4338         LASSERT(!dt_object_remote(dt));
4339         LASSERT(inode->i_op != NULL);
4340         LASSERT(inode->i_op->removexattr != NULL);
4341         LASSERT(handle != NULL);
4342
4343         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4344
4345         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4346                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4347
4348                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4349                                  &info->oti_ost_attrs);
4350                 if (!rc) {
4351                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4352
4353                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4354                         lustre_lma_swab(lma);
4355                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4356                                              sizeof(*lma), XATTR_REPLACE);
4357                         if (!rc)
4358                                 obj->oo_pfid_in_lma = 0;
4359                 }
4360         } else {
4361                 ll_vfs_dq_init(inode);
4362                 dentry->d_inode = inode;
4363                 dentry->d_sb = inode->i_sb;
4364                 rc = inode->i_op->removexattr(dentry, name);
4365         }
4366
4367         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4368
4369         if (rc == 0 &&
4370             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4371              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4372                 osd_oxc_del(obj, name);
4373
4374         return rc;
4375 }
4376
4377 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4378                            __u64 start, __u64 end)
4379 {
4380         struct osd_object       *obj    = osd_dt_obj(dt);
4381         struct inode            *inode  = obj->oo_inode;
4382         struct osd_thread_info  *info   = osd_oti_get(env);
4383         struct dentry           *dentry = &info->oti_obj_dentry;
4384         struct file             *file   = &info->oti_file;
4385         int                     rc;
4386
4387         ENTRY;
4388
4389         dentry->d_inode = inode;
4390         dentry->d_sb = inode->i_sb;
4391         file->f_path.dentry = dentry;
4392         file->f_mapping = inode->i_mapping;
4393         file->f_op = inode->i_fop;
4394         set_file_inode(file, inode);
4395
4396         rc = ll_vfs_fsync_range(file, start, end, 0);
4397
4398         RETURN(rc);
4399 }
4400
4401 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4402 {
4403         return 0;
4404 }
4405
4406 /*
4407  * Index operations.
4408  */
4409
4410 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4411                            const struct dt_index_features *feat)
4412 {
4413         struct iam_descr *descr;
4414
4415         if (osd_object_is_root(o))
4416                 return feat == &dt_directory_features;
4417
4418         LASSERT(o->oo_dir != NULL);
4419
4420         descr = o->oo_dir->od_container.ic_descr;
4421         if (feat == &dt_directory_features) {
4422                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4423                         return 1;
4424                 else
4425                         return 0;
4426         } else {
4427                 return
4428                         feat->dif_keysize_min <= descr->id_key_size &&
4429                         descr->id_key_size <= feat->dif_keysize_max &&
4430                         feat->dif_recsize_min <= descr->id_rec_size &&
4431                         descr->id_rec_size <= feat->dif_recsize_max &&
4432                         !(feat->dif_flags & (DT_IND_VARKEY |
4433                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4434                         ergo(feat->dif_flags & DT_IND_UPDATE,
4435                              1 /* XXX check that object (and file system) is
4436                                 * writable */);
4437         }
4438 }
4439
4440 static int osd_iam_container_init(const struct lu_env *env,
4441                                   struct osd_object *obj,
4442                                   struct osd_directory *dir)
4443 {
4444         struct iam_container *bag = &dir->od_container;
4445         int result;
4446
4447         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4448         if (result != 0)
4449                 return result;
4450
4451         result = iam_container_setup(bag);
4452         if (result == 0)
4453                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4454         else
4455                 iam_container_fini(bag);
4456
4457         return result;
4458 }
4459
4460
4461 /*
4462  * Concurrency: no external locking is necessary.
4463  */
4464 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4465                          const struct dt_index_features *feat)
4466 {
4467         int                      result;
4468         int                      skip_iam = 0;
4469         struct osd_object       *obj = osd_dt_obj(dt);
4470
4471         LINVRNT(osd_invariant(obj));
4472
4473         if (osd_object_is_root(obj)) {
4474                 dt->do_index_ops = &osd_index_ea_ops;
4475                 result = 0;
4476         } else if (feat == &dt_directory_features) {
4477                 dt->do_index_ops = &osd_index_ea_ops;
4478                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4479                         result = 0;
4480                 else
4481                         result = -ENOTDIR;
4482                 skip_iam = 1;
4483         } else if (unlikely(feat == &dt_otable_features)) {
4484                 dt->do_index_ops = &osd_otable_ops;
4485                 return 0;
4486         } else if (unlikely(feat == &dt_acct_features)) {
4487                 dt->do_index_ops = &osd_acct_index_ops;
4488                 result = 0;
4489                 skip_iam = 1;
4490         } else if (!osd_has_index(obj)) {
4491                 struct osd_directory *dir;
4492
4493                 OBD_ALLOC_PTR(dir);
4494                 if (dir != NULL) {
4495
4496                         spin_lock(&obj->oo_guard);
4497                         if (obj->oo_dir == NULL)
4498                                 obj->oo_dir = dir;
4499                         else
4500                                 /*
4501                                  * Concurrent thread allocated container data.
4502                                  */
4503                                 OBD_FREE_PTR(dir);
4504                         spin_unlock(&obj->oo_guard);
4505                         /*
4506                          * Now, that we have container data, serialize its
4507                          * initialization.
4508                          */
4509                         down_write(&obj->oo_ext_idx_sem);
4510                         /*
4511                          * recheck under lock.
4512                          */
4513                         if (!osd_has_index(obj))
4514                                 result = osd_iam_container_init(env, obj,
4515                                                                 obj->oo_dir);
4516                         else
4517                                 result = 0;
4518                         up_write(&obj->oo_ext_idx_sem);
4519                 } else {
4520                         result = -ENOMEM;
4521                 }
4522         } else {
4523                 result = 0;
4524         }
4525
4526         if (result == 0 && skip_iam == 0) {
4527                 if (!osd_iam_index_probe(env, obj, feat))
4528                         result = -ENOTDIR;
4529         }
4530         LINVRNT(osd_invariant(obj));
4531
4532         return result;
4533 }
4534
4535 static int osd_otable_it_attr_get(const struct lu_env *env,
4536                                  struct dt_object *dt,
4537                                  struct lu_attr *attr)
4538 {
4539         attr->la_valid = 0;
4540         return 0;
4541 }
4542
4543 static const struct dt_object_operations osd_obj_ops = {
4544         .do_read_lock           = osd_read_lock,
4545         .do_write_lock          = osd_write_lock,
4546         .do_read_unlock         = osd_read_unlock,
4547         .do_write_unlock        = osd_write_unlock,
4548         .do_write_locked        = osd_write_locked,
4549         .do_attr_get            = osd_attr_get,
4550         .do_declare_attr_set    = osd_declare_attr_set,
4551         .do_attr_set            = osd_attr_set,
4552         .do_ah_init             = osd_ah_init,
4553         .do_declare_create      = osd_declare_create,
4554         .do_create              = osd_create,
4555         .do_declare_destroy     = osd_declare_destroy,
4556         .do_destroy             = osd_destroy,
4557         .do_index_try           = osd_index_try,
4558         .do_declare_ref_add     = osd_declare_ref_add,
4559         .do_ref_add             = osd_ref_add,
4560         .do_declare_ref_del     = osd_declare_ref_del,
4561         .do_ref_del             = osd_ref_del,
4562         .do_xattr_get           = osd_xattr_get,
4563         .do_declare_xattr_set   = osd_declare_xattr_set,
4564         .do_xattr_set           = osd_xattr_set,
4565         .do_declare_xattr_del   = osd_declare_xattr_del,
4566         .do_xattr_del           = osd_xattr_del,
4567         .do_xattr_list          = osd_xattr_list,
4568         .do_object_sync         = osd_object_sync,
4569         .do_invalidate          = osd_invalidate,
4570 };
4571
4572 static const struct dt_object_operations osd_obj_otable_it_ops = {
4573         .do_attr_get    = osd_otable_it_attr_get,
4574         .do_index_try   = osd_index_try,
4575 };
4576
4577 static int osd_index_declare_iam_delete(const struct lu_env *env,
4578                                         struct dt_object *dt,
4579                                         const struct dt_key *key,
4580                                         struct thandle *handle)
4581 {
4582         struct osd_thandle    *oh;
4583
4584         oh = container_of0(handle, struct osd_thandle, ot_super);
4585         LASSERT(oh->ot_handle == NULL);
4586
4587         /* Recycle  may cause additional three blocks to be changed. */
4588         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4589                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4590
4591         return 0;
4592 }
4593
4594 /**
4595  *      delete a (key, value) pair from index \a dt specified by \a key
4596  *
4597  *      \param  dt      osd index object
4598  *      \param  key     key for index
4599  *      \param  rec     record reference
4600  *      \param  handle  transaction handler
4601  *
4602  *      \retval  0  success
4603  *      \retval -ve   failure
4604  */
4605 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4606                                 const struct dt_key *key,
4607                                 struct thandle *handle)
4608 {
4609         struct osd_thread_info *oti = osd_oti_get(env);
4610         struct osd_object      *obj = osd_dt_obj(dt);
4611         struct osd_thandle     *oh;
4612         struct iam_path_descr  *ipd;
4613         struct iam_container   *bag = &obj->oo_dir->od_container;
4614         int                     rc;
4615         ENTRY;
4616
4617         if (!dt_object_exists(dt))
4618                 RETURN(-ENOENT);
4619
4620         LINVRNT(osd_invariant(obj));
4621         LASSERT(!dt_object_remote(dt));
4622         LASSERT(bag->ic_object == obj->oo_inode);
4623         LASSERT(handle != NULL);
4624
4625         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4626
4627         ipd = osd_idx_ipd_get(env, bag);
4628         if (unlikely(ipd == NULL))
4629                 RETURN(-ENOMEM);
4630
4631         oh = container_of0(handle, struct osd_thandle, ot_super);
4632         LASSERT(oh->ot_handle != NULL);
4633         LASSERT(oh->ot_handle->h_transaction != NULL);
4634
4635         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4636                 /* swab quota uid/gid provided by caller */
4637                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4638                 key = (const struct dt_key *)&oti->oti_quota_id;
4639         }
4640
4641         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4642         osd_ipd_put(env, bag, ipd);
4643         LINVRNT(osd_invariant(obj));
4644         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4645         RETURN(rc);
4646 }
4647
4648 static int osd_index_declare_ea_delete(const struct lu_env *env,
4649                                        struct dt_object *dt,
4650                                        const struct dt_key *key,
4651                                        struct thandle *handle)
4652 {
4653         struct osd_thandle *oh;
4654         struct inode       *inode;
4655         int                 rc, credits;
4656         ENTRY;
4657
4658         LASSERT(!dt_object_remote(dt));
4659         LASSERT(handle != NULL);
4660
4661         oh = container_of0(handle, struct osd_thandle, ot_super);
4662         LASSERT(oh->ot_handle == NULL);
4663
4664         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4665         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4666
4667         inode = osd_dt_obj(dt)->oo_inode;
4668         if (inode == NULL)
4669                 RETURN(-ENOENT);
4670
4671         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4672                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4673                                    NULL, OSD_QID_BLK);
4674         RETURN(rc);
4675 }
4676
4677 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4678                                           struct dt_rec *fid)
4679 {
4680         struct osd_fid_pack *rec;
4681         int                  rc = -ENODATA;
4682
4683         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4684                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4685                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4686                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4687                         rc = -EINVAL;
4688         }
4689         return rc;
4690 }
4691
4692 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4693                           const struct lu_fid *fid)
4694 {
4695         struct seq_server_site  *ss = osd_seq_site(osd);
4696         ENTRY;
4697
4698         /* FID seqs not in FLDB, must be local seq */
4699         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4700                 RETURN(0);
4701
4702         /* If FLD is not being initialized yet, it only happens during the
4703          * initialization, likely during mgs initialization, and we assume
4704          * this is local FID. */
4705         if (ss == NULL || ss->ss_server_fld == NULL)
4706                 RETURN(0);
4707
4708         /* Only check the local FLDB here */
4709         if (osd_seq_exists(env, osd, fid_seq(fid)))
4710                 RETURN(0);
4711
4712         RETURN(1);
4713 }
4714
4715 /**
4716  * Index delete function for interoperability mode (b11826).
4717  * It will remove the directory entry added by osd_index_ea_insert().
4718  * This entry is needed to maintain name->fid mapping.
