lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
 499                               "maximum entry limit\n",
 500                               osd_name(osd), parent->i_ino, PFID(fid),
 501                               rc == -ENOSPC ? "has reached" : "is approaching");
 502                 else
 503                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
 504                               "maximum entry limit\n",
 505                               osd_name(osd), parent->i_ino,
 506                               rc == -ENOSPC ? "has reached" : "is approaching");
 507
 508                 /* ignore such error now */
 509                 if (rc == -ENOBUFS)
 510                         rc = 0;
 511         }
 512
 513         return rc;
 514 }
 515
 516
 517 struct inode *
 518 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 519              struct osd_inode_id *id, struct lu_fid *fid)
 520 {
 521         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 522         struct inode *inode;
 523         int rc;
 524
 525         inode = osd_iget(info, dev, id);
 526         if (IS_ERR(inode))
 527                 return inode;
 528
 529         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 530         if (!rc) {
 531                 *fid = loa->loa_lma.lma_self_fid;
 532         } else if (rc == -ENODATA) {
 533                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 534                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 535                 else
 536                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 537         } else {
 538                 iput(inode);
 539                 inode = ERR_PTR(rc);
 540         }
 541         return inode;
 542 }
 543
 544 static struct inode *osd_iget_check(struct osd_thread_info *info,
 545                                     struct osd_device *dev,
 546                                     const struct lu_fid *fid,
 547                                     struct osd_inode_id *id,
 548                                     bool trusted)
 549 {
 550         struct inode *inode;
 551         int rc = 0;
 552
 553         ENTRY;
 554
 555         /*
 556          * The cached OI mapping is trustable. If we cannot locate the inode
 557          * via the cached OI mapping, then return the failure to the caller
 558          * directly without further OI checking.
 559          */
 560
 561 again:
 562         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_18_23, contains self-fid
 711  * \retval -v: other failure cases
 712  */
 713 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 714                  struct dentry *dentry, struct lu_fid *fid)
 715 {
 716         struct filter_fid *ff = &info->oti_ff;
 717         struct ost_id *ostid = &info->oti_ostid;
 718         int rc;
 719
 720         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 721         if (rc == sizeof(struct filter_fid_18_23)) {
 722                 struct filter_fid_18_23 *ff_old = (void *)ff;
 723
 724                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 725                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 726                 /*
 727                  * XXX: use 0 as the index for compatibility, the caller will
 728                  * handle index related issues when necessary.
 729                  */
 730                 if (!rc)
 731                         ostid_to_fid(fid, ostid, 0);
 732         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 733                 rc = 1;
 734         } else if (rc >= 0) {
 735                 rc = -EINVAL;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int osd_lma_self_repair(struct osd_thread_info *info,
 742                                struct osd_device *osd, struct inode *inode,
 743                                const struct lu_fid *fid, __u32 compat)
 744 {
 745         handle_t *jh;
 746         int rc;
 747
 748         LASSERT(current->journal_info == NULL);
 749
 750         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 751                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 752         if (IS_ERR(jh)) {
 753                 rc = PTR_ERR(jh);
 754                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 755                       osd_name(osd), rc);
 756                 return rc;
 757         }
 758
 759         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 760         if (rc != 0)
 761                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 762                       osd_name(osd), rc);
 763         ldiskfs_journal_stop(jh);
 764         return rc;
 765 }
 766
 767 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 768 {
 769         struct osd_thread_info *info = osd_oti_get(env);
 770         struct osd_device *osd = osd_obj2dev(obj);
 771         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 772         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 773         struct inode *inode = obj->oo_inode;
 774         struct dentry *dentry = &info->oti_obj_dentry;
 775         struct lu_fid *fid = NULL;
 776         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 777         int rc;
 778
 779         ENTRY;
 780
 781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 782                              (void *)loa, sizeof(*loa));
 783         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 784                 fid = &lma->lma_self_fid;
 785                 rc = osd_get_idif(info, inode, dentry, fid);
 786                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 787                         /*
 788                          * For the given OST-object, if it has neither LMA nor
 789                          * FID in XATTR_NAME_FID, then the given FID (which is
 790                          * contained in the @obj, from client RPC for locating
 791                          * the OST-object) is trusted. We use it to generate
 792                          * the LMA.
 793                          */
 794                         osd_lma_self_repair(info, osd, inode, rfid,
 795                                             LMAC_FID_ON_OST);
 796                         RETURN(0);
 797                 }
 798         }
 799
 800         if (rc < 0)
 801                 RETURN(rc);
 802
 803         if (rc > 0) {
 804                 rc = 0;
 805                 lustre_lma_swab(lma);
 806                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 807                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 808                               S_ISREG(inode->i_mode)))) {
 809                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 810                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 811                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 812                               PFID(rfid), inode->i_ino);
 813                         rc = -EOPNOTSUPP;
 814                 } else {
 815                         fid = &lma->lma_self_fid;
 816                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 817                             osd->od_is_ost)
 818                                 obj->oo_pfid_in_lma = 1;
 819                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 820                             !osd->od_is_ost)
 821                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 822                 }
 823         }
 824
 825         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 826                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 827                         struct ost_id   *oi   = &info->oti_ostid;
 828                         struct lu_fid   *fid1 = &info->oti_fid3;
 829                         __u32            idx  = fid_idif_ost_idx(rfid);
 830
 831                         /*
 832                          * For old IDIF, the OST index is not part of the IDIF,
 833                          * Means that different OSTs may have the same IDIFs.
 834                          * Under such case, we need to make some compatible
 835                          * check to make sure to trigger OI scrub properly.
 836                          */
 837                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 838                                 /* Given @rfid is new, LMA is old. */
 839                                 fid_to_ostid(fid, oi);
 840                                 ostid_to_fid(fid1, oi, idx);
 841                                 if (lu_fid_eq(fid1, rfid)) {
 842                                         if (osd->od_index_in_idif)
 843                                                 osd_lma_self_repair(info, osd,
 844                                                         inode, rfid,
 845                                                         LMAC_FID_ON_OST);
 846                                         RETURN(0);
 847                                 }
 848                         }
 849                 }
 850
 851                 rc = -EREMCHG;
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 struct osd_check_lmv_buf {
 858 #ifdef HAVE_DIR_CONTEXT
 859         /* please keep it as first member */
 860         struct dir_context ctx;
 861 #endif
 862         struct osd_thread_info *oclb_info;
 863         struct osd_device *oclb_dev;
 864         struct osd_idmap_cache *oclb_oic;
 865         int oclb_items;
 866         bool oclb_found;
 867 };
 868
 869 /**
 870  * It is called internally by ->readdir() to filter out the
 871  * local slave object's FID of the striped directory.
 872  *
 873  * \retval      1 found the local slave's FID
 874  * \retval      0 continue to check next item
 875  * \retval      -ve for failure
 876  */
 877 #ifdef HAVE_FILLDIR_USE_CTX
 878 static int osd_stripe_dir_filldir(struct dir_context *buf,
 879 #else
 880 static int osd_stripe_dir_filldir(void *buf,
 881 #endif
 882                                   const char *name, int namelen,
 883                                   loff_t offset, __u64 ino, unsigned int d_type)
 884 {
 885         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 886         struct osd_thread_info *oti = oclb->oclb_info;
 887         struct lu_fid *fid = &oti->oti_fid3;
 888         struct osd_inode_id *id = &oti->oti_id3;
 889         struct osd_device *dev = oclb->oclb_dev;
 890         struct osd_idmap_cache *oic = oclb->oclb_oic;
 891         struct inode *inode;
 892
 893         oclb->oclb_items++;
 894
 895         if (name[0] == '.')
 896                 return 0;
 897
 898         fid_zero(fid);
 899         sscanf(name + 1, SFID, RFID(fid));
 900         if (!fid_is_sane(fid))
 901                 return 0;
 902
 903         if (osd_remote_fid(oti->oti_env, dev, fid))
 904                 return 0;
 905
 906         osd_id_gen(id, ino, OSD_OII_NOGEN);
 907         inode = osd_iget(oti, dev, id);
 908         if (IS_ERR(inode))
 909                 return PTR_ERR(inode);
 910
 911         iput(inode);
 912         osd_add_oi_cache(oti, dev, id, fid);
 913         oic->oic_fid = *fid;
 914         oic->oic_lid = *id;
 915         oic->oic_dev = dev;
 916         osd_oii_insert(dev, oic, true);
 917         oclb->oclb_found = true;
 918
 919         return 1;
 920 }
 921
 922 /*
 923  * When lookup item under striped directory, we need to locate the master
 924  * MDT-object of the striped directory firstly, then the client will send
 925  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 926  * and the item's name. If the system is restored from MDT file level backup,
 927  * then before the OI scrub completely built the OI files, the OI mappings of
 928  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 929  * not a problem for the master MDT-object. Because when locate the master
 930  * MDT-object, we will do name based lookup (for the striped directory itself)
 931  * firstly, during such process we can setup the correct OI mapping for the
 932  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 933  * the client will not trigger name based lookup on the MDT to locate the slave
 934  * MDT-object before locating item under the striped directory, then when
 935  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 936  * is invalid and does not know what the right OI mapping is, then the MDT has
 937  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 938  * the OI file, related OI mapping is unknown yet, please try again later. And
 939  * then client will re-try the RPC again and again until related OI mapping has
 940  * been updated. That is quite inefficient.
 941  *
 942  * To resolve above trouble, we will handle it as the following two cases:
 943  *
 944  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 945  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 946  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 947  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 948  *    directly. Please check osd_fid_lookup().
 949  *
 950  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 951  *    Under such case, during lookup the master MDT-object, we will lookup the
 952  *    slave MDT-object via readdir against the master MDT-object, because the
 953  *    slave MDT-objects information are stored as sub-directories with the name
 954  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 955  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 956  *    the correct OI mapping for the slave MDT-object.
