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
2165 /*
2166  * Concurrency: shouldn't matter.
2167  */
2168 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2169 {
2170         int rc;
2171
2172         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2173
2174         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2175
2176         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2177                rc);
2178
2179         return rc;
2180 }
2181
2182 /**
2183  * Start commit for OSD device.
2184  *
2185  * An implementation of dt_commit_async method for OSD device.
2186  * Asychronously starts underlayng fs sync and thereby a transaction
2187  * commit.
2188  *
2189  * \param env environment
2190  * \param d dt device
2191  *
2192  * \see dt_device_operations
2193  */
2194 static int osd_commit_async(const struct lu_env *env,
2195                             struct dt_device *d)
2196 {
2197         struct super_block *s = osd_sb(osd_dt_dev(d));
2198         ENTRY;
2199
2200         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2201         RETURN(s->s_op->sync_fs(s, 0));
2202 }
2203
2204 /* Our own copy of the set readonly functions if present, or NU if not. */
2205 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2206 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2207 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2208
2209 /*
2210  * Concurrency: shouldn't matter.
2211  */
2212 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2213 {
2214         struct super_block *sb = osd_sb(osd_dt_dev(d));
2215         struct block_device *dev = sb->s_bdev;
2216         int rc = -EOPNOTSUPP;
2217         ENTRY;
2218
2219         if (priv_dev_set_rdonly) {
2220                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2221
2222                 rc = 0;
2223                 CERROR("*** setting %s read-only ***\n",
2224                        osd_dt_dev(d)->od_svname);
2225
2226                 if (sb->s_op->freeze_fs) {
2227                         rc = sb->s_op->freeze_fs(sb);
2228                         if (rc)
2229                                 goto out;
2230                 }
2231
2232                 if (jdev && (jdev != dev)) {
2233                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2234                                (long)jdev);
2235                         priv_dev_set_rdonly(jdev);
2236                 }
2237                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2238                 priv_dev_set_rdonly(dev);
2239
2240                 if (sb->s_op->unfreeze_fs)
2241                         sb->s_op->unfreeze_fs(sb);
2242         }
2243
2244 out:
2245         if (rc)
2246                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2247                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2248
2249         RETURN(rc);
2250 }
2251
2252 /**
2253  * Note: we do not count into QUOTA here.
2254  * If we mount with --data_journal we may need more.
2255  */
2256 const int osd_dto_credits_noquota[DTO_NR] = {
2257         /**
2258          * Insert.
2259          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2260          * SINGLEDATA_TRANS_BLOCKS(8)
2261          * XXX Note: maybe iam need more, since iam have more level than
2262          *           EXT3 htree.
2263          */
2264         [DTO_INDEX_INSERT]  = 16,
2265         /**
2266          * Delete
2267          * just modify a single entry, probably merge few within a block
2268          */
2269         [DTO_INDEX_DELETE]  = 1,
2270         /**
2271          * Used for OI scrub
2272          */
2273         [DTO_INDEX_UPDATE]  = 16,
2274         /**
2275          * 4(inode, inode bits, groups, GDT)
2276          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2277          */
2278         [DTO_OBJECT_CREATE] = 4,
2279         /**
2280          * 4(inode, inode bits, groups, GDT)
2281          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2282          */
2283         [DTO_OBJECT_DELETE] = 4,
2284         /**
2285          * Attr set credits (inode)
2286          */
2287         [DTO_ATTR_SET_BASE] = 1,
2288         /**
2289          * Xattr set. The same as xattr of EXT3.
2290          * DATA_TRANS_BLOCKS(14)
2291          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2292          * are also counted in. Do not know why?
2293          */
2294         [DTO_XATTR_SET]     = 14,
2295         /**
2296          * credits for inode change during write.
2297          */
2298         [DTO_WRITE_BASE]    = 3,
2299         /**
2300          * credits for single block write.
2301          */
2302         [DTO_WRITE_BLOCK]   = 14,
2303         /**
2304          * Attr set credits for chown.
2305          * This is extra credits for setattr, and it is null without quota
2306          */
2307         [DTO_ATTR_SET_CHOWN] = 0
2308 };
2309
2310 static const struct dt_device_operations osd_dt_ops = {
2311         .dt_root_get       = osd_root_get,
2312         .dt_statfs         = osd_statfs,
2313         .dt_trans_create   = osd_trans_create,
2314         .dt_trans_start    = osd_trans_start,
2315         .dt_trans_stop     = osd_trans_stop,
2316         .dt_trans_cb_add   = osd_trans_cb_add,
2317         .dt_conf_get       = osd_conf_get,
2318         .dt_sync           = osd_sync,
2319         .dt_ro             = osd_ro,
2320         .dt_commit_async   = osd_commit_async,
2321 };
2322
2323 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2324                           unsigned role)
2325 {
2326         struct osd_object *obj = osd_dt_obj(dt);
2327         struct osd_thread_info *oti = osd_oti_get(env);
2328
2329         LINVRNT(osd_invariant(obj));
2330
2331         LASSERT(obj->oo_owner != env);
2332         down_read_nested(&obj->oo_sem, role);
2333
2334         LASSERT(obj->oo_owner == NULL);
2335         oti->oti_r_locks++;
2336 }
2337
2338 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2339                            unsigned role)
2340 {
2341         struct osd_object *obj = osd_dt_obj(dt);
2342         struct osd_thread_info *oti = osd_oti_get(env);
2343
2344         LINVRNT(osd_invariant(obj));
2345
2346         LASSERT(obj->oo_owner != env);
2347         down_write_nested(&obj->oo_sem, role);
2348
2349         LASSERT(obj->oo_owner == NULL);
2350         obj->oo_owner = env;
2351         oti->oti_w_locks++;
2352 }
2353
2354 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2355 {
2356         struct osd_object *obj = osd_dt_obj(dt);
2357         struct osd_thread_info *oti = osd_oti_get(env);
2358
2359         LINVRNT(osd_invariant(obj));
2360
2361         LASSERT(oti->oti_r_locks > 0);
2362         oti->oti_r_locks--;
2363         up_read(&obj->oo_sem);
2364 }
2365
2366 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2367 {
2368         struct osd_object *obj = osd_dt_obj(dt);
2369         struct osd_thread_info *oti = osd_oti_get(env);
2370
2371         LINVRNT(osd_invariant(obj));
2372
2373         LASSERT(obj->oo_owner == env);
2374         LASSERT(oti->oti_w_locks > 0);
2375         oti->oti_w_locks--;
2376         obj->oo_owner = NULL;
2377         up_write(&obj->oo_sem);
2378 }
2379
2380 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2381 {
2382         struct osd_object *obj = osd_dt_obj(dt);
2383
2384         LINVRNT(osd_invariant(obj));
2385
2386         return obj->oo_owner == env;
2387 }
2388
2389 static struct timespec *osd_inode_time(const struct lu_env *env,
2390                                        struct inode *inode, __u64 seconds)
2391 {
2392         struct osd_thread_info  *oti = osd_oti_get(env);
2393         struct timespec         *t   = &oti->oti_time;
2394
2395         t->tv_sec = seconds;
2396         t->tv_nsec = 0;
2397         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2398         return t;
2399 }
2400
2401 static void osd_inode_getattr(const struct lu_env *env,
2402                               struct inode *inode, struct lu_attr *attr)
2403 {
2404         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2405                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2406                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2407                            LA_BLKSIZE | LA_TYPE;
2408
2409         attr->la_atime   = LTIME_S(inode->i_atime);
2410         attr->la_mtime   = LTIME_S(inode->i_mtime);
2411         attr->la_ctime   = LTIME_S(inode->i_ctime);
2412         attr->la_mode    = inode->i_mode;
2413         attr->la_size    = i_size_read(inode);
2414         attr->la_blocks  = inode->i_blocks;
2415         attr->la_uid     = i_uid_read(inode);
2416         attr->la_gid     = i_gid_read(inode);
2417         attr->la_projid  = i_projid_read(inode);
2418         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2419         attr->la_nlink   = inode->i_nlink;
2420         attr->la_rdev    = inode->i_rdev;
2421         attr->la_blksize = 1 << inode->i_blkbits;
2422         attr->la_blkbits = inode->i_blkbits;
2423         /*
2424          * Ext4 did not transfer inherit flags from raw inode
2425          * to inode flags, and ext4 internally test raw inode
2426          * @i_flags directly. Instead of patching ext4, we do it here.
2427          */
2428         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2429                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2430 }
2431
2432 static int osd_attr_get(const struct lu_env *env,
2433                         struct dt_object *dt,
2434                         struct lu_attr *attr)
2435 {
2436         struct osd_object *obj = osd_dt_obj(dt);
2437
2438         if (unlikely(!dt_object_exists(dt)))
2439                 return -ENOENT;
2440         if (unlikely(obj->oo_destroyed))
2441                 return -ENOENT;
2442
2443         LASSERT(!dt_object_remote(dt));
2444         LINVRNT(osd_invariant(obj));
2445
2446         spin_lock(&obj->oo_guard);
2447         osd_inode_getattr(env, obj->oo_inode, attr);
2448         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2449                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2450         spin_unlock(&obj->oo_guard);
2451
2452         return 0;
2453 }
2454
2455 static int osd_declare_attr_qid(const struct lu_env *env,
2456                                 struct osd_object *obj,
2457                                 struct osd_thandle *oh, long long bspace,
2458                                 qid_t old_id, qid_t new_id, bool enforce,
2459                                 unsigned type)
2460 {
2461         int rc;
2462         struct osd_thread_info *info = osd_oti_get(env);
2463         struct lquota_id_info  *qi = &info->oti_qi;
2464
2465         qi->lqi_type = type;
2466         /* inode accounting */
2467         qi->lqi_is_blk = false;
2468
2469         /* one more inode for the new id ... */
2470         qi->lqi_id.qid_uid = new_id;
2471         qi->lqi_space      = 1;
2472         /* Reserve credits for the new id */
2473         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2474         if (rc == -EDQUOT || rc == -EINPROGRESS)
2475                 rc = 0;
2476         if (rc)
2477                 RETURN(rc);
2478
2479         /* and one less inode for the current id */
2480         qi->lqi_id.qid_uid = old_id;
2481         qi->lqi_space      = -1;
2482         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2483         if (rc == -EDQUOT || rc == -EINPROGRESS)
2484                 rc = 0;
2485         if (rc)
2486                 RETURN(rc);
2487
2488         /* block accounting */
2489         qi->lqi_is_blk = true;
2490
2491         /* more blocks for the new id ... */
2492         qi->lqi_id.qid_uid = new_id;
2493         qi->lqi_space      = bspace;
2494         /*
2495          * Credits for the new uid has been reserved, re-use "obj"
2496          * to save credit reservation.
2497          */
2498         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2499         if (rc == -EDQUOT || rc == -EINPROGRESS)
2500                 rc = 0;
2501         if (rc)
2502                 RETURN(rc);
2503
2504         /* and finally less blocks for the current uid */
2505         qi->lqi_id.qid_uid = old_id;
2506         qi->lqi_space      = -bspace;
2507         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2508         if (rc == -EDQUOT || rc == -EINPROGRESS)
2509                 rc = 0;
2510
2511         RETURN(rc);
2512 }
2513
2514 static int osd_declare_attr_set(const struct lu_env *env,
2515                                 struct dt_object *dt,
2516                                 const struct lu_attr *attr,
2517                                 struct thandle *handle)
2518 {
2519         struct osd_thandle     *oh;
2520         struct osd_object      *obj;
2521         qid_t                   uid;
2522         qid_t                   gid;
2523         long long               bspace;
2524         int                     rc = 0;
2525         bool                    enforce;
2526         ENTRY;
2527
2528         LASSERT(dt != NULL);
2529         LASSERT(handle != NULL);
2530
2531         obj = osd_dt_obj(dt);
2532         LASSERT(osd_invariant(obj));
2533
2534         oh = container_of0(handle, struct osd_thandle, ot_super);
2535         LASSERT(oh->ot_handle == NULL);
2536
2537         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2538                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2539
2540         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2541                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2542
2543         if (attr == NULL || obj->oo_inode == NULL)
2544                 RETURN(rc);
2545
2546         bspace   = obj->oo_inode->i_blocks;
2547         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2548         bspace   = toqb(bspace);
2549
2550         /* Changing ownership is always preformed by super user, it should not
2551          * fail with EDQUOT.
2552          *
2553          * We still need to call the osd_declare_qid() to calculate the journal
2554          * credits for updating quota accounting files and to trigger quota
2555          * space adjustment once the operation is completed.*/
2556         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2557                 /* USERQUOTA */
2558                 uid = i_uid_read(obj->oo_inode);
2559                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2560                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2561                                           attr->la_uid, enforce, USRQUOTA);
2562                 if (rc)
2563                         RETURN(rc);
2564
2565                 gid = i_gid_read(obj->oo_inode);
2566                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2567                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2568                                           i_gid_read(obj->oo_inode),
2569                                           attr->la_gid, enforce, GRPQUOTA);
2570                 if (rc)
2571                         RETURN(rc);
2572
2573         }
2574 #ifdef HAVE_PROJECT_QUOTA
2575         if (attr->la_valid & LA_PROJID) {
2576                 __u32 projid = i_projid_read(obj->oo_inode);
2577                 enforce = (attr->la_valid & LA_PROJID) &&
2578                                         (attr->la_projid != projid);
2579                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2580                                           (qid_t)projid, (qid_t)attr->la_projid,
2581                                           enforce, PRJQUOTA);
2582                 if (rc)
2583                         RETURN(rc);
2584         }
2585 #endif
2586         RETURN(rc);
2587 }
2588
2589 static int osd_inode_setattr(const struct lu_env *env,
2590                              struct inode *inode, const struct lu_attr *attr)
2591 {
2592         __u64 bits = attr->la_valid;
2593
2594         /* Only allow set size for regular file */
2595         if (!S_ISREG(inode->i_mode))
2596                 bits &= ~(LA_SIZE | LA_BLOCKS);
2597
2598         if (bits == 0)
2599                 return 0;
2600
2601         if (bits & LA_ATIME)
2602                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2603         if (bits & LA_CTIME)
2604                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2605         if (bits & LA_MTIME)
2606                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2607         if (bits & LA_SIZE) {
2608                 spin_lock(&inode->i_lock);
2609                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2610                 i_size_write(inode, attr->la_size);
2611                 spin_unlock(&inode->i_lock);
2612         }
2613
2614         /* OSD should not change "i_blocks" which is used by quota.
2615          * "i_blocks" should be changed by ldiskfs only. */
2616         if (bits & LA_MODE)
2617                 inode->i_mode = (inode->i_mode & S_IFMT) |
2618                                 (attr->la_mode & ~S_IFMT);
2619         if (bits & LA_UID)
2620                 i_uid_write(inode, attr->la_uid);
2621         if (bits & LA_GID)
2622                 i_gid_write(inode, attr->la_gid);
2623         if (bits & LA_PROJID)
2624                 i_projid_write(inode, attr->la_projid);
2625         if (bits & LA_NLINK)
2626                 set_nlink(inode, attr->la_nlink);
2627         if (bits & LA_RDEV)
2628                 inode->i_rdev = attr->la_rdev;
2629
2630         if (bits & LA_FLAGS) {
2631                 /* always keep S_NOCMTIME */
2632                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2633                                  S_NOCMTIME;
2634                 /*
2635                  * Ext4 did not transfer inherit flags from
2636                  * @inode->i_flags to raw inode i_flags when writing
2637                  * flags, we do it explictly here.
2638                  */
2639                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2640                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2641                 else
2642                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2643         }
2644         return 0;
2645 }
2646
2647 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2648 {
2649         int rc;
2650
2651         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2652             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2653                 struct iattr    iattr;
2654
2655                 ll_vfs_dq_init(inode);
2656                 iattr.ia_valid = 0;
2657                 if (attr->la_valid & LA_UID)
2658                         iattr.ia_valid |= ATTR_UID;
2659                 if (attr->la_valid & LA_GID)
2660                         iattr.ia_valid |= ATTR_GID;
2661                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2662                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2663
2664                 rc = ll_vfs_dq_transfer(inode, &iattr);
2665                 if (rc) {
2666                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2667                                "enforcement enabled on the ldiskfs "
2668                                "filesystem?\n", inode->i_sb->s_id, rc);
2669                         return rc;
2670                 }
2671         }
2672
2673         /* Handle project id transfer here properly */
2674         if (attr->la_valid & LA_PROJID &&
2675             attr->la_projid != i_projid_read(inode)) {
2676 #ifdef HAVE_PROJECT_QUOTA
2677                 rc = ldiskfs_transfer_project(inode, attr->la_projid);
2678 #else
2679                 rc = -ENOTSUPP;
2680 #endif
2681                 if (rc) {
2682                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2683                                "enforcement enabled on the ldiskfs "
2684                                "filesystem?\n", inode->i_sb->s_id, rc);
2685                         return rc;
2686                 }
2687         }
2688         return 0;
2689 }
2690
2691 static int osd_attr_set(const struct lu_env *env,
2692                         struct dt_object *dt,
2693                         const struct lu_attr *attr,
2694                         struct thandle *handle)
2695 {
2696         struct osd_object *obj = osd_dt_obj(dt);
2697         struct inode      *inode;
2698         int rc;
2699
2700         if (!dt_object_exists(dt))
2701                 return -ENOENT;
2702
2703         LASSERT(handle != NULL);
2704         LASSERT(!dt_object_remote(dt));
2705         LASSERT(osd_invariant(obj));
2706
2707         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2708
2709         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2710             !osd_obj2dev(obj)->od_is_ost) {
2711                 struct osd_thread_info  *oti  = osd_oti_get(env);
2712                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2713                 struct lu_fid           *fid1 = &oti->oti_fid;
2714                 struct osd_inode_id     *id   = &oti->oti_id;
2715                 struct iam_path_descr   *ipd;
2716                 struct iam_container    *bag;
2717                 struct osd_thandle      *oh;
2718                 int                      rc;
2719
2720                 fid_cpu_to_be(fid1, fid0);
2721                 memset(id, 1, sizeof(*id));
2722                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2723                                   fid0)->oi_dir.od_container;
2724                 ipd = osd_idx_ipd_get(env, bag);
2725                 if (unlikely(ipd == NULL))
2726                         RETURN(-ENOMEM);
2727
2728                 oh = container_of0(handle, struct osd_thandle, ot_super);
2729                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2730                                 (const struct iam_rec *)id, ipd);
2731                 osd_ipd_put(env, bag, ipd);
2732                 return(rc > 0 ? 0 : rc);
2733         }
2734
2735         inode = obj->oo_inode;
2736
2737         rc = osd_quota_transfer(inode, attr);
2738         if (rc)
2739                 return rc;
2740
2741         spin_lock(&obj->oo_guard);
2742         rc = osd_inode_setattr(env, inode, attr);
2743         spin_unlock(&obj->oo_guard);
2744         if (rc != 0)
2745                 GOTO(out, rc);
2746
2747         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2748
2749         if (!(attr->la_valid & LA_FLAGS))
2750                 GOTO(out, rc);
2751
2752         /* Let's check if there are extra flags need to be set into LMA */
2753         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2754                 struct osd_thread_info *info = osd_oti_get(env);
2755                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2756
2757                 LASSERT(!obj->oo_pfid_in_lma);
2758
2759                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2760                                  &info->oti_ost_attrs);
2761                 if (rc)
2762                         GOTO(out, rc);
2763
2764                 lma->lma_incompat |=
2765                         lustre_to_lma_flags(attr->la_flags);
2766                 lustre_lma_swab(lma);
2767                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2768                                      lma, sizeof(*lma), XATTR_REPLACE);
2769                 if (rc != 0) {
2770                         struct osd_device *osd = osd_obj2dev(obj);
2771
2772                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2773                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2774                               lma->lma_incompat, rc);
2775                 } else {
2776                         obj->oo_lma_flags =
2777                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2778                 }
2779                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2780         }
2781 out:
2782         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2783
2784         return rc;
2785 }
2786
2787 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2788                                            struct osd_object *obj,
2789                                            const char *name, const int namelen)
2790 {
2791         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2792 }
2793
2794 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2795                       umode_t mode, struct dt_allocation_hint *hint,
2796                       struct thandle *th)
2797 {
2798         int result;
2799         struct osd_device  *osd = osd_obj2dev(obj);
2800         struct osd_thandle *oth;
2801         struct dt_object   *parent = NULL;
2802         struct inode       *inode;
2803
2804         LINVRNT(osd_invariant(obj));
2805         LASSERT(obj->oo_inode == NULL);
2806         LASSERT(obj->oo_hl_head == NULL);
2807
2808         if (S_ISDIR(mode) && ldiskfs_pdo) {
2809                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2810                 if (obj->oo_hl_head == NULL)
2811                         return -ENOMEM;
2812         }
2813
2814         oth = container_of(th, struct osd_thandle, ot_super);
2815         LASSERT(oth->ot_handle->h_transaction != NULL);
2816
2817         if (hint != NULL && hint->dah_parent != NULL &&
2818             !dt_object_remote(hint->dah_parent))
2819                 parent = hint->dah_parent;
2820
2821         inode = ldiskfs_create_inode(oth->ot_handle,
2822                                      parent ? osd_dt_obj(parent)->oo_inode :
2823                                               osd_sb(osd)->s_root->d_inode,
2824                                      mode);
2825         if (!IS_ERR(inode)) {
2826                 /* Do not update file c/mtime in ldiskfs. */
2827                 inode->i_flags |= S_NOCMTIME;
2828
2829                 /* For new created object, it must be consistent,
2830                  * and it is unnecessary to scrub against it. */
2831                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2832
2833                 obj->oo_inode = inode;
2834                 result = 0;
2835         } else {
2836                 if (obj->oo_hl_head != NULL) {
2837                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2838                         obj->oo_hl_head = NULL;
2839                 }
2840                 result = PTR_ERR(inode);
2841         }
2842         LINVRNT(osd_invariant(obj));
2843         return result;
2844 }
2845
2846 enum {
2847         OSD_NAME_LEN = 255
2848 };
2849
2850 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2851                      struct lu_attr *attr,
2852                      struct dt_allocation_hint *hint,
2853                      struct dt_object_format *dof,
2854                      struct thandle *th)
2855 {
2856         int result;
2857         struct osd_thandle *oth;
2858         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2859
2860         LASSERT(S_ISDIR(attr->la_mode));
2861
2862         oth = container_of(th, struct osd_thandle, ot_super);
2863         LASSERT(oth->ot_handle->h_transaction != NULL);
2864         result = osd_mkfile(info, obj, mode, hint, th);
2865
2866         return result;
2867 }
2868
2869 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2870                         struct lu_attr *attr,
2871                         struct dt_allocation_hint *hint,
2872                         struct dt_object_format *dof,
2873                         struct thandle *th)
2874 {
2875         int result;
2876         struct osd_thandle *oth;
2877         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2878
2879         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2880
2881         LASSERT(S_ISREG(attr->la_mode));
2882
2883         oth = container_of(th, struct osd_thandle, ot_super);
2884         LASSERT(oth->ot_handle->h_transaction != NULL);
2885
2886         result = osd_mkfile(info, obj, mode, hint, th);
2887         if (result == 0) {
2888                 LASSERT(obj->oo_inode != NULL);
2889                 if (feat->dif_flags & DT_IND_VARKEY)
2890                         result = iam_lvar_create(obj->oo_inode,
2891                                                  feat->dif_keysize_max,
2892                                                  feat->dif_ptrsize,
2893                                                  feat->dif_recsize_max,
2894                                                  oth->ot_handle);
2895                 else
2896                         result = iam_lfix_create(obj->oo_inode,
2897                                                  feat->dif_keysize_max,
2898                                                  feat->dif_ptrsize,
2899                                                  feat->dif_recsize_max,
2900                                                  oth->ot_handle);
2901
2902         }
2903         return result;
2904 }
2905
2906 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2907                      struct lu_attr *attr,
2908                      struct dt_allocation_hint *hint,
2909                      struct dt_object_format *dof,
2910                      struct thandle *th)
2911 {
2912         LASSERT(S_ISREG(attr->la_mode));
2913         return osd_mkfile(info, obj, (attr->la_mode &
2914                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2915 }
2916
2917 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2918                      struct lu_attr *attr,
2919                      struct dt_allocation_hint *hint,
2920                      struct dt_object_format *dof,
2921                      struct thandle *th)
2922 {
2923         LASSERT(S_ISLNK(attr->la_mode));
2924         return osd_mkfile(info, obj, (attr->la_mode &
2925                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2926 }
2927
2928 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2929                      struct lu_attr *attr,
2930                      struct dt_allocation_hint *hint,
2931                      struct dt_object_format *dof,
2932                      struct thandle *th)
2933 {
2934         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2935         int result;
2936
2937         LINVRNT(osd_invariant(obj));
2938         LASSERT(obj->oo_inode == NULL);
2939         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2940                 S_ISFIFO(mode) || S_ISSOCK(mode));
2941
2942         result = osd_mkfile(info, obj, mode, hint, th);
2943         if (result == 0) {
2944                 LASSERT(obj->oo_inode != NULL);
2945                 /*
2946                  * This inode should be marked dirty for i_rdev.  Currently
2947                  * that is done in the osd_attr_init().