4719  *
4720  * \param key,  key i.e. file entry to be deleted
4721  *
4722  * \retval   0, on success
4723  * \retval -ve, on error
4724  */
4725 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4726                                const struct dt_key *key, struct thandle *handle)
4727 {
4728         struct osd_object          *obj = osd_dt_obj(dt);
4729         struct inode               *dir = obj->oo_inode;
4730         struct dentry              *dentry;
4731         struct osd_thandle         *oh;
4732         struct ldiskfs_dir_entry_2 *de = NULL;
4733         struct buffer_head         *bh;
4734         struct htree_lock          *hlock = NULL;
4735         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
4736         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4737         int                        rc;
4738         ENTRY;
4739
4740         if (!dt_object_exists(dt))
4741                 RETURN(-ENOENT);
4742
4743         LINVRNT(osd_invariant(obj));
4744         LASSERT(!dt_object_remote(dt));
4745         LASSERT(handle != NULL);
4746
4747         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4748
4749         oh = container_of(handle, struct osd_thandle, ot_super);
4750         LASSERT(oh->ot_handle != NULL);
4751         LASSERT(oh->ot_handle->h_transaction != NULL);
4752
4753         ll_vfs_dq_init(dir);
4754         dentry = osd_child_dentry_get(env, obj,
4755                                       (char *)key, strlen((char *)key));
4756
4757         if (obj->oo_hl_head != NULL) {
4758                 hlock = osd_oti_get(env)->oti_hlock;
4759                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4760                                    dir, LDISKFS_HLOCK_DEL);
4761         } else {
4762                 down_write(&obj->oo_ext_idx_sem);
4763         }
4764
4765         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4766         if (!IS_ERR(bh)) {
4767                 /* If this is not the ".." entry, it might be a remote DNE
4768                  * entry and  we need to check if the FID is for a remote
4769                  * MDT.  If the FID is  not in the directory entry (e.g.
4770                  * upgraded 1.8 filesystem without dirdata enabled) then
4771                  * we need to get the FID from the LMA. For a remote directory
4772                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4773                  * case.
4774                  *
4775                  * Delete the entry before the agent inode in order to
4776                  * simplify error handling.  At worst an error after deleting
4777                  * the entry first might leak the agent inode afterward. The
4778                  * reverse would need filesystem abort in case of error deleting
4779                  * the entry after the agent had been removed, or leave a
4780                  * dangling entry pointing at a random inode. */
4781                 if (strcmp((char *)key, dotdot) != 0) {
4782                         LASSERT(de != NULL);
4783                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4784                         if (rc == -ENODATA) {
4785                                 /* can't get FID, postpone to the end of the
4786                                  * transaction when iget() is safe */
4787                                 osd_schedule_agent_inode_removal(env, oh,
4788                                                 le32_to_cpu(de->inode));
4789                         } else if (rc == 0 &&
4790                                    unlikely(osd_remote_fid(env, osd, fid))) {
4791                                 osd_schedule_agent_inode_removal(env, oh,
4792                                                 le32_to_cpu(de->inode));
4793                         }
4794                 }
4795                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4796                 brelse(bh);
4797         } else {
4798                 rc = PTR_ERR(bh);
4799         }
4800         if (hlock != NULL)
4801                 ldiskfs_htree_unlock(hlock);
4802         else
4803                 up_write(&obj->oo_ext_idx_sem);
4804
4805         GOTO(out, rc);
4806
4807 out:
4808         LASSERT(osd_invariant(obj));
4809         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4810         RETURN(rc);
4811 }
4812
4813 /**
4814  *      Lookup index for \a key and copy record to \a rec.
4815  *
4816  *      \param  dt      osd index object
4817  *      \param  key     key for index
4818  *      \param  rec     record reference
4819  *
4820  *      \retval  +ve  success : exact mach
4821  *      \retval  0    return record with key not greater than \a key
4822  *      \retval -ve   failure
4823  */
4824 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4825                                 struct dt_rec *rec, const struct dt_key *key)
4826 {
4827         struct osd_object      *obj = osd_dt_obj(dt);
4828         struct iam_path_descr  *ipd;
4829         struct iam_container   *bag = &obj->oo_dir->od_container;
4830         struct osd_thread_info *oti = osd_oti_get(env);
4831         struct iam_iterator    *it = &oti->oti_idx_it;
4832         struct iam_rec         *iam_rec;
4833         int                     rc;
4834         ENTRY;
4835
4836         if (!dt_object_exists(dt))
4837                 RETURN(-ENOENT);
4838
4839         LASSERT(osd_invariant(obj));
4840         LASSERT(!dt_object_remote(dt));
4841         LASSERT(bag->ic_object == obj->oo_inode);
4842
4843         ipd = osd_idx_ipd_get(env, bag);
4844         if (IS_ERR(ipd))
4845                 RETURN(-ENOMEM);
4846
4847         /* got ipd now we can start iterator. */
4848         iam_it_init(it, bag, 0, ipd);
4849
4850         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4851                 /* swab quota uid/gid provided by caller */
4852                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4853                 key = (const struct dt_key *)&oti->oti_quota_id;
4854         }
4855
4856         rc = iam_it_get(it, (struct iam_key *)key);
4857         if (rc >= 0) {
4858                 if (S_ISDIR(obj->oo_inode->i_mode))
4859                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4860                 else
4861                         iam_rec = (struct iam_rec *) rec;
4862
4863                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4864
4865                 if (S_ISDIR(obj->oo_inode->i_mode))
4866                         osd_fid_unpack((struct lu_fid *) rec,
4867                                        (struct osd_fid_pack *)iam_rec);
4868                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4869                         osd_quota_unpack(obj, rec);
4870         }
4871
4872         iam_it_put(it);
4873         iam_it_fini(it);
4874         osd_ipd_put(env, bag, ipd);
4875
4876         LINVRNT(osd_invariant(obj));
4877
4878         RETURN(rc);
4879 }
4880
4881 static int osd_index_declare_iam_insert(const struct lu_env *env,
4882                                         struct dt_object *dt,
4883                                         const struct dt_rec *rec,
4884                                         const struct dt_key *key,
4885                                         struct thandle *handle)
4886 {
4887         struct osd_thandle *oh;
4888
4889         LASSERT(handle != NULL);
4890
4891         oh = container_of0(handle, struct osd_thandle, ot_super);
4892         LASSERT(oh->ot_handle == NULL);
4893
4894         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4895                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4896
4897         return 0;
4898 }
4899
4900 /**
4901  *      Inserts (key, value) pair in \a dt index object.
4902  *
4903  *      \param  dt      osd index object
4904  *      \param  key     key for index
4905  *      \param  rec     record reference
4906  *      \param  th      transaction handler
4907  *
4908  *      \retval  0  success
4909  *      \retval -ve failure
4910  */
4911 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4912                                 const struct dt_rec *rec,
4913                                 const struct dt_key *key, struct thandle *th,
4914                                 int ignore_quota)
4915 {
4916         struct osd_object     *obj = osd_dt_obj(dt);
4917         struct iam_path_descr *ipd;
4918         struct osd_thandle    *oh;
4919         struct iam_container  *bag;
4920         struct osd_thread_info *oti = osd_oti_get(env);
4921         struct iam_rec         *iam_rec;
4922         int                     rc;
4923         ENTRY;
4924
4925         if (!dt_object_exists(dt))
4926                 RETURN(-ENOENT);
4927
4928         LINVRNT(osd_invariant(obj));
4929         LASSERT(!dt_object_remote(dt));
4930
4931         bag = &obj->oo_dir->od_container;
4932         LASSERT(bag->ic_object == obj->oo_inode);
4933         LASSERT(th != NULL);
4934
4935         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4936
4937         ipd = osd_idx_ipd_get(env, bag);
4938         if (unlikely(ipd == NULL))
4939                 RETURN(-ENOMEM);
4940
4941         oh = container_of0(th, struct osd_thandle, ot_super);
4942         LASSERT(oh->ot_handle != NULL);
4943         LASSERT(oh->ot_handle->h_transaction != NULL);
4944         if (S_ISDIR(obj->oo_inode->i_mode)) {
4945                 iam_rec = (struct iam_rec *)oti->oti_ldp;
4946                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4947         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4948                 /* pack quota uid/gid */
4949                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4950                 key = (const struct dt_key *)&oti->oti_quota_id;
4951                 /* pack quota record */
4952                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4953                 iam_rec = (struct iam_rec *)rec;
4954         } else {
4955                 iam_rec = (struct iam_rec *)rec;
4956         }
4957
4958         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4959                         iam_rec, ipd);
4960         osd_ipd_put(env, bag, ipd);
4961         LINVRNT(osd_invariant(obj));
4962         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4963         RETURN(rc);
4964 }
4965
4966 /**
4967  * Calls ldiskfs_add_entry() to add directory entry
4968  * into the directory. This is required for
4969  * interoperability mode (b11826)
4970  *
4971  * \retval   0, on success
4972  * \retval -ve, on error
4973  */
4974 static int __osd_ea_add_rec(struct osd_thread_info *info,
4975                             struct osd_object *pobj, struct inode  *cinode,
4976                             const char *name, const struct lu_fid *fid,
4977                             struct htree_lock *hlock, struct thandle *th)
4978 {
4979         struct ldiskfs_dentry_param *ldp;
4980         struct dentry               *child;
4981         struct osd_thandle          *oth;
4982         int                          rc;
4983
4984         oth = container_of(th, struct osd_thandle, ot_super);
4985         LASSERT(oth->ot_handle != NULL);
4986         LASSERT(oth->ot_handle->h_transaction != NULL);
4987         LASSERT(pobj->oo_inode);
4988
4989         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4990         if (unlikely(pobj->oo_inode ==
4991                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4992                 ldp->edp_magic = 0;
4993         else
4994                 osd_get_ldiskfs_dirent_param(ldp, fid);
4995         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4996         child->d_fsdata = (void *)ldp;
4997         ll_vfs_dq_init(pobj->oo_inode);
4998         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4999                                    child, cinode, hlock);
5000         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5001                 struct ldiskfs_dir_entry_2      *de;
5002                 struct buffer_head              *bh;
5003                 int                              rc1;
5004
5005                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5006                                             NULL, hlock);
5007                 if (!IS_ERR(bh)) {
5008                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5009                                                                bh);
5010                         if (rc1 == 0) {
5011                                 if (S_ISDIR(cinode->i_mode))
5012                                         de->file_type = LDISKFS_DIRENT_LUFID |
5013                                                         LDISKFS_FT_REG_FILE;
5014                                 else
5015                                         de->file_type = LDISKFS_DIRENT_LUFID |
5016                                                         LDISKFS_FT_DIR;
5017                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5018                                                               NULL, bh);
5019                         }
5020                         brelse(bh);
5021                 }
5022         }
5023
5024         RETURN(rc);
5025 }
5026
5027 /**
5028  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5029  * into the directory.Also sets flags into osd object to
5030  * indicate dot and dotdot are created. This is required for
5031  * interoperability mode (b11826)
5032  *
5033  * \param dir   directory for dot and dotdot fixup.
5034  * \param obj   child object for linking
5035  *
5036  * \retval   0, on success
5037  * \retval -ve, on error
5038  */
5039 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5040                               struct osd_object *dir,
5041                               struct inode *parent_dir, const char *name,
5042                               const struct lu_fid *dot_fid,
5043                               const struct lu_fid *dot_dot_fid,
5044                               struct thandle *th)
5045 {
5046         struct inode                *inode = dir->oo_inode;
5047         struct osd_thandle          *oth;
5048         int result = 0;
5049
5050         oth = container_of(th, struct osd_thandle, ot_super);
5051         LASSERT(oth->ot_handle->h_transaction != NULL);
5052         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5053
5054         if (strcmp(name, dot) == 0) {
5055                 if (dir->oo_compat_dot_created) {
5056                         result = -EEXIST;
5057                 } else {
5058                         LASSERT(inode->i_ino == parent_dir->i_ino);
5059                         dir->oo_compat_dot_created = 1;
5060                         result = 0;
5061                 }
5062         } else if (strcmp(name, dotdot) == 0) {
5063                 if (!dir->oo_compat_dot_created)
5064                         return -EINVAL;
5065                 /* in case of rename, dotdot is already created */
5066                 if (dir->oo_compat_dotdot_created) {
5067                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5068                                                 dot_dot_fid, NULL, th);
5069                 }
5070
5071                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5072                         struct lu_fid tfid = *dot_dot_fid;
5073
5074                         tfid.f_oid--;
5075                         result = osd_add_dot_dotdot_internal(info,
5076                                         dir->oo_inode, parent_dir, dot_fid,
5077                                         &tfid, oth);
5078                 } else {
5079                         result = osd_add_dot_dotdot_internal(info,
5080                                         dir->oo_inode, parent_dir, dot_fid,
5081                                         dot_dot_fid, oth);
5082                 }
5083
5084                 if (result == 0)
5085                         dir->oo_compat_dotdot_created = 1;
5086         }
5087
5088         return result;
5089 }
5090
5091
5092 /**
5093  * It will call the appropriate osd_add* function and return the
5094  * value, return by respective functions.