 957  */
 958 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 959                          struct inode *inode, struct osd_idmap_cache *oic)
 960 {
 961         struct lu_buf *buf = &oti->oti_big_buf;
 962         struct dentry *dentry = &oti->oti_obj_dentry;
 963         struct file *filp = &oti->oti_file;
 964         const struct file_operations *fops;
 965         struct lmv_mds_md_v1 *lmv1;
 966         struct osd_check_lmv_buf oclb = {
 967 #ifdef HAVE_DIR_CONTEXT
 968                 .ctx.actor = osd_stripe_dir_filldir,
 969 #endif
 970                 .oclb_info = oti,
 971                 .oclb_dev = dev,
 972                 .oclb_oic = oic,
 973                 .oclb_found = false,
 974         };
 975         int rc = 0;
 976
 977         ENTRY;
 978
 979 again:
 980         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 981                              buf->lb_len);
 982         if (rc == -ERANGE) {
 983                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 984                 if (rc > 0) {
 985                         lu_buf_realloc(buf, rc);
 986                         if (buf->lb_buf == NULL)
 987                                 GOTO(out, rc = -ENOMEM);
 988
 989                         goto again;
 990                 }
 991         }
 992
 993         if (unlikely(rc == 0 || rc == -ENODATA))
 994                 GOTO(out, rc = 0);
 995
 996         if (rc < 0)
 997                 GOTO(out, rc);
 998
 999         if (unlikely(buf->lb_buf == NULL)) {
1000                 lu_buf_realloc(buf, rc);
1001                 if (buf->lb_buf == NULL)
1002                         GOTO(out, rc = -ENOMEM);
1003
1004                 goto again;
1005         }
1006
1007         lmv1 = buf->lb_buf;
1008         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1009                 GOTO(out, rc = 0);
1010
1011         fops = inode->i_fop;
1012         dentry->d_inode = inode;
1013         dentry->d_sb = inode->i_sb;
1014         filp->f_pos = 0;
1015         filp->f_path.dentry = dentry;
1016         filp->f_mode = FMODE_64BITHASH;
1017         filp->f_mapping = inode->i_mapping;
1018         filp->f_op = fops;
1019         filp->private_data = NULL;
1020         set_file_inode(filp, inode);
1021
1022         do {
1023                 oclb.oclb_items = 0;
1024 #ifdef HAVE_DIR_CONTEXT
1025                 oclb.ctx.pos = filp->f_pos;
1026 #ifdef HAVE_ITERATE_SHARED
1027                 rc = fops->iterate_shared(filp, &oclb.ctx);
1028 #else
1029                 rc = fops->iterate(filp, &oclb.ctx);
1030 #endif
1031                 filp->f_pos = oclb.ctx.pos;
1032 #else
1033                 rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
1034 #endif
1035         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1036                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1037         fops->release(inode, filp);
1038
1039 out:
1040         if (rc < 0)
1041                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1042                        DFID": rc = %d\n", osd_ino2name(inode),
1043                        inode->i_ino, inode->i_generation,
1044                        PFID(&oic->oic_fid), rc);
1045         else
1046                 rc = 0;
1047
1048         RETURN(rc);
1049 }
1050
1051 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1052                           const struct lu_fid *fid,
1053                           const struct lu_object_conf *conf)
1054 {
1055         struct osd_thread_info *info;
1056         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1057         struct osd_device *dev;
1058         struct osd_idmap_cache *oic;
1059         struct osd_inode_id *id;
1060         struct inode *inode = NULL;
1061         struct lustre_scrub *scrub;
1062         struct scrub_file *sf;
1063         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1064         __u32 saved_ino;
1065         __u32 saved_gen;
1066         int result = 0;
1067         int rc1 = 0;
1068         bool remote = false;
1069         bool trusted = true;
1070         bool updated = false;
1071         bool checked = false;
1072
1073         ENTRY;
1074
1075         LINVRNT(osd_invariant(obj));
1076         LASSERT(obj->oo_inode == NULL);
1077         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1078
1079         dev = osd_dev(ldev);
1080         scrub = &dev->od_scrub.os_scrub;
1081         sf = &scrub->os_file;
1082         info = osd_oti_get(env);
1083         LASSERT(info);
1084         oic = &info->oti_cache;
1085
1086         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1087                 RETURN(-ENOENT);
1088
1089         /*
1090          * For the object is created as locking anchor, or for the object to
1091          * be created on disk. No need to osd_oi_lookup() at here because FID
1092          * shouldn't never be re-used, if it's really a duplicate FID from
1093          * unexpected reason, we should be able to detect it later by calling
1094          * do_create->osd_oi_insert().
1095          */
1096         if (conf && conf->loc_flags & LOC_F_NEW)
1097                 GOTO(out, result = 0);
1098
1099         /* Search order: 1. per-thread cache. */
1100         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1101                 id = &oic->oic_lid;
1102                 goto iget;
1103         }
1104
1105         id = &info->oti_id;
1106         if (!list_empty(&scrub->os_inconsistent_items)) {
1107                 /* Search order: 2. OI scrub pending list. */
1108                 result = osd_oii_lookup(dev, fid, id);
1109                 if (!result)
1110                         goto iget;
1111         }
1112
1113         /*
1114          * The OI mapping in the OI file can be updated by the OI scrub
1115          * when we locate the inode via FID. So it may be not trustable.
1116          */
1117         trusted = false;
1118
1119         /* Search order: 3. OI files. */
1120         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1121         if (result == -ENOENT) {
1122                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1123                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1124                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1125                                       sf->sf_oi_bitmap))
1126                         GOTO(out, result = 0);
1127
1128                 goto trigger;
1129         }
1130
1131         if (result)
1132                 GOTO(out, result);
1133
1134 iget:
1135         obj->oo_inode = NULL;
1136         /* for later passes through checks, not true on first pass */
1137         if (!IS_ERR_OR_NULL(inode))
1138                 iput(inode);
1139
1140         inode = osd_iget_check(info, dev, fid, id, trusted);
1141         if (!IS_ERR(inode)) {
1142                 obj->oo_inode = inode;
1143                 result = 0;
1144                 if (remote)
1145                         goto trigger;
1146
1147                 goto check_lma;
1148         }
1149
1150         result = PTR_ERR(inode);
1151         if (result == -ENOENT || result == -ESTALE)
1152                 GOTO(out, result = 0);
1153
1154         if (result != -EREMCHG)
1155                 GOTO(out, result);
1156
1157 trigger:
1158         /*
1159          * We still have chance to get the valid inode: for the
1160          * object which is referenced by remote name entry, the
1161          * object on the local MDT will be linked under the dir
1162          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1163          *
1164          * We do not know whether the object for the given FID
1165          * is referenced by some remote name entry or not, and
1166          * especially for DNE II, a multiple-linked object may
1167          * have many name entries reside on many MDTs.
1168          *
1169          * To simplify the operation, OSD will not distinguish
1170          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1171          * only happened for the RPC from other MDT during the
1172          * OI scrub, or for the client side RPC with FID only,
1173          * such as FID to path, or from old connected client.
1174          */
1175         if (!remote) {
1176                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1177                 if (!rc1) {
1178                         remote = true;
1179                         trusted = true;
1180                         flags |= SS_AUTO_PARTIAL;
1181                         flags &= ~SS_AUTO_FULL;
1182                         goto iget;
1183                 }
1184         }
1185
1186         if (thread_is_running(&scrub->os_thread)) {
1187                 if (scrub->os_partial_scan && !scrub->os_in_join)
1188                         goto join;
1189
1190                 if (IS_ERR_OR_NULL(inode) || result)
1191                         GOTO(out, result = -EINPROGRESS);
1192
1193                 LASSERT(remote);
1194                 LASSERT(obj->oo_inode == inode);
1195
1196                 osd_add_oi_cache(info, dev, id, fid);
1197                 osd_oii_insert(dev, oic, true);
1198                 goto found;
1199         }
1200
1201         if (dev->od_auto_scrub_interval == AS_NEVER) {
1202                 if (!remote)
1203                         GOTO(out, result = -EREMCHG);
1204
1205                 LASSERT(!result);
1206                 LASSERT(obj->oo_inode == inode);
1207
1208                 osd_add_oi_cache(info, dev, id, fid);
1209                 goto found;
1210         }
1211
1212 join:
1213         rc1 = osd_scrub_start(env, dev, flags);
1214         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1215                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1216         if (rc1 && rc1 != -EALREADY)
1217                 GOTO(out, result = -EREMCHG);
1218
1219         if (IS_ERR_OR_NULL(inode) || result)
1220                 GOTO(out, result = -EINPROGRESS);
1221
1222         LASSERT(remote);
1223         LASSERT(obj->oo_inode == inode);
1224
1225         osd_add_oi_cache(info, dev, id, fid);
1226         osd_oii_insert(dev, oic, true);
1227         goto found;
1228
1229 check_lma:
1230         checked = true;
1231         if (unlikely(obj->oo_header))
1232                 goto found;
1233
1234         result = osd_check_lma(env, obj);
1235         if (!result)
1236                 goto found;
1237
1238         LASSERTF(id->oii_ino == inode->i_ino &&
1239                  id->oii_gen == inode->i_generation,
1240                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1241                  PFID(fid), id->oii_ino, id->oii_gen,
1242                  inode->i_ino, inode->i_generation);
1243
1244         saved_ino = inode->i_ino;
1245         saved_gen = inode->i_generation;
1246
1247         if (unlikely(result == -ENODATA)) {
1248                 /*
1249                  * If the OI scrub updated the OI mapping by race, it
1250                  * must be valid. Trust the inode that has no LMA EA.
1251                  */
1252                 if (updated)
1253                         goto found;
1254
1255                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1256                 if (!result) {
1257                         /*
1258                          * The OI mapping is still there, the inode is still
1259                          * valid. It is just becaues the inode has no LMA EA.
1260                          */
1261                         if (saved_ino == id->oii_ino &&
1262                             saved_gen == id->oii_gen)
1263                                 goto found;
1264
1265                         /*
1266                          * It is the OI scrub updated the OI mapping by race.
1267                          * The new OI mapping must be valid.
1268                          */
1269                         trusted = true;
1270                         updated = true;
1271                         goto iget;
1272                 }
1273
1274                 /*
1275                  * "result == -ENOENT" means that the OI mappinghas been
1276                  * removed by race, so the inode belongs to other object.
1277                  *
1278                  * Others error can be returned  directly.