2948                  */
2949                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2950                                    attr->la_rdev);
2951         }
2952         LINVRNT(osd_invariant(obj));
2953         return result;
2954 }
2955
2956 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2957                               struct lu_attr *,
2958                               struct dt_allocation_hint *hint,
2959                               struct dt_object_format *dof,
2960                               struct thandle *);
2961
2962 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2963 {
2964         osd_obj_type_f result;
2965
2966         switch (type) {
2967         case DFT_DIR:
2968                 result = osd_mkdir;
2969                 break;
2970         case DFT_REGULAR:
2971                 result = osd_mkreg;
2972                 break;
2973         case DFT_SYM:
2974                 result = osd_mksym;
2975                 break;
2976         case DFT_NODE:
2977                 result = osd_mknod;
2978                 break;
2979         case DFT_INDEX:
2980                 result = osd_mk_index;
2981                 break;
2982
2983         default:
2984                 LBUG();
2985                 break;
2986         }
2987         return result;
2988 }
2989
2990
2991 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2992                         struct dt_object *parent, struct dt_object *child,
2993                         umode_t child_mode)
2994 {
2995         LASSERT(ah);
2996
2997         ah->dah_parent = parent;
2998         ah->dah_mode = child_mode;
2999
3000         if (parent != NULL && !dt_object_remote(parent)) {
3001                 /* will help to find FID->ino at dt_insert("..") */
3002                 struct osd_object *pobj = osd_dt_obj(parent);
3003                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3004         }
3005 }
3006
3007 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3008                           struct lu_attr *attr, struct dt_object_format *dof)
3009 {
3010         struct inode   *inode = obj->oo_inode;
3011         __u64           valid = attr->la_valid;
3012         int             result;
3013
3014         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3015
3016         if (dof->dof_type != DFT_NODE)
3017                 attr->la_valid &= ~LA_RDEV;
3018         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
3019                 attr->la_valid &= ~LA_ATIME;
3020         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
3021                 attr->la_valid &= ~LA_CTIME;
3022         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
3023                 attr->la_valid &= ~LA_MTIME;
3024
3025         result = osd_quota_transfer(inode, attr);
3026         if (result)
3027                 return;
3028
3029         if (attr->la_valid != 0) {
3030                 result = osd_inode_setattr(info->oti_env, inode, attr);
3031                 /*
3032                  * The osd_inode_setattr() should always succeed here.  The
3033                  * only error that could be returned is EDQUOT when we are
3034                  * trying to change the UID or GID of the inode. However, this
3035                  * should not happen since quota enforcement is no longer
3036                  * enabled on ldiskfs (lquota takes care of it).
3037                  */
3038                 LASSERTF(result == 0, "%d\n", result);
3039                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3040         }
3041
3042         attr->la_valid = valid;
3043 }
3044
3045 /**
3046  * Helper function for osd_create()
3047  *
3048  * \retval 0, on success
3049  */
3050 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3051                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3052                         struct dt_object_format *dof, struct thandle *th)
3053 {
3054         int     result;
3055         __u32   umask;
3056
3057         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3058
3059         /* we drop umask so that permissions we pass are not affected */
3060         umask = current->fs->umask;
3061         current->fs->umask = 0;
3062
3063         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3064                                                   th);
3065         if (likely(obj->oo_inode != NULL)) {
3066                 LASSERT(obj->oo_inode->i_state & I_NEW);
3067
3068                 /* Unlock the inode before attr initialization to avoid
3069                  * unnecessary dqget operations. LU-6378 */
3070                 unlock_new_inode(obj->oo_inode);
3071         }
3072
3073         if (likely(result == 0)) {
3074                 osd_attr_init(info, obj, attr, dof);
3075                 osd_object_init0(obj);
3076         }
3077
3078         /* restore previous umask value */
3079         current->fs->umask = umask;
3080
3081         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3082
3083         return result;
3084 }
3085
3086 /**
3087  * Helper function for osd_create()
3088  *
3089  * \retval 0, on success
3090  */
3091 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3092                            const struct lu_fid *fid, struct thandle *th)
3093 {
3094         struct osd_thread_info *info = osd_oti_get(env);
3095         struct osd_inode_id    *id   = &info->oti_id;
3096         struct osd_device      *osd  = osd_obj2dev(obj);
3097         struct osd_thandle     *oh;
3098         int                     rc;
3099
3100         LASSERT(obj->oo_inode != NULL);
3101
3102         oh = container_of0(th, struct osd_thandle, ot_super);
3103         LASSERT(oh->ot_handle);
3104         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3105
3106         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3107         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3108                            OI_CHECK_FLD, NULL);
3109         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3110
3111         return rc;
3112 }
3113
3114 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3115                    u64 seq, struct lu_seq_range *range)
3116 {
3117         struct seq_server_site  *ss = osd_seq_site(osd);
3118
3119         if (fid_seq_is_idif(seq)) {
3120                 fld_range_set_ost(range);
3121                 range->lsr_index = idif_ost_idx(seq);
3122                 return 0;
3123         }
3124
3125         if (!fid_seq_in_fldb(seq)) {
3126                 fld_range_set_mdt(range);
3127                 if (ss != NULL)
3128                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3129                          * at all */
3130                         range->lsr_index = ss->ss_node_id;
3131                 return 0;
3132         }
3133
3134         LASSERT(ss != NULL);
3135         fld_range_set_any(range);
3136         /* OSD will only do local fld lookup */
3137         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3138 }
3139
3140 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3141                               struct lu_attr *attr,
3142                               struct dt_allocation_hint *hint,
3143                               struct dt_object_format *dof,
3144                               struct thandle *handle)
3145 {
3146         struct osd_thandle *oh;
3147         int rc;
3148         ENTRY;
3149
3150         LASSERT(handle != NULL);
3151
3152         oh = container_of0(handle, struct osd_thandle, ot_super);
3153         LASSERT(oh->ot_handle == NULL);
3154
3155         /* EA object consumes more credits than regular object: osd_mk_index
3156          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3157          * leaf_node, could involves the block, block bitmap, groups, GDT
3158          * change for each block, so add 4 * 2 credits in that case. */
3159         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3160                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3161                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3162         /* Reuse idle OI block may cause additional one OI block
3163          * to be changed. */
3164         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3165                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3166
3167         if (!attr)
3168                 RETURN(0);
3169
3170         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3171                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3172                                    NULL, OSD_QID_INODE);
3173         if (rc != 0)
3174                 RETURN(rc);
3175
3176         /* will help to find FID->ino mapping at dt_insert() */
3177         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3178                                    osd_dt_obj(dt));
3179
3180         RETURN(rc);
3181 }
3182
3183 /**
3184  * Called to destroy on-disk representation of the object
3185  *
3186  * Concurrency: must be locked
3187  */
3188 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3189                                struct thandle *th)
3190 {
3191         struct osd_object  *obj = osd_dt_obj(dt);
3192         struct inode       *inode = obj->oo_inode;
3193         struct osd_thandle *oh;
3194         int                 rc;
3195         ENTRY;
3196
3197         if (inode == NULL)
3198                 RETURN(-ENOENT);
3199
3200         oh = container_of0(th, struct osd_thandle, ot_super);
3201         LASSERT(oh->ot_handle == NULL);
3202
3203         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3204                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3205
3206         /* For removing agent entry */
3207         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3208                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3209
3210         /* Recycle idle OI leaf may cause additional three OI blocks
3211          * to be changed. */
3212         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3213                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3214                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3215         /* one less inode */
3216         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3217                                    i_projid_read(inode), -1, oh, obj, NULL,
3218                                    OSD_QID_INODE);
3219         if (rc)
3220                 RETURN(rc);
3221         /* data to be truncated */
3222         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3223                                    i_projid_read(inode), 0, oh, obj, NULL,
3224                                    OSD_QID_BLK);
3225         if (rc)
3226                 RETURN(rc);
3227
3228         /* will help to find FID->ino when this object is being
3229          * added to PENDING/ */
3230         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3231
3232         RETURN(rc);
3233 }
3234
3235 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3236                        struct thandle *th)
3237 {
3238         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3239         struct osd_object *obj = osd_dt_obj(dt);
3240         struct inode *inode = obj->oo_inode;
3241         struct osd_device *osd = osd_obj2dev(obj);
3242         struct osd_thandle *oh;
3243         int result;
3244         ENTRY;
3245
3246         oh = container_of0(th, struct osd_thandle, ot_super);
3247         LASSERT(oh->ot_handle);
3248         LASSERT(inode);
3249         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3250
3251         if (unlikely(fid_is_acct(fid)))
3252                 RETURN(-EPERM);
3253
3254         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3255                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3256                 if (result != 0)
3257                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3258                                osd_name(osd), PFID(fid), result);
3259         }
3260
3261         if (S_ISDIR(inode->i_mode)) {
3262                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3263                         inode->i_nlink == 2);
3264
3265                 spin_lock(&obj->oo_guard);
3266                 clear_nlink(inode);
3267                 spin_unlock(&obj->oo_guard);
3268                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3269         }
3270
3271         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3272
3273         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3274
3275         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3276                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3277                                        oh->ot_handle, OI_CHECK_FLD);
3278
3279         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3280         /* XXX: add to ext3 orphan list */
3281         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3282
3283         /* not needed in the cache anymore */
3284         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3285         obj->oo_destroyed = 1;
3286
3287         RETURN(0);
3288 }
3289
3290 /**
3291  * Put the fid into lustre_mdt_attrs, and then place the structure
3292  * inode's ea. This fid should not be altered during the life time
3293  * of the inode.
3294  *
3295  * \retval +ve, on success
3296  * \retval -ve, on error
3297  *
3298  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3299  */
3300 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3301                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3302 {
3303         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3304         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3305         int rc;
3306         ENTRY;
3307
3308         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3309                 RETURN(0);
3310
3311         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3312                 rc = -ENOMEM;
3313
3314         lustre_loa_init(loa, fid, compat, incompat);
3315         lustre_loa_swab(loa, false);
3316
3317         /* For the OST device with 256 bytes inode size by default,
3318          * the PFID EA will be stored together with LMA EA to avoid
3319          * performance trouble. Otherwise the PFID EA can be stored
3320          * independently. LU-8998 */
3321         if ((compat & LMAC_FID_ON_OST) &&
3322             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3323                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3324                                      sizeof(*loa), XATTR_CREATE);
3325         else
3326                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3327                                      sizeof(*lma), XATTR_CREATE);
3328         /* LMA may already exist, but we need to check that all the
3329          * desired compat/incompat flags have been added. */
3330         if (unlikely(rc == -EEXIST)) {
3331                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3332                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3333                 if (rc < 0)
3334                         RETURN(rc);
3335
3336                 if (rc < sizeof(*lma))
3337                         RETURN(-EINVAL);
3338
3339                 lustre_loa_swab(loa, true);
3340                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3341                     ((compat == 0 && incompat == 0) ||
3342                      (!(~lma->lma_compat & compat) &&
3343                       !(~lma->lma_incompat & incompat))))
3344                         RETURN(0);
3345
3346                 lma->lma_self_fid = *fid;
3347                 lma->lma_compat |= compat;
3348                 lma->lma_incompat |= incompat;
3349                 if (rc == sizeof(*lma)) {
3350                         lustre_lma_swab(lma);
3351                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3352                                              sizeof(*lma), XATTR_REPLACE);
3353                 } else {
3354                         lustre_loa_swab(loa, false);
3355                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3356                                              sizeof(*loa), XATTR_REPLACE);
3357                 }
3358         }
3359
3360         RETURN(rc);
3361 }
3362
3363 /**
3364  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3365  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3366  * To have compatilibility with 1.8 ldiskfs driver we need to have
3367  * magic number at start of fid data.
3368  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3369  * its inmemory API.
3370  */
3371 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3372                                          const struct lu_fid *fid)
3373 {
3374         if (!fid_is_namespace_visible(fid) ||
3375             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3376                 param->edp_magic = 0;
3377                 return;
3378         }
3379
3380         param->edp_magic = LDISKFS_LUFID_MAGIC;
3381         param->edp_len =  sizeof(struct lu_fid) + 1;
3382         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3383 }
3384
3385 /**
3386  * Try to read the fid from inode ea into dt_rec.
3387  *
3388  * \param fid object fid.
3389  *
3390  * \retval 0 on success
3391  */
3392 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3393                           __u32 ino, struct lu_fid *fid,
3394                           struct osd_inode_id *id)
3395 {
3396         struct osd_thread_info *info  = osd_oti_get(env);
3397         struct inode           *inode;
3398         ENTRY;
3399
3400         osd_id_gen(id, ino, OSD_OII_NOGEN);
3401         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3402         if (IS_ERR(inode))
3403                 RETURN(PTR_ERR(inode));
3404
3405         iput(inode);
3406         RETURN(0);
3407 }
3408
3409 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3410                                         struct inode *dir,
3411                                         struct inode *parent_dir,
3412                                         const struct lu_fid *dot_fid,
3413                                         const struct lu_fid *dot_dot_fid,
3414                                         struct osd_thandle *oth)
3415 {
3416         struct ldiskfs_dentry_param *dot_ldp;
3417         struct ldiskfs_dentry_param *dot_dot_ldp;
3418         __u32 saved_nlink = dir->i_nlink;
3419         int rc;
3420
3421         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3422         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3423
3424         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3425         dot_ldp->edp_magic = 0;
3426
3427         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3428                                     dir, dot_ldp, dot_dot_ldp);
3429         /* The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3430          * the subseqent ref_add() will increase the dir->i_nlink
3431          * as 3. That is incorrect for new created directory.
3432          *
3433          * It looks like hack, because we want to make the OSD API
3434          * to be order-independent for new created directory object
3435          * between dt_insert(..) and ref_add() operations.
3436          *
3437          * Here, we only restore the in-RAM dir-inode's nlink attr,
3438          * becuase if the nlink attr is not 2, then there will be
3439          * ref_add() called following the dt_insert(..), such call
3440          * will make both the in-RAM and on-disk dir-inode's nlink
3441          * attr to be set as 2. LU-7447 */
3442         set_nlink(dir, saved_nlink);
3443         return rc;
3444 }
3445
3446 /**
3447  * Create an local agent inode for remote entry
3448  */
3449 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3450                                                   struct osd_device *osd,
3451                                                   struct osd_object *pobj,
3452                                                   const struct lu_fid *fid,
3453                                                   __u32 type,
3454                                                   struct thandle *th)
3455 {
3456         struct osd_thread_info  *info = osd_oti_get(env);
3457         struct inode            *local;
3458         struct osd_thandle      *oh;
3459         int                     rc;
3460         ENTRY;
3461
3462         LASSERT(th);
3463         oh = container_of(th, struct osd_thandle, ot_super);
3464         LASSERT(oh->ot_handle->h_transaction != NULL);
3465
3466         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
3467         if (IS_ERR(local)) {
3468                 CERROR("%s: create local error %d\n", osd_name(osd),
3469                        (int)PTR_ERR(local));
3470                 RETURN(local);
3471         }
3472
3473         /* restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3474          * correct gid on remote file, not agent here */
3475         local->i_gid = current_fsgid();
3476         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3477         unlock_new_inode(local);
3478
3479         /* Agent inode should not have project ID */
3480 #ifdef  HAVE_PROJECT_QUOTA
3481         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3482             i_projid_read(pobj->oo_inode) != 0) {
3483                 rc = ldiskfs_transfer_project(local, 0);
3484                 if (rc) {
3485                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3486                                "quota enforcement enabled on the ldiskfs "
3487                                "filesystem?\n", local->i_sb->s_id, rc);
3488                         RETURN(ERR_PTR(rc));
3489                 }
3490         }
3491 #endif
3492         /* Set special LMA flag for local agent inode */
3493         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3494         if (rc != 0) {
3495                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3496                        osd_name(osd), PFID(fid), rc);
3497                 RETURN(ERR_PTR(rc));
3498         }
3499
3500         if (!S_ISDIR(type))
3501                 RETURN(local);
3502
3503         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3504                                          lu_object_fid(&pobj->oo_dt.do_lu),
3505                                          fid, oh);
3506         if (rc != 0) {
3507                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3508                         osd_name(osd), PFID(fid), rc);
3509                 RETURN(ERR_PTR(rc));
3510         }
3511
3512         RETURN(local);
3513 }
3514
3515 /**
3516  * when direntry is deleted, we have to take care of possible agent inode
3517  * referenced by that. unfortunately we can't do this at that point:
3518  * iget() within a running transaction leads to deadlock and we better do
3519  * not call that every delete declaration to save performance. so we put
3520  * a potention agent inode on a list and process that once the transaction
3521  * is over. Notice it's not any worse in terms of real orphans as regular
3522  * object destroy doesn't put inodes on the on-disk orphan list. this should
3523  * be addressed separately
3524  */
3525 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3526                                             struct osd_thandle *oh,
3527                                             __u32 ino)
3528 {
3529         struct osd_device      *osd = osd_dt_dev(oh->ot_super.th_dev);
3530         struct osd_obj_orphan *oor;
3531
3532         OBD_ALLOC_PTR(oor);
3533         if (oor == NULL)
3534                 return -ENOMEM;
3535
3536         oor->oor_ino = ino;
3537         oor->oor_env = (struct lu_env *)env;
3538         spin_lock(&osd->od_osfs_lock);
3539         list_add(&oor->oor_list, &osd->od_orphan_list);
3540         spin_unlock(&osd->od_osfs_lock);
3541
3542         oh->ot_remove_agents = 1;
3543
3544         return 0;
3545
3546 }
3547
3548 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3549                                                 struct osd_device *osd)
3550 {
3551         struct osd_thread_info *info = osd_oti_get(env);
3552         struct osd_obj_orphan *oor, *tmp;
3553         struct osd_inode_id id;
3554         struct list_head list;
3555         struct inode *inode;
3556         struct lu_fid fid;
3557         handle_t *jh;
3558         __u32 ino;
3559
3560         INIT_LIST_HEAD(&list);
3561
3562         spin_lock(&osd->od_osfs_lock);
3563         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3564                 if (oor->oor_env == env) {
3565                         list_del(&oor->oor_list);
3566                         list_add(&oor->oor_list, &list);
3567                 }
3568         }
3569         spin_unlock(&osd->od_osfs_lock);
3570
3571         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3572
3573                 ino = oor->oor_ino;
3574
3575                 list_del(&oor->oor_list);
3576                 OBD_FREE_PTR(oor);
3577
3578                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3579                 inode = osd_iget_fid(info, osd, &id, &fid);
3580                 if (IS_ERR(inode))
3581                         continue;
3582
3583                 if (!osd_remote_fid(env, osd, &fid)) {
3584                         iput(inode);
3585                         continue;
3586                 }
3587
3588                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3589                 clear_nlink(inode);
3590                 mark_inode_dirty(inode);
3591                 ldiskfs_journal_stop(jh);
3592                 iput(inode);
3593         }
3594
3595         return 0;
3596 }
3597
3598 /**
3599  * OSD layer object create function for OST objects (b=11826).