5095  */
5096 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5097                           struct inode *cinode, const char *name,
5098                           const struct lu_fid *fid, struct thandle *th)
5099 {
5100         struct osd_thread_info *info   = osd_oti_get(env);
5101         struct htree_lock      *hlock;
5102         int                     rc;
5103
5104         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5105
5106         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
5107                                                    name[2] =='\0'))) {
5108                 if (hlock != NULL) {
5109                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5110                                            pobj->oo_inode, 0);
5111                 } else {
5112                         down_write(&pobj->oo_ext_idx_sem);
5113                 }
5114
5115                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5116                                         lu_object_fid(&pobj->oo_dt.do_lu),
5117                                         fid, th);
5118         } else {
5119                 if (hlock != NULL) {
5120                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5121                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5122                 } else {
5123                         down_write(&pobj->oo_ext_idx_sem);
5124                 }
5125
5126                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5127                         struct lu_fid *tfid = &info->oti_fid;
5128
5129                         *tfid = *fid;
5130                         tfid->f_ver = ~0;
5131                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5132                                               tfid, hlock, th);
5133                 } else {
5134                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5135                                               hlock, th);
5136                 }
5137         }
5138         if (hlock != NULL)
5139                 ldiskfs_htree_unlock(hlock);
5140         else
5141                 up_write(&pobj->oo_ext_idx_sem);
5142
5143         return rc;
5144 }
5145
5146 static int
5147 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5148                       struct osd_idmap_cache *oic)
5149 {
5150         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5151         struct lu_fid *fid = &oic->oic_fid;
5152         struct osd_inode_id *id = &oic->oic_lid;
5153         struct inode *inode = NULL;
5154         int once = 0;
5155         bool insert;
5156         int rc;
5157         ENTRY;
5158
5159         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5160                 RETURN(0);
5161
5162         if (thread_is_running(&scrub->os_thread) &&
5163             scrub->os_pos_current > id->oii_ino)
5164                 RETURN(0);
5165
5166         if (dev->od_auto_scrub_interval == AS_NEVER ||
5167             cfs_time_before(ktime_get_real_seconds(),
5168                             scrub->os_file.sf_time_last_complete +
5169                             dev->od_auto_scrub_interval))
5170                 RETURN(0);
5171
5172 again:
5173         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5174         if (rc == -ENOENT) {
5175                 __u32 gen = id->oii_gen;
5176
5177                 insert = true;
5178                 if (inode != NULL)
5179                         goto trigger;
5180
5181                 inode = osd_iget(oti, dev, id);
5182                 /* The inode has been removed (by race maybe). */
5183                 if (IS_ERR(inode)) {
5184                         rc = PTR_ERR(inode);
5185
5186                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5187                 }
5188
5189                 /* The OI mapping is lost. */
5190                 if (gen != OSD_OII_NOGEN)
5191                         goto trigger;
5192
5193                 /* The inode may has been reused by others, we do not know,
5194                  * leave it to be handled by subsequent osd_fid_lookup(). */
5195                 GOTO(out, rc = 0);
5196         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5197                 GOTO(out, rc);
5198         }
5199
5200         insert = false;
5201
5202 trigger:
5203         if (thread_is_running(&scrub->os_thread)) {
5204                 if (inode == NULL) {
5205                         inode = osd_iget(oti, dev, id);
5206                         /* The inode has been removed (by race maybe). */
5207                         if (IS_ERR(inode)) {
5208                                 rc = PTR_ERR(inode);
5209
5210                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5211                         }
5212                 }
5213
5214                 rc = osd_oii_insert(dev, oic, insert);
5215                 /* There is race condition between osd_oi_lookup and OI scrub.
5216                  * The OI scrub finished just after osd_oi_lookup() failure.
5217                  * Under such case, it is unnecessary to trigger OI scrub again,
5218                  * but try to call osd_oi_lookup() again. */
5219                 if (unlikely(rc == -EAGAIN))
5220                         goto again;
5221
5222                 if (!S_ISDIR(inode->i_mode))
5223                         rc = 0;
5224                 else
5225                         rc = osd_check_lmv(oti, dev, inode, oic);
5226
5227                 GOTO(out, rc);
5228         }
5229
5230         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5231                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5232                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5233                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5234                        "%s: trigger partial OI scrub for RPC inconsistency "
5235                        "checking FID "DFID": rc = %d\n",
5236                        osd_dev2name(dev), PFID(fid), rc);
5237                 if (rc == 0 || rc == -EALREADY)
5238                         goto again;
5239         }
5240
5241         GOTO(out, rc);
5242
5243 out:
5244         if (inode)
5245                 iput(inode);
5246
5247         RETURN(rc);
5248 }
5249
5250 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5251                                struct osd_device *dev,
5252                                struct osd_idmap_cache *oic,
5253                                struct lu_fid *fid, __u32 ino)
5254 {
5255         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5256         struct inode *inode;
5257         int rc;
5258
5259         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5260         inode = osd_iget(oti, dev, &oic->oic_lid);
5261         if (IS_ERR(inode)) {
5262                 fid_zero(&oic->oic_fid);
5263                 return PTR_ERR(inode);
5264         }
5265
5266         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5267         iput(inode);
5268         if (rc != 0)
5269                 fid_zero(&oic->oic_fid);
5270         else
5271                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5272         return rc;
5273 }
5274
5275 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5276                       struct osd_inode_id *id, const struct lu_fid *fid)
5277 {
5278         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5279                id->oii_ino, id->oii_gen, info);
5280         info->oti_cache.oic_lid = *id;
5281         info->oti_cache.oic_fid = *fid;
5282         info->oti_cache.oic_dev = osd;
5283 }
5284
5285 /**
5286  * Get parent FID from the linkEA.
5287  *
5288  * For a directory which parent resides on remote MDT, to satisfy the
5289  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5290  * the other hand, to make the lookup(..) on the directory can return
5291  * the real parent FID, we append the real parent FID after its ".."
5292  * name entry in the /REMOTE_PARENT_DIR.
5293  *
5294  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5295  * backup. So after the restore, we cannot get the right parent FID from
5296  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5297  * we have stored the real parent FID in the directory object's linkEA,
5298  * we can parse the linkEA for the real parent FID.
5299  *
5300  * \param[in] env       pointer to the thread context
5301  * \param[in] obj       pointer to the object to be handled
5302  * \param[out]fid       pointer to the buffer to hold the parent FID
5303  *
5304  * \retval              0 for getting the real parent FID successfully
5305  * \retval              negative error number on failure
5306  */
5307 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5308                                     struct osd_object *obj,
5309                                     struct lu_fid *fid)
5310 {
5311         struct osd_thread_info  *oti    = osd_oti_get(env);
5312         struct lu_buf           *buf    = &oti->oti_big_buf;
5313         struct dentry           *dentry = &oti->oti_obj_dentry;
5314         struct inode            *inode  = obj->oo_inode;
5315         struct linkea_data       ldata  = { NULL };
5316         int                      rc;
5317         ENTRY;
5318
5319         fid_zero(fid);
5320         if (!S_ISDIR(inode->i_mode))
5321                 RETURN(-EIO);
5322
5323 again:
5324         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5325                              buf->lb_buf, buf->lb_len);
5326         if (rc == -ERANGE) {
5327                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5328                                      NULL, 0);
5329                 if (rc > 0) {
5330                         lu_buf_realloc(buf, rc);
5331                         if (buf->lb_buf == NULL)
5332                                 RETURN(-ENOMEM);
5333
5334                         goto again;
5335                 }
5336         }
5337
5338         if (unlikely(rc == 0))
5339                 RETURN(-ENODATA);
5340
5341         if (rc < 0)
5342                 RETURN(rc);
5343
5344         if (unlikely(buf->lb_buf == NULL)) {
5345                 lu_buf_realloc(buf, rc);
5346                 if (buf->lb_buf == NULL)
5347                         RETURN(-ENOMEM);
5348
5349                 goto again;
5350         }
5351
5352         ldata.ld_buf = buf;
5353         rc = linkea_init_with_rec(&ldata);
5354         if (!rc) {
5355                 linkea_first_entry(&ldata);
5356                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5357         }
5358
5359         RETURN(rc);
5360 }
5361
5362 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5363                                     struct inode *inode,
5364                                     const struct lu_fid *pfid,
5365                                     const char *name, const int namelen)
5366 {
5367         struct osd_thread_info *oti = osd_oti_get(env);
5368         struct lu_buf *buf = &oti->oti_big_buf;
5369         struct dentry *dentry = &oti->oti_obj_dentry;
5370         struct linkea_data ldata = { NULL };
5371         struct lu_name cname = { .ln_name = name,
5372                                  .ln_namelen = namelen };
5373         int rc;
5374         ENTRY;
5375
5376 again:
5377         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5378                              buf->lb_buf, buf->lb_len);
5379         if (rc == -ERANGE)
5380                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5381
5382         if (rc < 0)
5383                 RETURN(rc);
5384
5385         if (unlikely(rc == 0))
5386                 RETURN(-ENODATA);
5387
5388         if (buf->lb_len < rc) {
5389                 lu_buf_realloc(buf, rc);
5390                 if (buf->lb_buf == NULL)
5391                         RETURN(-ENOMEM);
5392
5393                 goto again;
5394         }
5395
5396         ldata.ld_buf = buf;
5397         rc = linkea_init_with_rec(&ldata);
5398         if (!rc)
5399                 rc = linkea_links_find(&ldata, &cname, pfid);
5400
5401         RETURN(rc);
5402 }
5403
5404 /**
5405  * Calls ->lookup() to find dentry. From dentry get inode and
5406  * read inode's ea to get fid. This is required for  interoperability
5407  * mode (b11826)
5408  *
5409  * \retval   0, on success
5410  * \retval -ve, on error
5411  */
5412 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5413                              struct dt_rec *rec, const struct dt_key *key)
5414 {
5415         struct inode                    *dir    = obj->oo_inode;
5416         struct dentry                   *dentry;
5417         struct ldiskfs_dir_entry_2      *de;
5418         struct buffer_head              *bh;
5419         struct lu_fid                   *fid = (struct lu_fid *) rec;
5420         struct htree_lock               *hlock = NULL;
5421         int                             ino;
5422         int                             rc;
5423         ENTRY;
5424
5425         LASSERT(dir->i_op != NULL);
5426         LASSERT(dir->i_op->lookup != NULL);
5427
5428         dentry = osd_child_dentry_get(env, obj,
5429                                       (char *)key, strlen((char *)key));
5430
5431         if (obj->oo_hl_head != NULL) {
5432                 hlock = osd_oti_get(env)->oti_hlock;
5433                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5434                                    dir, LDISKFS_HLOCK_LOOKUP);
5435         } else {
5436                 down_read(&obj->oo_ext_idx_sem);
5437         }
5438
5439         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5440         if (!IS_ERR(bh)) {
5441                 struct osd_thread_info *oti = osd_oti_get(env);
5442                 struct osd_inode_id *id = &oti->oti_id;
5443                 struct osd_idmap_cache *oic = &oti->oti_cache;
5444                 struct osd_device *dev = osd_obj2dev(obj);
5445
5446                 ino = le32_to_cpu(de->inode);
5447                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5448                         brelse(bh);
5449                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5450                         GOTO(out, rc);
5451                 }
5452
5453                 rc = osd_get_fid_from_dentry(de, rec);
5454
5455                 /* done with de, release bh */
5456                 brelse(bh);
5457                 if (rc != 0) {
5458                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
5459                                 const char *name = (const char *)key;
5460
5461                                 /* If the parent is on remote MDT, and there
5462                                  * is no FID-in-dirent, then we have to get
5463                                  * the parent FID from the linkEA.  */
5464                                 if (likely(strlen(name) == 2 &&
5465                                            name[0] == '.' && name[1] == '.'))
5466                                         rc = osd_get_pfid_from_linkea(env, obj,
5467                                                                       fid);
5468                         } else {
5469                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5470                         }
5471                 } else {
5472                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5473                 }
5474
5475                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5476                         fid_zero(&oic->oic_fid);
5477
5478                         GOTO(out, rc);
5479                 }
5480
5481                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5482                 rc = osd_consistency_check(oti, dev, oic);
5483                 if (rc == -ENOENT)
5484                         fid_zero(&oic->oic_fid);
5485                 else
5486                         /* Other error should not affect lookup result. */
5487                         rc = 0;
5488         } else {
5489                 rc = PTR_ERR(bh);
5490         }
5491
5492         GOTO(out, rc);
5493
5494 out:
5495         if (hlock != NULL)
5496                 ldiskfs_htree_unlock(hlock);
5497         else
5498                 up_read(&obj->oo_ext_idx_sem);
5499         return rc;
5500 }
5501
5502 static int osd_index_declare_ea_insert(const struct lu_env *env,
5503                                        struct dt_object *dt,
5504                                        const struct dt_rec *rec,
5505                                        const struct dt_key *key,
5506                                        struct thandle *handle)
5507 {
5508         struct osd_thandle      *oh;
5509         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5510         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5511         const struct lu_fid     *fid    = rec1->rec_fid;
5512         int                      credits, rc = 0;
5513         struct osd_idmap_cache  *idc;
5514         ENTRY;
5515
5516         LASSERT(!dt_object_remote(dt));
5517         LASSERT(handle != NULL);
5518         LASSERT(fid != NULL);
5519         LASSERT(rec1->rec_type != 0);
5520
5521         oh = container_of0(handle, struct osd_thandle, ot_super);
5522         LASSERT(oh->ot_handle == NULL);
5523
5524         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5525
5526         /* we can't call iget() while a transactions is running
5527          * (this can lead to a deadlock), but we need to know
5528          * inum and object type. so we find this information at
5529          * declaration and cache in per-thread info */
5530         idc = osd_idc_find_or_init(env, osd, fid);
5531         if (IS_ERR(idc))
5532                 RETURN(PTR_ERR(idc));
5533         if (idc->oic_remote) {
5534                 /* a reference to remote inode is represented by an
5535                  * agent inode which we have to create */
5536                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5537                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5538         }
5539
5540         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5541
5542         if (osd_dt_obj(dt)->oo_inode != NULL) {
5543                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5544
5545                 /* We ignore block quota on meta pool (MDTs), so needn't
5546                  * calculate how many blocks will be consumed by this index
5547                  * insert */
5548                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5549                                            i_gid_read(inode),
5550                                            i_projid_read(inode), 0,
5551                                            oh, osd_dt_obj(dt), NULL,
5552                                            OSD_QID_BLK);
5553                 if (rc)
5554                         RETURN(rc);
5555
5556 #ifdef HAVE_PROJECT_QUOTA
5557                 /* Reserve credits for local agent inode to transfer
5558                  * to 0, quota enforcement is ignored in this case.