1279                  */
1280                 if (result == -ENOENT) {
1281                         LASSERT(trusted);
1282
1283                         obj->oo_inode = NULL;
1284                         result = 0;
1285                 }
1286         }
1287
1288         if (result != -EREMCHG)
1289                 GOTO(out, result);
1290
1291         LASSERT(!updated);
1292
1293         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1294         /*
1295          * "result == -ENOENT" means the cached OI mapping has been removed
1296          * from the OI file by race, above inode belongs to other object.
1297          */
1298         if (result == -ENOENT) {
1299                 LASSERT(trusted);
1300
1301                 obj->oo_inode = NULL;
1302                 GOTO(out, result = 0);
1303         }
1304
1305         if (result)
1306                 GOTO(out, result);
1307
1308         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1309                 result = -EREMCHG;
1310                 goto trigger;
1311         }
1312
1313         /*
1314          * It is the OI scrub updated the OI mapping by race.
1315          * The new OI mapping must be valid.
1316          */
1317         trusted = true;
1318         updated = true;
1319         goto iget;
1320
1321 found:
1322         if (!checked) {
1323                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1324                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1325
1326                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1327                 if (!result) {
1328                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1329                             dev->od_is_ost)
1330                                 obj->oo_pfid_in_lma = 1;
1331                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1332                             !dev->od_is_ost)
1333                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1334                 } else if (result != -ENODATA) {
1335                         GOTO(out, result);
1336                 }
1337         }
1338
1339         obj->oo_compat_dot_created = 1;
1340         obj->oo_compat_dotdot_created = 1;
1341
1342         if (S_ISDIR(inode->i_mode) &&
1343             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1344                 osd_check_lmv(info, dev, inode, oic);
1345
1346         result = osd_attach_jinode(inode);
1347         if (result)
1348                 GOTO(out, result);
1349
1350         if (!ldiskfs_pdo)
1351                 GOTO(out, result = 0);
1352
1353         LASSERT(!obj->oo_hl_head);
1354         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1355
1356         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1357
1358 out:
1359         if (result || !obj->oo_inode) {
1360                 if (!IS_ERR_OR_NULL(inode))
1361                         iput(inode);
1362
1363                 obj->oo_inode = NULL;
1364                 if (trusted)
1365                         fid_zero(&oic->oic_fid);
1366         }
1367
1368         LINVRNT(osd_invariant(obj));
1369         return result;
1370 }
1371
1372 /*
1373  * Concurrency: shouldn't matter.
1374  */
1375 static void osd_object_init0(struct osd_object *obj)
1376 {
1377         LASSERT(obj->oo_inode != NULL);
1378         obj->oo_dt.do_body_ops = &osd_body_ops;
1379         obj->oo_dt.do_lu.lo_header->loh_attr |=
1380                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1381 }
1382
1383 /*
1384  * Concurrency: no concurrent access is possible that early in object
1385  * life-cycle.
1386  */
1387 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1388                            const struct lu_object_conf *conf)
1389 {
1390         struct osd_object *obj = osd_obj(l);
1391         int result;
1392
1393         LINVRNT(osd_invariant(obj));
1394
1395         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1396                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1397                 l->lo_header->loh_attr |= LOHA_EXISTS;
1398                 return 0;
1399         }
1400
1401         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1402         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1403         if (result == 0 && obj->oo_inode != NULL) {
1404                 struct osd_thread_info *oti = osd_oti_get(env);
1405                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1406
1407                 osd_object_init0(obj);
1408                 if (unlikely(obj->oo_header))
1409                         return 0;
1410
1411                 result = osd_get_lma(oti, obj->oo_inode,
1412                                      &oti->oti_obj_dentry, loa);
1413                 if (!result) {
1414                         /*
1415                          * Convert LMAI flags to lustre LMA flags
1416                          * and cache it to oo_lma_flags
1417                          */
1418                         obj->oo_lma_flags =
1419                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1420                 } else if (result == -ENODATA) {
1421                         result = 0;
1422                 }
1423         }
1424
1425         LINVRNT(osd_invariant(obj));
1426         return result;
1427 }
1428
1429 /*
1430  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1431  * The left part is for value, binary mode.
1432  */
1433 struct osd_xattr_entry {
1434         struct list_head        oxe_list;
1435         size_t                  oxe_len;
1436         size_t                  oxe_namelen;
1437         bool                    oxe_exist;
1438         struct rcu_head         oxe_rcu;
1439         char                    oxe_buf[0];
1440 };
1441
1442 static int osd_oxc_get(struct osd_object *obj, const char *name,
1443                        struct lu_buf *buf)
1444 {
1445         struct osd_xattr_entry *tmp;
1446         struct osd_xattr_entry *oxe = NULL;
1447         size_t namelen = strlen(name);
1448         int rc;
1449
1450         ENTRY;
1451
1452         rcu_read_lock();
1453         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1454                 if (namelen == tmp->oxe_namelen &&
1455                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1456                         oxe = tmp;
1457                         break;
1458                 }
1459         }
1460
1461         if (oxe == NULL)
1462                 GOTO(out, rc = -ENOENT);
1463
1464         if (!oxe->oxe_exist)
1465                 GOTO(out, rc = -ENODATA);
1466
1467         /* vallen */
1468         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1469         LASSERT(rc > 0);
1470
1471         if (buf->lb_buf == NULL)
1472                 GOTO(out, rc);
1473
1474         if (buf->lb_len < rc)
1475                 GOTO(out, rc = -ERANGE);
1476
1477         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1478         EXIT;
1479 out:
1480         rcu_read_unlock();
1481
1482         return rc;
1483 }
1484
1485 static void osd_oxc_free(struct rcu_head *head)
1486 {
1487         struct osd_xattr_entry *oxe;
1488
1489         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1490         OBD_FREE(oxe, oxe->oxe_len);
1491 }
1492
1493 static void osd_oxc_add(struct osd_object *obj, const char *name,
1494                         const char *buf, int buflen)
1495 {
1496         struct osd_xattr_entry *oxe;
1497         struct osd_xattr_entry *old = NULL;
1498         struct osd_xattr_entry *tmp;
1499         size_t namelen = strlen(name);
1500         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1501
1502         OBD_ALLOC(oxe, len);
1503         if (oxe == NULL)
1504                 return;
1505
1506         INIT_LIST_HEAD(&oxe->oxe_list);
1507         oxe->oxe_len = len;
1508         oxe->oxe_namelen = namelen;
1509         memcpy(oxe->oxe_buf, name, namelen);
1510         if (buflen > 0) {
1511                 LASSERT(buf != NULL);
1512                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1513                 oxe->oxe_exist = true;
1514         } else {
1515                 oxe->oxe_exist = false;
1516         }
1517
1518         /* this should be rarely called, just remove old and add new */
1519         spin_lock(&obj->oo_guard);
1520         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1521                 if (namelen == tmp->oxe_namelen &&
1522                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1523                         old = tmp;
1524                         break;
1525                 }
1526         }
1527         if (old != NULL) {
1528                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1529                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1530         } else {
1531                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1532         }
1533         spin_unlock(&obj->oo_guard);
1534 }
1535
1536 static void osd_oxc_del(struct osd_object *obj, const char *name)
1537 {
1538         struct osd_xattr_entry *oxe;
1539         size_t namelen = strlen(name);
1540
1541         spin_lock(&obj->oo_guard);
1542         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1543                 if (namelen == oxe->oxe_namelen &&
1544                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1545                         list_del_rcu(&oxe->oxe_list);
1546                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1547                         break;
1548                 }
1549         }
1550         spin_unlock(&obj->oo_guard);
1551 }
1552
1553 static void osd_oxc_fini(struct osd_object *obj)
1554 {
1555         struct osd_xattr_entry *oxe, *next;
1556
1557         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1558                 list_del(&oxe->oxe_list);
1559                 OBD_FREE(oxe, oxe->oxe_len);
1560         }
1561 }
1562
1563 /*
1564  * Concurrency: no concurrent access is possible that late in object
1565  * life-cycle.
1566  */
1567 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1568 {
1569         struct osd_object *obj = osd_obj(l);
1570         struct lu_object_header *h = obj->oo_header;
1571
1572         LINVRNT(osd_invariant(obj));
1573
1574         osd_oxc_fini(obj);
1575         dt_object_fini(&obj->oo_dt);
1576         if (obj->oo_hl_head != NULL)
1577                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1578         OBD_FREE_PTR(obj);
1579         if (unlikely(h)) {
1580                 lu_object_header_fini(h);
1581                 OBD_FREE_PTR(h);
1582         }
1583 }
1584
1585 /*
1586  * Concurrency: no concurrent access is possible that late in object
1587  * life-cycle.
1588  */
1589 static void osd_index_fini(struct osd_object *o)
1590 {
1591         struct iam_container *bag;
1592
1593         if (o->oo_dir != NULL) {
1594                 bag = &o->oo_dir->od_container;
1595                 if (o->oo_inode != NULL) {
1596                         if (bag->ic_object == o->oo_inode)
1597                                 iam_container_fini(bag);
1598                 }
1599                 OBD_FREE_PTR(o->oo_dir);
1600                 o->oo_dir = NULL;
1601         }
1602 }
1603
1604 /*
1605  * Concurrency: no concurrent access is possible that late in object
1606  * life-cycle (for all existing callers, that is. New callers have to provide
1607  * their own locking.)
1608  */
1609 static int osd_inode_unlinked(const struct inode *inode)
1610 {
1611         return inode->i_nlink == 0;
1612 }
1613
1614 enum {
1615         OSD_TXN_OI_DELETE_CREDITS    = 20,
1616         OSD_TXN_INODE_DELETE_CREDITS = 20
1617 };
1618
1619 /*
1620  * Journal
1621  */
1622
1623 #if OSD_THANDLE_STATS
1624 /**
1625  * Set time when the handle is allocated
1626  */
1627 static void osd_th_alloced(struct osd_thandle *oth)
1628 {
1629         oth->oth_alloced = ktime_get();
1630 }
1631
1632 /**
1633  * Set time when the handle started
1634  */
1635 static void osd_th_started(struct osd_thandle *oth)
1636 {
1637         oth->oth_started = ktime_get();
1638 }
1639
1640 /**
1641  * Check whether the we deal with this handle for too long.