3600  *
3601  * The FID is inserted into inode xattr here.
3602  *
3603  * \retval   0, on success
3604  * \retval -ve, on error
3605  */
3606 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3607                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3608                       struct dt_object_format *dof, struct thandle *th)
3609 {
3610         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3611         struct osd_object       *obj    = osd_dt_obj(dt);
3612         struct osd_thread_info  *info   = osd_oti_get(env);
3613         int                      result, on_ost = 0;
3614
3615         ENTRY;
3616
3617         if (dt_object_exists(dt))
3618                 RETURN(-EEXIST);
3619
3620         LINVRNT(osd_invariant(obj));
3621         LASSERT(!dt_object_remote(dt));
3622         LASSERT(osd_is_write_locked(env, obj));
3623         LASSERT(th != NULL);
3624
3625         if (unlikely(fid_is_acct(fid)))
3626                 /* Quota files can't be created from the kernel any more,
3627                  * 'tune2fs -O quota' will take care of creating them */
3628                 RETURN(-EPERM);
3629
3630         result = __osd_create(info, obj, attr, hint, dof, th);
3631         if (result == 0) {
3632                 if (fid_is_idif(fid) &&
3633                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3634                         struct lu_fid *tfid = &info->oti_fid;
3635                         struct ost_id *oi   = &info->oti_ostid;
3636
3637                         fid_to_ostid(fid, oi);
3638                         ostid_to_fid(tfid, oi, 0);
3639                         on_ost = 1;
3640                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3641                                                 LMAC_FID_ON_OST, 0);
3642                 } else {
3643                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3644                                                fid, OI_CHECK_FLD);
3645                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3646                                                 on_ost ? LMAC_FID_ON_OST : 0,
3647                                                 0);
3648                 }
3649                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3650                         obj->oo_dt.do_body_ops = &osd_body_ops;
3651         }
3652
3653         if (result == 0)
3654                 result = __osd_oi_insert(env, obj, fid, th);
3655
3656         /* a small optimization - dt_insert() isn't usually applied
3657          * to OST objects, so we don't need to cache OI mapping for
3658          * OST objects */
3659         if (result == 0 && on_ost == 0) {
3660                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3661                 result = osd_idc_find_and_init(env, osd, obj);
3662                 LASSERT(result == 0);
3663         }
3664
3665         LASSERT(ergo(result == 0,
3666                      dt_object_exists(dt) && !dt_object_remote(dt)));
3667         LINVRNT(osd_invariant(obj));
3668         RETURN(result);
3669 }
3670
3671 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
3672                                struct thandle *handle)
3673 {
3674         struct osd_thandle *oh;
3675
3676         /* it's possible that object doesn't exist yet */
3677         LASSERT(handle != NULL);
3678
3679         oh = container_of0(handle, struct osd_thandle, ot_super);
3680         LASSERT(oh->ot_handle == NULL);
3681
3682         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
3683                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3684
3685         return 0;
3686 }
3687
3688 /*
3689  * Concurrency: @dt is write locked.
3690  */
3691 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
3692                        struct thandle *th)
3693 {
3694         struct osd_object  *obj = osd_dt_obj(dt);
3695         struct inode       *inode = obj->oo_inode;
3696         struct osd_thandle *oh;
3697         int                 rc = 0;
3698
3699         if (!dt_object_exists(dt) || obj->oo_destroyed)
3700                 return -ENOENT;
3701
3702         LINVRNT(osd_invariant(obj));
3703         LASSERT(!dt_object_remote(dt));
3704         LASSERT(osd_is_write_locked(env, obj));
3705         LASSERT(th != NULL);
3706
3707         oh = container_of0(th, struct osd_thandle, ot_super);
3708         LASSERT(oh->ot_handle != NULL);
3709
3710         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
3711
3712         CDEBUG(D_INODE, DFID" increase nlink %d\n",
3713                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3714         /*
3715          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
3716          * (65000) subdirectories by storing "1" in i_nlink if the link count
3717          * would otherwise overflow. Directory tranversal tools understand
3718          * that (st_nlink == 1) indicates that the filesystem dose not track
3719          * hard links count on the directory, and will not abort subdirectory
3720          * scanning early once (st_nlink - 2) subdirs have been found.
3721          *
3722          * This also has to properly handle the case of inodes with nlink == 0
3723          * in case they are being linked into the PENDING directory
3724          */
3725         spin_lock(&obj->oo_guard);
3726         if (unlikely(inode->i_nlink == 0))
3727                 /* inc_nlink from 0 may cause WARN_ON */
3728                 set_nlink(inode, 1);
3729         else {
3730                 ldiskfs_inc_count(oh->ot_handle, inode);
3731                 if (!S_ISDIR(inode->i_mode))
3732                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
3733         }
3734         spin_unlock(&obj->oo_guard);
3735
3736         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3737         LINVRNT(osd_invariant(obj));
3738
3739         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
3740
3741         return rc;
3742 }
3743
3744 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
3745                                struct thandle *handle)
3746 {
3747         struct osd_thandle *oh;
3748
3749         if (!dt_object_exists(dt))
3750                 return -ENOENT;
3751
3752         LASSERT(!dt_object_remote(dt));
3753         LASSERT(handle != NULL);
3754
3755         oh = container_of0(handle, struct osd_thandle, ot_super);
3756         LASSERT(oh->ot_handle == NULL);
3757
3758         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
3759                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
3760
3761         return 0;
3762 }
3763
3764 /*
3765  * Concurrency: @dt is write locked.
3766  */
3767 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
3768                        struct thandle *th)
3769 {
3770         struct osd_object       *obj = osd_dt_obj(dt);
3771         struct inode            *inode = obj->oo_inode;
3772         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3773         struct osd_thandle      *oh;
3774
3775         if (!dt_object_exists(dt))
3776                 return -ENOENT;
3777
3778         LINVRNT(osd_invariant(obj));
3779         LASSERT(!dt_object_remote(dt));
3780         LASSERT(osd_is_write_locked(env, obj));
3781         LASSERT(th != NULL);
3782
3783         oh = container_of0(th, struct osd_thandle, ot_super);
3784         LASSERT(oh->ot_handle != NULL);
3785
3786         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
3787
3788         spin_lock(&obj->oo_guard);
3789         /* That can be result of upgrade from old Lustre version and
3790          * applied only to local files.  Just skip this ref_del call.
3791          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
3792         if (inode->i_nlink == 0) {
3793                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
3794                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
3795                              ", maybe an upgraded file? (LU-3915)\n",
3796                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
3797                 spin_unlock(&obj->oo_guard);
3798                 return 0;
3799         }
3800
3801         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
3802                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
3803
3804         ldiskfs_dec_count(oh->ot_handle, inode);
3805         spin_unlock(&obj->oo_guard);
3806
3807         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3808         LINVRNT(osd_invariant(obj));
3809
3810         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
3811
3812         return 0;
3813 }
3814
3815 /*
3816  * Concurrency: @dt is read locked.
3817  */
3818 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
3819                          struct lu_buf *buf, const char *name)
3820 {
3821         struct osd_object      *obj    = osd_dt_obj(dt);
3822         struct inode           *inode  = obj->oo_inode;
3823         struct osd_thread_info *info   = osd_oti_get(env);
3824         struct dentry          *dentry = &info->oti_obj_dentry;
3825         bool                    cache_xattr = false;
3826         int                     rc;
3827
3828         LASSERT(buf);
3829
3830         /* version get is not real XATTR but uses xattr API */
3831         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3832                 dt_obj_version_t *ver = buf->lb_buf;
3833
3834                 /* for version we are just using xattr API but change inode
3835                  * field instead */
3836                 if (buf->lb_len == 0)
3837                         return sizeof(dt_obj_version_t);
3838
3839                 if (buf->lb_len < sizeof(dt_obj_version_t))
3840                         return -ERANGE;
3841
3842                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
3843                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
3844
3845                 *ver = LDISKFS_I(inode)->i_fs_version;
3846
3847                 return sizeof(dt_obj_version_t);
3848         }
3849
3850         if (!dt_object_exists(dt))
3851                 return -ENOENT;
3852
3853         LASSERT(!dt_object_remote(dt));
3854         LASSERT(inode->i_op != NULL);
3855         LASSERT(inode->i_op->getxattr != NULL);
3856
3857         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
3858             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
3859                 cache_xattr = true;
3860
3861         if (cache_xattr) {
3862                 rc = osd_oxc_get(obj, name, buf);
3863                 if (rc != -ENOENT)
3864                         return rc;
3865         }
3866
3867         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
3868                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3869                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
3870                 struct filter_fid *ff;
3871                 struct ost_layout *ol;
3872
3873                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
3874                 if (rc)
3875                         return rc;
3876
3877                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
3878
3879                 rc = sizeof(*ff);
3880                 if (buf->lb_len == 0 || !buf->lb_buf)
3881                         return rc;
3882
3883                 if (buf->lb_len < rc)
3884                         return -ERANGE;
3885
3886                 ff = buf->lb_buf;
3887                 ol = &ff->ff_layout;
3888                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
3889                                                   PFID_STRIPE_IDX_BITS);
3890                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
3891                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
3892                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
3893                 if (lma->lma_compat & LMAC_COMP_INFO) {
3894                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
3895                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
3896                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
3897                 } else {
3898                         ol->ol_comp_start = 0;
3899                         ol->ol_comp_end = 0;
3900                         ol->ol_comp_id = 0;
3901                 }
3902         } else {
3903                 rc = __osd_xattr_get(inode, dentry, name,
3904                                      buf->lb_buf, buf->lb_len);
3905         }
3906
3907         if (cache_xattr) {
3908                 if (rc == -ENOENT || rc == -ENODATA)
3909                         osd_oxc_add(obj, name, NULL, 0);
3910                 else if (rc > 0 && buf->lb_buf != NULL)
3911                         osd_oxc_add(obj, name, buf->lb_buf, rc);
3912         }
3913
3914         return rc;
3915 }
3916
3917 static int osd_declare_xattr_set(const struct lu_env *env,
3918                                  struct dt_object *dt,
3919                                  const struct lu_buf *buf, const char *name,
3920                                  int fl, struct thandle *handle)
3921 {
3922         struct osd_thandle *oh;
3923         int credits = 0;
3924         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
3925
3926         LASSERT(handle != NULL);
3927
3928         oh = container_of0(handle, struct osd_thandle, ot_super);
3929         LASSERT(oh->ot_handle == NULL);
3930
3931         if (strcmp(name, XATTR_NAME_LMA) == 0) {
3932                 /* For non-upgrading case, the LMA is set first and
3933                  * usually fit inode. But for upgrade case, the LMA
3934                  * may be in another separated EA block. */
3935                 if (dt_object_exists(dt)) {
3936                         if (fl == LU_XATTR_REPLACE)
3937                                 credits = 1;
3938                         else
3939                                 goto upgrade;
3940                 }
3941         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3942                 credits = 1;
3943         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
3944                 /* We may need to delete the old PFID EA. */
3945                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
3946                 if (fl == LU_XATTR_REPLACE)
3947                         credits += 1;
3948                 else
3949                         goto upgrade;
3950         } else {
3951                 /* If some name entry resides on remote MDT, then will create
3952                  * agent entry under remote parent. On the other hand, if the
3953                  * remote entry will be removed, then related agent entry may
3954                  * need to be removed from the remote parent. So there may be
3955                  * kinds of cases, let's declare enough credits. The credits
3956                  * for create agent entry is enough for remove case. */
3957                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
3958                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3959                         if (dt_object_exists(dt))
3960                                 credits += 1; /* For updating LMA */
3961                 }
3962
3963 upgrade:
3964                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
3965
3966                 if (buf != NULL) {
3967                         ssize_t buflen;
3968
3969                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
3970                                 /* learn xattr size from osd_xattr_get if
3971                                    attribute has not been read yet */
3972                                 buflen = __osd_xattr_get(
3973                                     osd_dt_obj(dt)->oo_inode,
3974                                     &osd_oti_get(env)->oti_obj_dentry,
3975                                     name, NULL, 0);
3976                                 if (buflen < 0)
3977                                         buflen = 0;
3978                         } else {
3979                                 buflen = buf->lb_len;
3980                         }
3981
3982                         if (buflen > sb->s_blocksize) {
3983                                 credits += osd_calc_bkmap_credits(
3984                                     sb, NULL, 0, -1,
3985                                     (buflen + sb->s_blocksize - 1) >>
3986                                     sb->s_blocksize_bits);
3987                         }
3988                 }
3989                 /*
3990                  * xattr set may involve inode quota change, reserve credits for
3991                  * dquot_initialize()
3992                  */
3993                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3994         }
3995
3996         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
3997
3998         return 0;
3999 }
4000
4001 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4002                               const struct lu_buf *buf, int fl,
4003                               struct thandle *handle)
4004 {
4005         struct osd_thread_info *info = osd_oti_get(env);
4006         struct dentry *dentry = &info->oti_obj_dentry;
4007         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4008         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4009         struct inode *inode = obj->oo_inode;
4010         struct filter_fid *ff = buf->lb_buf;
4011         struct ost_layout *ol = &ff->ff_layout;
4012         int flags = XATTR_REPLACE;
4013         int rc;
4014         ENTRY;
4015
4016         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4017                 RETURN(-EINVAL);
4018
4019         rc = osd_get_lma(info, inode, dentry, loa);
4020         if (rc == -ENODATA) {
4021                 /* Usually for upgarding from old device */
4022                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4023                                 LMAC_FID_ON_OST, 0);
4024                 flags = XATTR_CREATE;
4025         } else if (rc) {
4026                 RETURN(rc);
4027         }
4028
4029         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4030                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4031                         RETURN(-EEXIST);
4032
4033                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4034                         /* Separate PFID EA from LMA */
4035                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4036                         lustre_lma_swab(lma);
4037                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4038                                              sizeof(*lma), XATTR_REPLACE);
4039                         if (!rc) {
4040                                 obj->oo_pfid_in_lma = 0;
4041                                 rc = LU_XATTR_CREATE;
4042                         }
4043
4044                         RETURN(rc);
4045                 }
4046         } else {
4047                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4048                         RETURN(fl);
4049
4050                 /* Old client does not send stripe information,
4051                  * then store the PFID EA on disk separatedly. */
4052                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4053                              ol->ol_stripe_size == 0))
4054                         RETURN(fl);
4055
4056                 /* Remove old PFID EA entry firstly. */
4057                 ll_vfs_dq_init(inode);
4058                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
4059                 if (rc == -ENODATA) {
4060                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4061                                 RETURN(rc);
4062                 } else if (rc) {
4063                         RETURN(rc);
4064                 }
4065         }
4066
4067         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4068         if (likely(ol->ol_stripe_size != 0)) {
4069                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4070                                              PFID_STRIPE_IDX_BITS;
4071                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4072                 lma->lma_compat |= LMAC_STRIPE_INFO;
4073                 if (ol->ol_comp_id != 0) {
4074                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4075                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4076                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4077                         lma->lma_compat |= LMAC_COMP_INFO;
4078                 }
4079         }
4080
4081         lustre_loa_swab(loa, false);
4082
4083         /* Store the PFID EA inside LMA. */
4084         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4085                              flags);
4086         if (!rc)
4087                 obj->oo_pfid_in_lma = 1;
4088
4089         RETURN(rc);
4090 }
4091
4092 /*
4093  * In DNE environment, the object (in spite of regular file or directory)
4094  * and its name entry may reside on different MDTs. Under such case, we will
4095  * create an agent entry on the MDT where the object resides. The agent entry
4096  * references the object locally, that makes the object to be visible to the
4097  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4098  * such as 'tar' can handle the object properly.
4099  *
4100  * We handle the agent entry during set linkEA that is the common interface
4101  * for both regular file and directroy, can handle kinds of cases, such as
4102  * create/link/unlink/rename, and so on.
4103  *
4104  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4105  *
4106  * XXX: There are two known issues:
4107  * 1. For one object, we will create at most one agent entry even if there
4108  *    may be more than one cross-MDTs hard links on the object. So the local
4109  *    e2fsck may claim that the object's nlink is larger than the name entries
4110  *    that reference such inode. And in further, the e2fsck will fix the nlink
4111  *    attribute to match the local references. Then it will cause the object's
4112  *    nlink attribute to be inconsistent with the global references. it is bad
4113  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4114  *    other hand, the global namespace LFSCK can repair the object's attribute
4115  *    according to the linkEA.
4116  * 2. There may be too many hard links on the object as to its linkEA overflow,
4117  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4118  *    case happened, then at this point, we do not know whether there are some
4119  *    cross-MDTs reference. But there are local references, it guarantees that
4120  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4121  */
4122 static int osd_xattr_handle_linkea(const struct lu_env *env,
4123                                    struct osd_device *osd,
4124                                    struct osd_object *obj,
4125                                    const struct lu_buf *buf,
4126                                    struct thandle *handle)
4127 {
4128         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4129         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4130         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4131         struct lu_name tmpname;
4132         struct osd_thandle *oh;
4133         int rc;
4134         bool remote = false;
4135         ENTRY;
4136
4137         oh = container_of0(handle, struct osd_thandle, ot_super);
4138         LASSERT(oh->ot_handle != NULL);
4139
4140         rc = linkea_init_with_rec(&ldata);
4141         if (!rc) {
4142                 linkea_first_entry(&ldata);
4143                 while (ldata.ld_lee != NULL && !remote) {
4144                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4145                                             &tmpname, tfid);
4146                         if (osd_remote_fid(env, osd, tfid) > 0)
4147                                 remote = true;
4148                         else
4149                                 linkea_next_entry(&ldata);
4150                 }
4151         } else if (rc == -ENODATA) {
4152                 rc = 0;
4153         } else {
4154                 RETURN(rc);
4155         }
4156
4157         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4158                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4159                 if (rc)
4160                         CERROR("%s: failed to remove agent entry for "DFID
4161                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4162         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4163                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4164                 if (rc)
4165                         CERROR("%s: failed to create agent entry for "DFID
4166                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4167         }
4168
4169         RETURN(rc);
4170 }
4171
4172 /*
4173  * Concurrency: @dt is write locked.
4174  */
4175 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4176                          const struct lu_buf *buf, const char *name, int fl,
4177                          struct thandle *handle)
4178 {
4179         struct osd_object *obj = osd_dt_obj(dt);
4180         struct osd_device *osd = osd_obj2dev(obj);
4181         struct inode *inode = obj->oo_inode;
4182         struct osd_thread_info *info = osd_oti_get(env);
4183         int fs_flags = 0;
4184         int len;
4185         int rc;
4186         ENTRY;
4187
4188         LASSERT(handle);
4189         LASSERT(buf);
4190
4191         /* version set is not real XATTR */
4192         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4193                 dt_obj_version_t *version = buf->lb_buf;
4194
4195                 /* for version we are just using xattr API but change inode
4196                  * field instead */
4197                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4198
4199                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4200                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4201
4202                 LDISKFS_I(inode)->i_fs_version = *version;
4203                 /* Version is set after all inode operations are finished,
4204                  * so we should mark it dirty here */
4205                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4206
4207                 RETURN(0);
4208         }
4209
4210         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4211                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4212
4213         len = buf->lb_len;
4214         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4215
4216         /* For the OST device with 256 bytes inode size by default,
4217          * the PFID EA will be stored together with LMA EA to avoid
4218          * performance trouble. Otherwise the PFID EA can be stored
4219          * independently. LU-8998 */
4220         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4221             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4222                 LASSERT(buf->lb_buf);
4223
4224                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4225                 if (fl <= 0)
4226                         RETURN(fl);
4227         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4228                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4229                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4230
4231                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4232                 if (rc)
4233                         RETURN(rc);
4234
4235                 lma->lma_incompat |= LMAI_STRIPED;
4236                 lustre_lma_swab(lma);
4237                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4238                                      sizeof(*lma), XATTR_REPLACE);
4239                 if (rc != 0)
4240                         RETURN(rc);
4241         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4242                 LASSERT(!osd->od_is_ost);
4243
4244                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4245                 if (rc)
4246                         RETURN(rc);
4247         }
4248
4249         if (fl & LU_XATTR_REPLACE)
4250                 fs_flags |= XATTR_REPLACE;
4251
4252         if (fl & LU_XATTR_CREATE)
4253                 fs_flags |= XATTR_CREATE;
4254
4255         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4256         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4257
4258         if (rc == 0 &&
4259             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4260              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4261                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4262
4263         return rc;
4264 }
4265
4266 /*
4267  * Concurrency: @dt is read locked.