5559                  */
5560                 if (idc->oic_remote &&
5561                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5562                     i_projid_read(inode) != 0)
5563                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5564                                                   0, i_projid_read(inode),
5565                                                   0, false, PRJQUOTA);
5566 #endif
5567         }
5568
5569
5570         RETURN(rc);
5571 }
5572
5573 /**
5574  * Index add function for interoperability mode (b11826).
5575  * It will add the directory entry.This entry is needed to
5576  * maintain name->fid mapping.
5577  *
5578  * \param key it is key i.e. file entry to be inserted
5579  * \param rec it is value of given key i.e. fid
5580  *
5581  * \retval   0, on success
5582  * \retval -ve, on error
5583  */
5584 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5585                                const struct dt_rec *rec,
5586                                const struct dt_key *key, struct thandle *th,
5587                                int ignore_quota)
5588 {
5589         struct osd_object       *obj = osd_dt_obj(dt);
5590         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5591         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5592         const struct lu_fid     *fid    = rec1->rec_fid;
5593         const char              *name = (const char *)key;
5594         struct osd_thread_info  *oti   = osd_oti_get(env);
5595         struct inode            *child_inode = NULL;
5596         struct osd_idmap_cache  *idc;
5597         int                     rc;
5598         ENTRY;
5599
5600         if (!dt_object_exists(dt))
5601                 RETURN(-ENOENT);
5602
5603         LASSERT(osd_invariant(obj));
5604         LASSERT(!dt_object_remote(dt));
5605         LASSERT(th != NULL);
5606
5607         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5608
5609         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5610
5611         idc = osd_idc_find(env, osd, fid);
5612         if (unlikely(idc == NULL)) {
5613                 /* this dt_insert() wasn't declared properly, so
5614                  * FID is missing in OI cache. we better do not
5615                  * lookup FID in FLDB/OI and don't risk to deadlock,
5616                  * but in some special cases (lfsck testing, etc)
5617                  * it's much simpler than fixing a caller */
5618                 CERROR("%s: "DFID" wasn't declared for insert\n",
5619                        osd_name(osd), PFID(fid));
5620                 dump_stack();
5621                 idc = osd_idc_find_or_init(env, osd, fid);
5622                 if (IS_ERR(idc))
5623                         RETURN(PTR_ERR(idc));
5624         }
5625
5626         if (idc->oic_remote) {
5627                 /* Insert remote entry */
5628                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5629                         child_inode =
5630                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
5631                 } else {
5632                         child_inode = osd_create_local_agent_inode(env, osd,
5633                                         obj, fid, rec1->rec_type & S_IFMT, th);
5634                         if (IS_ERR(child_inode))
5635                                 RETURN(PTR_ERR(child_inode));
5636                 }
5637         } else {
5638                 /* Insert local entry */
5639                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5640                         /* for a reason OI cache wasn't filled properly */
5641                         CERROR("%s: OIC for "DFID" isn't filled\n",
5642                                osd_name(osd), PFID(fid));
5643                         RETURN(-EINVAL);
5644                 }
5645                 child_inode = oti->oti_inode;
5646                 if (unlikely(child_inode == NULL)) {
5647                         struct ldiskfs_inode_info *lii;
5648                         OBD_ALLOC_PTR(lii);
5649                         if (lii == NULL)
5650                                 RETURN(-ENOMEM);
5651                         child_inode = oti->oti_inode = &lii->vfs_inode;
5652                 }
5653                 child_inode->i_sb = osd_sb(osd);
5654                 child_inode->i_ino = idc->oic_lid.oii_ino;
5655                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5656         }
5657
5658         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5659
5660         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5661                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5662
5663         if (child_inode && child_inode != oti->oti_inode)
5664                 iput(child_inode);
5665         LASSERT(osd_invariant(obj));
5666         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5667         RETURN(rc);
5668 }
5669
5670 /**
5671  *  Initialize osd Iterator for given osd index object.
5672  *
5673  *  \param  dt      osd index object
5674  */
5675
5676 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5677                                      struct dt_object *dt,
5678                                      __u32 unused)
5679 {
5680         struct osd_it_iam      *it;
5681         struct osd_object      *obj = osd_dt_obj(dt);
5682         struct lu_object       *lo  = &dt->do_lu;
5683         struct iam_path_descr  *ipd;
5684         struct iam_container   *bag = &obj->oo_dir->od_container;
5685
5686         if (!dt_object_exists(dt))
5687                 return ERR_PTR(-ENOENT);
5688
5689         OBD_ALLOC_PTR(it);
5690         if (it == NULL)
5691                 return ERR_PTR(-ENOMEM);
5692
5693         ipd = osd_it_ipd_get(env, bag);
5694         if (likely(ipd != NULL)) {
5695                 it->oi_obj = obj;
5696                 it->oi_ipd = ipd;
5697                 lu_object_get(lo);
5698                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5699                 return (struct dt_it *)it;
5700         } else {
5701                 OBD_FREE_PTR(it);
5702                 return ERR_PTR(-ENOMEM);
5703         }
5704 }
5705
5706 /**
5707  * free given Iterator.
5708  */
5709 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5710 {
5711         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5712         struct osd_object       *obj = it->oi_obj;
5713
5714         iam_it_fini(&it->oi_it);
5715         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5716         osd_object_put(env, obj);
5717         OBD_FREE_PTR(it);
5718 }
5719
5720 /**
5721  *  Move Iterator to record specified by \a key
5722  *
5723  *  \param  di      osd iterator
5724  *  \param  key     key for index
5725  *
5726  *  \retval +ve  di points to record with least key not larger than key
5727  *  \retval  0   di points to exact matched key
5728  *  \retval -ve  failure
5729  */
5730
5731 static int osd_it_iam_get(const struct lu_env *env,
5732                           struct dt_it *di, const struct dt_key *key)
5733 {
5734         struct osd_thread_info  *oti = osd_oti_get(env);
5735         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5736
5737         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5738                 /* swab quota uid/gid */
5739                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5740                 key = (struct dt_key *)&oti->oti_quota_id;
5741         }
5742
5743         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5744 }
5745
5746 /**
5747  *  Release Iterator
5748  *
5749  *  \param  di      osd iterator
5750  */
5751 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5752 {
5753         struct osd_it_iam *it = (struct osd_it_iam *)di;
5754
5755         iam_it_put(&it->oi_it);
5756 }
5757
5758 /**
5759  *  Move iterator by one record
5760  *
5761  *  \param  di      osd iterator
5762  *
5763  *  \retval +1   end of container reached
5764  *  \retval  0   success
5765  *  \retval -ve  failure
5766  */
5767
5768 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5769 {
5770         struct osd_it_iam *it = (struct osd_it_iam *)di;
5771
5772         return iam_it_next(&it->oi_it);
5773 }
5774
5775 /**
5776  * Return pointer to the key under iterator.
5777  */
5778
5779 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5780                                  const struct dt_it *di)
5781 {
5782         struct osd_thread_info *oti = osd_oti_get(env);
5783         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5784         struct osd_object      *obj = it->oi_obj;
5785         struct dt_key          *key;
5786
5787         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5788
5789         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5790                 /* swab quota uid/gid */
5791                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5792                 key = (struct dt_key *)&oti->oti_quota_id;
5793         }
5794
5795         return key;
5796 }
5797
5798 /**
5799  * Return size of key under iterator (in bytes)
5800  */
5801
5802 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5803 {
5804         struct osd_it_iam *it = (struct osd_it_iam *)di;
5805
5806         return iam_it_key_size(&it->oi_it);
5807 }
5808
5809 static inline void
5810 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5811 {
5812         /* check if file type is required */
5813         if (ent->lde_attrs & LUDA_TYPE) {
5814                 struct luda_type *lt;
5815                 int align = sizeof(*lt) - 1;
5816
5817                 len = (len + align) & ~align;
5818                 lt = (struct luda_type *)(ent->lde_name + len);
5819                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5820         }
5821
5822         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5823 }
5824
5825 /**
5826  * build lu direct from backend fs dirent.
5827  */
5828
5829 static inline void
5830 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5831                    char *name, __u16 namelen, __u16 type, __u32 attr)
5832 {
5833         ent->lde_attrs = attr | LUDA_FID;
5834         fid_cpu_to_le(&ent->lde_fid, fid);
5835
5836         ent->lde_hash = cpu_to_le64(offset);
5837         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5838
5839         strncpy(ent->lde_name, name, namelen);
5840         ent->lde_name[namelen] = '\0';
5841         ent->lde_namelen = cpu_to_le16(namelen);
5842
5843         /* append lustre attributes */
5844         osd_it_append_attrs(ent, namelen, type);
5845 }
5846
5847 /**
5848  * Return pointer to the record under iterator.
5849  */
5850 static int osd_it_iam_rec(const struct lu_env *env,
5851                           const struct dt_it *di,
5852                           struct dt_rec *dtrec, __u32 attr)
5853 {
5854         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5855         struct osd_thread_info *info = osd_oti_get(env);
5856         ENTRY;
5857
5858         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5859                 const struct osd_fid_pack *rec;
5860                 struct lu_fid             *fid = &info->oti_fid;
5861                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5862                 char                      *name;
5863                 int                        namelen;
5864                 __u64                      hash;
5865                 int                        rc;
5866
5867                 name = (char *)iam_it_key_get(&it->oi_it);
5868                 if (IS_ERR(name))
5869                         RETURN(PTR_ERR(name));
5870
5871                 namelen = iam_it_key_size(&it->oi_it);
5872
5873                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5874                 if (IS_ERR(rec))
5875                         RETURN(PTR_ERR(rec));
5876
5877                 rc = osd_fid_unpack(fid, rec);
5878                 if (rc)
5879                         RETURN(rc);
5880
5881                 hash = iam_it_store(&it->oi_it);
5882
5883                 /* IAM does not store object type in IAM index (dir) */
5884                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5885                                    0, LUDA_FID);
5886         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5887                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5888                            (struct iam_rec *)dtrec);
5889                 osd_quota_unpack(it->oi_obj, dtrec);
5890         } else {
5891                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5892                            (struct iam_rec *)dtrec);
5893         }
5894
5895         RETURN(0);
5896 }
5897
5898 /**
5899  * Returns cookie for current Iterator position.
5900  */
5901 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5902 {
5903         struct osd_it_iam *it = (struct osd_it_iam *)di;
5904
5905         return iam_it_store(&it->oi_it);
5906 }
5907
5908 /**
5909  * Restore iterator from cookie.
5910  *
5911  * \param  di      osd iterator
5912  * \param  hash    Iterator location cookie
5913  *
5914  * \retval +ve  di points to record with least key not larger than key.
5915  * \retval  0   di points to exact matched key
5916  * \retval -ve  failure
5917  */
5918
5919 static int osd_it_iam_load(const struct lu_env *env,
5920                            const struct dt_it *di, __u64 hash)
5921 {
5922         struct osd_it_iam *it = (struct osd_it_iam *)di;
5923
5924         return iam_it_load(&it->oi_it, hash);
5925 }
5926
5927 static const struct dt_index_operations osd_index_iam_ops = {
5928         .dio_lookup         = osd_index_iam_lookup,
5929         .dio_declare_insert = osd_index_declare_iam_insert,
5930         .dio_insert         = osd_index_iam_insert,
5931         .dio_declare_delete = osd_index_declare_iam_delete,
5932         .dio_delete         = osd_index_iam_delete,
5933         .dio_it     = {
5934                 .init     = osd_it_iam_init,
5935                 .fini     = osd_it_iam_fini,
5936                 .get      = osd_it_iam_get,
5937                 .put      = osd_it_iam_put,
5938                 .next     = osd_it_iam_next,
5939                 .key      = osd_it_iam_key,
5940                 .key_size = osd_it_iam_key_size,
5941                 .rec      = osd_it_iam_rec,
5942                 .store    = osd_it_iam_store,
5943                 .load     = osd_it_iam_load
5944         }
5945 };
5946
5947
5948 /**
5949  * Creates or initializes iterator context.