1642  */
1643 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1644                                 ktime_t alloced, ktime_t started,
1645                                 ktime_t closed)
1646 {
1647         ktime_t now = ktime_get();
1648
1649         LASSERT(dev != NULL);
1650
1651         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1652                             ktime_us_delta(started, alloced));
1653         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1654                             ktime_us_delta(closed, started));
1655         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1656                             ktime_us_delta(now, closed));
1657
1658         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1659                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1660                                 oth, now, alloced, started, closed);
1661                 libcfs_debug_dumpstack(NULL);
1662         }
1663 }
1664
1665 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1666 {                                                                       \
1667         ktime_t __closed = ktime_get();                                 \
1668         ktime_t __alloced = oth->oth_alloced;                           \
1669         ktime_t __started = oth->oth_started;                           \
1670                                                                         \
1671         expr;                                                           \
1672         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1673 }
1674
1675 #else /* OSD_THANDLE_STATS */
1676
1677 #define osd_th_alloced(h)                  do {} while(0)
1678 #define osd_th_started(h)                  do {} while(0)
1679 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1680
1681 #endif /* OSD_THANDLE_STATS */
1682
1683 /*
1684  * Concurrency: doesn't access mutable data.
1685  */
1686 static int osd_param_is_not_sane(const struct osd_device *dev,
1687                                  const struct thandle *th)
1688 {
1689         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1690
1691         return oh->ot_credits > osd_transaction_size(dev);
1692 }
1693
1694 /*
1695  * Concurrency: shouldn't matter.
1696  */
1697 static void osd_trans_commit_cb(struct super_block *sb,
1698                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1699 {
1700         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1701         struct thandle *th = &oh->ot_super;
1702         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1703         struct dt_txn_commit_cb *dcb, *tmp;
1704
1705         LASSERT(oh->ot_handle == NULL);
1706
1707         if (error)
1708                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1709
1710         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1711         /* call per-transaction callbacks if any */
1712         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1713                                  dcb_linkage) {
1714                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1715                          "commit callback entry: magic=%x name='%s'\n",
1716                          dcb->dcb_magic, dcb->dcb_name);
1717                 list_del_init(&dcb->dcb_linkage);
1718                 dcb->dcb_func(NULL, th, dcb, error);
1719         }
1720
1721         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1722         lu_device_put(lud);
1723         th->th_dev = NULL;
1724
1725         OBD_FREE_PTR(oh);
1726 }
1727
1728 #ifndef HAVE_SB_START_WRITE
1729 # define sb_start_write(sb) do {} while (0)
1730 # define sb_end_write(sb) do {} while (0)
1731 #endif
1732
1733 static struct thandle *osd_trans_create(const struct lu_env *env,
1734                                         struct dt_device *d)
1735 {
1736         struct osd_thread_info *oti = osd_oti_get(env);
1737         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1738         struct osd_thandle *oh;
1739         struct thandle *th;
1740
1741         ENTRY;
1742
1743         if (d->dd_rdonly) {
1744                 CERROR("%s: someone try to start transaction under "
1745                        "readonly mode, should be disabled.\n",
1746                        osd_name(osd_dt_dev(d)));
1747                 dump_stack();
1748                 RETURN(ERR_PTR(-EROFS));
1749         }
1750
1751         /* on pending IO in this thread should left from prev. request */
1752         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1753
1754         sb_start_write(osd_sb(osd_dt_dev(d)));
1755
1756         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1757         if (!oh) {
1758                 sb_end_write(osd_sb(osd_dt_dev(d)));
1759                 RETURN(ERR_PTR(-ENOMEM));
1760         }
1761
1762         oh->ot_quota_trans = &oti->oti_quota_trans;
1763         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1764         th = &oh->ot_super;
1765         th->th_dev = d;
1766         th->th_result = 0;
1767         oh->ot_credits = 0;
1768         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1769         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1770         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1771         osd_th_alloced(oh);
1772
1773         memset(oti->oti_declare_ops, 0,
1774                sizeof(oti->oti_declare_ops));
1775         memset(oti->oti_declare_ops_cred, 0,
1776                sizeof(oti->oti_declare_ops_cred));
1777         memset(oti->oti_declare_ops_used, 0,
1778                sizeof(oti->oti_declare_ops_used));
1779
1780         oti->oti_ins_cache_depth++;
1781
1782         RETURN(th);
1783 }
1784
1785 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1786 {
1787         struct osd_thread_info *oti = osd_oti_get(env);
1788         struct osd_thandle *oh;
1789
1790         oh = container_of0(th, struct osd_thandle, ot_super);
1791         LASSERT(oh != NULL);
1792
1793         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1794               oti->oti_declare_ops[OSD_OT_CREATE],
1795               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1796               oti->oti_declare_ops_used[OSD_OT_CREATE],
1797               oti->oti_declare_ops[OSD_OT_DESTROY],
1798               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1799               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1800         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1801               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1802               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1803               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1805               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1806               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1807         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1808               oti->oti_declare_ops[OSD_OT_WRITE],
1809               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1810               oti->oti_declare_ops_used[OSD_OT_WRITE],
1811               oti->oti_declare_ops[OSD_OT_PUNCH],
1812               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1813               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1814               oti->oti_declare_ops[OSD_OT_QUOTA],
1815               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1816               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1817         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_INSERT],
1819               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1820               oti->oti_declare_ops_used[OSD_OT_INSERT],
1821               oti->oti_declare_ops[OSD_OT_DELETE],
1822               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1823               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1824         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1825               oti->oti_declare_ops[OSD_OT_REF_ADD],
1826               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1827               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops[OSD_OT_REF_DEL],
1829               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1830               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1831 }
1832
1833 /*
1834  * Concurrency: shouldn't matter.
1835  */
1836 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1837                            struct thandle *th)
1838 {
1839         struct osd_thread_info *oti = osd_oti_get(env);
1840         struct osd_device *dev = osd_dt_dev(d);
1841         handle_t *jh;
1842         struct osd_thandle *oh;
1843         int rc;
1844
1845         ENTRY;
1846
1847         LASSERT(current->journal_info == NULL);
1848
1849         oh = container_of0(th, struct osd_thandle, ot_super);
1850         LASSERT(oh != NULL);
1851         LASSERT(oh->ot_handle == NULL);
1852
1853         rc = dt_txn_hook_start(env, d, th);
1854         if (rc != 0)
1855                 GOTO(out, rc);
1856
1857         if (unlikely(osd_param_is_not_sane(dev, th))) {
1858                 static unsigned long last_printed;
1859                 static int last_credits;
1860
1861                 /*
1862                  * don't make noise on a tiny testing systems
1863                  * actual credits misuse will be caught anyway
1864                  */
1865                 if (last_credits != oh->ot_credits &&
1866                     time_after(jiffies, last_printed +
1867                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1868                     osd_transaction_size(dev) > 512) {
1869                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1870                               oh->ot_credits, osd_transaction_size(dev));
1871                         osd_trans_dump_creds(env, th);
1872                         libcfs_debug_dumpstack(NULL);
1873                         last_credits = oh->ot_credits;
1874                         last_printed = jiffies;
1875                 }
1876                 /*
1877                  * XXX Limit the credits to 'max_transaction_buffers', and
1878                  *     let the underlying filesystem to catch the error if
1879                  *     we really need so many credits.
1880                  *
1881                  *     This should be removed when we can calculate the
1882                  *     credits precisely.
1883                  */
1884                 oh->ot_credits = osd_transaction_size(dev);
1885         } else if (ldiskfs_track_declares_assert != 0) {
1886                 /*
1887                  * reserve few credits to prevent an assertion in JBD
1888                  * our debugging mechanism will be able to detected
1889                  * overuse. this can help to debug single-update
1890                  * transactions
1891                  */
1892                 oh->ot_credits += 10;
1893                 if (unlikely(osd_param_is_not_sane(dev, th)))
1894                         oh->ot_credits = osd_transaction_size(dev);
1895         }
1896
1897         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1898                 GOTO(out, rc = -EIO);
1899
1900         /*
1901          * XXX temporary stuff. Some abstraction layer should
1902          * be used.
1903          */
1904         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1905         osd_th_started(oh);
1906         if (!IS_ERR(jh)) {
1907                 oh->ot_handle = jh;
1908                 LASSERT(oti->oti_txns == 0);
1909
1910                 lu_device_get(&d->dd_lu_dev);
1911                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1912                               "osd-tx", th);
1913                 oti->oti_txns++;
1914                 rc = 0;
1915         } else {
1916                 rc = PTR_ERR(jh);
1917         }
1918 out:
1919         RETURN(rc);
1920 }
1921
1922 static int osd_seq_exists(const struct lu_env *env,
1923                           struct osd_device *osd, u64 seq)
1924 {
1925         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1926         struct seq_server_site *ss = osd_seq_site(osd);
1927         int rc;
1928
1929         ENTRY;
1930
1931         LASSERT(ss != NULL);
1932         LASSERT(ss->ss_server_fld != NULL);
1933
1934         rc = osd_fld_lookup(env, osd, seq, range);
1935         if (rc != 0) {
1936                 if (rc != -ENOENT)
1937                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1938                                osd_name(osd), seq, rc);
1939                 RETURN(0);
1940         }
1941
1942         RETURN(ss->ss_node_id == range->lsr_index);
1943 }
1944
1945 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1946 {
1947         struct dt_txn_commit_cb *dcb;
1948         struct dt_txn_commit_cb *tmp;
1949
1950         /* call per-transaction stop callbacks if any */
1951         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1952                                  dcb_linkage) {
1953                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1954                          "commit callback entry: magic=%x name='%s'\n",
1955                          dcb->dcb_magic, dcb->dcb_name);
1956                 list_del_init(&dcb->dcb_linkage);
1957                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1958         }
1959 }
1960
1961 /*
1962  * Concurrency: shouldn't matter.