4268  */
4269 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4270                           const struct lu_buf *buf)
4271 {
4272         struct osd_object      *obj    = osd_dt_obj(dt);
4273         struct inode           *inode  = obj->oo_inode;
4274         struct osd_thread_info *info   = osd_oti_get(env);
4275         struct dentry          *dentry = &info->oti_obj_dentry;
4276
4277         if (!dt_object_exists(dt))
4278                 return -ENOENT;
4279
4280         LASSERT(!dt_object_remote(dt));
4281         LASSERT(inode->i_op != NULL);
4282         LASSERT(inode->i_op->listxattr != NULL);
4283
4284         dentry->d_inode = inode;
4285         dentry->d_sb = inode->i_sb;
4286         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4287 }
4288
4289 static int osd_declare_xattr_del(const struct lu_env *env,
4290                                  struct dt_object *dt, const char *name,
4291                                  struct thandle *handle)
4292 {
4293         struct osd_thandle *oh;
4294         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4295
4296         LASSERT(!dt_object_remote(dt));
4297         LASSERT(handle != NULL);
4298
4299         oh = container_of0(handle, struct osd_thandle, ot_super);
4300         LASSERT(oh->ot_handle == NULL);
4301
4302         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4303                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4304         /*
4305          * xattr del may involve inode quota change, reserve credits for
4306          * dquot_initialize()
4307          */
4308         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4309
4310         return 0;
4311 }
4312
4313 /*
4314  * Concurrency: @dt is write locked.
4315  */
4316 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4317                          const char *name, struct thandle *handle)
4318 {
4319         struct osd_object      *obj    = osd_dt_obj(dt);
4320         struct inode           *inode  = obj->oo_inode;
4321         struct osd_thread_info *info   = osd_oti_get(env);
4322         struct dentry          *dentry = &info->oti_obj_dentry;
4323         int                     rc;
4324
4325         if (!dt_object_exists(dt))
4326                 return -ENOENT;
4327
4328         LASSERT(!dt_object_remote(dt));
4329         LASSERT(inode->i_op != NULL);
4330         LASSERT(inode->i_op->removexattr != NULL);
4331         LASSERT(handle != NULL);
4332
4333         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4334
4335         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4336                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4337
4338                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4339                                  &info->oti_ost_attrs);
4340                 if (!rc) {
4341                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4342
4343                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4344                         lustre_lma_swab(lma);
4345                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4346                                              sizeof(*lma), XATTR_REPLACE);
4347                         if (!rc)
4348                                 obj->oo_pfid_in_lma = 0;
4349                 }
4350         } else {
4351                 ll_vfs_dq_init(inode);
4352                 dentry->d_inode = inode;
4353                 dentry->d_sb = inode->i_sb;
4354                 rc = inode->i_op->removexattr(dentry, name);
4355         }
4356
4357         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4358
4359         if (rc == 0 &&
4360             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4361              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4362                 osd_oxc_del(obj, name);
4363
4364         return rc;
4365 }
4366
4367 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4368                            __u64 start, __u64 end)
4369 {
4370         struct osd_object       *obj    = osd_dt_obj(dt);
4371         struct inode            *inode  = obj->oo_inode;
4372         struct osd_thread_info  *info   = osd_oti_get(env);
4373         struct dentry           *dentry = &info->oti_obj_dentry;
4374         struct file             *file   = &info->oti_file;
4375         int                     rc;
4376
4377         ENTRY;
4378
4379         dentry->d_inode = inode;
4380         dentry->d_sb = inode->i_sb;
4381         file->f_path.dentry = dentry;
4382         file->f_mapping = inode->i_mapping;
4383         file->f_op = inode->i_fop;
4384         set_file_inode(file, inode);
4385
4386         rc = ll_vfs_fsync_range(file, start, end, 0);
4387
4388         RETURN(rc);
4389 }
4390
4391 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4392 {
4393         return 0;
4394 }
4395
4396 /*
4397  * Index operations.
4398  */
4399
4400 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4401                            const struct dt_index_features *feat)
4402 {
4403         struct iam_descr *descr;
4404
4405         if (osd_object_is_root(o))
4406                 return feat == &dt_directory_features;
4407
4408         LASSERT(o->oo_dir != NULL);
4409
4410         descr = o->oo_dir->od_container.ic_descr;
4411         if (feat == &dt_directory_features) {
4412                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4413                         return 1;
4414                 else
4415                         return 0;
4416         } else {
4417                 return
4418                         feat->dif_keysize_min <= descr->id_key_size &&
4419                         descr->id_key_size <= feat->dif_keysize_max &&
4420                         feat->dif_recsize_min <= descr->id_rec_size &&
4421                         descr->id_rec_size <= feat->dif_recsize_max &&
4422                         !(feat->dif_flags & (DT_IND_VARKEY |
4423                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
4424                         ergo(feat->dif_flags & DT_IND_UPDATE,
4425                              1 /* XXX check that object (and file system) is
4426                                 * writable */);
4427         }
4428 }
4429
4430 static int osd_iam_container_init(const struct lu_env *env,
4431                                   struct osd_object *obj,
4432                                   struct osd_directory *dir)
4433 {
4434         struct iam_container *bag = &dir->od_container;
4435         int result;
4436
4437         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4438         if (result != 0)
4439                 return result;
4440
4441         result = iam_container_setup(bag);
4442         if (result == 0)
4443                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4444         else
4445                 iam_container_fini(bag);
4446
4447         return result;
4448 }
4449
4450
4451 /*
4452  * Concurrency: no external locking is necessary.
4453  */
4454 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4455                          const struct dt_index_features *feat)
4456 {
4457         int                      result;
4458         int                      skip_iam = 0;
4459         struct osd_object       *obj = osd_dt_obj(dt);
4460
4461         LINVRNT(osd_invariant(obj));
4462
4463         if (osd_object_is_root(obj)) {
4464                 dt->do_index_ops = &osd_index_ea_ops;
4465                 result = 0;
4466         } else if (feat == &dt_directory_features) {
4467                 dt->do_index_ops = &osd_index_ea_ops;
4468                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4469                         result = 0;
4470                 else
4471                         result = -ENOTDIR;
4472                 skip_iam = 1;
4473         } else if (unlikely(feat == &dt_otable_features)) {
4474                 dt->do_index_ops = &osd_otable_ops;
4475                 return 0;
4476         } else if (unlikely(feat == &dt_acct_features)) {
4477                 dt->do_index_ops = &osd_acct_index_ops;
4478                 result = 0;
4479                 skip_iam = 1;
4480         } else if (!osd_has_index(obj)) {
4481                 struct osd_directory *dir;
4482
4483                 OBD_ALLOC_PTR(dir);
4484                 if (dir != NULL) {
4485
4486                         spin_lock(&obj->oo_guard);
4487                         if (obj->oo_dir == NULL)
4488                                 obj->oo_dir = dir;
4489                         else
4490                                 /*
4491                                  * Concurrent thread allocated container data.
4492                                  */
4493                                 OBD_FREE_PTR(dir);
4494                         spin_unlock(&obj->oo_guard);
4495                         /*
4496                          * Now, that we have container data, serialize its
4497                          * initialization.
4498                          */
4499                         down_write(&obj->oo_ext_idx_sem);
4500                         /*
4501                          * recheck under lock.
4502                          */
4503                         if (!osd_has_index(obj))
4504                                 result = osd_iam_container_init(env, obj,
4505                                                                 obj->oo_dir);
4506                         else
4507                                 result = 0;
4508                         up_write(&obj->oo_ext_idx_sem);
4509                 } else {
4510                         result = -ENOMEM;
4511                 }
4512         } else {
4513                 result = 0;
4514         }
4515
4516         if (result == 0 && skip_iam == 0) {
4517                 if (!osd_iam_index_probe(env, obj, feat))
4518                         result = -ENOTDIR;
4519         }
4520         LINVRNT(osd_invariant(obj));
4521
4522         return result;
4523 }
4524
4525 static int osd_otable_it_attr_get(const struct lu_env *env,
4526                                  struct dt_object *dt,
4527                                  struct lu_attr *attr)
4528 {
4529         attr->la_valid = 0;
4530         return 0;
4531 }
4532
4533 static const struct dt_object_operations osd_obj_ops = {
4534         .do_read_lock           = osd_read_lock,
4535         .do_write_lock          = osd_write_lock,
4536         .do_read_unlock         = osd_read_unlock,
4537         .do_write_unlock        = osd_write_unlock,
4538         .do_write_locked        = osd_write_locked,
4539         .do_attr_get            = osd_attr_get,
4540         .do_declare_attr_set    = osd_declare_attr_set,
4541         .do_attr_set            = osd_attr_set,
4542         .do_ah_init             = osd_ah_init,
4543         .do_declare_create      = osd_declare_create,
4544         .do_create              = osd_create,
4545         .do_declare_destroy     = osd_declare_destroy,
4546         .do_destroy             = osd_destroy,
4547         .do_index_try           = osd_index_try,
4548         .do_declare_ref_add     = osd_declare_ref_add,
4549         .do_ref_add             = osd_ref_add,
4550         .do_declare_ref_del     = osd_declare_ref_del,
4551         .do_ref_del             = osd_ref_del,
4552         .do_xattr_get           = osd_xattr_get,
4553         .do_declare_xattr_set   = osd_declare_xattr_set,
4554         .do_xattr_set           = osd_xattr_set,
4555         .do_declare_xattr_del   = osd_declare_xattr_del,
4556         .do_xattr_del           = osd_xattr_del,
4557         .do_xattr_list          = osd_xattr_list,
4558         .do_object_sync         = osd_object_sync,
4559         .do_invalidate          = osd_invalidate,
4560 };
4561
4562 static const struct dt_object_operations osd_obj_otable_it_ops = {
4563         .do_attr_get    = osd_otable_it_attr_get,
4564         .do_index_try   = osd_index_try,
4565 };
4566
4567 static int osd_index_declare_iam_delete(const struct lu_env *env,
4568                                         struct dt_object *dt,
4569                                         const struct dt_key *key,
4570                                         struct thandle *handle)
4571 {
4572         struct osd_thandle    *oh;
4573
4574         oh = container_of0(handle, struct osd_thandle, ot_super);
4575         LASSERT(oh->ot_handle == NULL);
4576
4577         /* Recycle  may cause additional three blocks to be changed. */
4578         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4579                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4580
4581         return 0;
4582 }
4583
4584 /**
4585  *      delete a (key, value) pair from index \a dt specified by \a key
4586  *
4587  *      \param  dt      osd index object
4588  *      \param  key     key for index
4589  *      \param  rec     record reference
4590  *      \param  handle  transaction handler
4591  *
4592  *      \retval  0  success
4593  *      \retval -ve   failure
4594  */
4595 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4596                                 const struct dt_key *key,
4597                                 struct thandle *handle)
4598 {
4599         struct osd_thread_info *oti = osd_oti_get(env);
4600         struct osd_object      *obj = osd_dt_obj(dt);
4601         struct osd_thandle     *oh;
4602         struct iam_path_descr  *ipd;
4603         struct iam_container   *bag = &obj->oo_dir->od_container;
4604         int                     rc;
4605         ENTRY;
4606
4607         if (!dt_object_exists(dt))
4608                 RETURN(-ENOENT);
4609
4610         LINVRNT(osd_invariant(obj));
4611         LASSERT(!dt_object_remote(dt));
4612         LASSERT(bag->ic_object == obj->oo_inode);
4613         LASSERT(handle != NULL);
4614
4615         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4616
4617         ipd = osd_idx_ipd_get(env, bag);
4618         if (unlikely(ipd == NULL))
4619                 RETURN(-ENOMEM);
4620
4621         oh = container_of0(handle, struct osd_thandle, ot_super);
4622         LASSERT(oh->ot_handle != NULL);
4623         LASSERT(oh->ot_handle->h_transaction != NULL);
4624
4625         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4626                 /* swab quota uid/gid provided by caller */
4627                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4628                 key = (const struct dt_key *)&oti->oti_quota_id;
4629         }
4630
4631         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
4632         osd_ipd_put(env, bag, ipd);
4633         LINVRNT(osd_invariant(obj));
4634         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4635         RETURN(rc);
4636 }
4637
4638 static int osd_index_declare_ea_delete(const struct lu_env *env,
4639                                        struct dt_object *dt,
4640                                        const struct dt_key *key,
4641                                        struct thandle *handle)
4642 {
4643         struct osd_thandle *oh;
4644         struct inode       *inode;
4645         int                 rc, credits;
4646         ENTRY;
4647
4648         LASSERT(!dt_object_remote(dt));
4649         LASSERT(handle != NULL);
4650
4651         oh = container_of0(handle, struct osd_thandle, ot_super);
4652         LASSERT(oh->ot_handle == NULL);
4653
4654         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
4655         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
4656
4657         inode = osd_dt_obj(dt)->oo_inode;
4658         if (inode == NULL)
4659                 RETURN(-ENOENT);
4660
4661         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
4662                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
4663                                    NULL, OSD_QID_BLK);
4664         RETURN(rc);
4665 }
4666
4667 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
4668                                           struct dt_rec *fid)
4669 {
4670         struct osd_fid_pack *rec;
4671         int                  rc = -ENODATA;
4672
4673         if (de->file_type & LDISKFS_DIRENT_LUFID) {
4674                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
4675                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
4676                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
4677                         rc = -EINVAL;
4678         }
4679         return rc;
4680 }
4681
4682 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
4683                           const struct lu_fid *fid)
4684 {
4685         struct seq_server_site  *ss = osd_seq_site(osd);
4686         ENTRY;
4687
4688         /* FID seqs not in FLDB, must be local seq */
4689         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
4690                 RETURN(0);
4691
4692         /* If FLD is not being initialized yet, it only happens during the
4693          * initialization, likely during mgs initialization, and we assume
4694          * this is local FID. */
4695         if (ss == NULL || ss->ss_server_fld == NULL)
4696                 RETURN(0);
4697
4698         /* Only check the local FLDB here */
4699         if (osd_seq_exists(env, osd, fid_seq(fid)))
4700                 RETURN(0);
4701
4702         RETURN(1);
4703 }
4704
4705 /**
4706  * Index delete function for interoperability mode (b11826).
4707  * It will remove the directory entry added by osd_index_ea_insert().
4708  * This entry is needed to maintain name->fid mapping.
4709  *
4710  * \param key,  key i.e. file entry to be deleted
4711  *
4712  * \retval   0, on success
4713  * \retval -ve, on error
4714  */
4715 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
4716                                const struct dt_key *key, struct thandle *handle)
4717 {
4718         struct osd_object          *obj = osd_dt_obj(dt);
4719         struct inode               *dir = obj->oo_inode;
4720         struct dentry              *dentry;
4721         struct osd_thandle         *oh;
4722         struct ldiskfs_dir_entry_2 *de = NULL;
4723         struct buffer_head         *bh;
4724         struct htree_lock          *hlock = NULL;
4725         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
4726         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
4727         int                        rc;
4728         ENTRY;
4729
4730         if (!dt_object_exists(dt))
4731                 RETURN(-ENOENT);
4732
4733         LINVRNT(osd_invariant(obj));
4734         LASSERT(!dt_object_remote(dt));
4735         LASSERT(handle != NULL);
4736
4737         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4738
4739         oh = container_of(handle, struct osd_thandle, ot_super);
4740         LASSERT(oh->ot_handle != NULL);
4741         LASSERT(oh->ot_handle->h_transaction != NULL);
4742
4743         ll_vfs_dq_init(dir);
4744         dentry = osd_child_dentry_get(env, obj,
4745                                       (char *)key, strlen((char *)key));
4746
4747         if (obj->oo_hl_head != NULL) {
4748                 hlock = osd_oti_get(env)->oti_hlock;
4749                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4750                                    dir, LDISKFS_HLOCK_DEL);
4751         } else {
4752                 down_write(&obj->oo_ext_idx_sem);
4753         }
4754
4755         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4756         if (!IS_ERR(bh)) {
4757                 /* If this is not the ".." entry, it might be a remote DNE
4758                  * entry and  we need to check if the FID is for a remote
4759                  * MDT.  If the FID is  not in the directory entry (e.g.
4760                  * upgraded 1.8 filesystem without dirdata enabled) then
4761                  * we need to get the FID from the LMA. For a remote directory
4762                  * there HAS to be an LMA, it cannot be an IGIF inode in this
4763                  * case.
4764                  *
4765                  * Delete the entry before the agent inode in order to
4766                  * simplify error handling.  At worst an error after deleting
4767                  * the entry first might leak the agent inode afterward. The
4768                  * reverse would need filesystem abort in case of error deleting
4769                  * the entry after the agent had been removed, or leave a
4770                  * dangling entry pointing at a random inode. */
4771                 if (strcmp((char *)key, dotdot) != 0) {
4772                         LASSERT(de != NULL);
4773                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
4774                         if (rc == -ENODATA) {
4775                                 /* can't get FID, postpone to the end of the
4776                                  * transaction when iget() is safe */
4777                                 osd_schedule_agent_inode_removal(env, oh,
4778                                                 le32_to_cpu(de->inode));
4779                         } else if (rc == 0 &&
4780                                    unlikely(osd_remote_fid(env, osd, fid))) {
4781                                 osd_schedule_agent_inode_removal(env, oh,
4782                                                 le32_to_cpu(de->inode));
4783                         }
4784                 }
4785                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
4786                 brelse(bh);
4787         } else {
4788                 rc = PTR_ERR(bh);
4789         }
4790         if (hlock != NULL)
4791                 ldiskfs_htree_unlock(hlock);
4792         else
4793                 up_write(&obj->oo_ext_idx_sem);
4794
4795         GOTO(out, rc);
4796
4797 out:
4798         LASSERT(osd_invariant(obj));
4799         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
4800         RETURN(rc);
4801 }
4802
4803 /**
4804  *      Lookup index for \a key and copy record to \a rec.
4805  *
4806  *      \param  dt      osd index object
4807  *      \param  key     key for index
4808  *      \param  rec     record reference
4809  *
4810  *      \retval  +ve  success : exact mach
4811  *      \retval  0    return record with key not greater than \a key
4812  *      \retval -ve   failure
4813  */
4814 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
4815                                 struct dt_rec *rec, const struct dt_key *key)
4816 {
4817         struct osd_object      *obj = osd_dt_obj(dt);
4818         struct iam_path_descr  *ipd;
4819         struct iam_container   *bag = &obj->oo_dir->od_container;
4820         struct osd_thread_info *oti = osd_oti_get(env);
4821         struct iam_iterator    *it = &oti->oti_idx_it;
4822         struct iam_rec         *iam_rec;
4823         int                     rc;
4824         ENTRY;
4825
4826         if (!dt_object_exists(dt))
4827                 RETURN(-ENOENT);
4828
4829         LASSERT(osd_invariant(obj));
4830         LASSERT(!dt_object_remote(dt));
4831         LASSERT(bag->ic_object == obj->oo_inode);
4832
4833         ipd = osd_idx_ipd_get(env, bag);
4834         if (IS_ERR(ipd))
4835                 RETURN(-ENOMEM);
4836
4837         /* got ipd now we can start iterator. */
4838         iam_it_init(it, bag, 0, ipd);
4839
4840         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4841                 /* swab quota uid/gid provided by caller */
4842                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4843                 key = (const struct dt_key *)&oti->oti_quota_id;
4844         }
4845
4846         rc = iam_it_get(it, (struct iam_key *)key);
4847         if (rc >= 0) {
4848                 if (S_ISDIR(obj->oo_inode->i_mode))
4849                         iam_rec = (struct iam_rec *)oti->oti_ldp;
4850                 else
4851                         iam_rec = (struct iam_rec *) rec;
4852
4853                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
4854
4855                 if (S_ISDIR(obj->oo_inode->i_mode))
4856                         osd_fid_unpack((struct lu_fid *) rec,
4857                                        (struct osd_fid_pack *)iam_rec);
4858                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
4859                         osd_quota_unpack(obj, rec);
4860         }
4861
4862         iam_it_put(it);
4863         iam_it_fini(it);
4864         osd_ipd_put(env, bag, ipd);
4865
4866         LINVRNT(osd_invariant(obj));
4867
4868         RETURN(rc);
4869 }
4870
4871 static int osd_index_declare_iam_insert(const struct lu_env *env,
4872                                         struct dt_object *dt,
4873                                         const struct dt_rec *rec,
4874                                         const struct dt_key *key,
4875                                         struct thandle *handle)
4876 {
4877         struct osd_thandle *oh;
4878
4879         LASSERT(handle != NULL);
4880
4881         oh = container_of0(handle, struct osd_thandle, ot_super);
4882         LASSERT(oh->ot_handle == NULL);
4883
4884         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
4885                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
4886
4887         return 0;
4888 }
4889
4890 /**
4891  *      Inserts (key, value) pair in \a dt index object.