5950  *
5951  * \retval struct osd_it_ea, iterator structure on success
5952  *
5953  */
5954 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5955                                     struct dt_object *dt,
5956                                     __u32 attr)
5957 {
5958         struct osd_object       *obj  = osd_dt_obj(dt);
5959         struct osd_thread_info  *info = osd_oti_get(env);
5960         struct osd_it_ea        *oie;
5961         struct file             *file;
5962         struct lu_object        *lo   = &dt->do_lu;
5963         struct dentry           *obj_dentry;
5964         ENTRY;
5965
5966         if (!dt_object_exists(dt) || obj->oo_destroyed)
5967                 RETURN(ERR_PTR(-ENOENT));
5968
5969         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5970         if (oie == NULL)
5971                 RETURN(ERR_PTR(-ENOMEM));
5972         obj_dentry = &oie->oie_dentry;
5973
5974         obj_dentry->d_inode = obj->oo_inode;
5975         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5976         obj_dentry->d_name.hash = 0;
5977
5978         oie->oie_rd_dirent       = 0;
5979         oie->oie_it_dirent       = 0;
5980         oie->oie_dirent          = NULL;
5981         if (unlikely(!info->oti_it_ea_buf_used)) {
5982                 oie->oie_buf = info->oti_it_ea_buf;
5983                 info->oti_it_ea_buf_used = 1;
5984         } else {
5985                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5986                 if (oie->oie_buf == NULL)
5987                         RETURN(ERR_PTR(-ENOMEM));
5988         }
5989         oie->oie_obj             = obj;
5990
5991         file = &oie->oie_file;
5992
5993         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5994         if (attr & LUDA_64BITHASH)
5995                 file->f_mode    = FMODE_64BITHASH;
5996         else
5997                 file->f_mode    = FMODE_32BITHASH;
5998         file->f_path.dentry     = obj_dentry;
5999         file->f_mapping         = obj->oo_inode->i_mapping;
6000         file->f_op              = obj->oo_inode->i_fop;
6001         set_file_inode(file, obj->oo_inode);
6002
6003         lu_object_get(lo);
6004         RETURN((struct dt_it *) oie);
6005 }
6006
6007 /**
6008  * Destroy or finishes iterator context.
6009  *
6010  * \param di iterator structure to be destroyed
6011  */
6012 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6013 {
6014         struct osd_thread_info  *info = osd_oti_get(env);
6015         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
6016         struct osd_object       *obj    = oie->oie_obj;
6017         struct inode            *inode  = obj->oo_inode;
6018
6019         ENTRY;
6020         oie->oie_file.f_op->release(inode, &oie->oie_file);
6021         osd_object_put(env, obj);
6022         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6023                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6024         else
6025                 info->oti_it_ea_buf_used = 0;
6026         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6027         EXIT;
6028 }
6029
6030 /**
6031  * It position the iterator at given key, so that next lookup continues from
6032  * that key Or it is similar to dio_it->load() but based on a key,
6033  * rather than file position.
6034  *
6035  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6036  * to the beginning.
6037  *
6038  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6039  */
6040 static int osd_it_ea_get(const struct lu_env *env,
6041                          struct dt_it *di, const struct dt_key *key)
6042 {
6043         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
6044
6045         ENTRY;
6046         LASSERT(((const char *)key)[0] == '\0');
6047         it->oie_file.f_pos      = 0;
6048         it->oie_rd_dirent       = 0;
6049         it->oie_it_dirent       = 0;
6050         it->oie_dirent          = NULL;
6051
6052         RETURN(+1);
6053 }
6054
6055 /**
6056  * Does nothing
6057  */
6058 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6059 {
6060 }
6061
6062 struct osd_filldir_cbs {
6063 #ifdef HAVE_DIR_CONTEXT
6064         struct dir_context ctx;
6065 #endif
6066         struct osd_it_ea  *it;
6067 };
6068 /**
6069  * It is called internally by ->readdir(). It fills the
6070  * iterator's in-memory data structure with required
6071  * information i.e. name, namelen, rec_size etc.
6072  *
6073  * \param buf in which information to be filled in.
6074  * \param name name of the file in given dir
6075  *
6076  * \retval 0 on success
6077  * \retval 1 on buffer full
6078  */
6079 #ifdef HAVE_FILLDIR_USE_CTX
6080 static int osd_ldiskfs_filldir(struct dir_context *buf,
6081 #else
6082 static int osd_ldiskfs_filldir(void *buf,
6083 #endif
6084                                const char *name, int namelen,
6085                                loff_t offset, __u64 ino, unsigned d_type)
6086 {
6087         struct osd_it_ea        *it   =
6088                 ((struct osd_filldir_cbs *)buf)->it;
6089         struct osd_object       *obj  = it->oie_obj;
6090         struct osd_it_ea_dirent *ent  = it->oie_dirent;
6091         struct lu_fid           *fid  = &ent->oied_fid;
6092         struct osd_fid_pack     *rec;
6093         ENTRY;
6094
6095         /* this should never happen */
6096         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6097                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6098                 RETURN(-EIO);
6099         }
6100
6101         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
6102             OSD_IT_EA_BUFSIZE)
6103                 RETURN(1);
6104
6105         /* "." is just the object itself. */
6106         if (namelen == 1 && name[0] == '.') {
6107                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6108         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6109                 rec = (struct osd_fid_pack*) (name + namelen + 1);
6110                 if (osd_fid_unpack(fid, rec) != 0)
6111                         fid_zero(fid);
6112         } else {
6113                 fid_zero(fid);
6114         }
6115         d_type &= ~LDISKFS_DIRENT_LUFID;
6116
6117         /* NOT export local root. */
6118         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6119                 ino = obj->oo_inode->i_ino;
6120                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6121         }
6122
6123         ent->oied_ino     = ino;
6124         ent->oied_off     = offset;
6125         ent->oied_namelen = namelen;
6126         ent->oied_type    = d_type;
6127
6128         memcpy(ent->oied_name, name, namelen);
6129
6130         it->oie_rd_dirent++;
6131         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6132         RETURN(0);
6133 }
6134
6135 /**
6136  * Calls ->readdir() to load a directory entry at a time
6137  * and stored it in iterator's in-memory data structure.
6138  *
6139  * \param di iterator's in memory structure
6140  *
6141  * \retval   0 on success
6142  * \retval -ve on error
6143  * \retval +1 reach the end of entry
6144  */
6145 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6146                                const struct dt_it *di)
6147 {
6148         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
6149         struct osd_object  *obj   = it->oie_obj;
6150         struct inode       *inode = obj->oo_inode;
6151         struct htree_lock  *hlock = NULL;
6152         struct file        *filp  = &it->oie_file;
6153         int                 rc = 0;
6154         struct osd_filldir_cbs buf = {
6155 #ifdef HAVE_DIR_CONTEXT
6156                 .ctx.actor = osd_ldiskfs_filldir,
6157 #endif
6158                 .it = it
6159         };
6160
6161         ENTRY;
6162         it->oie_dirent = it->oie_buf;
6163         it->oie_rd_dirent = 0;
6164
6165         if (obj->oo_hl_head != NULL) {
6166                 hlock = osd_oti_get(env)->oti_hlock;
6167                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6168                                    inode, LDISKFS_HLOCK_READDIR);
6169         } else {
6170                 down_read(&obj->oo_ext_idx_sem);
6171         }
6172
6173 #ifdef HAVE_DIR_CONTEXT
6174         buf.ctx.pos = filp->f_pos;
6175         rc = inode->i_fop->iterate(filp, &buf.ctx);
6176         filp->f_pos = buf.ctx.pos;
6177 #else
6178         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6179 #endif
6180
6181         if (hlock != NULL)
6182                 ldiskfs_htree_unlock(hlock);
6183         else
6184                 up_read(&obj->oo_ext_idx_sem);
6185
6186         if (it->oie_rd_dirent == 0) {
6187                 /*If it does not get any dirent, it means it has been reached
6188                  *to the end of the dir */
6189                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6190                 if (rc == 0)
6191                         rc = 1;
6192         } else {
6193                 it->oie_dirent = it->oie_buf;
6194                 it->oie_it_dirent = 1;
6195         }
6196
6197         RETURN(rc);
6198 }
6199
6200 /**
6201  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6202  * to load a directory entry at a time and stored it in
6203  * iterator's in-memory data structure.
6204  *
6205  * \param di iterator's in memory structure
6206  *
6207  * \retval +ve iterator reached to end
6208  * \retval   0 iterator not reached to end
6209  * \retval -ve on error
6210  */
6211 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6212 {
6213         struct osd_it_ea *it = (struct osd_it_ea *)di;
6214         int rc;
6215
6216         ENTRY;
6217
6218         if (it->oie_it_dirent < it->oie_rd_dirent) {
6219                 it->oie_dirent =
6220                         (void *) it->oie_dirent +
6221                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6222                                        it->oie_dirent->oied_namelen);
6223                 it->oie_it_dirent++;
6224                 RETURN(0);
6225         } else {
6226                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6227                         rc = +1;
6228                 else
6229                         rc = osd_ldiskfs_it_fill(env, di);
6230         }
6231
6232         RETURN(rc);
6233 }
6234
6235 /**
6236  * Returns the key at current position from iterator's in memory structure.
6237  *
6238  * \param di iterator's in memory structure
6239  *
6240  * \retval key i.e. struct dt_key on success
6241  */
6242 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6243                                     const struct dt_it *di)
6244 {
6245         struct osd_it_ea *it = (struct osd_it_ea *)di;
6246
6247         return (struct dt_key *)it->oie_dirent->oied_name;
6248 }
6249
6250 /**
6251  * Returns the key's size at current position from iterator's in memory structure.
6252  *
6253  * \param di iterator's in memory structure
6254  *
6255  * \retval key_size i.e. struct dt_key on success
6256  */
6257 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6258 {
6259         struct osd_it_ea *it = (struct osd_it_ea *)di;
6260
6261         return it->oie_dirent->oied_namelen;
6262 }
6263
6264 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6265 {
6266         if (LDISKFS_DIR_REC_LEN(de) >=
6267             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6268                 return true;
6269
6270         return false;
6271 }
6272
6273 static inline bool
6274 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6275                      unsigned blocksize, bool dotdot)
6276 {
6277         if (dotdot)
6278                 return osd_dotdot_has_space(de);
6279
6280         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6281             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6282                 return true;
6283
6284         return false;
6285 }
6286
6287 static int
6288 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6289                     handle_t *jh, struct dentry *dentry,
6290                     const struct lu_fid *fid, struct buffer_head *bh,
6291                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6292                     bool dotdot)
6293 {
6294         struct inode                *dir        = dentry->d_parent->d_inode;
6295         struct inode                *inode      = dentry->d_inode;
6296         struct osd_fid_pack         *rec;
6297         struct ldiskfs_dentry_param *ldp;
6298         int                          namelen    = dentry->d_name.len;
6299         int                          rc;
6300         struct osd_thread_info     *info        = osd_oti_get(env);
6301         ENTRY;
6302
6303         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6304                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6305                 RETURN(0);
6306
6307         /* There is enough space to hold the FID-in-dirent. */
6308         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6309                 rc = ldiskfs_journal_get_write_access(jh, bh);
6310                 if (rc != 0)
6311                         RETURN(rc);
6312
6313                 de->name[namelen] = 0;
6314                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6315                 rec->fp_len = sizeof(struct lu_fid) + 1;
6316                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6317                 de->file_type |= LDISKFS_DIRENT_LUFID;
6318                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6319
6320                 RETURN(rc);
6321         }
6322
6323         LASSERT(!dotdot);
6324
6325         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6326         if (rc != 0)
6327                 RETURN(rc);
6328
6329         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6330         osd_get_ldiskfs_dirent_param(ldp, fid);
6331         dentry->d_fsdata = (void *)ldp;
6332         ll_vfs_dq_init(dir);
6333         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6334         /* It is too bad, we cannot reinsert the name entry back.
6335          * That means we lose it! */
6336         if (rc != 0)
6337                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6338                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6339                        osd_ino2name(inode),
6340                        dir->i_ino, dir->i_generation, namelen,
6341                        dentry->d_name.name, PFID(fid), rc);
6342
6343         RETURN(rc);
6344 }
6345
6346 static int
6347 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6348                         struct osd_it_ea *it, struct lu_fid *fid,
6349                         struct osd_inode_id *id, __u32 *attr)
6350 {
6351         struct osd_thread_info     *info        = osd_oti_get(env);
6352         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6353         struct osd_device          *dev         = osd_obj2dev(obj);
6354         struct super_block         *sb          = osd_sb(dev);
6355         const char                 *devname     = osd_name(dev);
6356         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6357         struct inode               *dir         = obj->oo_inode;
6358         struct htree_lock          *hlock       = NULL;
6359         struct buffer_head         *bh          = NULL;
6360         handle_t                   *jh          = NULL;
6361         struct ldiskfs_dir_entry_2 *de;
6362         struct dentry              *dentry;
6363         struct inode               *inode;
6364         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6365         int                         credits;
6366         int                         rc;
6367         bool                        dotdot      = false;
6368         bool                        dirty       = false;
6369         ENTRY;
6370
6371         if (ent->oied_name[0] == '.') {
6372                 if (ent->oied_namelen == 1)
6373                         RETURN(0);
6374
6375                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6376                         dotdot = true;
6377         }
6378
6379         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6380         inode = osd_iget(info, dev, id);
6381         if (IS_ERR(inode)) {
6382                 rc = PTR_ERR(inode);
6383                 if (rc == -ENOENT || rc == -ESTALE) {
6384                         /* Maybe dangling name entry, or
6385                          * corrupted directory entry. */
6386                         *attr |= LUDA_UNKNOWN;
6387                         rc = 0;
6388                 } else {
6389                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6390                                "check_repair, dir = %lu/%u, name = %.*s, "
6391                                "ino = %llu, rc = %d\n",
6392                                devname, dir->i_ino, dir->i_generation,
6393                                ent->oied_namelen, ent->oied_name,
6394                                ent->oied_ino, rc);
6395                 }
6396
6397                 RETURN(rc);
6398         }
6399
6400         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6401                                            ent->oied_namelen);
6402         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6403         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6404                 lma = NULL;
6405         else if (rc != 0)
6406                 GOTO(out, rc);
6407
6408         /* We need to ensure that the name entry is still valid.