1963  */
1964 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1965                           struct thandle *th)
1966 {
1967         struct osd_thread_info *oti = osd_oti_get(env);
1968         struct osd_thandle *oh;
1969         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1970         struct osd_device *osd = osd_dt_dev(th->th_dev);
1971         struct qsd_instance *qsd = osd_def_qsd(osd);
1972         struct lquota_trans *qtrans;
1973         struct list_head truncates = LIST_HEAD_INIT(truncates);
1974         int rc = 0, remove_agents = 0;
1975
1976         ENTRY;
1977
1978         oh = container_of0(th, struct osd_thandle, ot_super);
1979
1980         remove_agents = oh->ot_remove_agents;
1981
1982         qtrans = oh->ot_quota_trans;
1983         oh->ot_quota_trans = NULL;
1984
1985         /* move locks to local list, stop tx, execute truncates */
1986         list_splice(&oh->ot_trunc_locks, &truncates);
1987
1988         if (oh->ot_handle != NULL) {
1989                 int rc2;
1990
1991                 handle_t *hdl = oh->ot_handle;
1992
1993                 /*
1994                  * add commit callback
1995                  * notice we don't do this in osd_trans_start()
1996                  * as underlying transaction can change during truncate
1997                  */
1998                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1999                                              &oh->ot_jcb);
2000
2001                 LASSERT(oti->oti_txns == 1);
2002                 oti->oti_txns--;
2003
2004                 rc = dt_txn_hook_stop(env, th);
2005                 if (rc != 0)
2006                         CERROR("%s: failed in transaction hook: rc = %d\n",
2007                                osd_name(osd), rc);
2008
2009                 osd_trans_stop_cb(oh, rc);
2010                 /* hook functions might modify th_sync */
2011                 hdl->h_sync = th->th_sync;
2012
2013                 oh->ot_handle = NULL;
2014                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2015                 if (rc2 != 0)
2016                         CERROR("%s: failed to stop transaction: rc = %d\n",
2017                                osd_name(osd), rc2);
2018                 if (!rc)
2019                         rc = rc2;
2020
2021                 osd_process_truncates(&truncates);
2022         } else {
2023                 osd_trans_stop_cb(oh, th->th_result);
2024                 OBD_FREE_PTR(oh);
2025         }
2026
2027         osd_trunc_unlock_all(&truncates);
2028
2029         /* inform the quota slave device that the transaction is stopping */
2030         qsd_op_end(env, qsd, qtrans);
2031
2032         /*
2033          * as we want IO to journal and data IO be concurrent, we don't block
2034          * awaiting data IO completion in osd_do_bio(), instead we wait here
2035          * once transaction is submitted to the journal. all reqular requests
2036          * don't do direct IO (except read/write), thus this wait_event becomes
2037          * no-op for them.
2038          *
2039          * IMPORTANT: we have to wait till any IO submited by the thread is
2040          * completed otherwise iobuf may be corrupted by different request
2041          */
2042         wait_event(iobuf->dr_wait,
2043                        atomic_read(&iobuf->dr_numreqs) == 0);
2044         osd_fini_iobuf(osd, iobuf);
2045         if (!rc)
2046                 rc = iobuf->dr_error;
2047
2048         if (unlikely(remove_agents != 0))
2049                 osd_process_scheduled_agent_removals(env, osd);
2050
2051         oti->oti_ins_cache_depth--;
2052         /* reset OI cache for safety */
2053         if (oti->oti_ins_cache_depth == 0)
2054                 oti->oti_ins_cache_used = 0;
2055
2056         sb_end_write(osd_sb(osd));
2057
2058         RETURN(rc);
2059 }
2060
2061 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2062 {
2063         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2064                                                ot_super);
2065
2066         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2067         LASSERT(&dcb->dcb_func != NULL);
2068         if (dcb->dcb_flags & DCB_TRANS_STOP)
2069                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2070         else
2071                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2072
2073         return 0;
2074 }
2075
2076 /*
2077  * Called just before object is freed. Releases all resources except for
2078  * object itself (that is released by osd_object_free()).
2079  *
2080  * Concurrency: no concurrent access is possible that late in object
2081  * life-cycle.
2082  */
2083 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2084 {
2085         struct osd_object *obj = osd_obj(l);
2086         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2087         struct inode *inode = obj->oo_inode;
2088         __u64 projid;
2089         qid_t uid;
2090         qid_t gid;
2091
2092         LINVRNT(osd_invariant(obj));
2093
2094         /*
2095          * If object is unlinked remove fid->ino mapping from object index.
2096          */
2097
2098         osd_index_fini(obj);
2099
2100         if (!inode)
2101                 return;
2102
2103         uid = i_uid_read(inode);
2104         gid = i_gid_read(inode);
2105         projid = i_projid_read(inode);
2106
2107         obj->oo_inode = NULL;
2108         iput(inode);
2109
2110         /* do not rebalance quota if the caller needs to release memory
2111          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2112          * transaction and risk to deadlock - LU-12178 */
2113         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2114                 return;
2115
2116         if (!obj->oo_header && qsd) {
2117                 struct osd_thread_info *info = osd_oti_get(env);
2118                 struct lquota_id_info *qi = &info->oti_qi;
2119
2120                 /* Release granted quota to master if necessary */
2121                 qi->lqi_id.qid_uid = uid;
2122                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2123
2124                 qi->lqi_id.qid_uid = gid;
2125                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2126
2127                 qi->lqi_id.qid_uid = projid;
2128                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2129         }
2130 }
2131
2132 /*
2133  * Concurrency: ->loo_object_release() is called under site spin-lock.
2134  */
2135 static void osd_object_release(const struct lu_env *env,
2136                                struct lu_object *l)
2137 {
2138         struct osd_object *o = osd_obj(l);
2139
2140         /*
2141          * nobody should be releasing a non-destroyed object with nlink=0
2142          * the API allows this, but ldiskfs doesn't like and then report
2143          * this inode as deleted
2144          */
2145         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2146                   o->oo_inode->i_nlink == 0));
2147 }
2148
2149 /*
2150  * Concurrency: shouldn't matter.
2151  */
2152 static int osd_object_print(const struct lu_env *env, void *cookie,
2153                             lu_printer_t p, const struct lu_object *l)
2154 {
2155         struct osd_object *o = osd_obj(l);
2156         struct iam_descr *d;
2157
2158         if (o->oo_dir != NULL)
2159                 d = o->oo_dir->od_container.ic_descr;
2160         else
2161                 d = NULL;
2162         return (*p)(env, cookie,
2163                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2164                     o, o->oo_inode,
2165                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2166                     o->oo_inode ? o->oo_inode->i_generation : 0,
2167                     d ? d->id_ops->id_name : "plain");
2168 }
2169
2170 /*
2171  * Concurrency: shouldn't matter.
2172  */
2173 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2174                 struct obd_statfs *sfs)
2175 {
2176         struct osd_device *osd = osd_dt_dev(d);
2177         struct super_block *sb = osd_sb(osd);
2178         struct kstatfs *ksfs;
2179         __u64 reserved;
2180         int result = 0;
2181
2182         if (unlikely(osd->od_mnt == NULL))
2183                 return -EINPROGRESS;
2184
2185         /* osd_lproc.c call this without env, allocate ksfs for that case */
2186         if (unlikely(env == NULL)) {
2187                 OBD_ALLOC_PTR(ksfs);
2188                 if (ksfs == NULL)
2189                         return -ENOMEM;
2190         } else {
2191                 ksfs = &osd_oti_get(env)->oti_ksfs;
2192         }
2193
2194         result = sb->s_op->statfs(sb->s_root, ksfs);
2195         if (result)
2196                 goto out;
2197
2198         statfs_pack(sfs, ksfs);
2199         if (unlikely(sb->s_flags & MS_RDONLY))
2200                 sfs->os_state |= OS_STATE_READONLY;
2201         if (ldiskfs_has_feature_extents(sb))
2202                 sfs->os_maxbytes = sb->s_maxbytes;
2203         else
2204                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2205
2206         /*
2207          * Reserve some space so to avoid fragmenting the filesystem too much.
2208          * Fragmentation not only impacts performance, but can also increase
2209          * metadata overhead significantly, causing grant calculation to be
2210          * wrong.
2211          *
2212          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2213          */
2214         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2215         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2216         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2217                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2218
2219         sfs->os_blocks -= reserved;
2220         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2221         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2222
2223 out:
2224         if (unlikely(env == NULL))
2225                 OBD_FREE_PTR(ksfs);
2226         return result;
2227 }
2228
2229 /**
2230  * Estimate space needed for file creations. We assume the largest filename
2231  * which is 2^64 - 1, hence a filename of 20 chars.
2232  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2233  */
2234 #ifdef __LDISKFS_DIR_REC_LEN
2235 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2236 #else
2237 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2238 #endif
2239
2240 /*
2241  * Concurrency: doesn't access mutable data.
2242  */
2243 static void osd_conf_get(const struct lu_env *env,
2244                          const struct dt_device *dev,
2245                          struct dt_device_param *param)
2246 {
2247         struct osd_device *d = osd_dt_dev(dev);
2248         struct super_block *sb = osd_sb(d);
2249         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2250         const char *name;
2251         int ea_overhead;
2252
2253         /*
2254          * XXX should be taken from not-yet-existing fs abstraction layer.
2255          */
2256         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2257         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2258         param->ddp_symlink_max  = sb->s_blocksize;
2259         param->ddp_mount_type   = LDD_MT_LDISKFS;
2260         if (ldiskfs_has_feature_extents(sb))
2261                 param->ddp_maxbytes = sb->s_maxbytes;
2262         else
2263                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2264         /*
2265          * inode are statically allocated, so per-inode space consumption
2266          * is the space consumed by the directory entry
2267          */
2268         param->ddp_inodespace     = PER_OBJ_USAGE;
2269         /*
2270          * EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2271          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2272          * maximum length to not under count the actual number of extents
2273          * needed for writing a file.
2274          */
2275         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2276         /* worst-case extent insertion metadata overhead */
2277         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2278         param->ddp_mntopts = 0;
2279         if (test_opt(sb, XATTR_USER))
2280                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2281         if (test_opt(sb, POSIX_ACL))
2282                 param->ddp_mntopts |= MNTOPT_ACL;
2283
2284         /*
2285          * LOD might calculate the max stripe count based on max_ea_size,
2286          * so we need take account in the overhead as well,
2287          * xattr_header + magic + xattr_entry_head
2288          */
2289         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2290                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2291
2292 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2293         if (ldiskfs_has_feature_ea_inode(sb))
2294                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2295                                                                 ea_overhead;
2296         else
2297 #endif
2298                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2299
2300         if (param->ddp_max_ea_size > OSD_MAX_EA_SIZE)
2301                 param->ddp_max_ea_size = OSD_MAX_EA_SIZE;
2302
2303         /*
2304          * Preferred RPC size for efficient disk IO.  4MB shows good
2305          * all-around performance for ldiskfs, but use bigalloc chunk size
2306          * by default if larger.