4892  *
4893  *      \param  dt      osd index object
4894  *      \param  key     key for index
4895  *      \param  rec     record reference
4896  *      \param  th      transaction handler
4897  *
4898  *      \retval  0  success
4899  *      \retval -ve failure
4900  */
4901 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
4902                                 const struct dt_rec *rec,
4903                                 const struct dt_key *key, struct thandle *th,
4904                                 int ignore_quota)
4905 {
4906         struct osd_object     *obj = osd_dt_obj(dt);
4907         struct iam_path_descr *ipd;
4908         struct osd_thandle    *oh;
4909         struct iam_container  *bag;
4910         struct osd_thread_info *oti = osd_oti_get(env);
4911         struct iam_rec         *iam_rec;
4912         int                     rc;
4913         ENTRY;
4914
4915         if (!dt_object_exists(dt))
4916                 RETURN(-ENOENT);
4917
4918         LINVRNT(osd_invariant(obj));
4919         LASSERT(!dt_object_remote(dt));
4920
4921         bag = &obj->oo_dir->od_container;
4922         LASSERT(bag->ic_object == obj->oo_inode);
4923         LASSERT(th != NULL);
4924
4925         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4926
4927         ipd = osd_idx_ipd_get(env, bag);
4928         if (unlikely(ipd == NULL))
4929                 RETURN(-ENOMEM);
4930
4931         oh = container_of0(th, struct osd_thandle, ot_super);
4932         LASSERT(oh->ot_handle != NULL);
4933         LASSERT(oh->ot_handle->h_transaction != NULL);
4934         if (S_ISDIR(obj->oo_inode->i_mode)) {
4935                 iam_rec = (struct iam_rec *)oti->oti_ldp;
4936                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
4937         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
4938                 /* pack quota uid/gid */
4939                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4940                 key = (const struct dt_key *)&oti->oti_quota_id;
4941                 /* pack quota record */
4942                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
4943                 iam_rec = (struct iam_rec *)rec;
4944         } else {
4945                 iam_rec = (struct iam_rec *)rec;
4946         }
4947
4948         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
4949                         iam_rec, ipd);
4950         osd_ipd_put(env, bag, ipd);
4951         LINVRNT(osd_invariant(obj));
4952         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4953         RETURN(rc);
4954 }
4955
4956 /**
4957  * Calls ldiskfs_add_entry() to add directory entry
4958  * into the directory. This is required for
4959  * interoperability mode (b11826)
4960  *
4961  * \retval   0, on success
4962  * \retval -ve, on error
4963  */
4964 static int __osd_ea_add_rec(struct osd_thread_info *info,
4965                             struct osd_object *pobj, struct inode  *cinode,
4966                             const char *name, const struct lu_fid *fid,
4967                             struct htree_lock *hlock, struct thandle *th)
4968 {
4969         struct ldiskfs_dentry_param *ldp;
4970         struct dentry               *child;
4971         struct osd_thandle          *oth;
4972         int                          rc;
4973
4974         oth = container_of(th, struct osd_thandle, ot_super);
4975         LASSERT(oth->ot_handle != NULL);
4976         LASSERT(oth->ot_handle->h_transaction != NULL);
4977         LASSERT(pobj->oo_inode);
4978
4979         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
4980         if (unlikely(pobj->oo_inode ==
4981                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
4982                 ldp->edp_magic = 0;
4983         else
4984                 osd_get_ldiskfs_dirent_param(ldp, fid);
4985         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
4986         child->d_fsdata = (void *)ldp;
4987         ll_vfs_dq_init(pobj->oo_inode);
4988         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
4989                                    child, cinode, hlock);
4990         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
4991                 struct ldiskfs_dir_entry_2      *de;
4992                 struct buffer_head              *bh;
4993                 int                              rc1;
4994
4995                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
4996                                             NULL, hlock);
4997                 if (!IS_ERR(bh)) {
4998                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
4999                                                                bh);
5000                         if (rc1 == 0) {
5001                                 if (S_ISDIR(cinode->i_mode))
5002                                         de->file_type = LDISKFS_DIRENT_LUFID |
5003                                                         LDISKFS_FT_REG_FILE;
5004                                 else
5005                                         de->file_type = LDISKFS_DIRENT_LUFID |
5006                                                         LDISKFS_FT_DIR;
5007                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5008                                                               NULL, bh);
5009                         }
5010                         brelse(bh);
5011                 }
5012         }
5013
5014         RETURN(rc);
5015 }
5016
5017 /**
5018  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5019  * into the directory.Also sets flags into osd object to
5020  * indicate dot and dotdot are created. This is required for
5021  * interoperability mode (b11826)
5022  *
5023  * \param dir   directory for dot and dotdot fixup.
5024  * \param obj   child object for linking
5025  *
5026  * \retval   0, on success
5027  * \retval -ve, on error
5028  */
5029 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5030                               struct osd_object *dir,
5031                               struct inode *parent_dir, const char *name,
5032                               const struct lu_fid *dot_fid,
5033                               const struct lu_fid *dot_dot_fid,
5034                               struct thandle *th)
5035 {
5036         struct inode                *inode = dir->oo_inode;
5037         struct osd_thandle          *oth;
5038         int result = 0;
5039
5040         oth = container_of(th, struct osd_thandle, ot_super);
5041         LASSERT(oth->ot_handle->h_transaction != NULL);
5042         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5043
5044         if (strcmp(name, dot) == 0) {
5045                 if (dir->oo_compat_dot_created) {
5046                         result = -EEXIST;
5047                 } else {
5048                         LASSERT(inode->i_ino == parent_dir->i_ino);
5049                         dir->oo_compat_dot_created = 1;
5050                         result = 0;
5051                 }
5052         } else if (strcmp(name, dotdot) == 0) {
5053                 if (!dir->oo_compat_dot_created)
5054                         return -EINVAL;
5055                 /* in case of rename, dotdot is already created */
5056                 if (dir->oo_compat_dotdot_created) {
5057                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5058                                                 dot_dot_fid, NULL, th);
5059                 }
5060
5061                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5062                         struct lu_fid tfid = *dot_dot_fid;
5063
5064                         tfid.f_oid--;
5065                         result = osd_add_dot_dotdot_internal(info,
5066                                         dir->oo_inode, parent_dir, dot_fid,
5067                                         &tfid, oth);
5068                 } else {
5069                         result = osd_add_dot_dotdot_internal(info,
5070                                         dir->oo_inode, parent_dir, dot_fid,
5071                                         dot_dot_fid, oth);
5072                 }
5073
5074                 if (result == 0)
5075                         dir->oo_compat_dotdot_created = 1;
5076         }
5077
5078         return result;
5079 }
5080
5081
5082 /**
5083  * It will call the appropriate osd_add* function and return the
5084  * value, return by respective functions.
5085  */
5086 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5087                           struct inode *cinode, const char *name,
5088                           const struct lu_fid *fid, struct thandle *th)
5089 {
5090         struct osd_thread_info *info   = osd_oti_get(env);
5091         struct htree_lock      *hlock;
5092         int                     rc;
5093
5094         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5095
5096         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
5097                                                    name[2] =='\0'))) {
5098                 if (hlock != NULL) {
5099                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5100                                            pobj->oo_inode, 0);
5101                 } else {
5102                         down_write(&pobj->oo_ext_idx_sem);
5103                 }
5104
5105                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5106                                         lu_object_fid(&pobj->oo_dt.do_lu),
5107                                         fid, th);
5108         } else {
5109                 if (hlock != NULL) {
5110                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5111                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5112                 } else {
5113                         down_write(&pobj->oo_ext_idx_sem);
5114                 }
5115
5116                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5117                         struct lu_fid *tfid = &info->oti_fid;
5118
5119                         *tfid = *fid;
5120                         tfid->f_ver = ~0;
5121                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5122                                               tfid, hlock, th);
5123                 } else {
5124                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5125                                               hlock, th);
5126                 }
5127         }
5128         if (hlock != NULL)
5129                 ldiskfs_htree_unlock(hlock);
5130         else
5131                 up_write(&pobj->oo_ext_idx_sem);
5132
5133         return rc;
5134 }
5135
5136 static int
5137 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5138                       struct osd_idmap_cache *oic)
5139 {
5140         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5141         struct lu_fid *fid = &oic->oic_fid;
5142         struct osd_inode_id *id = &oic->oic_lid;
5143         struct inode *inode = NULL;
5144         int once = 0;
5145         bool insert;
5146         int rc;
5147         ENTRY;
5148
5149         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5150                 RETURN(0);
5151
5152         if (thread_is_running(&scrub->os_thread) &&
5153             scrub->os_pos_current > id->oii_ino)
5154                 RETURN(0);
5155
5156         if (dev->od_auto_scrub_interval == AS_NEVER ||
5157             cfs_time_before(ktime_get_real_seconds(),
5158                             scrub->os_file.sf_time_last_complete +
5159                             dev->od_auto_scrub_interval))
5160                 RETURN(0);
5161
5162 again:
5163         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5164         if (rc == -ENOENT) {
5165                 __u32 gen = id->oii_gen;
5166
5167                 insert = true;
5168                 if (inode != NULL)
5169                         goto trigger;
5170
5171                 inode = osd_iget(oti, dev, id);
5172                 /* The inode has been removed (by race maybe). */
5173                 if (IS_ERR(inode)) {
5174                         rc = PTR_ERR(inode);
5175
5176                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5177                 }
5178
5179                 /* The OI mapping is lost. */
5180                 if (gen != OSD_OII_NOGEN)
5181                         goto trigger;
5182
5183                 /* The inode may has been reused by others, we do not know,
5184                  * leave it to be handled by subsequent osd_fid_lookup(). */
5185                 GOTO(out, rc = 0);
5186         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5187                 GOTO(out, rc);
5188         }
5189
5190         insert = false;
5191
5192 trigger:
5193         if (thread_is_running(&scrub->os_thread)) {
5194                 if (inode == NULL) {
5195                         inode = osd_iget(oti, dev, id);
5196                         /* The inode has been removed (by race maybe). */
5197                         if (IS_ERR(inode)) {
5198                                 rc = PTR_ERR(inode);
5199
5200                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5201                         }
5202                 }
5203
5204                 rc = osd_oii_insert(dev, oic, insert);
5205                 /* There is race condition between osd_oi_lookup and OI scrub.
5206                  * The OI scrub finished just after osd_oi_lookup() failure.
5207                  * Under such case, it is unnecessary to trigger OI scrub again,
5208                  * but try to call osd_oi_lookup() again. */
5209                 if (unlikely(rc == -EAGAIN))
5210                         goto again;
5211
5212                 if (!S_ISDIR(inode->i_mode))
5213                         rc = 0;
5214                 else
5215                         rc = osd_check_lmv(oti, dev, inode, oic);
5216
5217                 GOTO(out, rc);
5218         }
5219
5220         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5221                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5222                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5223                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5224                        "%s: trigger partial OI scrub for RPC inconsistency "
5225                        "checking FID "DFID": rc = %d\n",
5226                        osd_dev2name(dev), PFID(fid), rc);
5227                 if (rc == 0 || rc == -EALREADY)
5228                         goto again;
5229         }
5230
5231         GOTO(out, rc);
5232
5233 out:
5234         if (inode)
5235                 iput(inode);
5236
5237         RETURN(rc);
5238 }
5239
5240 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5241                                struct osd_device *dev,
5242                                struct osd_idmap_cache *oic,
5243                                struct lu_fid *fid, __u32 ino)
5244 {
5245         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5246         struct inode *inode;
5247         int rc;
5248
5249         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5250         inode = osd_iget(oti, dev, &oic->oic_lid);
5251         if (IS_ERR(inode)) {
5252                 fid_zero(&oic->oic_fid);
5253                 return PTR_ERR(inode);
5254         }
5255
5256         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5257         iput(inode);
5258         if (rc != 0)
5259                 fid_zero(&oic->oic_fid);
5260         else
5261                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5262         return rc;
5263 }
5264
5265 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5266                       struct osd_inode_id *id, const struct lu_fid *fid)
5267 {
5268         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5269                id->oii_ino, id->oii_gen, info);
5270         info->oti_cache.oic_lid = *id;
5271         info->oti_cache.oic_fid = *fid;
5272         info->oti_cache.oic_dev = osd;
5273 }
5274
5275 /**
5276  * Get parent FID from the linkEA.
5277  *
5278  * For a directory which parent resides on remote MDT, to satisfy the
5279  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5280  * the other hand, to make the lookup(..) on the directory can return
5281  * the real parent FID, we append the real parent FID after its ".."
5282  * name entry in the /REMOTE_PARENT_DIR.
5283  *
5284  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5285  * backup. So after the restore, we cannot get the right parent FID from
5286  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5287  * we have stored the real parent FID in the directory object's linkEA,
5288  * we can parse the linkEA for the real parent FID.
5289  *
5290  * \param[in] env       pointer to the thread context
5291  * \param[in] obj       pointer to the object to be handled
5292  * \param[out]fid       pointer to the buffer to hold the parent FID
5293  *
5294  * \retval              0 for getting the real parent FID successfully
5295  * \retval              negative error number on failure
5296  */
5297 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5298                                     struct osd_object *obj,
5299                                     struct lu_fid *fid)
5300 {
5301         struct osd_thread_info  *oti    = osd_oti_get(env);
5302         struct lu_buf           *buf    = &oti->oti_big_buf;
5303         struct dentry           *dentry = &oti->oti_obj_dentry;
5304         struct inode            *inode  = obj->oo_inode;
5305         struct linkea_data       ldata  = { NULL };
5306         int                      rc;
5307         ENTRY;
5308
5309         fid_zero(fid);
5310         if (!S_ISDIR(inode->i_mode))
5311                 RETURN(-EIO);
5312
5313 again:
5314         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5315                              buf->lb_buf, buf->lb_len);
5316         if (rc == -ERANGE) {
5317                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5318                                      NULL, 0);
5319                 if (rc > 0) {
5320                         lu_buf_realloc(buf, rc);
5321                         if (buf->lb_buf == NULL)
5322                                 RETURN(-ENOMEM);
5323
5324                         goto again;
5325                 }
5326         }
5327
5328         if (unlikely(rc == 0))
5329                 RETURN(-ENODATA);
5330
5331         if (rc < 0)
5332                 RETURN(rc);
5333
5334         if (unlikely(buf->lb_buf == NULL)) {
5335                 lu_buf_realloc(buf, rc);
5336                 if (buf->lb_buf == NULL)
5337                         RETURN(-ENOMEM);
5338
5339                 goto again;
5340         }
5341
5342         ldata.ld_buf = buf;
5343         rc = linkea_init_with_rec(&ldata);
5344         if (!rc) {
5345                 linkea_first_entry(&ldata);
5346                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5347         }
5348
5349         RETURN(rc);
5350 }
5351
5352 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5353                                     struct inode *inode,
5354                                     const struct lu_fid *pfid,
5355                                     const char *name, const int namelen)
5356 {
5357         struct osd_thread_info *oti = osd_oti_get(env);
5358         struct lu_buf *buf = &oti->oti_big_buf;
5359         struct dentry *dentry = &oti->oti_obj_dentry;
5360         struct linkea_data ldata = { NULL };
5361         struct lu_name cname = { .ln_name = name,
5362                                  .ln_namelen = namelen };
5363         int rc;
5364         ENTRY;
5365
5366 again:
5367         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5368                              buf->lb_buf, buf->lb_len);
5369         if (rc == -ERANGE)
5370                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5371
5372         if (rc < 0)
5373                 RETURN(rc);
5374
5375         if (unlikely(rc == 0))
5376                 RETURN(-ENODATA);
5377
5378         if (buf->lb_len < rc) {
5379                 lu_buf_realloc(buf, rc);
5380                 if (buf->lb_buf == NULL)
5381                         RETURN(-ENOMEM);
5382
5383                 goto again;
5384         }
5385
5386         ldata.ld_buf = buf;
5387         rc = linkea_init_with_rec(&ldata);
5388         if (!rc)
5389                 rc = linkea_links_find(&ldata, &cname, pfid);
5390
5391         RETURN(rc);
5392 }
5393
5394 /**
5395  * Calls ->lookup() to find dentry. From dentry get inode and
5396  * read inode's ea to get fid. This is required for  interoperability
5397  * mode (b11826)
5398  *
5399  * \retval   0, on success
5400  * \retval -ve, on error
5401  */
5402 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5403                              struct dt_rec *rec, const struct dt_key *key)
5404 {
5405         struct inode                    *dir    = obj->oo_inode;
5406         struct dentry                   *dentry;
5407         struct ldiskfs_dir_entry_2      *de;
5408         struct buffer_head              *bh;
5409         struct lu_fid                   *fid = (struct lu_fid *) rec;
5410         struct htree_lock               *hlock = NULL;
5411         int                             ino;
5412         int                             rc;
5413         ENTRY;
5414
5415         LASSERT(dir->i_op != NULL);
5416         LASSERT(dir->i_op->lookup != NULL);
5417
5418         dentry = osd_child_dentry_get(env, obj,
5419                                       (char *)key, strlen((char *)key));
5420
5421         if (obj->oo_hl_head != NULL) {
5422                 hlock = osd_oti_get(env)->oti_hlock;
5423                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5424                                    dir, LDISKFS_HLOCK_LOOKUP);
5425         } else {
5426                 down_read(&obj->oo_ext_idx_sem);
5427         }
5428
5429         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5430         if (!IS_ERR(bh)) {
5431                 struct osd_thread_info *oti = osd_oti_get(env);
5432                 struct osd_inode_id *id = &oti->oti_id;
5433                 struct osd_idmap_cache *oic = &oti->oti_cache;
5434                 struct osd_device *dev = osd_obj2dev(obj);
5435
5436                 ino = le32_to_cpu(de->inode);
5437                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5438                         brelse(bh);
5439                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5440                         GOTO(out, rc);
5441                 }
5442
5443                 rc = osd_get_fid_from_dentry(de, rec);
5444
5445                 /* done with de, release bh */
5446                 brelse(bh);
5447                 if (rc != 0) {
5448                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
5449                                 const char *name = (const char *)key;
5450
5451                                 /* If the parent is on remote MDT, and there
5452                                  * is no FID-in-dirent, then we have to get
5453                                  * the parent FID from the linkEA.  */
5454                                 if (likely(strlen(name) == 2 &&
5455                                            name[0] == '.' && name[1] == '.'))
5456                                         rc = osd_get_pfid_from_linkea(env, obj,
5457                                                                       fid);
5458                         } else {
5459                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5460                         }
5461                 } else {
5462                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5463                 }
5464
5465                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5466                         fid_zero(&oic->oic_fid);
5467
5468                         GOTO(out, rc);
5469                 }
5470
5471                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5472                 rc = osd_consistency_check(oti, dev, oic);
5473                 if (rc == -ENOENT)
5474                         fid_zero(&oic->oic_fid);
5475                 else
5476                         /* Other error should not affect lookup result. */
5477                         rc = 0;
5478         } else {
5479                 rc = PTR_ERR(bh);
5480         }
5481
5482         GOTO(out, rc);
5483
5484 out:
5485         if (hlock != NULL)
5486                 ldiskfs_htree_unlock(hlock);
5487         else
5488                 up_read(&obj->oo_ext_idx_sem);
5489         return rc;
5490 }
5491
5492 static int osd_index_declare_ea_insert(const struct lu_env *env,
5493                                        struct dt_object *dt,
5494                                        const struct dt_rec *rec,
5495                                        const struct dt_key *key,
5496                                        struct thandle *handle)
5497 {
5498         struct osd_thandle      *oh;
5499         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
5500         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5501         const struct lu_fid     *fid    = rec1->rec_fid;
5502         int                      credits, rc = 0;
5503         struct osd_idmap_cache  *idc;
5504         ENTRY;
5505
5506         LASSERT(!dt_object_remote(dt));
5507         LASSERT(handle != NULL);
5508         LASSERT(fid != NULL);
5509         LASSERT(rec1->rec_type != 0);
5510
5511         oh = container_of0(handle, struct osd_thandle, ot_super);
5512         LASSERT(oh->ot_handle == NULL);
5513
5514         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5515
5516         /* we can't call iget() while a transactions is running
5517          * (this can lead to a deadlock), but we need to know
5518          * inum and object type. so we find this information at
5519          * declaration and cache in per-thread info */
5520         idc = osd_idc_find_or_init(env, osd, fid);
5521         if (IS_ERR(idc))
5522                 RETURN(PTR_ERR(idc));
5523         if (idc->oic_remote) {
5524                 /* a reference to remote inode is represented by an
5525                  * agent inode which we have to create */
5526                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5527                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5528         }
5529
5530         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5531
5532         if (osd_dt_obj(dt)->oo_inode != NULL) {
5533                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5534
5535                 /* We ignore block quota on meta pool (MDTs), so needn't
5536                  * calculate how many blocks will be consumed by this index
5537                  * insert */
5538                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5539                                            i_gid_read(inode),
5540                                            i_projid_read(inode), 0,
5541                                            oh, osd_dt_obj(dt), NULL,
5542                                            OSD_QID_BLK);
5543                 if (rc)
5544                         RETURN(rc);
5545
5546 #ifdef HAVE_PROJECT_QUOTA
5547                 /* Reserve credits for local agent inode to transfer
5548                  * to 0, quota enforcement is ignored in this case.
5549                  */
5550                 if (idc->oic_remote &&
5551                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5552                     i_projid_read(inode) != 0)
5553                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5554                                                   0, i_projid_read(inode),
5555                                                   0, false, PRJQUOTA);
5556 #endif
5557         }
5558
5559
5560         RETURN(rc);
5561 }
5562
5563 /**
5564  * Index add function for interoperability mode (b11826).
5565  * It will add the directory entry.This entry is needed to
5566  * maintain name->fid mapping.