6409          * Because it may be removed or renamed by other already.
6410          *
6411          * The unlink or rename operation will start journal before PDO lock,
6412          * so to avoid deadlock, here we need to start journal handle before
6413          * related PDO lock also. But because we do not know whether there
6414          * will be something to be repaired before PDO lock, we just start
6415          * journal without conditions.
6416          *
6417          * We may need to remove the name entry firstly, then insert back.
6418          * One credit is for user quota file update.
6419          * One credit is for group quota file update.
6420          * Two credits are for dirty inode. */
6421         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6422                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6423
6424         if (dev->od_dirent_journal != 0) {
6425
6426 again:
6427                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6428                 if (IS_ERR(jh)) {
6429                         rc = PTR_ERR(jh);
6430                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6431                                "check_repair, dir = %lu/%u, credits = %d, "
6432                                "name = %.*s, ino = %llu: rc = %d\n",
6433                                devname, dir->i_ino, dir->i_generation, credits,
6434                                ent->oied_namelen, ent->oied_name,
6435                                ent->oied_ino, rc);
6436
6437                         GOTO(out_inode, rc);
6438                 }
6439
6440                 if (obj->oo_hl_head != NULL) {
6441                         hlock = osd_oti_get(env)->oti_hlock;
6442                         /* "0" means exclusive lock for the whole directory.
6443                          * We need to prevent others access such name entry
6444                          * during the delete + insert. Neither HLOCK_ADD nor
6445                          * HLOCK_DEL cannot guarantee the atomicity. */
6446                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6447                 } else {
6448                         down_write(&obj->oo_ext_idx_sem);
6449                 }
6450         } else {
6451                 if (obj->oo_hl_head != NULL) {
6452                         hlock = osd_oti_get(env)->oti_hlock;
6453                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6454                                            LDISKFS_HLOCK_LOOKUP);
6455                 } else {
6456                         down_read(&obj->oo_ext_idx_sem);
6457                 }
6458         }
6459
6460         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6461         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6462                 *attr |= LUDA_IGNORE;
6463
6464                 GOTO(out, rc = 0);
6465         }
6466
6467         /* For dotdot entry, if there is not enough space to hold the
6468          * FID-in-dirent, just keep them there. It only happens when the
6469          * device upgraded from 1.8 or restored from MDT file-level backup.
6470          * For the whole directory, only dotdot entry have no FID-in-dirent
6471          * and needs to get FID from LMA when readdir, it will not affect the
6472          * performance much. */
6473         if (dotdot && !osd_dotdot_has_space(de)) {
6474                 *attr |= LUDA_UNKNOWN;
6475
6476                 GOTO(out, rc = 0);
6477         }
6478
6479         if (lma != NULL) {
6480                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6481                         GOTO(out, rc = 0);
6482
6483                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6484                         struct lu_fid *tfid = &lma->lma_self_fid;
6485
6486                         if (likely(dotdot &&
6487                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6488                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6489                                 /* It must be REMOTE_PARENT_DIR and as the
6490                                  * 'dotdot' entry of remote directory */
6491                                 *attr |= LUDA_IGNORE;
6492                         } else {
6493                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6494                                        "parent directory, but got %.*s under "
6495                                        "dir = %lu/%u with the FID "DFID"\n",
6496                                        devname, ent->oied_namelen,
6497                                        ent->oied_name, dir->i_ino,
6498                                        dir->i_generation, PFID(tfid));
6499
6500                                 *attr |= LUDA_UNKNOWN;
6501                         }
6502
6503                         GOTO(out, rc = 0);
6504                 }
6505         }
6506
6507         if (!fid_is_zero(fid)) {
6508                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6509                                               ent->oied_namelen);
6510                 if (rc == -ENOENT ||
6511                     (rc == -ENODATA &&
6512                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6513                         /* linkEA does not recognize the dirent entry,
6514                          * it may because the dirent entry corruption
6515                          * and points to other's inode. */
6516                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6517                                "recognize the dirent, dir = %lu/%u, "
6518                                " name = %.*s, ino = %llu, "
6519                                DFID": rc = %d\n", devname, dir->i_ino,
6520                                dir->i_generation, ent->oied_namelen,
6521                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6522                         *attr |= LUDA_UNKNOWN;
6523
6524                         GOTO(out, rc = 0);
6525                 }
6526
6527                 if (rc && rc != -ENODATA) {
6528                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6529                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6530                                DFID": rc = %d\n", devname, dir->i_ino,
6531                                dir->i_generation, ent->oied_namelen,
6532                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6533                         *attr |= LUDA_UNKNOWN;
6534
6535                         GOTO(out, rc = 0);
6536                 }
6537         }
6538
6539         if (lma != NULL) {
6540                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6541                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6542                 if (*attr & LUDA_VERIFY_DRYRUN) {
6543                         *fid = lma->lma_self_fid;
6544                         *attr |= LUDA_REPAIR;
6545
6546                         GOTO(out, rc = 0);
6547                 }
6548
6549                 if (jh == NULL) {
6550                         brelse(bh);
6551                         dev->od_dirent_journal = 1;
6552                         if (hlock != NULL) {
6553                                 ldiskfs_htree_unlock(hlock);
6554                                 hlock = NULL;
6555                         } else {
6556                                 up_read(&obj->oo_ext_idx_sem);
6557                         }
6558
6559                         goto again;
6560                 }
6561
6562                 *fid = lma->lma_self_fid;
6563                 dirty = true;
6564                 /* Update or append the FID-in-dirent. */
6565                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6566                                          bh, de, hlock, dotdot);
6567                 if (rc == 0)
6568                         *attr |= LUDA_REPAIR;
6569                 else
6570                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6571                                "the dirent, dir = %lu/%u, name = %.*s, "
6572                                "ino = %llu, "DFID": rc = %d\n",
6573                                devname, dir->i_ino, dir->i_generation,
6574                                ent->oied_namelen, ent->oied_name,
6575                                ent->oied_ino, PFID(fid), rc);
6576         } else {
6577                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6578                 if (*attr & LUDA_VERIFY_DRYRUN) {
6579                         if (fid_is_sane(fid)) {
6580                                 *attr |= LUDA_REPAIR;
6581                         } else if (dev->od_index == 0) {
6582                                 lu_igif_build(fid, inode->i_ino,
6583                                               inode->i_generation);
6584                                 *attr |= LUDA_UPGRADE;
6585                         }
6586
6587                         GOTO(out, rc = 0);
6588                 }
6589
6590                 if (jh == NULL) {
6591                         brelse(bh);
6592                         dev->od_dirent_journal = 1;
6593                         if (hlock != NULL) {
6594                                 ldiskfs_htree_unlock(hlock);
6595                                 hlock = NULL;
6596                         } else {
6597                                 up_read(&obj->oo_ext_idx_sem);
6598                         }
6599
6600                         goto again;
6601                 }
6602
6603                 dirty = true;
6604                 if (unlikely(fid_is_sane(fid))) {
6605                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6606                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6607                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6608                         if (rc == 0)
6609                                 *attr |= LUDA_REPAIR;
6610                         else
6611                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6612                                        "update dirent, dir = %lu/%u, "
6613                                        "name = %.*s, ino = %llu, "
6614                                        DFID": rc = %d\n",
6615                                        devname, dir->i_ino, dir->i_generation,
6616                                        ent->oied_namelen, ent->oied_name,
6617                                        ent->oied_ino, PFID(fid), rc);
6618                 } else if (dev->od_index == 0) {
6619                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6620                         /* It is probably IGIF object. Only aappend the
6621                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6622                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6623                                                  bh, de, hlock, dotdot);
6624                         if (rc == 0)
6625                                 *attr |= LUDA_UPGRADE;
6626                         else
6627                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6628                                        "after the dirent, dir = %lu/%u, "
6629                                        "name = %.*s, ino = %llu, "
6630                                        DFID": rc = %d\n",
6631                                        devname, dir->i_ino, dir->i_generation,
6632                                        ent->oied_namelen, ent->oied_name,
6633                                        ent->oied_ino, PFID(fid), rc);
6634                 }
6635         }
6636
6637         GOTO(out, rc);
6638
6639 out:
6640         if (!IS_ERR(bh))
6641                 brelse(bh);
6642         if (hlock != NULL) {
6643                 ldiskfs_htree_unlock(hlock);
6644         } else {
6645                 if (dev->od_dirent_journal != 0)
6646                         up_write(&obj->oo_ext_idx_sem);
6647                 else
6648                         up_read(&obj->oo_ext_idx_sem);
6649         }
6650
6651         if (jh != NULL)
6652                 ldiskfs_journal_stop(jh);
6653
6654 out_inode:
6655         iput(inode);
6656         if (rc >= 0 && !dirty)
6657                 dev->od_dirent_journal = 0;
6658
6659         return rc;
6660 }
6661
6662 /**
6663  * Returns the value at current position from iterator's in memory structure.
6664  *
6665  * \param di struct osd_it_ea, iterator's in memory structure
6666  * \param attr attr requested for dirent.
6667  * \param lde lustre dirent
6668  *
6669  * \retval   0 no error and \param lde has correct lustre dirent.
6670  * \retval -ve on error
6671  */
6672 static inline int osd_it_ea_rec(const struct lu_env *env,
6673                                 const struct dt_it *di,
6674                                 struct dt_rec *dtrec, __u32 attr)
6675 {
6676         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6677         struct osd_object      *obj   = it->oie_obj;
6678         struct osd_device      *dev   = osd_obj2dev(obj);
6679         struct osd_thread_info *oti   = osd_oti_get(env);
6680         struct osd_inode_id    *id    = &oti->oti_id;
6681         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6682         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6683         __u32                   ino   = it->oie_dirent->oied_ino;
6684         int                     rc    = 0;
6685         ENTRY;
6686
6687         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6688
6689         if (attr & LUDA_VERIFY) {
6690                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6691                         attr |= LUDA_IGNORE;
6692                         /* If the parent is on remote MDT, and there
6693                          * is no FID-in-dirent, then we have to get
6694                          * the parent FID from the linkEA.  */
6695                         if (!fid_is_sane(fid) &&
6696                             it->oie_dirent->oied_namelen == 2 &&
6697                             it->oie_dirent->oied_name[0] == '.' &&
6698                             it->oie_dirent->oied_name[1] == '.')
6699                                 osd_get_pfid_from_linkea(env, obj, fid);
6700                 } else {
6701                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6702                                                      &attr);
6703                 }
6704
6705                 if (!fid_is_sane(fid))
6706                         attr |= LUDA_UNKNOWN;
6707         } else {
6708                 attr &= ~LU_DIRENT_ATTRS_MASK;
6709                 if (!fid_is_sane(fid)) {
6710                         bool is_dotdot = false;
6711                         if (it->oie_dirent->oied_namelen == 2 &&
6712                             it->oie_dirent->oied_name[0] == '.' &&
6713                             it->oie_dirent->oied_name[1] == '.')
6714                                 is_dotdot = true;
6715                         /* If the parent is on remote MDT, and there
6716                          * is no FID-in-dirent, then we have to get
6717                          * the parent FID from the linkEA.  */
6718                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6719                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6720                         } else {
6721                                 if (is_dotdot == false &&
6722                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6723                                         RETURN(-ENOENT);
6724
6725                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6726                         }
6727                 } else {
6728                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6729                 }
6730         }
6731
6732         /* Pack the entry anyway, at least the offset is right. */
6733         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6734                            it->oie_dirent->oied_name,
6735                            it->oie_dirent->oied_namelen,
6736                            it->oie_dirent->oied_type, attr);
6737
6738         if (rc < 0)
6739                 RETURN(rc);
6740
6741         if (osd_remote_fid(env, dev, fid))
6742                 RETURN(0);
6743
6744         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6745                 osd_add_oi_cache(oti, dev, id, fid);
6746
6747         RETURN(rc > 0 ? 0 : rc);
6748 }
6749
6750 /**
6751  * Returns the record size size at current position.
6752  *
6753  * This function will return record(lu_dirent) size in bytes.
6754  *
6755  * \param[in] env       execution environment
6756  * \param[in] di        iterator's in memory structure
6757  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6758  *                      calculate the lu_dirent size.
6759  *
6760  * \retval      record size(in bytes & in memory) of the current lu_dirent
6761  *              entry.
6762  */
6763 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6764                               __u32 attr)
6765 {
6766         struct osd_it_ea *it = (struct osd_it_ea *)di;
6767
6768         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6769 }
6770
6771 /**
6772  * Returns a cookie for current position of the iterator head, so that
6773  * user can use this cookie to load/start the iterator next time.
6774  *
6775  * \param di iterator's in memory structure
6776  *
6777  * \retval cookie for current position, on success
6778  */
6779 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6780 {
6781         struct osd_it_ea *it = (struct osd_it_ea *)di;
6782
6783         return it->oie_dirent->oied_off;
6784 }
6785
6786 /**
6787  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6788  * to load a directory entry at a time and stored it i inn,
6789  * in iterator's in-memory data structure.