2307          */
2308 #if defined(LDISKFS_CLUSTER_SIZE)
2309         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2310                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2311         else
2312 #endif
2313                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2314
2315         param->ddp_t10_cksum_type = 0;
2316         if (bi) {
2317                 unsigned short interval = blk_integrity_interval(bi);
2318                 name = blk_integrity_name(bi);
2319                 /*
2320                  * Expected values:
2321                  * T10-DIF-TYPE1-CRC
2322                  * T10-DIF-TYPE3-CRC
2323                  * T10-DIF-TYPE1-IP
2324                  * T10-DIF-TYPE3-IP
2325                  */
2326                 if (strncmp(name, "T10-DIF-TYPE",
2327                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2328                         /* also skip "1/3-" at end */
2329                         const int type_off = sizeof("T10-DIF-TYPE.");
2330                         char type_number = name[type_off - 2];
2331
2332                         if (interval != 512 && interval != 4096) {
2333                                 CERROR("%s: unsupported T10PI sector size %u\n",
2334                                        d->od_svname, interval);
2335                         } else if (type_number != '1' && type_number != '3') {
2336                                 CERROR("%s: unsupported T10PI type %s\n",
2337                                        d->od_svname, name);
2338                         } else if (strcmp(name + type_off, "CRC") == 0) {
2339                                 d->od_t10_type = type_number == '1' ?
2340                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2341                                 param->ddp_t10_cksum_type = interval == 512 ?
2342                                         OBD_CKSUM_T10CRC512 :
2343                                         OBD_CKSUM_T10CRC4K;
2344                         } else if (strcmp(name + type_off, "IP") == 0) {
2345                                 d->od_t10_type = type_number == '1' ?
2346                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2347                                 param->ddp_t10_cksum_type = interval == 512 ?
2348                                         OBD_CKSUM_T10IP512 :
2349                                         OBD_CKSUM_T10IP4K;
2350                         } else {
2351                                 CERROR("%s: unsupported checksum type of "
2352                                        "T10PI type '%s'",
2353                                        d->od_svname, name);
2354                         }
2355
2356                 } else {
2357                         CERROR("%s: unsupported T10PI type '%s'",
2358                                d->od_svname, name);
2359                 }
2360         }
2361 }
2362
2363 /*
2364  * Concurrency: shouldn't matter.
2365  */
2366 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2367 {
2368         int rc;
2369         struct super_block *s = osd_sb(osd_dt_dev(d));
2370         ENTRY;
2371
2372         down_read(&s->s_umount);
2373         rc = s->s_op->sync_fs(s, 1);
2374         up_read(&s->s_umount);
2375
2376         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2377                rc);
2378
2379         return rc;
2380 }
2381
2382 /**
2383  * Start commit for OSD device.
2384  *
2385  * An implementation of dt_commit_async method for OSD device.
2386  * Asychronously starts underlayng fs sync and thereby a transaction
2387  * commit.
2388  *
2389  * \param env environment
2390  * \param d dt device
2391  *
2392  * \see dt_device_operations
2393  */
2394 static int osd_commit_async(const struct lu_env *env,
2395                             struct dt_device *d)
2396 {
2397         struct super_block *s = osd_sb(osd_dt_dev(d));
2398         int rc;
2399
2400         ENTRY;
2401
2402         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2403         down_read(&s->s_umount);
2404         rc = s->s_op->sync_fs(s, 0);
2405         up_read(&s->s_umount);
2406
2407         RETURN(rc);
2408 }
2409
2410 /* Our own copy of the set readonly functions if present, or NU if not. */
2411 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2412 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2413 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2414
2415 /*
2416  * Concurrency: shouldn't matter.
2417  */
2418 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2419 {
2420         struct super_block *sb = osd_sb(osd_dt_dev(d));
2421         struct block_device *dev = sb->s_bdev;
2422         int rc = -EOPNOTSUPP;
2423
2424         ENTRY;
2425
2426         if (priv_dev_set_rdonly) {
2427                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2428
2429                 rc = 0;
2430                 CERROR("*** setting %s read-only ***\n",
2431                        osd_dt_dev(d)->od_svname);
2432
2433                 if (sb->s_op->freeze_fs) {
2434                         rc = sb->s_op->freeze_fs(sb);
2435                         if (rc)
2436                                 goto out;
2437                 }
2438
2439                 if (jdev && (jdev != dev)) {
2440                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2441                                (long)jdev);
2442                         priv_dev_set_rdonly(jdev);
2443                 }
2444                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2445                 priv_dev_set_rdonly(dev);
2446
2447                 if (sb->s_op->unfreeze_fs)
2448                         sb->s_op->unfreeze_fs(sb);
2449         }
2450
2451 out:
2452         if (rc)
2453                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2454                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2455
2456         RETURN(rc);
2457 }
2458
2459 /**
2460  * Note: we do not count into QUOTA here.
2461  * If we mount with --data_journal we may need more.
2462  */
2463 const int osd_dto_credits_noquota[DTO_NR] = {
2464         /**
2465          * Insert.
2466          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2467          * SINGLEDATA_TRANS_BLOCKS(8)
2468          * XXX Note: maybe iam need more, since iam have more level than
2469          *           EXT3 htree.
2470          */
2471         [DTO_INDEX_INSERT]  = 16,
2472         /**
2473          * Delete
2474          * just modify a single entry, probably merge few within a block
2475          */
2476         [DTO_INDEX_DELETE]  = 1,
2477         /**
2478          * Used for OI scrub
2479          */
2480         [DTO_INDEX_UPDATE]  = 16,
2481         /**
2482          * 4(inode, inode bits, groups, GDT)
2483          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2484          */
2485         [DTO_OBJECT_CREATE] = 4,
2486         /**
2487          * 4(inode, inode bits, groups, GDT)
2488          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2489          */
2490         [DTO_OBJECT_DELETE] = 4,
2491         /**
2492          * Attr set credits (inode)
2493          */
2494         [DTO_ATTR_SET_BASE] = 1,
2495         /**
2496          * Xattr set. The same as xattr of EXT3.
2497          * DATA_TRANS_BLOCKS(14)
2498          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2499          * are also counted in. Do not know why?
2500          */
2501         [DTO_XATTR_SET]     = 14,
2502         /**
2503          * credits for inode change during write.
2504          */
2505         [DTO_WRITE_BASE]    = 3,
2506         /**
2507          * credits for single block write.
2508          */
2509         [DTO_WRITE_BLOCK]   = 14,
2510         /**
2511          * Attr set credits for chown.
2512          * This is extra credits for setattr, and it is null without quota
2513          */
2514         [DTO_ATTR_SET_CHOWN] = 0
2515 };
2516
2517 static const struct dt_device_operations osd_dt_ops = {
2518         .dt_root_get       = osd_root_get,
2519         .dt_statfs         = osd_statfs,
2520         .dt_trans_create   = osd_trans_create,
2521         .dt_trans_start    = osd_trans_start,
2522         .dt_trans_stop     = osd_trans_stop,
2523         .dt_trans_cb_add   = osd_trans_cb_add,
2524         .dt_conf_get       = osd_conf_get,
2525         .dt_sync           = osd_sync,
2526         .dt_ro             = osd_ro,
2527         .dt_commit_async   = osd_commit_async,
2528 };
2529
2530 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2531                           unsigned int role)
2532 {
2533         struct osd_object *obj = osd_dt_obj(dt);
2534         struct osd_thread_info *oti = osd_oti_get(env);
2535
2536         LINVRNT(osd_invariant(obj));
2537
2538         LASSERT(obj->oo_owner != env);
2539         down_read_nested(&obj->oo_sem, role);
2540
2541         LASSERT(obj->oo_owner == NULL);
2542         oti->oti_r_locks++;
2543 }
2544
2545 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2546                            unsigned int role)
2547 {
2548         struct osd_object *obj = osd_dt_obj(dt);
2549         struct osd_thread_info *oti = osd_oti_get(env);
2550
2551         LINVRNT(osd_invariant(obj));
2552
2553         LASSERT(obj->oo_owner != env);
2554         down_write_nested(&obj->oo_sem, role);
2555
2556         LASSERT(obj->oo_owner == NULL);
2557         obj->oo_owner = env;
2558         oti->oti_w_locks++;
2559 }
2560
2561 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2562 {
2563         struct osd_object *obj = osd_dt_obj(dt);
2564         struct osd_thread_info *oti = osd_oti_get(env);
2565
2566         LINVRNT(osd_invariant(obj));
2567
2568         LASSERT(oti->oti_r_locks > 0);
2569         oti->oti_r_locks--;
2570         up_read(&obj->oo_sem);
2571 }
2572
2573 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2574 {
2575         struct osd_object *obj = osd_dt_obj(dt);
2576         struct osd_thread_info *oti = osd_oti_get(env);
2577
2578         LINVRNT(osd_invariant(obj));
2579
2580         LASSERT(obj->oo_owner == env);
2581         LASSERT(oti->oti_w_locks > 0);
2582         oti->oti_w_locks--;
2583         obj->oo_owner = NULL;
2584         up_write(&obj->oo_sem);
2585 }
2586
2587 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2588 {
2589         struct osd_object *obj = osd_dt_obj(dt);
2590
2591         LINVRNT(osd_invariant(obj));
2592
2593         return obj->oo_owner == env;
2594 }
2595
2596 static struct timespec *osd_inode_time(const struct lu_env *env,
2597                                        struct inode *inode, __u64 seconds)
2598 {
2599         struct osd_thread_info *oti = osd_oti_get(env);
2600         struct timespec *t = &oti->oti_time;
2601
2602         t->tv_sec = seconds;
2603         t->tv_nsec = 0;
2604         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2605         return t;
2606 }
2607
2608 static void osd_inode_getattr(const struct lu_env *env,
2609                               struct inode *inode, struct lu_attr *attr)
2610 {
2611         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2612                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2613                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2614                            LA_BLKSIZE | LA_TYPE;
2615
2616         attr->la_atime = inode->i_atime.tv_sec;
2617         attr->la_mtime = inode->i_mtime.tv_sec;
2618         attr->la_ctime = inode->i_ctime.tv_sec;
2619         attr->la_mode    = inode->i_mode;
2620         attr->la_size    = i_size_read(inode);
2621         attr->la_blocks  = inode->i_blocks;
2622         attr->la_uid     = i_uid_read(inode);
2623         attr->la_gid     = i_gid_read(inode);
2624         attr->la_projid  = i_projid_read(inode);
2625         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2626         attr->la_nlink   = inode->i_nlink;
2627         attr->la_rdev    = inode->i_rdev;
2628         attr->la_blksize = 1 << inode->i_blkbits;
2629         attr->la_blkbits = inode->i_blkbits;
2630         /*
2631          * Ext4 did not transfer inherit flags from raw inode
2632          * to inode flags, and ext4 internally test raw inode
2633          * @i_flags directly. Instead of patching ext4, we do it here.