5567  *
5568  * \param key it is key i.e. file entry to be inserted
5569  * \param rec it is value of given key i.e. fid
5570  *
5571  * \retval   0, on success
5572  * \retval -ve, on error
5573  */
5574 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5575                                const struct dt_rec *rec,
5576                                const struct dt_key *key, struct thandle *th,
5577                                int ignore_quota)
5578 {
5579         struct osd_object       *obj = osd_dt_obj(dt);
5580         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
5581         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
5582         const struct lu_fid     *fid    = rec1->rec_fid;
5583         const char              *name = (const char *)key;
5584         struct osd_thread_info  *oti   = osd_oti_get(env);
5585         struct inode            *child_inode = NULL;
5586         struct osd_idmap_cache  *idc;
5587         int                     rc;
5588         ENTRY;
5589
5590         if (!dt_object_exists(dt))
5591                 RETURN(-ENOENT);
5592
5593         LASSERT(osd_invariant(obj));
5594         LASSERT(!dt_object_remote(dt));
5595         LASSERT(th != NULL);
5596
5597         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5598
5599         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
5600
5601         idc = osd_idc_find(env, osd, fid);
5602         if (unlikely(idc == NULL)) {
5603                 /* this dt_insert() wasn't declared properly, so
5604                  * FID is missing in OI cache. we better do not
5605                  * lookup FID in FLDB/OI and don't risk to deadlock,
5606                  * but in some special cases (lfsck testing, etc)
5607                  * it's much simpler than fixing a caller */
5608                 CERROR("%s: "DFID" wasn't declared for insert\n",
5609                        osd_name(osd), PFID(fid));
5610                 dump_stack();
5611                 idc = osd_idc_find_or_init(env, osd, fid);
5612                 if (IS_ERR(idc))
5613                         RETURN(PTR_ERR(idc));
5614         }
5615
5616         if (idc->oic_remote) {
5617                 /* Insert remote entry */
5618                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
5619                         child_inode =
5620                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
5621                 } else {
5622                         child_inode = osd_create_local_agent_inode(env, osd,
5623                                         obj, fid, rec1->rec_type & S_IFMT, th);
5624                         if (IS_ERR(child_inode))
5625                                 RETURN(PTR_ERR(child_inode));
5626                 }
5627         } else {
5628                 /* Insert local entry */
5629                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
5630                         /* for a reason OI cache wasn't filled properly */
5631                         CERROR("%s: OIC for "DFID" isn't filled\n",
5632                                osd_name(osd), PFID(fid));
5633                         RETURN(-EINVAL);
5634                 }
5635                 child_inode = oti->oti_inode;
5636                 if (unlikely(child_inode == NULL)) {
5637                         struct ldiskfs_inode_info *lii;
5638                         OBD_ALLOC_PTR(lii);
5639                         if (lii == NULL)
5640                                 RETURN(-ENOMEM);
5641                         child_inode = oti->oti_inode = &lii->vfs_inode;
5642                 }
5643                 child_inode->i_sb = osd_sb(osd);
5644                 child_inode->i_ino = idc->oic_lid.oii_ino;
5645                 child_inode->i_mode = rec1->rec_type & S_IFMT;
5646         }
5647
5648         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
5649
5650         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
5651                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
5652
5653         if (child_inode && child_inode != oti->oti_inode)
5654                 iput(child_inode);
5655         LASSERT(osd_invariant(obj));
5656         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5657         RETURN(rc);
5658 }
5659
5660 /**
5661  *  Initialize osd Iterator for given osd index object.
5662  *
5663  *  \param  dt      osd index object
5664  */
5665
5666 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
5667                                      struct dt_object *dt,
5668                                      __u32 unused)
5669 {
5670         struct osd_it_iam      *it;
5671         struct osd_object      *obj = osd_dt_obj(dt);
5672         struct lu_object       *lo  = &dt->do_lu;
5673         struct iam_path_descr  *ipd;
5674         struct iam_container   *bag = &obj->oo_dir->od_container;
5675
5676         if (!dt_object_exists(dt))
5677                 return ERR_PTR(-ENOENT);
5678
5679         OBD_ALLOC_PTR(it);
5680         if (it == NULL)
5681                 return ERR_PTR(-ENOMEM);
5682
5683         ipd = osd_it_ipd_get(env, bag);
5684         if (likely(ipd != NULL)) {
5685                 it->oi_obj = obj;
5686                 it->oi_ipd = ipd;
5687                 lu_object_get(lo);
5688                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
5689                 return (struct dt_it *)it;
5690         } else {
5691                 OBD_FREE_PTR(it);
5692                 return ERR_PTR(-ENOMEM);
5693         }
5694 }
5695
5696 /**
5697  * free given Iterator.
5698  */
5699 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
5700 {
5701         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
5702         struct osd_object       *obj = it->oi_obj;
5703
5704         iam_it_fini(&it->oi_it);
5705         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
5706         osd_object_put(env, obj);
5707         OBD_FREE_PTR(it);
5708 }
5709
5710 /**
5711  *  Move Iterator to record specified by \a key
5712  *
5713  *  \param  di      osd iterator
5714  *  \param  key     key for index
5715  *
5716  *  \retval +ve  di points to record with least key not larger than key
5717  *  \retval  0   di points to exact matched key
5718  *  \retval -ve  failure
5719  */
5720
5721 static int osd_it_iam_get(const struct lu_env *env,
5722                           struct dt_it *di, const struct dt_key *key)
5723 {
5724         struct osd_thread_info  *oti = osd_oti_get(env);
5725         struct osd_it_iam       *it = (struct osd_it_iam *)di;
5726
5727         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5728                 /* swab quota uid/gid */
5729                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5730                 key = (struct dt_key *)&oti->oti_quota_id;
5731         }
5732
5733         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
5734 }
5735
5736 /**
5737  *  Release Iterator
5738  *
5739  *  \param  di      osd iterator
5740  */
5741 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
5742 {
5743         struct osd_it_iam *it = (struct osd_it_iam *)di;
5744
5745         iam_it_put(&it->oi_it);
5746 }
5747
5748 /**
5749  *  Move iterator by one record
5750  *
5751  *  \param  di      osd iterator
5752  *
5753  *  \retval +1   end of container reached
5754  *  \retval  0   success
5755  *  \retval -ve  failure
5756  */
5757
5758 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
5759 {
5760         struct osd_it_iam *it = (struct osd_it_iam *)di;
5761
5762         return iam_it_next(&it->oi_it);
5763 }
5764
5765 /**
5766  * Return pointer to the key under iterator.
5767  */
5768
5769 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
5770                                  const struct dt_it *di)
5771 {
5772         struct osd_thread_info *oti = osd_oti_get(env);
5773         struct osd_it_iam      *it = (struct osd_it_iam *)di;
5774         struct osd_object      *obj = it->oi_obj;
5775         struct dt_key          *key;
5776
5777         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
5778
5779         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
5780                 /* swab quota uid/gid */
5781                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
5782                 key = (struct dt_key *)&oti->oti_quota_id;
5783         }
5784
5785         return key;
5786 }
5787
5788 /**
5789  * Return size of key under iterator (in bytes)
5790  */
5791
5792 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
5793 {
5794         struct osd_it_iam *it = (struct osd_it_iam *)di;
5795
5796         return iam_it_key_size(&it->oi_it);
5797 }
5798
5799 static inline void
5800 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
5801 {
5802         /* check if file type is required */
5803         if (ent->lde_attrs & LUDA_TYPE) {
5804                 struct luda_type *lt;
5805                 int align = sizeof(*lt) - 1;
5806
5807                 len = (len + align) & ~align;
5808                 lt = (struct luda_type *)(ent->lde_name + len);
5809                 lt->lt_type = cpu_to_le16(DTTOIF(type));
5810         }
5811
5812         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
5813 }
5814
5815 /**
5816  * build lu direct from backend fs dirent.
5817  */
5818
5819 static inline void
5820 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
5821                    char *name, __u16 namelen, __u16 type, __u32 attr)
5822 {
5823         ent->lde_attrs = attr | LUDA_FID;
5824         fid_cpu_to_le(&ent->lde_fid, fid);
5825
5826         ent->lde_hash = cpu_to_le64(offset);
5827         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
5828
5829         strncpy(ent->lde_name, name, namelen);
5830         ent->lde_name[namelen] = '\0';
5831         ent->lde_namelen = cpu_to_le16(namelen);
5832
5833         /* append lustre attributes */
5834         osd_it_append_attrs(ent, namelen, type);
5835 }
5836
5837 /**
5838  * Return pointer to the record under iterator.
5839  */
5840 static int osd_it_iam_rec(const struct lu_env *env,
5841                           const struct dt_it *di,
5842                           struct dt_rec *dtrec, __u32 attr)
5843 {
5844         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
5845         struct osd_thread_info *info = osd_oti_get(env);
5846         ENTRY;
5847
5848         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
5849                 const struct osd_fid_pack *rec;
5850                 struct lu_fid             *fid = &info->oti_fid;
5851                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
5852                 char                      *name;
5853                 int                        namelen;
5854                 __u64                      hash;
5855                 int                        rc;
5856
5857                 name = (char *)iam_it_key_get(&it->oi_it);
5858                 if (IS_ERR(name))
5859                         RETURN(PTR_ERR(name));
5860
5861                 namelen = iam_it_key_size(&it->oi_it);
5862
5863                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
5864                 if (IS_ERR(rec))
5865                         RETURN(PTR_ERR(rec));
5866
5867                 rc = osd_fid_unpack(fid, rec);
5868                 if (rc)
5869                         RETURN(rc);
5870
5871                 hash = iam_it_store(&it->oi_it);
5872
5873                 /* IAM does not store object type in IAM index (dir) */
5874                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
5875                                    0, LUDA_FID);
5876         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
5877                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5878                            (struct iam_rec *)dtrec);
5879                 osd_quota_unpack(it->oi_obj, dtrec);
5880         } else {
5881                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
5882                            (struct iam_rec *)dtrec);
5883         }
5884
5885         RETURN(0);
5886 }
5887
5888 /**
5889  * Returns cookie for current Iterator position.
5890  */
5891 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
5892 {
5893         struct osd_it_iam *it = (struct osd_it_iam *)di;
5894
5895         return iam_it_store(&it->oi_it);
5896 }
5897
5898 /**
5899  * Restore iterator from cookie.
5900  *
5901  * \param  di      osd iterator
5902  * \param  hash    Iterator location cookie
5903  *
5904  * \retval +ve  di points to record with least key not larger than key.
5905  * \retval  0   di points to exact matched key
5906  * \retval -ve  failure
5907  */
5908
5909 static int osd_it_iam_load(const struct lu_env *env,
5910                            const struct dt_it *di, __u64 hash)
5911 {
5912         struct osd_it_iam *it = (struct osd_it_iam *)di;
5913
5914         return iam_it_load(&it->oi_it, hash);
5915 }
5916
5917 static const struct dt_index_operations osd_index_iam_ops = {
5918         .dio_lookup         = osd_index_iam_lookup,
5919         .dio_declare_insert = osd_index_declare_iam_insert,
5920         .dio_insert         = osd_index_iam_insert,
5921         .dio_declare_delete = osd_index_declare_iam_delete,
5922         .dio_delete         = osd_index_iam_delete,
5923         .dio_it     = {
5924                 .init     = osd_it_iam_init,
5925                 .fini     = osd_it_iam_fini,
5926                 .get      = osd_it_iam_get,
5927                 .put      = osd_it_iam_put,
5928                 .next     = osd_it_iam_next,
5929                 .key      = osd_it_iam_key,
5930                 .key_size = osd_it_iam_key_size,
5931                 .rec      = osd_it_iam_rec,
5932                 .store    = osd_it_iam_store,
5933                 .load     = osd_it_iam_load
5934         }
5935 };
5936
5937
5938 /**
5939  * Creates or initializes iterator context.
5940  *
5941  * \retval struct osd_it_ea, iterator structure on success
5942  *
5943  */
5944 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
5945                                     struct dt_object *dt,
5946                                     __u32 attr)
5947 {
5948         struct osd_object       *obj  = osd_dt_obj(dt);
5949         struct osd_thread_info  *info = osd_oti_get(env);
5950         struct osd_it_ea        *oie;
5951         struct file             *file;
5952         struct lu_object        *lo   = &dt->do_lu;
5953         struct dentry           *obj_dentry;
5954         ENTRY;
5955
5956         if (!dt_object_exists(dt) || obj->oo_destroyed)
5957                 RETURN(ERR_PTR(-ENOENT));
5958
5959         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
5960         if (oie == NULL)
5961                 RETURN(ERR_PTR(-ENOMEM));
5962         obj_dentry = &oie->oie_dentry;
5963
5964         obj_dentry->d_inode = obj->oo_inode;
5965         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
5966         obj_dentry->d_name.hash = 0;
5967
5968         oie->oie_rd_dirent       = 0;
5969         oie->oie_it_dirent       = 0;
5970         oie->oie_dirent          = NULL;
5971         if (unlikely(!info->oti_it_ea_buf_used)) {
5972                 oie->oie_buf = info->oti_it_ea_buf;
5973                 info->oti_it_ea_buf_used = 1;
5974         } else {
5975                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
5976                 if (oie->oie_buf == NULL)
5977                         RETURN(ERR_PTR(-ENOMEM));
5978         }
5979         oie->oie_obj             = obj;
5980
5981         file = &oie->oie_file;
5982
5983         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
5984         if (attr & LUDA_64BITHASH)
5985                 file->f_mode    = FMODE_64BITHASH;
5986         else
5987                 file->f_mode    = FMODE_32BITHASH;
5988         file->f_path.dentry     = obj_dentry;
5989         file->f_mapping         = obj->oo_inode->i_mapping;
5990         file->f_op              = obj->oo_inode->i_fop;
5991         set_file_inode(file, obj->oo_inode);
5992
5993         lu_object_get(lo);
5994         RETURN((struct dt_it *) oie);
5995 }
5996
5997 /**
5998  * Destroy or finishes iterator context.
5999  *
6000  * \param di iterator structure to be destroyed
6001  */
6002 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6003 {
6004         struct osd_thread_info  *info = osd_oti_get(env);
6005         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
6006         struct osd_object       *obj    = oie->oie_obj;
6007         struct inode            *inode  = obj->oo_inode;
6008
6009         ENTRY;
6010         oie->oie_file.f_op->release(inode, &oie->oie_file);
6011         osd_object_put(env, obj);
6012         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6013                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6014         else
6015                 info->oti_it_ea_buf_used = 0;
6016         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6017         EXIT;
6018 }
6019
6020 /**
6021  * It position the iterator at given key, so that next lookup continues from
6022  * that key Or it is similar to dio_it->load() but based on a key,
6023  * rather than file position.
6024  *
6025  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6026  * to the beginning.
6027  *
6028  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6029  */
6030 static int osd_it_ea_get(const struct lu_env *env,
6031                          struct dt_it *di, const struct dt_key *key)
6032 {
6033         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
6034
6035         ENTRY;
6036         LASSERT(((const char *)key)[0] == '\0');
6037         it->oie_file.f_pos      = 0;
6038         it->oie_rd_dirent       = 0;
6039         it->oie_it_dirent       = 0;
6040         it->oie_dirent          = NULL;
6041
6042         RETURN(+1);
6043 }
6044
6045 /**
6046  * Does nothing
6047  */
6048 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6049 {
6050 }
6051
6052 struct osd_filldir_cbs {
6053 #ifdef HAVE_DIR_CONTEXT
6054         struct dir_context ctx;
6055 #endif
6056         struct osd_it_ea  *it;
6057 };
6058 /**
6059  * It is called internally by ->readdir(). It fills the
6060  * iterator's in-memory data structure with required
6061  * information i.e. name, namelen, rec_size etc.
6062  *
6063  * \param buf in which information to be filled in.
6064  * \param name name of the file in given dir
6065  *
6066  * \retval 0 on success
6067  * \retval 1 on buffer full
6068  */
6069 #ifdef HAVE_FILLDIR_USE_CTX
6070 static int osd_ldiskfs_filldir(struct dir_context *buf,
6071 #else
6072 static int osd_ldiskfs_filldir(void *buf,
6073 #endif
6074                                const char *name, int namelen,
6075                                loff_t offset, __u64 ino, unsigned d_type)
6076 {
6077         struct osd_it_ea        *it   =
6078                 ((struct osd_filldir_cbs *)buf)->it;
6079         struct osd_object       *obj  = it->oie_obj;
6080         struct osd_it_ea_dirent *ent  = it->oie_dirent;
6081         struct lu_fid           *fid  = &ent->oied_fid;
6082         struct osd_fid_pack     *rec;
6083         ENTRY;
6084
6085         /* this should never happen */
6086         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6087                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6088                 RETURN(-EIO);
6089         }
6090
6091         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
6092             OSD_IT_EA_BUFSIZE)
6093                 RETURN(1);
6094
6095         /* "." is just the object itself. */
6096         if (namelen == 1 && name[0] == '.') {
6097                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6098         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6099                 rec = (struct osd_fid_pack*) (name + namelen + 1);
6100                 if (osd_fid_unpack(fid, rec) != 0)
6101                         fid_zero(fid);
6102         } else {
6103                 fid_zero(fid);
6104         }
6105         d_type &= ~LDISKFS_DIRENT_LUFID;
6106
6107         /* NOT export local root. */
6108         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6109                 ino = obj->oo_inode->i_ino;
6110                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6111         }
6112
6113         ent->oied_ino     = ino;
6114         ent->oied_off     = offset;
6115         ent->oied_namelen = namelen;
6116         ent->oied_type    = d_type;
6117
6118         memcpy(ent->oied_name, name, namelen);
6119
6120         it->oie_rd_dirent++;
6121         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
6122         RETURN(0);
6123 }
6124
6125 /**
6126  * Calls ->readdir() to load a directory entry at a time
6127  * and stored it in iterator's in-memory data structure.
6128  *
6129  * \param di iterator's in memory structure
6130  *
6131  * \retval   0 on success
6132  * \retval -ve on error
6133  * \retval +1 reach the end of entry
6134  */
6135 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6136                                const struct dt_it *di)
6137 {
6138         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
6139         struct osd_object  *obj   = it->oie_obj;
6140         struct inode       *inode = obj->oo_inode;
6141         struct htree_lock  *hlock = NULL;
6142         struct file        *filp  = &it->oie_file;
6143         int                 rc = 0;
6144         struct osd_filldir_cbs buf = {
6145 #ifdef HAVE_DIR_CONTEXT
6146                 .ctx.actor = osd_ldiskfs_filldir,
6147 #endif
6148                 .it = it
6149         };
6150
6151         ENTRY;
6152         it->oie_dirent = it->oie_buf;
6153         it->oie_rd_dirent = 0;
6154
6155         if (obj->oo_hl_head != NULL) {
6156                 hlock = osd_oti_get(env)->oti_hlock;
6157                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6158                                    inode, LDISKFS_HLOCK_READDIR);
6159         } else {
6160                 down_read(&obj->oo_ext_idx_sem);
6161         }
6162
6163 #ifdef HAVE_DIR_CONTEXT
6164         buf.ctx.pos = filp->f_pos;
6165         rc = inode->i_fop->iterate(filp, &buf.ctx);
6166         filp->f_pos = buf.ctx.pos;
6167 #else
6168         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6169 #endif
6170
6171         if (hlock != NULL)
6172                 ldiskfs_htree_unlock(hlock);
6173         else
6174                 up_read(&obj->oo_ext_idx_sem);
6175
6176         if (it->oie_rd_dirent == 0) {
6177                 /*If it does not get any dirent, it means it has been reached
6178                  *to the end of the dir */
6179                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6180                 if (rc == 0)
6181                         rc = 1;
6182         } else {
6183                 it->oie_dirent = it->oie_buf;
6184                 it->oie_it_dirent = 1;
6185         }
6186
6187         RETURN(rc);
6188 }
6189
6190 /**
6191  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6192  * to load a directory entry at a time and stored it in
6193  * iterator's in-memory data structure.
6194  *
6195  * \param di iterator's in memory structure
6196  *
6197  * \retval +ve iterator reached to end
6198  * \retval   0 iterator not reached to end
6199  * \retval -ve on error
6200  */
6201 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6202 {
6203         struct osd_it_ea *it = (struct osd_it_ea *)di;
6204         int rc;
6205
6206         ENTRY;
6207
6208         if (it->oie_it_dirent < it->oie_rd_dirent) {
6209                 it->oie_dirent =
6210                         (void *) it->oie_dirent +
6211                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6212                                        it->oie_dirent->oied_namelen);
6213                 it->oie_it_dirent++;
6214                 RETURN(0);
6215         } else {
6216                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6217                         rc = +1;
6218                 else
6219                         rc = osd_ldiskfs_it_fill(env, di);
6220         }
6221
6222         RETURN(rc);
6223 }
6224
6225 /**
6226  * Returns the key at current position from iterator's in memory structure.
6227  *
6228  * \param di iterator's in memory structure
6229  *
6230  * \retval key i.e. struct dt_key on success
6231  */
6232 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6233                                     const struct dt_it *di)
6234 {
6235         struct osd_it_ea *it = (struct osd_it_ea *)di;
6236
6237         return (struct dt_key *)it->oie_dirent->oied_name;
6238 }
6239
6240 /**
6241  * Returns the key's size at current position from iterator's in memory structure.
6242  *
6243  * \param di iterator's in memory structure
6244  *
6245  * \retval key_size i.e. struct dt_key on success
6246  */
6247 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6248 {
6249         struct osd_it_ea *it = (struct osd_it_ea *)di;
6250
6251         return it->oie_dirent->oied_namelen;
6252 }
6253
6254 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6255 {
6256         if (LDISKFS_DIR_REC_LEN(de) >=
6257             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6258                 return true;
6259
6260         return false;
6261 }
6262
6263 static inline bool
6264 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6265                      unsigned blocksize, bool dotdot)
6266 {
6267         if (dotdot)
6268                 return osd_dotdot_has_space(de);
6269
6270         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6271             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6272                 return true;
6273
6274         return false;
6275 }
6276
6277 static int
6278 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6279                     handle_t *jh, struct dentry *dentry,
6280                     const struct lu_fid *fid, struct buffer_head *bh,
6281                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6282                     bool dotdot)
6283 {
6284         struct inode                *dir        = dentry->d_parent->d_inode;
6285         struct inode                *inode      = dentry->d_inode;
6286         struct osd_fid_pack         *rec;
6287         struct ldiskfs_dentry_param *ldp;
6288         int                          namelen    = dentry->d_name.len;
6289         int                          rc;
6290         struct osd_thread_info     *info        = osd_oti_get(env);
6291         ENTRY;
6292
6293         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
6294                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
6295                 RETURN(0);
6296
6297         /* There is enough space to hold the FID-in-dirent. */
6298         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6299                 rc = ldiskfs_journal_get_write_access(jh, bh);
6300                 if (rc != 0)
6301                         RETURN(rc);
6302
6303                 de->name[namelen] = 0;
6304                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6305                 rec->fp_len = sizeof(struct lu_fid) + 1;
6306                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6307                 de->file_type |= LDISKFS_DIRENT_LUFID;
6308                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6309
6310                 RETURN(rc);
6311         }
6312
6313         LASSERT(!dotdot);
6314
6315         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6316         if (rc != 0)
6317                 RETURN(rc);
6318
6319         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6320         osd_get_ldiskfs_dirent_param(ldp, fid);
6321         dentry->d_fsdata = (void *)ldp;
6322         ll_vfs_dq_init(dir);
6323         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6324         /* It is too bad, we cannot reinsert the name entry back.