6790  *
6791  * \param di struct osd_it_ea, iterator's in memory structure
6792  *
6793  * \retval +ve on success
6794  * \retval -ve on error
6795  */
6796 static int osd_it_ea_load(const struct lu_env *env,
6797                           const struct dt_it *di, __u64 hash)
6798 {
6799         struct osd_it_ea *it = (struct osd_it_ea *)di;
6800         int rc;
6801
6802         ENTRY;
6803         it->oie_file.f_pos = hash;
6804
6805         rc =  osd_ldiskfs_it_fill(env, di);
6806         if (rc > 0)
6807                 rc = -ENODATA;
6808
6809         if (rc == 0)
6810                 rc = +1;
6811
6812         RETURN(rc);
6813 }
6814
6815 /**
6816  * Index lookup function for interoperability mode (b11826).
6817  *
6818  * \param key,  key i.e. file name to be searched
6819  *
6820  * \retval +ve, on success
6821  * \retval -ve, on error
6822  */
6823 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6824                                struct dt_rec *rec, const struct dt_key *key)
6825 {
6826         struct osd_object *obj = osd_dt_obj(dt);
6827         int rc = 0;
6828
6829         ENTRY;
6830
6831         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6832         LINVRNT(osd_invariant(obj));
6833
6834         rc = osd_ea_lookup_rec(env, obj, rec, key);
6835         if (rc == 0)
6836                 rc = +1;
6837         RETURN(rc);
6838 }
6839
6840 /**
6841  * Index and Iterator operations for interoperability
6842  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6843  */
6844 static const struct dt_index_operations osd_index_ea_ops = {
6845         .dio_lookup         = osd_index_ea_lookup,
6846         .dio_declare_insert = osd_index_declare_ea_insert,
6847         .dio_insert         = osd_index_ea_insert,
6848         .dio_declare_delete = osd_index_declare_ea_delete,
6849         .dio_delete         = osd_index_ea_delete,
6850         .dio_it     = {
6851                 .init     = osd_it_ea_init,
6852                 .fini     = osd_it_ea_fini,
6853                 .get      = osd_it_ea_get,
6854                 .put      = osd_it_ea_put,
6855                 .next     = osd_it_ea_next,
6856                 .key      = osd_it_ea_key,
6857                 .key_size = osd_it_ea_key_size,
6858                 .rec      = osd_it_ea_rec,
6859                 .rec_size = osd_it_ea_rec_size,
6860                 .store    = osd_it_ea_store,
6861                 .load     = osd_it_ea_load
6862         }
6863 };
6864
6865 static void *osd_key_init(const struct lu_context *ctx,
6866                           struct lu_context_key *key)
6867 {
6868         struct osd_thread_info *info;
6869
6870         OBD_ALLOC_PTR(info);
6871         if (info == NULL)
6872                 return ERR_PTR(-ENOMEM);
6873
6874         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6875         if (info->oti_it_ea_buf == NULL)
6876                 goto out_free_info;
6877
6878         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6879
6880         info->oti_hlock = ldiskfs_htree_lock_alloc();
6881         if (info->oti_hlock == NULL)
6882                 goto out_free_ea;
6883
6884         return info;
6885
6886  out_free_ea:
6887         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6888  out_free_info:
6889         OBD_FREE_PTR(info);
6890         return ERR_PTR(-ENOMEM);
6891 }
6892
6893 static void osd_key_fini(const struct lu_context *ctx,
6894                          struct lu_context_key *key, void* data)
6895 {
6896         struct osd_thread_info *info = data;
6897         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6898         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6899
6900         if (info->oti_inode != NULL)
6901                 OBD_FREE_PTR(lli);
6902         if (info->oti_hlock != NULL)
6903                 ldiskfs_htree_lock_free(info->oti_hlock);
6904         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6905         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6906         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6907         lu_buf_free(&info->oti_big_buf);
6908         if (idc != NULL) {
6909                 LASSERT(info->oti_ins_cache_size > 0);
6910                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6911                 info->oti_ins_cache = NULL;
6912                 info->oti_ins_cache_size = 0;
6913         }
6914         OBD_FREE_PTR(info);
6915 }
6916
6917 static void osd_key_exit(const struct lu_context *ctx,
6918                          struct lu_context_key *key, void *data)
6919 {
6920         struct osd_thread_info *info = data;
6921
6922         LASSERT(info->oti_r_locks == 0);
6923         LASSERT(info->oti_w_locks == 0);
6924         LASSERT(info->oti_txns    == 0);
6925 }
6926
6927 /* type constructor/destructor: osd_type_init, osd_type_fini */
6928 LU_TYPE_INIT_FINI(osd, &osd_key);
6929
6930 struct lu_context_key osd_key = {
6931         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6932         .lct_init = osd_key_init,
6933         .lct_fini = osd_key_fini,
6934         .lct_exit = osd_key_exit
6935 };
6936
6937
6938 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6939                            const char *name, struct lu_device *next)
6940 {
6941         struct osd_device *osd = osd_dev(d);
6942
6943         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6944             >= sizeof(osd->od_svname))
6945                 return -E2BIG;
6946         return osd_procfs_init(osd, name);
6947 }
6948
6949 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6950 {
6951         struct seq_server_site  *ss = osd_seq_site(osd);
6952         int                     rc;
6953         ENTRY;
6954
6955         if (osd->od_is_ost || osd->od_cl_seq != NULL)
6956                 RETURN(0);
6957
6958         if (unlikely(ss == NULL))
6959                 RETURN(-ENODEV);
6960
6961         OBD_ALLOC_PTR(osd->od_cl_seq);
6962         if (osd->od_cl_seq == NULL)
6963                 RETURN(-ENOMEM);
6964
6965         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6966                              osd->od_svname, ss->ss_server_seq);
6967         if (rc != 0) {
6968                 OBD_FREE_PTR(osd->od_cl_seq);
6969                 osd->od_cl_seq = NULL;
6970                 RETURN(rc);
6971         }
6972
6973         if (ss->ss_node_id == 0) {
6974                 /* If the OSD on the sequence controller(MDT0), then allocate
6975                  * sequence here, otherwise allocate sequence after connected
6976                  * to MDT0 (see mdt_register_lwp_callback()). */
6977                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6978                                    &osd->od_cl_seq->lcs_space, env);
6979         }
6980
6981         RETURN(rc);
6982 }
6983
6984 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6985 {
6986         if (osd->od_cl_seq == NULL)
6987                 return;
6988
6989         seq_client_fini(osd->od_cl_seq);
6990         OBD_FREE_PTR(osd->od_cl_seq);
6991         osd->od_cl_seq = NULL;
6992 }
6993
6994 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6995 {
6996         ENTRY;
6997
6998         /* shutdown quota slave instance associated with the device */
6999         if (o->od_quota_slave != NULL) {
7000                 qsd_fini(env, o->od_quota_slave);
7001                 o->od_quota_slave = NULL;
7002         }
7003
7004         osd_fid_fini(env, o);
7005         osd_scrub_cleanup(env, o);
7006
7007         RETURN(0);
7008 }
7009
7010 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7011 {
7012         ENTRY;
7013
7014         if (o->od_mnt != NULL) {
7015                 shrink_dcache_sb(osd_sb(o));
7016                 osd_sync(env, &o->od_dt_dev);
7017
7018                 mntput(o->od_mnt);
7019                 o->od_mnt = NULL;
7020         }
7021
7022         EXIT;
7023 }
7024
7025 static int osd_mount(const struct lu_env *env,
7026                      struct osd_device *o, struct lustre_cfg *cfg)
7027 {
7028         const char              *name  = lustre_cfg_string(cfg, 0);
7029         const char              *dev  = lustre_cfg_string(cfg, 1);
7030         const char              *opts;
7031         unsigned long            page, s_flags, lmd_flags = 0;
7032         struct page             *__page;
7033         struct file_system_type *type;
7034         char                    *options = NULL;
7035         char                    *str;
7036         struct osd_thread_info  *info = osd_oti_get(env);
7037         struct lu_fid           *fid = &info->oti_fid;
7038         struct inode            *inode;
7039         int                      rc = 0, force_over_512tb = 0;
7040         ENTRY;
7041
7042         if (o->od_mnt != NULL)
7043                 RETURN(0);
7044
7045         if (strlen(dev) >= sizeof(o->od_mntdev))
7046                 RETURN(-E2BIG);
7047         strcpy(o->od_mntdev, dev);
7048
7049         str = lustre_cfg_string(cfg, 2);
7050         s_flags = simple_strtoul(str, NULL, 0);
7051         str = strstr(str, ":");
7052         if (str)
7053                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7054         opts = lustre_cfg_string(cfg, 3);
7055 #ifdef __BIG_ENDIAN
7056         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7057                 CERROR("%s: device %s extents feature is not guaranteed to "
7058                        "work on big-endian systems. Use \"bigendian_extents\" "
7059                        "mount option to override.\n", name, dev);
7060                 RETURN(-EINVAL);
7061         }
7062 #endif
7063 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7064         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7065                 CWARN("force_over_128tb option is deprecated. "
7066                       "Filesystems less than 512TB can be created without any "
7067                       "force options. Use force_over_512tb option for "
7068                       "filesystems greater than 512TB.\n");
7069         }
7070 #endif
7071 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7072         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7073                 CWARN("force_over_256tb option is deprecated. "
7074                       "Filesystems less than 512TB can be created without any "
7075                       "force options. Use force_over_512tb option for "
7076                       "filesystems greater than 512TB.\n");
7077         }
7078 #endif
7079
7080         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7081                 force_over_512tb = 1;
7082
7083         __page = alloc_page(GFP_KERNEL);
7084         if (__page == NULL)
7085                 GOTO(out, rc = -ENOMEM);
7086         page = (unsigned long)page_address(__page);
7087         options = (char *)page;
7088         *options = '\0';
7089         if (opts != NULL) {
7090                 /* strip out the options for back compatiblity */
7091                 static char *sout[] = {
7092                         "mballoc",
7093                         "iopen",
7094                         "noiopen",
7095                         "iopen_nopriv",
7096                         "extents",
7097                         "noextents",
7098                         /* strip out option we processed in osd */
7099                         "bigendian_extents",
7100                         "force_over_128tb",
7101                         "force_over_256tb",
7102                         "force_over_512tb",
7103                         NULL
7104                 };
7105                 strcat(options, opts);
7106                 for (rc = 0, str = options; sout[rc]; ) {
7107                         char *op = strstr(str, sout[rc]);
7108                         if (op == NULL) {
7109                                 rc++;
7110                                 str = options;
7111                                 continue;
7112                         }
7113                         if (op == options || *(op - 1) == ',') {
7114                                 str = op + strlen(sout[rc]);
7115                                 if (*str == ',' || *str == '\0') {
7116                                         *str == ',' ? str++ : str;
7117                                         memmove(op, str, strlen(str) + 1);
7118                                 }
7119                         }
7120                         for (str = op; *str != ',' && *str != '\0'; str++)
7121                                 ;
7122                 }
7123         } else {
7124                 strncat(options, "user_xattr,acl", 14);
7125         }
7126
7127         /* Glom up mount options */
7128         if (*options != '\0')
7129                 strcat(options, ",");
7130         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7131
7132         type = get_fs_type("ldiskfs");
7133         if (!type) {
7134                 CERROR("%s: cannot find ldiskfs module\n", name);
7135                 GOTO(out, rc = -ENODEV);
7136         }
7137
7138         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7139         module_put(type->owner);
7140
7141         if (IS_ERR(o->od_mnt)) {
7142                 rc = PTR_ERR(o->od_mnt);
7143                 o->od_mnt = NULL;
7144                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7145                 GOTO(out, rc);
7146         }
7147
7148         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7149                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7150                                  force_over_512tb == 0) {
7151                 CERROR("%s: device %s LDISKFS does not support filesystems "
7152                        "greater than 512TB and can cause data corruption. "
7153                        "Use \"force_over_512tb\" mount option to override.\n",
7154                        name, dev);
7155                 GOTO(out_mnt, rc = -EINVAL);
7156         }
7157
7158         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7159                 if (priv_dev_set_rdonly) {
7160                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7161                         o->od_dt_dev.dd_rdonly = 1;
7162                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7163                                       name);
7164                 } else {
7165                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7166                                       name);
7167
7168                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7169                 }
7170         } else if (priv_dev_check_rdonly &&
7171                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7172                 CERROR("%s: underlying device %s is marked as "
7173                        "read-only. Setup failed\n", name, dev);
7174
7175                 GOTO(out_mnt, rc = -EROFS);
7176         }
7177
7178         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7179                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7180                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7181                 GOTO(out_mnt, rc = -EINVAL);
7182         }
7183
7184 #ifdef LDISKFS_MOUNT_DIRDATA
7185         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7186                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7187                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7188         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7189                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7190                       "enabling the dirdata feature. If you do not want to "