2634          */
2635         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2636                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2637 }
2638
2639 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2640                         struct lu_attr *attr)
2641 {
2642         struct osd_object *obj = osd_dt_obj(dt);
2643
2644         if (unlikely(!dt_object_exists(dt)))
2645                 return -ENOENT;
2646         if (unlikely(obj->oo_destroyed))
2647                 return -ENOENT;
2648
2649         LASSERT(!dt_object_remote(dt));
2650         LINVRNT(osd_invariant(obj));
2651
2652         spin_lock(&obj->oo_guard);
2653         osd_inode_getattr(env, obj->oo_inode, attr);
2654         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2655                 attr->la_valid |= LA_FLAGS;
2656                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2657         }
2658         spin_unlock(&obj->oo_guard);
2659
2660         return 0;
2661 }
2662
2663 static int osd_declare_attr_qid(const struct lu_env *env,
2664                                 struct osd_object *obj,
2665                                 struct osd_thandle *oh, long long bspace,
2666                                 qid_t old_id, qid_t new_id, bool enforce,
2667                                 unsigned int type, bool ignore_edquot)
2668 {
2669         int rc;
2670         struct osd_thread_info *info = osd_oti_get(env);
2671         struct lquota_id_info  *qi = &info->oti_qi;
2672
2673         qi->lqi_type = type;
2674         /* inode accounting */
2675         qi->lqi_is_blk = false;
2676
2677         /* one more inode for the new id ... */
2678         qi->lqi_id.qid_uid = new_id;
2679         qi->lqi_space      = 1;
2680         /* Reserve credits for the new id */
2681         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2682         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2683                 rc = 0;
2684         if (rc)
2685                 RETURN(rc);
2686
2687         /* and one less inode for the current id */
2688         qi->lqi_id.qid_uid = old_id;
2689         qi->lqi_space = -1;
2690         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2691         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2692                 rc = 0;
2693         if (rc)
2694                 RETURN(rc);
2695
2696         /* block accounting */
2697         qi->lqi_is_blk = true;
2698
2699         /* more blocks for the new id ... */
2700         qi->lqi_id.qid_uid = new_id;
2701         qi->lqi_space      = bspace;
2702         /*
2703          * Credits for the new uid has been reserved, re-use "obj"
2704          * to save credit reservation.
2705          */
2706         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2707         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2708                 rc = 0;
2709         if (rc)
2710                 RETURN(rc);
2711
2712         /* and finally less blocks for the current uid */
2713         qi->lqi_id.qid_uid = old_id;
2714         qi->lqi_space      = -bspace;
2715         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2716         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2717                 rc = 0;
2718
2719         RETURN(rc);
2720 }
2721
2722 static int osd_declare_attr_set(const struct lu_env *env,
2723                                 struct dt_object *dt,
2724                                 const struct lu_attr *attr,
2725                                 struct thandle *handle)
2726 {
2727         struct osd_thandle *oh;
2728         struct osd_object *obj;
2729         qid_t uid;
2730         qid_t gid;
2731         long long bspace;
2732         int rc = 0;
2733         bool enforce;
2734
2735         ENTRY;
2736
2737         LASSERT(dt != NULL);
2738         LASSERT(handle != NULL);
2739
2740         obj = osd_dt_obj(dt);
2741         LASSERT(osd_invariant(obj));
2742
2743         oh = container_of0(handle, struct osd_thandle, ot_super);
2744         LASSERT(oh->ot_handle == NULL);
2745
2746         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2747                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2748
2749         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2750                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2751
2752         if (attr == NULL || obj->oo_inode == NULL)
2753                 RETURN(rc);
2754
2755         bspace   = obj->oo_inode->i_blocks << 9;
2756         bspace   = toqb(bspace);
2757
2758         /*
2759          * Changing ownership is always preformed by super user, it should not
2760          * fail with EDQUOT unless required explicitly.
2761          *
2762          * We still need to call the osd_declare_qid() to calculate the journal
2763          * credits for updating quota accounting files and to trigger quota
2764          * space adjustment once the operation is completed.
2765          */
2766         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2767                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2768
2769                 if (!ignore_edquot)
2770                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2771                                "(the quota space is %lld)\n",
2772                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2773                                attr->la_gid, bspace);
2774
2775                 /* USERQUOTA */
2776                 uid = i_uid_read(obj->oo_inode);
2777                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2778                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2779                                           attr->la_uid, enforce, USRQUOTA,
2780                                           true);
2781                 if (rc)
2782                         RETURN(rc);
2783
2784                 gid = i_gid_read(obj->oo_inode);
2785                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2786                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2787                                           i_gid_read(obj->oo_inode),
2788                                           attr->la_gid, enforce, GRPQUOTA,
2789                                           ignore_edquot);
2790                 if (rc)
2791                         RETURN(rc);
2792
2793         }
2794 #ifdef HAVE_PROJECT_QUOTA
2795         if (attr->la_valid & LA_PROJID) {
2796                 __u32 projid = i_projid_read(obj->oo_inode);
2797
2798                 enforce = (attr->la_valid & LA_PROJID) &&
2799                                         (attr->la_projid != projid);
2800                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2801                                           (qid_t)projid, (qid_t)attr->la_projid,
2802                                           enforce, PRJQUOTA, true);
2803                 if (rc)
2804                         RETURN(rc);
2805         }
2806 #endif
2807         RETURN(rc);
2808 }
2809
2810 static int osd_inode_setattr(const struct lu_env *env,
2811                              struct inode *inode, const struct lu_attr *attr)
2812 {
2813         __u64 bits = attr->la_valid;
2814
2815         /* Only allow set size for regular file */
2816         if (!S_ISREG(inode->i_mode))
2817                 bits &= ~(LA_SIZE | LA_BLOCKS);
2818
2819         if (bits == 0)
2820                 return 0;
2821
2822         if (bits & LA_ATIME)
2823                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2824         if (bits & LA_CTIME)
2825                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2826         if (bits & LA_MTIME)
2827                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2828         if (bits & LA_SIZE) {
2829                 spin_lock(&inode->i_lock);
2830                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2831                 i_size_write(inode, attr->la_size);
2832                 spin_unlock(&inode->i_lock);
2833         }
2834
2835         /*
2836          * OSD should not change "i_blocks" which is used by quota.
2837          * "i_blocks" should be changed by ldiskfs only.
2838          */
2839         if (bits & LA_MODE)
2840                 inode->i_mode = (inode->i_mode & S_IFMT) |
2841                                 (attr->la_mode & ~S_IFMT);
2842         if (bits & LA_UID)
2843                 i_uid_write(inode, attr->la_uid);
2844         if (bits & LA_GID)
2845                 i_gid_write(inode, attr->la_gid);
2846         if (bits & LA_PROJID)
2847                 i_projid_write(inode, attr->la_projid);
2848         if (bits & LA_NLINK)
2849                 set_nlink(inode, attr->la_nlink);
2850         if (bits & LA_RDEV)
2851                 inode->i_rdev = attr->la_rdev;
2852
2853         if (bits & LA_FLAGS) {
2854                 /* always keep S_NOCMTIME */
2855                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2856                                  S_NOCMTIME;
2857                 /*
2858                  * Ext4 did not transfer inherit flags from
2859                  * @inode->i_flags to raw inode i_flags when writing
2860                  * flags, we do it explictly here.