6325          * That means we lose it! */
6326         if (rc != 0)
6327                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6328                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6329                        osd_ino2name(inode),
6330                        dir->i_ino, dir->i_generation, namelen,
6331                        dentry->d_name.name, PFID(fid), rc);
6332
6333         RETURN(rc);
6334 }
6335
6336 static int
6337 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6338                         struct osd_it_ea *it, struct lu_fid *fid,
6339                         struct osd_inode_id *id, __u32 *attr)
6340 {
6341         struct osd_thread_info     *info        = osd_oti_get(env);
6342         struct lustre_mdt_attrs    *lma         = &info->oti_ost_attrs.loa_lma;
6343         struct osd_device          *dev         = osd_obj2dev(obj);
6344         struct super_block         *sb          = osd_sb(dev);
6345         const char                 *devname     = osd_name(dev);
6346         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
6347         struct inode               *dir         = obj->oo_inode;
6348         struct htree_lock          *hlock       = NULL;
6349         struct buffer_head         *bh          = NULL;
6350         handle_t                   *jh          = NULL;
6351         struct ldiskfs_dir_entry_2 *de;
6352         struct dentry              *dentry;
6353         struct inode               *inode;
6354         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6355         int                         credits;
6356         int                         rc;
6357         bool                        dotdot      = false;
6358         bool                        dirty       = false;
6359         ENTRY;
6360
6361         if (ent->oied_name[0] == '.') {
6362                 if (ent->oied_namelen == 1)
6363                         RETURN(0);
6364
6365                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6366                         dotdot = true;
6367         }
6368
6369         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6370         inode = osd_iget(info, dev, id);
6371         if (IS_ERR(inode)) {
6372                 rc = PTR_ERR(inode);
6373                 if (rc == -ENOENT || rc == -ESTALE) {
6374                         /* Maybe dangling name entry, or
6375                          * corrupted directory entry. */
6376                         *attr |= LUDA_UNKNOWN;
6377                         rc = 0;
6378                 } else {
6379                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6380                                "check_repair, dir = %lu/%u, name = %.*s, "
6381                                "ino = %llu, rc = %d\n",
6382                                devname, dir->i_ino, dir->i_generation,
6383                                ent->oied_namelen, ent->oied_name,
6384                                ent->oied_ino, rc);
6385                 }
6386
6387                 RETURN(rc);
6388         }
6389
6390         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6391                                            ent->oied_namelen);
6392         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6393         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6394                 lma = NULL;
6395         else if (rc != 0)
6396                 GOTO(out, rc);
6397
6398         /* We need to ensure that the name entry is still valid.
6399          * Because it may be removed or renamed by other already.
6400          *
6401          * The unlink or rename operation will start journal before PDO lock,
6402          * so to avoid deadlock, here we need to start journal handle before
6403          * related PDO lock also. But because we do not know whether there
6404          * will be something to be repaired before PDO lock, we just start
6405          * journal without conditions.
6406          *
6407          * We may need to remove the name entry firstly, then insert back.
6408          * One credit is for user quota file update.
6409          * One credit is for group quota file update.
6410          * Two credits are for dirty inode. */
6411         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6412                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6413
6414         if (dev->od_dirent_journal != 0) {
6415
6416 again:
6417                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6418                 if (IS_ERR(jh)) {
6419                         rc = PTR_ERR(jh);
6420                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6421                                "check_repair, dir = %lu/%u, credits = %d, "
6422                                "name = %.*s, ino = %llu: rc = %d\n",
6423                                devname, dir->i_ino, dir->i_generation, credits,
6424                                ent->oied_namelen, ent->oied_name,
6425                                ent->oied_ino, rc);
6426
6427                         GOTO(out_inode, rc);
6428                 }
6429
6430                 if (obj->oo_hl_head != NULL) {
6431                         hlock = osd_oti_get(env)->oti_hlock;
6432                         /* "0" means exclusive lock for the whole directory.
6433                          * We need to prevent others access such name entry
6434                          * during the delete + insert. Neither HLOCK_ADD nor
6435                          * HLOCK_DEL cannot guarantee the atomicity. */
6436                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6437                 } else {
6438                         down_write(&obj->oo_ext_idx_sem);
6439                 }
6440         } else {
6441                 if (obj->oo_hl_head != NULL) {
6442                         hlock = osd_oti_get(env)->oti_hlock;
6443                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6444                                            LDISKFS_HLOCK_LOOKUP);
6445                 } else {
6446                         down_read(&obj->oo_ext_idx_sem);
6447                 }
6448         }
6449
6450         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6451         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6452                 *attr |= LUDA_IGNORE;
6453
6454                 GOTO(out, rc = 0);
6455         }
6456
6457         /* For dotdot entry, if there is not enough space to hold the
6458          * FID-in-dirent, just keep them there. It only happens when the
6459          * device upgraded from 1.8 or restored from MDT file-level backup.
6460          * For the whole directory, only dotdot entry have no FID-in-dirent
6461          * and needs to get FID from LMA when readdir, it will not affect the
6462          * performance much. */
6463         if (dotdot && !osd_dotdot_has_space(de)) {
6464                 *attr |= LUDA_UNKNOWN;
6465
6466                 GOTO(out, rc = 0);
6467         }
6468
6469         if (lma != NULL) {
6470                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6471                         GOTO(out, rc = 0);
6472
6473                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6474                         struct lu_fid *tfid = &lma->lma_self_fid;
6475
6476                         if (likely(dotdot &&
6477                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6478                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6479                                 /* It must be REMOTE_PARENT_DIR and as the
6480                                  * 'dotdot' entry of remote directory */
6481                                 *attr |= LUDA_IGNORE;
6482                         } else {
6483                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6484                                        "parent directory, but got %.*s under "
6485                                        "dir = %lu/%u with the FID "DFID"\n",
6486                                        devname, ent->oied_namelen,
6487                                        ent->oied_name, dir->i_ino,
6488                                        dir->i_generation, PFID(tfid));
6489
6490                                 *attr |= LUDA_UNKNOWN;
6491                         }
6492
6493                         GOTO(out, rc = 0);
6494                 }
6495         }
6496
6497         if (!fid_is_zero(fid)) {
6498                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6499                                               ent->oied_namelen);
6500                 if (rc == -ENOENT ||
6501                     (rc == -ENODATA &&
6502                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6503                         /* linkEA does not recognize the dirent entry,
6504                          * it may because the dirent entry corruption
6505                          * and points to other's inode. */
6506                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6507                                "recognize the dirent, dir = %lu/%u, "
6508                                " name = %.*s, ino = %llu, "
6509                                DFID": rc = %d\n", devname, dir->i_ino,
6510                                dir->i_generation, ent->oied_namelen,
6511                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6512                         *attr |= LUDA_UNKNOWN;
6513
6514                         GOTO(out, rc = 0);
6515                 }
6516
6517                 if (rc && rc != -ENODATA) {
6518                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6519                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6520                                DFID": rc = %d\n", devname, dir->i_ino,
6521                                dir->i_generation, ent->oied_namelen,
6522                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6523                         *attr |= LUDA_UNKNOWN;
6524
6525                         GOTO(out, rc = 0);
6526                 }
6527         }
6528
6529         if (lma != NULL) {
6530                 /* linkEA recognizes the dirent entry, the FID-in-LMA is
6531                  * valid, trusted, in spite of fid_is_sane(fid) or not. */
6532                 if (*attr & LUDA_VERIFY_DRYRUN) {
6533                         *fid = lma->lma_self_fid;
6534                         *attr |= LUDA_REPAIR;
6535
6536                         GOTO(out, rc = 0);
6537                 }
6538
6539                 if (jh == NULL) {
6540                         brelse(bh);
6541                         dev->od_dirent_journal = 1;
6542                         if (hlock != NULL) {
6543                                 ldiskfs_htree_unlock(hlock);
6544                                 hlock = NULL;
6545                         } else {
6546                                 up_read(&obj->oo_ext_idx_sem);
6547                         }
6548
6549                         goto again;
6550                 }
6551
6552                 *fid = lma->lma_self_fid;
6553                 dirty = true;
6554                 /* Update or append the FID-in-dirent. */
6555                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6556                                          bh, de, hlock, dotdot);
6557                 if (rc == 0)
6558                         *attr |= LUDA_REPAIR;
6559                 else
6560                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
6561                                "the dirent, dir = %lu/%u, name = %.*s, "
6562                                "ino = %llu, "DFID": rc = %d\n",
6563                                devname, dir->i_ino, dir->i_generation,
6564                                ent->oied_namelen, ent->oied_name,
6565                                ent->oied_ino, PFID(fid), rc);
6566         } else {
6567                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
6568                 if (*attr & LUDA_VERIFY_DRYRUN) {
6569                         if (fid_is_sane(fid)) {
6570                                 *attr |= LUDA_REPAIR;
6571                         } else if (dev->od_index == 0) {
6572                                 lu_igif_build(fid, inode->i_ino,
6573                                               inode->i_generation);
6574                                 *attr |= LUDA_UPGRADE;
6575                         }
6576
6577                         GOTO(out, rc = 0);
6578                 }
6579
6580                 if (jh == NULL) {
6581                         brelse(bh);
6582                         dev->od_dirent_journal = 1;
6583                         if (hlock != NULL) {
6584                                 ldiskfs_htree_unlock(hlock);
6585                                 hlock = NULL;
6586                         } else {
6587                                 up_read(&obj->oo_ext_idx_sem);
6588                         }
6589
6590                         goto again;
6591                 }
6592
6593                 dirty = true;
6594                 if (unlikely(fid_is_sane(fid))) {
6595                         /* FID-in-dirent exists, but FID-in-LMA is lost.
6596                          * Trust the FID-in-dirent, and add FID-in-LMA. */
6597                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
6598                         if (rc == 0)
6599                                 *attr |= LUDA_REPAIR;
6600                         else
6601                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
6602                                        "update dirent, dir = %lu/%u, "
6603                                        "name = %.*s, ino = %llu, "
6604                                        DFID": rc = %d\n",
6605                                        devname, dir->i_ino, dir->i_generation,
6606                                        ent->oied_namelen, ent->oied_name,
6607                                        ent->oied_ino, PFID(fid), rc);
6608                 } else if (dev->od_index == 0) {
6609                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
6610                         /* It is probably IGIF object. Only aappend the
6611                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
6612                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
6613                                                  bh, de, hlock, dotdot);
6614                         if (rc == 0)
6615                                 *attr |= LUDA_UPGRADE;
6616                         else
6617                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
6618                                        "after the dirent, dir = %lu/%u, "
6619                                        "name = %.*s, ino = %llu, "
6620                                        DFID": rc = %d\n",
6621                                        devname, dir->i_ino, dir->i_generation,
6622                                        ent->oied_namelen, ent->oied_name,
6623                                        ent->oied_ino, PFID(fid), rc);
6624                 }
6625         }
6626
6627         GOTO(out, rc);
6628
6629 out:
6630         if (!IS_ERR(bh))
6631                 brelse(bh);
6632         if (hlock != NULL) {
6633                 ldiskfs_htree_unlock(hlock);
6634         } else {
6635                 if (dev->od_dirent_journal != 0)
6636                         up_write(&obj->oo_ext_idx_sem);
6637                 else
6638                         up_read(&obj->oo_ext_idx_sem);
6639         }
6640
6641         if (jh != NULL)
6642                 ldiskfs_journal_stop(jh);
6643
6644 out_inode:
6645         iput(inode);
6646         if (rc >= 0 && !dirty)
6647                 dev->od_dirent_journal = 0;
6648
6649         return rc;
6650 }
6651
6652 /**
6653  * Returns the value at current position from iterator's in memory structure.
6654  *
6655  * \param di struct osd_it_ea, iterator's in memory structure
6656  * \param attr attr requested for dirent.
6657  * \param lde lustre dirent
6658  *
6659  * \retval   0 no error and \param lde has correct lustre dirent.
6660  * \retval -ve on error
6661  */
6662 static inline int osd_it_ea_rec(const struct lu_env *env,
6663                                 const struct dt_it *di,
6664                                 struct dt_rec *dtrec, __u32 attr)
6665 {
6666         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
6667         struct osd_object      *obj   = it->oie_obj;
6668         struct osd_device      *dev   = osd_obj2dev(obj);
6669         struct osd_thread_info *oti   = osd_oti_get(env);
6670         struct osd_inode_id    *id    = &oti->oti_id;
6671         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
6672         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
6673         __u32                   ino   = it->oie_dirent->oied_ino;
6674         int                     rc    = 0;
6675         ENTRY;
6676
6677         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
6678
6679         if (attr & LUDA_VERIFY) {
6680                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
6681                         attr |= LUDA_IGNORE;
6682                         /* If the parent is on remote MDT, and there
6683                          * is no FID-in-dirent, then we have to get
6684                          * the parent FID from the linkEA.  */
6685                         if (!fid_is_sane(fid) &&
6686                             it->oie_dirent->oied_namelen == 2 &&
6687                             it->oie_dirent->oied_name[0] == '.' &&
6688                             it->oie_dirent->oied_name[1] == '.')
6689                                 osd_get_pfid_from_linkea(env, obj, fid);
6690                 } else {
6691                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
6692                                                      &attr);
6693                 }
6694
6695                 if (!fid_is_sane(fid))
6696                         attr |= LUDA_UNKNOWN;
6697         } else {
6698                 attr &= ~LU_DIRENT_ATTRS_MASK;
6699                 if (!fid_is_sane(fid)) {
6700                         bool is_dotdot = false;
6701                         if (it->oie_dirent->oied_namelen == 2 &&
6702                             it->oie_dirent->oied_name[0] == '.' &&
6703                             it->oie_dirent->oied_name[1] == '.')
6704                                 is_dotdot = true;
6705                         /* If the parent is on remote MDT, and there
6706                          * is no FID-in-dirent, then we have to get
6707                          * the parent FID from the linkEA.  */
6708                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
6709                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
6710                         } else {
6711                                 if (is_dotdot == false &&
6712                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
6713                                         RETURN(-ENOENT);
6714
6715                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
6716                         }
6717                 } else {
6718                         osd_id_gen(id, ino, OSD_OII_NOGEN);
6719                 }
6720         }
6721
6722         /* Pack the entry anyway, at least the offset is right. */
6723         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
6724                            it->oie_dirent->oied_name,
6725                            it->oie_dirent->oied_namelen,
6726                            it->oie_dirent->oied_type, attr);
6727
6728         if (rc < 0)
6729                 RETURN(rc);
6730
6731         if (osd_remote_fid(env, dev, fid))
6732                 RETURN(0);
6733
6734         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
6735                 osd_add_oi_cache(oti, dev, id, fid);
6736
6737         RETURN(rc > 0 ? 0 : rc);
6738 }
6739
6740 /**
6741  * Returns the record size size at current position.
6742  *
6743  * This function will return record(lu_dirent) size in bytes.
6744  *
6745  * \param[in] env       execution environment
6746  * \param[in] di        iterator's in memory structure
6747  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
6748  *                      calculate the lu_dirent size.
6749  *
6750  * \retval      record size(in bytes & in memory) of the current lu_dirent
6751  *              entry.
6752  */
6753 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
6754                               __u32 attr)
6755 {
6756         struct osd_it_ea *it = (struct osd_it_ea *)di;
6757
6758         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
6759 }
6760
6761 /**
6762  * Returns a cookie for current position of the iterator head, so that
6763  * user can use this cookie to load/start the iterator next time.
6764  *
6765  * \param di iterator's in memory structure
6766  *
6767  * \retval cookie for current position, on success
6768  */
6769 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
6770 {
6771         struct osd_it_ea *it = (struct osd_it_ea *)di;
6772
6773         return it->oie_dirent->oied_off;
6774 }
6775
6776 /**
6777  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6778  * to load a directory entry at a time and stored it i inn,
6779  * in iterator's in-memory data structure.
6780  *
6781  * \param di struct osd_it_ea, iterator's in memory structure
6782  *
6783  * \retval +ve on success
6784  * \retval -ve on error
6785  */
6786 static int osd_it_ea_load(const struct lu_env *env,
6787                           const struct dt_it *di, __u64 hash)
6788 {
6789         struct osd_it_ea *it = (struct osd_it_ea *)di;
6790         int rc;
6791
6792         ENTRY;
6793         it->oie_file.f_pos = hash;
6794
6795         rc =  osd_ldiskfs_it_fill(env, di);
6796         if (rc > 0)
6797                 rc = -ENODATA;
6798
6799         if (rc == 0)
6800                 rc = +1;
6801
6802         RETURN(rc);
6803 }
6804
6805 /**
6806  * Index lookup function for interoperability mode (b11826).