7191                       "downgrade to Lustre-1.x again, you can enable it via "
7192                       "'tune2fs -O dirdata device'\n", name, dev);
7193 #endif
7194         inode = osd_sb(o)->s_root->d_inode;
7195         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7196         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7197         if (rc != 0) {
7198                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7199                 GOTO(out_mnt, rc);
7200         }
7201
7202         if (lmd_flags & LMD_FLG_NOSCRUB)
7203                 o->od_auto_scrub_interval = AS_NEVER;
7204
7205         GOTO(out, rc = 0);
7206
7207 out_mnt:
7208         mntput(o->od_mnt);
7209         o->od_mnt = NULL;
7210
7211 out:
7212         if (__page)
7213                 __free_page(__page);
7214
7215         return rc;
7216 }
7217
7218 static struct lu_device *osd_device_fini(const struct lu_env *env,
7219                                          struct lu_device *d)
7220 {
7221         struct osd_device *o = osd_dev(d);
7222         ENTRY;
7223
7224         osd_shutdown(env, o);
7225         osd_procfs_fini(o);
7226         osd_obj_map_fini(o);
7227         osd_umount(env, o);
7228
7229         RETURN(NULL);
7230 }
7231
7232 static int osd_device_init0(const struct lu_env *env,
7233                             struct osd_device *o,
7234                             struct lustre_cfg *cfg)
7235 {
7236         struct lu_device        *l = osd2lu_dev(o);
7237         struct osd_thread_info *info;
7238         int                     rc;
7239         int                     cplen = 0;
7240
7241         /* if the module was re-loaded, env can loose its keys */
7242         rc = lu_env_refill((struct lu_env *) env);
7243         if (rc)
7244                 GOTO(out, rc);
7245         info = osd_oti_get(env);
7246         LASSERT(info);
7247
7248         l->ld_ops = &osd_lu_ops;
7249         o->od_dt_dev.dd_ops = &osd_dt_ops;
7250
7251         spin_lock_init(&o->od_osfs_lock);
7252         mutex_init(&o->od_otable_mutex);
7253         INIT_LIST_HEAD(&o->od_orphan_list);
7254
7255         o->od_read_cache = 1;
7256         o->od_writethrough_cache = 1;
7257         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7258         o->od_auto_scrub_interval = AS_DEFAULT;
7259
7260         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7261                         sizeof(o->od_svname));
7262         if (cplen >= sizeof(o->od_svname)) {
7263                 rc = -E2BIG;
7264                 GOTO(out, rc);
7265         }
7266
7267         o->od_index = -1; /* -1 means index is invalid */
7268         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7269         if (rc == LDD_F_SV_TYPE_OST)
7270                 o->od_is_ost = 1;
7271
7272         o->od_full_scrub_ratio = OFSR_DEFAULT;
7273         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7274         rc = osd_mount(env, o, cfg);
7275         if (rc != 0)
7276                 GOTO(out, rc);
7277
7278         rc = osd_obj_map_init(env, o);
7279         if (rc != 0)
7280                 GOTO(out_mnt, rc);
7281
7282         rc = lu_site_init(&o->od_site, l);
7283         if (rc != 0)
7284                 GOTO(out_compat, rc);
7285         o->od_site.ls_bottom_dev = l;
7286
7287         rc = lu_site_init_finish(&o->od_site);
7288         if (rc != 0)
7289                 GOTO(out_site, rc);
7290
7291         INIT_LIST_HEAD(&o->od_ios_list);
7292         /* setup scrub, including OI files initialization */
7293         rc = osd_scrub_setup(env, o);
7294         if (rc < 0)
7295                 GOTO(out_site, rc);
7296
7297         rc = osd_procfs_init(o, o->od_svname);
7298         if (rc != 0) {
7299                 CERROR("%s: can't initialize procfs: rc = %d\n",
7300                        o->od_svname, rc);
7301                 GOTO(out_scrub, rc);
7302         }
7303
7304         LASSERT(l->ld_site->ls_linkage.next != NULL);
7305         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7306
7307         /* initialize quota slave instance */
7308         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7309                                      o->od_proc_entry);
7310         if (IS_ERR(o->od_quota_slave)) {
7311                 rc = PTR_ERR(o->od_quota_slave);
7312                 o->od_quota_slave = NULL;
7313                 GOTO(out_procfs, rc);
7314         }
7315
7316         RETURN(0);
7317
7318 out_procfs:
7319         osd_procfs_fini(o);
7320 out_scrub:
7321         osd_scrub_cleanup(env, o);
7322 out_site:
7323         lu_site_fini(&o->od_site);
7324 out_compat:
7325         osd_obj_map_fini(o);
7326 out_mnt:
7327         osd_umount(env, o);
7328 out:
7329         return rc;
7330 }
7331
7332 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7333                                           struct lu_device_type *t,
7334                                           struct lustre_cfg *cfg)
7335 {
7336         struct osd_device *o;
7337         int                rc;
7338
7339         OBD_ALLOC_PTR(o);
7340         if (o == NULL)
7341                 return ERR_PTR(-ENOMEM);
7342
7343         rc = dt_device_init(&o->od_dt_dev, t);
7344         if (rc == 0) {
7345                 /* Because the ctx might be revived in dt_device_init,
7346                  * refill the env here */
7347                 lu_env_refill((struct lu_env *)env);
7348                 rc = osd_device_init0(env, o, cfg);
7349                 if (rc)
7350                         dt_device_fini(&o->od_dt_dev);
7351         }
7352
7353         if (unlikely(rc != 0))
7354                 OBD_FREE_PTR(o);
7355
7356         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7357 }
7358
7359 static struct lu_device *osd_device_free(const struct lu_env *env,
7360                                          struct lu_device *d)
7361 {
7362         struct osd_device *o = osd_dev(d);
7363         ENTRY;
7364
7365         /* XXX: make osd top device in order to release reference */
7366         d->ld_site->ls_top_dev = d;
7367         lu_site_purge(env, d->ld_site, -1);
7368         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7369                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7370                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7371         }
7372         lu_site_fini(&o->od_site);
7373         dt_device_fini(&o->od_dt_dev);
7374         OBD_FREE_PTR(o);
7375         RETURN(NULL);
7376 }
7377
7378 static int osd_process_config(const struct lu_env *env,
7379                               struct lu_device *d, struct lustre_cfg *cfg)
7380 {
7381         struct osd_device               *o = osd_dev(d);
7382         int                             rc;
7383         ENTRY;
7384
7385         switch (cfg->lcfg_command) {
7386         case LCFG_SETUP:
7387                 rc = osd_mount(env, o, cfg);
7388                 break;
7389         case LCFG_CLEANUP:
7390                 lu_dev_del_linkage(d->ld_site, d);
7391                 rc = osd_shutdown(env, o);
7392                 break;
7393         case LCFG_PARAM:
7394                 LASSERT(&o->od_dt_dev);
7395                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7396                                               cfg, &o->od_dt_dev);
7397                 if (rc > 0 || rc == -ENOSYS) {
7398                         rc = class_process_proc_param(PARAM_OST,
7399                                                       lprocfs_osd_obd_vars,
7400                                                       cfg, &o->od_dt_dev);
7401                         if (rc > 0)
7402                                 rc = 0;
7403                 }
7404                 break;
7405         default:
7406                 rc = -ENOSYS;
7407         }
7408
7409         RETURN(rc);
7410 }
7411
7412 static int osd_recovery_complete(const struct lu_env *env,
7413                                  struct lu_device *d)
7414 {
7415         struct osd_device       *osd = osd_dev(d);
7416         int                      rc = 0;
7417         ENTRY;
7418
7419         if (osd->od_quota_slave == NULL)
7420                 RETURN(0);
7421
7422         /* start qsd instance on recovery completion, this notifies the quota
7423          * slave code that we are about to process new requests now */
7424         rc = qsd_start(env, osd->od_quota_slave);
7425         RETURN(rc);
7426 }
7427
7428 /*
7429  * we use exports to track all osd users
7430  */
7431 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7432                            struct obd_device *obd, struct obd_uuid *cluuid,
7433                            struct obd_connect_data *data, void *localdata)
7434 {
7435         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7436         struct lustre_handle  conn;
7437         int                   rc;
7438         ENTRY;
7439
7440         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7441
7442         rc = class_connect(&conn, obd, cluuid);
7443         if (rc)
7444                 RETURN(rc);
7445
7446         *exp = class_conn2export(&conn);
7447
7448         spin_lock(&osd->od_osfs_lock);
7449         osd->od_connects++;
7450         spin_unlock(&osd->od_osfs_lock);
7451
7452         RETURN(0);
7453 }
7454
7455 /*
7456  * once last export (we don't count self-export) disappeared
7457  * osd can be released
7458  */
7459 static int osd_obd_disconnect(struct obd_export *exp)
7460 {
7461         struct obd_device *obd = exp->exp_obd;
7462         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7463         int                rc, release = 0;
7464         ENTRY;
7465
7466         /* Only disconnect the underlying layers on the final disconnect. */
7467         spin_lock(&osd->od_osfs_lock);
7468         osd->od_connects--;
7469         if (osd->od_connects == 0)
7470                 release = 1;
7471         spin_unlock(&osd->od_osfs_lock);
7472
7473         rc = class_disconnect(exp); /* bz 9811 */
7474
7475         if (rc == 0 && release)
7476                 class_manual_cleanup(obd);
7477         RETURN(rc);
7478 }
7479
7480 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7481                        struct lu_device *dev)
7482 {
7483         struct osd_device       *osd    = osd_dev(dev);
7484         struct lr_server_data   *lsd    =
7485                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7486         int                      result = 0;
7487         ENTRY;
7488
7489         if (osd->od_quota_slave != NULL) {
7490                 /* set up quota slave objects */
7491                 result = qsd_prepare(env, osd->od_quota_slave);
7492                 if (result != 0)
7493                         RETURN(result);
7494         }
7495
7496         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7497 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7498                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7499                         osd->od_index_in_idif = 1;
7500                 } else {
7501                         osd->od_index_in_idif = 0;
7502                         result = osd_register_proc_index_in_idif(osd);
7503                         if (result != 0)
7504                                 RETURN(result);
7505                 }
7506 #else
7507                 osd->od_index_in_idif = 1;
7508 #endif
7509         }
7510
7511         result = osd_fid_init(env, osd);
7512
7513         RETURN(result);
7514 }
7515
7516 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7517                          struct lu_fid *fid, struct md_op_data *op_data)
7518 {
7519         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7520
7521         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7522 }
7523
7524 static const struct lu_object_operations osd_lu_obj_ops = {
7525         .loo_object_init      = osd_object_init,
7526         .loo_object_delete    = osd_object_delete,
7527         .loo_object_release   = osd_object_release,
7528         .loo_object_free      = osd_object_free,
7529         .loo_object_print     = osd_object_print,
7530         .loo_object_invariant = osd_object_invariant
7531 };
7532
7533 const struct lu_device_operations osd_lu_ops = {
7534         .ldo_object_alloc      = osd_object_alloc,
7535         .ldo_process_config    = osd_process_config,
7536         .ldo_recovery_complete = osd_recovery_complete,
7537         .ldo_prepare           = osd_prepare,
7538 };
7539
7540 static const struct lu_device_type_operations osd_device_type_ops = {
7541         .ldto_init = osd_type_init,
7542         .ldto_fini = osd_type_fini,
7543
7544         .ldto_start = osd_type_start,
7545         .ldto_stop  = osd_type_stop,
7546
7547         .ldto_device_alloc = osd_device_alloc,
7548         .ldto_device_free  = osd_device_free,
7549
7550         .ldto_device_init    = osd_device_init,
7551         .ldto_device_fini    = osd_device_fini
7552 };
7553
7554 static struct lu_device_type osd_device_type = {
7555         .ldt_tags     = LU_DEVICE_DT,
7556         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7557         .ldt_ops      = &osd_device_type_ops,
7558         .ldt_ctx_tags = LCT_LOCAL,
7559 };
7560
7561 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7562 {
7563         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7564         struct super_block *sb = osd_sb(osd);
7565
7566         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7567 }
7568
7569 /*
7570  * lprocfs legacy support.
7571  */
7572 static struct obd_ops osd_obd_device_ops = {
7573         .o_owner = THIS_MODULE,
7574         .o_connect      = osd_obd_connect,
7575         .o_disconnect   = osd_obd_disconnect,
7576         .o_fid_alloc    = osd_fid_alloc,
7577         .o_health_check = osd_health_check,
7578 };
7579
7580 static int __init osd_init(void)
7581 {
7582         int rc;
7583
7584         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7585 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7586         /* please, try to keep osd_thread_info smaller than a page */
7587         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7588 #endif
7589
7590         osd_oi_mod_init();
7591
7592         rc = lu_kmem_init(ldiskfs_caches);
7593         if (rc)
7594                 return rc;
7595
7596 #ifdef CONFIG_KALLSYMS
7597         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7598         priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7599         /* Clear readonly is unused at this time */
7600         /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7601 #endif
7602
7603         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7604                                  lprocfs_osd_module_vars,
7605                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7606         if (rc)
7607                 lu_kmem_fini(ldiskfs_caches);
7608         return rc;
7609 }
7610
7611 static void __exit osd_exit(void)
7612 {
7613         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7614         lu_kmem_fini(ldiskfs_caches);
7615 }
7616
7617 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7618 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7619 MODULE_VERSION(LUSTRE_VERSION_STRING);
7620 MODULE_LICENSE("GPL");
7621
7622 module_init(osd_init);
7623 module_exit(osd_exit);