6807  *
6808  * \param key,  key i.e. file name to be searched
6809  *
6810  * \retval +ve, on success
6811  * \retval -ve, on error
6812  */
6813 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
6814                                struct dt_rec *rec, const struct dt_key *key)
6815 {
6816         struct osd_object *obj = osd_dt_obj(dt);
6817         int rc = 0;
6818
6819         ENTRY;
6820
6821         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
6822         LINVRNT(osd_invariant(obj));
6823
6824         rc = osd_ea_lookup_rec(env, obj, rec, key);
6825         if (rc == 0)
6826                 rc = +1;
6827         RETURN(rc);
6828 }
6829
6830 /**
6831  * Index and Iterator operations for interoperability
6832  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
6833  */
6834 static const struct dt_index_operations osd_index_ea_ops = {
6835         .dio_lookup         = osd_index_ea_lookup,
6836         .dio_declare_insert = osd_index_declare_ea_insert,
6837         .dio_insert         = osd_index_ea_insert,
6838         .dio_declare_delete = osd_index_declare_ea_delete,
6839         .dio_delete         = osd_index_ea_delete,
6840         .dio_it     = {
6841                 .init     = osd_it_ea_init,
6842                 .fini     = osd_it_ea_fini,
6843                 .get      = osd_it_ea_get,
6844                 .put      = osd_it_ea_put,
6845                 .next     = osd_it_ea_next,
6846                 .key      = osd_it_ea_key,
6847                 .key_size = osd_it_ea_key_size,
6848                 .rec      = osd_it_ea_rec,
6849                 .rec_size = osd_it_ea_rec_size,
6850                 .store    = osd_it_ea_store,
6851                 .load     = osd_it_ea_load
6852         }
6853 };
6854
6855 static void *osd_key_init(const struct lu_context *ctx,
6856                           struct lu_context_key *key)
6857 {
6858         struct osd_thread_info *info;
6859
6860         OBD_ALLOC_PTR(info);
6861         if (info == NULL)
6862                 return ERR_PTR(-ENOMEM);
6863
6864         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6865         if (info->oti_it_ea_buf == NULL)
6866                 goto out_free_info;
6867
6868         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
6869
6870         info->oti_hlock = ldiskfs_htree_lock_alloc();
6871         if (info->oti_hlock == NULL)
6872                 goto out_free_ea;
6873
6874         return info;
6875
6876  out_free_ea:
6877         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6878  out_free_info:
6879         OBD_FREE_PTR(info);
6880         return ERR_PTR(-ENOMEM);
6881 }
6882
6883 static void osd_key_fini(const struct lu_context *ctx,
6884                          struct lu_context_key *key, void* data)
6885 {
6886         struct osd_thread_info *info = data;
6887         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
6888         struct osd_idmap_cache  *idc = info->oti_ins_cache;
6889
6890         if (info->oti_inode != NULL)
6891                 OBD_FREE_PTR(lli);
6892         if (info->oti_hlock != NULL)
6893                 ldiskfs_htree_lock_free(info->oti_hlock);
6894         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
6895         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
6896         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
6897         lu_buf_free(&info->oti_big_buf);
6898         if (idc != NULL) {
6899                 LASSERT(info->oti_ins_cache_size > 0);
6900                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
6901                 info->oti_ins_cache = NULL;
6902                 info->oti_ins_cache_size = 0;
6903         }
6904         OBD_FREE_PTR(info);
6905 }
6906
6907 static void osd_key_exit(const struct lu_context *ctx,
6908                          struct lu_context_key *key, void *data)
6909 {
6910         struct osd_thread_info *info = data;
6911
6912         LASSERT(info->oti_r_locks == 0);
6913         LASSERT(info->oti_w_locks == 0);
6914         LASSERT(info->oti_txns    == 0);
6915 }
6916
6917 /* type constructor/destructor: osd_type_init, osd_type_fini */
6918 LU_TYPE_INIT_FINI(osd, &osd_key);
6919
6920 struct lu_context_key osd_key = {
6921         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
6922         .lct_init = osd_key_init,
6923         .lct_fini = osd_key_fini,
6924         .lct_exit = osd_key_exit
6925 };
6926
6927
6928 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
6929                            const char *name, struct lu_device *next)
6930 {
6931         struct osd_device *osd = osd_dev(d);
6932
6933         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
6934             >= sizeof(osd->od_svname))
6935                 return -E2BIG;
6936         return osd_procfs_init(osd, name);
6937 }
6938
6939 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
6940 {
6941         struct seq_server_site  *ss = osd_seq_site(osd);
6942         int                     rc;
6943         ENTRY;
6944
6945         if (osd->od_is_ost || osd->od_cl_seq != NULL)
6946                 RETURN(0);
6947
6948         if (unlikely(ss == NULL))
6949                 RETURN(-ENODEV);
6950
6951         OBD_ALLOC_PTR(osd->od_cl_seq);
6952         if (osd->od_cl_seq == NULL)
6953                 RETURN(-ENOMEM);
6954
6955         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
6956                              osd->od_svname, ss->ss_server_seq);
6957         if (rc != 0) {
6958                 OBD_FREE_PTR(osd->od_cl_seq);
6959                 osd->od_cl_seq = NULL;
6960                 RETURN(rc);
6961         }
6962
6963         if (ss->ss_node_id == 0) {
6964                 /* If the OSD on the sequence controller(MDT0), then allocate
6965                  * sequence here, otherwise allocate sequence after connected
6966                  * to MDT0 (see mdt_register_lwp_callback()). */
6967                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
6968                                    &osd->od_cl_seq->lcs_space, env);
6969         }
6970
6971         RETURN(rc);
6972 }
6973
6974 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
6975 {
6976         if (osd->od_cl_seq == NULL)
6977                 return;
6978
6979         seq_client_fini(osd->od_cl_seq);
6980         OBD_FREE_PTR(osd->od_cl_seq);
6981         osd->od_cl_seq = NULL;
6982 }
6983
6984 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
6985 {
6986         ENTRY;
6987
6988         /* shutdown quota slave instance associated with the device */
6989         if (o->od_quota_slave != NULL) {
6990                 qsd_fini(env, o->od_quota_slave);
6991                 o->od_quota_slave = NULL;
6992         }
6993
6994         osd_fid_fini(env, o);
6995         osd_scrub_cleanup(env, o);
6996
6997         RETURN(0);
6998 }
6999
7000 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7001 {
7002         ENTRY;
7003
7004         if (o->od_mnt != NULL) {
7005                 shrink_dcache_sb(osd_sb(o));
7006                 osd_sync(env, &o->od_dt_dev);
7007
7008                 mntput(o->od_mnt);
7009                 o->od_mnt = NULL;
7010         }
7011
7012         EXIT;
7013 }
7014
7015 static int osd_mount(const struct lu_env *env,
7016                      struct osd_device *o, struct lustre_cfg *cfg)
7017 {
7018         const char              *name  = lustre_cfg_string(cfg, 0);
7019         const char              *dev  = lustre_cfg_string(cfg, 1);
7020         const char              *opts;
7021         unsigned long            page, s_flags, lmd_flags = 0;
7022         struct page             *__page;
7023         struct file_system_type *type;
7024         char                    *options = NULL;
7025         char                    *str;
7026         struct osd_thread_info  *info = osd_oti_get(env);
7027         struct lu_fid           *fid = &info->oti_fid;
7028         struct inode            *inode;
7029         int                      rc = 0, force_over_512tb = 0;
7030         ENTRY;
7031
7032         if (o->od_mnt != NULL)
7033                 RETURN(0);
7034
7035         if (strlen(dev) >= sizeof(o->od_mntdev))
7036                 RETURN(-E2BIG);
7037         strcpy(o->od_mntdev, dev);
7038
7039         str = lustre_cfg_string(cfg, 2);
7040         s_flags = simple_strtoul(str, NULL, 0);
7041         str = strstr(str, ":");
7042         if (str)
7043                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7044         opts = lustre_cfg_string(cfg, 3);
7045 #ifdef __BIG_ENDIAN
7046         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7047                 CERROR("%s: device %s extents feature is not guaranteed to "
7048                        "work on big-endian systems. Use \"bigendian_extents\" "
7049                        "mount option to override.\n", name, dev);
7050                 RETURN(-EINVAL);
7051         }
7052 #endif
7053 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7054         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7055                 CWARN("force_over_128tb option is deprecated. "
7056                       "Filesystems less than 512TB can be created without any "
7057                       "force options. Use force_over_512tb option for "
7058                       "filesystems greater than 512TB.\n");
7059         }
7060 #endif
7061 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7062         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7063                 CWARN("force_over_256tb option is deprecated. "
7064                       "Filesystems less than 512TB can be created without any "
7065                       "force options. Use force_over_512tb option for "
7066                       "filesystems greater than 512TB.\n");
7067         }
7068 #endif
7069
7070         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7071                 force_over_512tb = 1;
7072
7073         __page = alloc_page(GFP_KERNEL);
7074         if (__page == NULL)
7075                 GOTO(out, rc = -ENOMEM);
7076         page = (unsigned long)page_address(__page);
7077         options = (char *)page;
7078         *options = '\0';
7079         if (opts != NULL) {
7080                 /* strip out the options for back compatiblity */
7081                 static char *sout[] = {
7082                         "mballoc",
7083                         "iopen",
7084                         "noiopen",
7085                         "iopen_nopriv",
7086                         "extents",
7087                         "noextents",
7088                         /* strip out option we processed in osd */
7089                         "bigendian_extents",
7090                         "force_over_128tb",
7091                         "force_over_256tb",
7092                         "force_over_512tb",
7093                         NULL
7094                 };
7095                 strcat(options, opts);
7096                 for (rc = 0, str = options; sout[rc]; ) {
7097                         char *op = strstr(str, sout[rc]);
7098                         if (op == NULL) {
7099                                 rc++;
7100                                 str = options;
7101                                 continue;
7102                         }
7103                         if (op == options || *(op - 1) == ',') {
7104                                 str = op + strlen(sout[rc]);
7105                                 if (*str == ',' || *str == '\0') {
7106                                         *str == ',' ? str++ : str;
7107                                         memmove(op, str, strlen(str) + 1);
7108                                 }
7109                         }
7110                         for (str = op; *str != ',' && *str != '\0'; str++)
7111                                 ;
7112                 }
7113         } else {
7114                 strncat(options, "user_xattr,acl", 14);
7115         }
7116
7117         /* Glom up mount options */
7118         if (*options != '\0')
7119                 strcat(options, ",");
7120         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7121
7122         type = get_fs_type("ldiskfs");
7123         if (!type) {
7124                 CERROR("%s: cannot find ldiskfs module\n", name);
7125                 GOTO(out, rc = -ENODEV);
7126         }
7127
7128         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7129         module_put(type->owner);
7130
7131         if (IS_ERR(o->od_mnt)) {
7132                 rc = PTR_ERR(o->od_mnt);
7133                 o->od_mnt = NULL;
7134                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7135                 GOTO(out, rc);
7136         }
7137
7138         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7139                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7140                                  force_over_512tb == 0) {
7141                 CERROR("%s: device %s LDISKFS does not support filesystems "
7142                        "greater than 512TB and can cause data corruption. "
7143                        "Use \"force_over_512tb\" mount option to override.\n",
7144                        name, dev);
7145                 GOTO(out_mnt, rc = -EINVAL);
7146         }
7147
7148         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7149                 if (priv_dev_set_rdonly) {
7150                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7151                         o->od_dt_dev.dd_rdonly = 1;
7152                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7153                                       name);
7154                 } else {
7155                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7156                                       name);
7157
7158                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7159                 }
7160         } else if (priv_dev_check_rdonly &&
7161                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7162                 CERROR("%s: underlying device %s is marked as "
7163                        "read-only. Setup failed\n", name, dev);
7164
7165                 GOTO(out_mnt, rc = -EROFS);
7166         }
7167
7168         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
7169                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
7170                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7171                 GOTO(out_mnt, rc = -EINVAL);
7172         }
7173
7174 #ifdef LDISKFS_MOUNT_DIRDATA
7175         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
7176                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
7177                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7178         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7179                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7180                       "enabling the dirdata feature. If you do not want to "
7181                       "downgrade to Lustre-1.x again, you can enable it via "
7182                       "'tune2fs -O dirdata device'\n", name, dev);
7183 #endif
7184         inode = osd_sb(o)->s_root->d_inode;
7185         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7186         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7187         if (rc != 0) {
7188                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7189                 GOTO(out_mnt, rc);
7190         }
7191
7192         if (lmd_flags & LMD_FLG_NOSCRUB)
7193                 o->od_auto_scrub_interval = AS_NEVER;
7194
7195         GOTO(out, rc = 0);
7196
7197 out_mnt:
7198         mntput(o->od_mnt);
7199         o->od_mnt = NULL;
7200
7201 out:
7202         if (__page)
7203                 __free_page(__page);
7204
7205         return rc;
7206 }
7207
7208 static struct lu_device *osd_device_fini(const struct lu_env *env,
7209                                          struct lu_device *d)
7210 {
7211         struct osd_device *o = osd_dev(d);
7212         ENTRY;
7213
7214         osd_shutdown(env, o);
7215         osd_procfs_fini(o);
7216         osd_obj_map_fini(o);
7217         osd_umount(env, o);
7218
7219         RETURN(NULL);
7220 }
7221
7222 static int osd_device_init0(const struct lu_env *env,
7223                             struct osd_device *o,
7224                             struct lustre_cfg *cfg)
7225 {
7226         struct lu_device        *l = osd2lu_dev(o);
7227         struct osd_thread_info *info;
7228         int                     rc;
7229         int                     cplen = 0;
7230
7231         /* if the module was re-loaded, env can loose its keys */
7232         rc = lu_env_refill((struct lu_env *) env);
7233         if (rc)
7234                 GOTO(out, rc);
7235         info = osd_oti_get(env);
7236         LASSERT(info);
7237
7238         l->ld_ops = &osd_lu_ops;
7239         o->od_dt_dev.dd_ops = &osd_dt_ops;
7240
7241         spin_lock_init(&o->od_osfs_lock);
7242         mutex_init(&o->od_otable_mutex);
7243         INIT_LIST_HEAD(&o->od_orphan_list);
7244
7245         o->od_read_cache = 1;
7246         o->od_writethrough_cache = 1;
7247         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7248         o->od_auto_scrub_interval = AS_DEFAULT;
7249
7250         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7251                         sizeof(o->od_svname));
7252         if (cplen >= sizeof(o->od_svname)) {
7253                 rc = -E2BIG;
7254                 GOTO(out, rc);
7255         }
7256
7257         o->od_index = -1; /* -1 means index is invalid */
7258         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7259         if (rc == LDD_F_SV_TYPE_OST)
7260                 o->od_is_ost = 1;
7261
7262         o->od_full_scrub_ratio = OFSR_DEFAULT;
7263         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7264         rc = osd_mount(env, o, cfg);
7265         if (rc != 0)
7266                 GOTO(out, rc);
7267
7268         rc = osd_obj_map_init(env, o);
7269         if (rc != 0)
7270                 GOTO(out_mnt, rc);
7271
7272         rc = lu_site_init(&o->od_site, l);
7273         if (rc != 0)
7274                 GOTO(out_compat, rc);
7275         o->od_site.ls_bottom_dev = l;
7276
7277         rc = lu_site_init_finish(&o->od_site);
7278         if (rc != 0)
7279                 GOTO(out_site, rc);
7280
7281         INIT_LIST_HEAD(&o->od_ios_list);
7282         /* setup scrub, including OI files initialization */
7283         rc = osd_scrub_setup(env, o);
7284         if (rc < 0)
7285                 GOTO(out_site, rc);
7286
7287         rc = osd_procfs_init(o, o->od_svname);
7288         if (rc != 0) {
7289                 CERROR("%s: can't initialize procfs: rc = %d\n",
7290                        o->od_svname, rc);
7291                 GOTO(out_scrub, rc);
7292         }
7293
7294         LASSERT(l->ld_site->ls_linkage.next != NULL);
7295         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7296
7297         /* initialize quota slave instance */
7298         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
7299                                      o->od_proc_entry);
7300         if (IS_ERR(o->od_quota_slave)) {
7301                 rc = PTR_ERR(o->od_quota_slave);
7302                 o->od_quota_slave = NULL;
7303                 GOTO(out_procfs, rc);
7304         }
7305
7306         RETURN(0);
7307
7308 out_procfs:
7309         osd_procfs_fini(o);
7310 out_scrub:
7311         osd_scrub_cleanup(env, o);
7312 out_site:
7313         lu_site_fini(&o->od_site);
7314 out_compat:
7315         osd_obj_map_fini(o);
7316 out_mnt:
7317         osd_umount(env, o);
7318 out:
7319         return rc;
7320 }
7321
7322 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7323                                           struct lu_device_type *t,
7324                                           struct lustre_cfg *cfg)
7325 {
7326         struct osd_device *o;
7327         int                rc;
7328
7329         OBD_ALLOC_PTR(o);
7330         if (o == NULL)
7331                 return ERR_PTR(-ENOMEM);
7332
7333         rc = dt_device_init(&o->od_dt_dev, t);
7334         if (rc == 0) {
7335                 /* Because the ctx might be revived in dt_device_init,
7336                  * refill the env here */
7337                 lu_env_refill((struct lu_env *)env);
7338                 rc = osd_device_init0(env, o, cfg);
7339                 if (rc)
7340                         dt_device_fini(&o->od_dt_dev);
7341         }
7342
7343         if (unlikely(rc != 0))
7344                 OBD_FREE_PTR(o);
7345
7346         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7347 }
7348
7349 static struct lu_device *osd_device_free(const struct lu_env *env,
7350                                          struct lu_device *d)
7351 {
7352         struct osd_device *o = osd_dev(d);
7353         ENTRY;
7354
7355         /* XXX: make osd top device in order to release reference */
7356         d->ld_site->ls_top_dev = d;
7357         lu_site_purge(env, d->ld_site, -1);
7358         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7359                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7360                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7361         }
7362         lu_site_fini(&o->od_site);
7363         dt_device_fini(&o->od_dt_dev);
7364         OBD_FREE_PTR(o);
7365         RETURN(NULL);
7366 }
7367
7368 static int osd_process_config(const struct lu_env *env,
7369                               struct lu_device *d, struct lustre_cfg *cfg)
7370 {
7371         struct osd_device               *o = osd_dev(d);
7372         int                             rc;
7373         ENTRY;
7374
7375         switch (cfg->lcfg_command) {
7376         case LCFG_SETUP:
7377                 rc = osd_mount(env, o, cfg);
7378                 break;
7379         case LCFG_CLEANUP:
7380                 lu_dev_del_linkage(d->ld_site, d);
7381                 rc = osd_shutdown(env, o);
7382                 break;
7383         case LCFG_PARAM:
7384                 LASSERT(&o->od_dt_dev);
7385                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
7386                                               cfg, &o->od_dt_dev);
7387                 if (rc > 0 || rc == -ENOSYS) {
7388                         rc = class_process_proc_param(PARAM_OST,
7389                                                       lprocfs_osd_obd_vars,
7390                                                       cfg, &o->od_dt_dev);
7391                         if (rc > 0)
7392                                 rc = 0;
7393                 }
7394                 break;
7395         default:
7396                 rc = -ENOSYS;
7397         }
7398
7399         RETURN(rc);
7400 }
7401
7402 static int osd_recovery_complete(const struct lu_env *env,
7403                                  struct lu_device *d)
7404 {
7405         struct osd_device       *osd = osd_dev(d);
7406         int                      rc = 0;
7407         ENTRY;
7408
7409         if (osd->od_quota_slave == NULL)
7410                 RETURN(0);
7411
7412         /* start qsd instance on recovery completion, this notifies the quota
7413          * slave code that we are about to process new requests now */
7414         rc = qsd_start(env, osd->od_quota_slave);
7415         RETURN(rc);
7416 }
7417
7418 /*
7419  * we use exports to track all osd users
7420  */
7421 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7422                            struct obd_device *obd, struct obd_uuid *cluuid,
7423                            struct obd_connect_data *data, void *localdata)
7424 {
7425         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
7426         struct lustre_handle  conn;
7427         int                   rc;
7428         ENTRY;
7429
7430         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7431
7432         rc = class_connect(&conn, obd, cluuid);
7433         if (rc)
7434                 RETURN(rc);
7435
7436         *exp = class_conn2export(&conn);
7437
7438         spin_lock(&osd->od_osfs_lock);
7439         osd->od_connects++;
7440         spin_unlock(&osd->od_osfs_lock);
7441
7442         RETURN(0);
7443 }
7444
7445 /*
7446  * once last export (we don't count self-export) disappeared
7447  * osd can be released
7448  */
7449 static int osd_obd_disconnect(struct obd_export *exp)
7450 {
7451         struct obd_device *obd = exp->exp_obd;
7452         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7453         int                rc, release = 0;
7454         ENTRY;
7455
7456         /* Only disconnect the underlying layers on the final disconnect. */
7457         spin_lock(&osd->od_osfs_lock);
7458         osd->od_connects--;
7459         if (osd->od_connects == 0)
7460                 release = 1;
7461         spin_unlock(&osd->od_osfs_lock);
7462
7463         rc = class_disconnect(exp); /* bz 9811 */
7464
7465         if (rc == 0 && release)
7466                 class_manual_cleanup(obd);
7467         RETURN(rc);
7468 }
7469
7470 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
7471                        struct lu_device *dev)
7472 {
7473         struct osd_device       *osd    = osd_dev(dev);
7474         struct lr_server_data   *lsd    =
7475                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
7476         int                      result = 0;
7477         ENTRY;
7478
7479         if (osd->od_quota_slave != NULL) {
7480                 /* set up quota slave objects */
7481                 result = qsd_prepare(env, osd->od_quota_slave);
7482                 if (result != 0)
7483                         RETURN(result);
7484         }
7485
7486         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
7487 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
7488                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
7489                         osd->od_index_in_idif = 1;
7490                 } else {
7491                         osd->od_index_in_idif = 0;
7492                         result = osd_register_proc_index_in_idif(osd);
7493                         if (result != 0)
7494                                 RETURN(result);
7495                 }
7496 #else
7497                 osd->od_index_in_idif = 1;
7498 #endif
7499         }
7500
7501         result = osd_fid_init(env, osd);
7502
7503         RETURN(result);
7504 }
7505
7506 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
7507                          struct lu_fid *fid, struct md_op_data *op_data)
7508 {
7509         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
7510
7511         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
7512 }
7513
7514 static const struct lu_object_operations osd_lu_obj_ops = {
7515         .loo_object_init      = osd_object_init,
7516         .loo_object_delete    = osd_object_delete,
7517         .loo_object_release   = osd_object_release,
7518         .loo_object_free      = osd_object_free,
7519         .loo_object_print     = osd_object_print,
7520         .loo_object_invariant = osd_object_invariant
7521 };
7522
7523 const struct lu_device_operations osd_lu_ops = {
7524         .ldo_object_alloc      = osd_object_alloc,
7525         .ldo_process_config    = osd_process_config,
7526         .ldo_recovery_complete = osd_recovery_complete,
7527         .ldo_prepare           = osd_prepare,
7528 };
7529
7530 static const struct lu_device_type_operations osd_device_type_ops = {
7531         .ldto_init = osd_type_init,
7532         .ldto_fini = osd_type_fini,
7533
7534         .ldto_start = osd_type_start,
7535         .ldto_stop  = osd_type_stop,
7536
7537         .ldto_device_alloc = osd_device_alloc,
7538         .ldto_device_free  = osd_device_free,
7539
7540         .ldto_device_init    = osd_device_init,
7541         .ldto_device_fini    = osd_device_fini
7542 };
7543
7544 static struct lu_device_type osd_device_type = {
7545         .ldt_tags     = LU_DEVICE_DT,
7546         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
7547         .ldt_ops      = &osd_device_type_ops,
7548         .ldt_ctx_tags = LCT_LOCAL,
7549 };
7550
7551 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
7552 {
7553         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7554         struct super_block *sb = osd_sb(osd);
7555
7556         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
7557 }
7558
7559 /*
7560  * lprocfs legacy support.
7561  */
7562 static struct obd_ops osd_obd_device_ops = {
7563         .o_owner = THIS_MODULE,
7564         .o_connect      = osd_obd_connect,
7565         .o_disconnect   = osd_obd_disconnect,
7566         .o_fid_alloc    = osd_fid_alloc,
7567         .o_health_check = osd_health_check,
7568 };
7569
7570 static int __init osd_init(void)
7571 {
7572         int rc;
7573
7574         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
7575 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
7576         /* please, try to keep osd_thread_info smaller than a page */
7577         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
7578 #endif
7579
7580         osd_oi_mod_init();
7581
7582         rc = lu_kmem_init(ldiskfs_caches);
7583         if (rc)
7584                 return rc;
7585
7586 #ifdef CONFIG_KALLSYMS
7587         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
7588         priv_dev_check_rdonly = (void *)kallsyms_lookup_name("dev_check_rdonly");
7589         /* Clear readonly is unused at this time */
7590         /*priv_dev_clear_rdonly = (void *)kallsyms_lookup_name("dev_clear_rdonly");*/
7591 #endif
7592
7593         rc = class_register_type(&osd_obd_device_ops, NULL, true,
7594                                  lprocfs_osd_module_vars,
7595                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
7596         if (rc)
7597                 lu_kmem_fini(ldiskfs_caches);
7598         return rc;
7599 }
7600
7601 static void __exit osd_exit(void)
7602 {
7603         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
7604         lu_kmem_fini(ldiskfs_caches);
7605 }
7606
7607 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
7608 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
7609 MODULE_VERSION(LUSTRE_VERSION_STRING);
7610 MODULE_LICENSE("GPL");
7611
7612 module_init(osd_init);
7613 module_exit(osd_exit);