2861                  */
2862                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2863                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2864                 else
2865                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2866         }
2867         return 0;
2868 }
2869
2870 #ifdef HAVE_PROJECT_QUOTA
2871 static int osd_transfer_project(struct inode *inode, __u32 projid)
2872 {
2873         struct super_block *sb = inode->i_sb;
2874         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2875         int err;
2876         kprojid_t kprojid;
2877         struct ldiskfs_iloc iloc;
2878         struct ldiskfs_inode *raw_inode;
2879         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2880
2881         if (!ldiskfs_has_feature_project(sb)) {
2882                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2883                         == LDISKFS_DEF_PROJID);
2884                 if (projid != LDISKFS_DEF_PROJID)
2885                         return -EOPNOTSUPP;
2886                 else
2887                         return 0;
2888         }
2889
2890         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2891                 return -EOPNOTSUPP;
2892
2893         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2894         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2895                 return 0;
2896
2897         err = ldiskfs_get_inode_loc(inode, &iloc);
2898         if (err)
2899                 return err;
2900
2901         raw_inode = ldiskfs_raw_inode(&iloc);
2902         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2903                 err = -EOVERFLOW;
2904                 brelse(iloc.bh);
2905                 return err;
2906         }
2907         brelse(iloc.bh);
2908
2909         dquot_initialize(inode);
2910         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2911         if (transfer_to[PRJQUOTA]) {
2912                 err = __dquot_transfer(inode, transfer_to);
2913                 dqput(transfer_to[PRJQUOTA]);
2914                 if (err)
2915                         return err;
2916         }
2917
2918         return err;
2919 }
2920 #endif
2921
2922 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2923 {
2924         int rc;
2925
2926         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2927             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2928                 struct iattr iattr;
2929
2930                 ll_vfs_dq_init(inode);
2931                 iattr.ia_valid = 0;
2932                 if (attr->la_valid & LA_UID)
2933                         iattr.ia_valid |= ATTR_UID;
2934                 if (attr->la_valid & LA_GID)
2935                         iattr.ia_valid |= ATTR_GID;
2936                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2937                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2938
2939                 rc = ll_vfs_dq_transfer(inode, &iattr);
2940                 if (rc) {
2941                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2942                                "enforcement enabled on the ldiskfs "
2943                                "filesystem?\n", inode->i_sb->s_id, rc);
2944                         return rc;
2945                 }
2946         }
2947
2948         /* Handle project id transfer here properly */
2949         if (attr->la_valid & LA_PROJID &&
2950             attr->la_projid != i_projid_read(inode)) {
2951 #ifdef HAVE_PROJECT_QUOTA
2952                 rc = osd_transfer_project(inode, attr->la_projid);
2953 #else
2954                 rc = -ENOTSUPP;
2955 #endif
2956                 if (rc) {
2957                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2958                                "enforcement enabled on the ldiskfs "
2959                                "filesystem?\n", inode->i_sb->s_id, rc);
2960                         return rc;
2961                 }
2962         }
2963         return 0;
2964 }
2965
2966 static int osd_attr_set(const struct lu_env *env,
2967                         struct dt_object *dt,
2968                         const struct lu_attr *attr,
2969                         struct thandle *handle)
2970 {
2971         struct osd_object *obj = osd_dt_obj(dt);
2972         struct inode *inode;
2973         int rc;
2974
2975         if (!dt_object_exists(dt))
2976                 return -ENOENT;
2977
2978         LASSERT(handle != NULL);
2979         LASSERT(!dt_object_remote(dt));
2980         LASSERT(osd_invariant(obj));
2981
2982         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2983
2984         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2985             !osd_obj2dev(obj)->od_is_ost) {
2986                 struct osd_thread_info *oti = osd_oti_get(env);
2987                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2988                 struct lu_fid *fid1 = &oti->oti_fid;
2989                 struct osd_inode_id *id = &oti->oti_id;
2990                 struct iam_path_descr *ipd;
2991                 struct iam_container *bag;
2992                 struct osd_thandle *oh;
2993                 int rc;
2994
2995                 fid_cpu_to_be(fid1, fid0);
2996                 memset(id, 1, sizeof(*id));
2997                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2998                                   fid0)->oi_dir.od_container;
2999                 ipd = osd_idx_ipd_get(env, bag);
3000                 if (unlikely(ipd == NULL))
3001                         RETURN(-ENOMEM);
3002
3003                 oh = container_of0(handle, struct osd_thandle, ot_super);
3004                 rc = iam_update(oh->ot_handle, bag,
3005                                 (const struct iam_key *)fid1,
3006                                 (const struct iam_rec *)id, ipd);
3007                 osd_ipd_put(env, bag, ipd);
3008                 return(rc > 0 ? 0 : rc);
3009         }
3010
3011         inode = obj->oo_inode;
3012
3013         rc = osd_quota_transfer(inode, attr);
3014         if (rc)
3015                 return rc;
3016
3017         spin_lock(&obj->oo_guard);
3018         rc = osd_inode_setattr(env, inode, attr);
3019         spin_unlock(&obj->oo_guard);
3020         if (rc != 0)
3021                 GOTO(out, rc);
3022
3023         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3024
3025         if (!(attr->la_valid & LA_FLAGS))
3026                 GOTO(out, rc);
3027
3028         /* Let's check if there are extra flags need to be set into LMA */
3029         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3030                 struct osd_thread_info *info = osd_oti_get(env);
3031                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3032
3033                 LASSERT(!obj->oo_pfid_in_lma);
3034
3035                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3036                                  &info->oti_ost_attrs);
3037                 if (rc)
3038                         GOTO(out, rc);
3039
3040                 lma->lma_incompat |=
3041                         lustre_to_lma_flags(attr->la_flags);
3042                 lustre_lma_swab(lma);
3043                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3044                                      lma, sizeof(*lma), XATTR_REPLACE);
3045                 if (rc != 0) {
3046                         struct osd_device *osd = osd_obj2dev(obj);
3047
3048                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3049                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3050                               lma->lma_incompat, rc);
3051                 } else {
3052                         obj->oo_lma_flags =
3053                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3054                 }
3055                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3056         }
3057 out:
3058         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3059
3060         return rc;
3061 }
3062
3063 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3064                                            struct osd_object *obj,
3065                                            const char *name, const int namelen)
3066 {
3067         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3068 }
3069
3070 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3071                       umode_t mode, struct dt_allocation_hint *hint,
3072                       struct thandle *th, struct lu_attr *attr)
3073 {
3074         int result;
3075         struct osd_device *osd = osd_obj2dev(obj);
3076         struct osd_thandle *oth;
3077         struct dt_object *parent = NULL;
3078         struct inode *inode;
3079         uid_t owner[2] = {0, 0};
3080
3081         if (attr->la_valid & LA_UID)
3082                 owner[0] = attr->la_uid;
3083         if (attr->la_valid & LA_GID)
3084                 owner[1] = attr->la_gid;
3085
3086         LINVRNT(osd_invariant(obj));
3087         LASSERT(obj->oo_inode == NULL);
3088         LASSERT(obj->oo_hl_head == NULL);
3089
3090         if (S_ISDIR(mode) && ldiskfs_pdo) {
3091                 obj->oo_hl_head =
3092                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3093                 if (obj->oo_hl_head == NULL)
3094                         return -ENOMEM;
3095         }
3096
3097         oth = container_of(th, struct osd_thandle, ot_super);
3098         LASSERT(oth->ot_handle->h_transaction != NULL);
3099
3100         if (hint != NULL && hint->dah_parent != NULL &&
3101             !dt_object_remote(hint->dah_parent))
3102                 parent = hint->dah_parent;
3103
3104         inode = ldiskfs_create_inode(oth->ot_handle,
3105                                      parent ? osd_dt_obj(parent)->oo_inode :
3106                                               osd_sb(osd)->s_root->d_inode,
3107                                      mode, owner);
3108         if (!IS_ERR(inode)) {
3109                 /* Do not update file c/mtime in ldiskfs. */
3110                 inode->i_flags |= S_NOCMTIME;
3111
3112                 /*
3113                  * For new created object, it must be consistent,
3114                  * and it is unnecessary to scrub against it.
3115                  */
3116                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3117
3118                 obj->oo_inode = inode;
3119                 result = 0;
3120         } else {
3121                 if (obj->oo_hl_head != NULL) {
3122                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3123                         obj->oo_hl_head = NULL;
3124                 }
3125                 result = PTR_ERR(inode);
3126         }
3127         LINVRNT(osd_invariant(obj));
3128         return result;
3129 }
3130
3131 enum {
3132         OSD_NAME_LEN = 255
3133 };
3134
3135 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3136                      struct lu_attr *attr,
3137                      struct dt_allocation_hint *hint,
3138                      struct dt_object_format *dof,
3139                      struct thandle *th)
3140 {
3141         int result;
3142         struct osd_thandle *oth;
3143         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3144
3145         LASSERT(S_ISDIR(attr->la_mode));
3146
3147         oth = container_of(th, struct osd_thandle, ot_super);
3148         LASSERT(oth->ot_handle->h_transaction != NULL);
3149         result = osd_mkfile(info, obj, mode, hint, th, attr);
3150
3151         return result;
3152 }
3153
3154 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3155                         struct lu_attr *attr,
3156                         struct dt_allocation_hint *hint,
3157                         struct dt_object_format *dof,
3158                         struct thandle *th)
3159 {
3160         int result;
3161         struct osd_thandle *oth;
3162         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3163
3164         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3165
3166         LASSERT(S_ISREG(attr->la_mode));
3167
3168         oth = container_of(th, struct osd_thandle, ot_super);
3169         LASSERT(oth->ot_handle->h_transaction != NULL);
3170
3171         result = osd_mkfile(info, obj, mode, hint, th, attr);
3172         if (result == 0) {
3173                 LASSERT(obj->oo_inode != NULL);
3174                 if (feat->dif_flags & DT_IND_VARKEY)
3175                         result = iam_lvar_create(obj->oo_inode,
3176                                                  feat->dif_keysize_max,
3177                                                  feat->dif_ptrsize,
3178                                                  feat->dif_recsize_max,
3179                                                  oth->ot_handle);
3180                 else
3181                         result = iam_lfix_create(obj->oo_inode,
3182                                                  feat->dif_keysize_max,
3183                                                  feat->dif_ptrsize,
3184                                                  feat->dif_recsize_max,
3185                                                  oth->ot_handle);
3186         }
3187         return result;
3188 }
3189
3190 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3191                      struct lu_attr *attr,
3192                      struct dt_allocation_hint *hint,
3193                      struct dt_object_format *dof,
3194                      struct thandle *th)
3195 {
3196         LASSERT(S_ISREG(attr->la_mode));
3197         return osd_mkfile(info, obj, (attr->la_mode &
3198                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3199                           attr);
3200 }
3201
3202 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3203                      struct lu_attr *attr,
3204                      struct dt_allocation_hint *hint,
3205                      struct dt_object_format *dof,
3206                      struct thandle *th)
3207 {
3208         LASSERT(S_ISLNK(attr->la_mode));
3209         return osd_mkfile(info, obj, (attr->la_mode &
3210                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3211                           attr);
3212 }
3213
3214 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3215                      struct lu_attr *attr,
3216                      struct dt_allocation_hint *hint,
3217                      struct dt_object_format *dof,
3218                      struct thandle *th)
3219 {
3220         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3221         int result;
3222
3223         LINVRNT(osd_invariant(obj));
3224         LASSERT(obj->oo_inode == NULL);
3225         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3226                 S_ISFIFO(mode) || S_ISSOCK(mode));
3227
3228         result = osd_mkfile(info, obj, mode, hint, th, attr);
3229         if (result == 0) {
3230                 LASSERT(obj->oo_inode != NULL);
3231                 /*
3232                  * This inode should be marked dirty for i_rdev.  Currently
3233                  * that is done in the osd_attr_init().
3234                  */
3235                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3236                                    attr->la_rdev);
3237         }
3238         LINVRNT(osd_invariant(obj));
3239         return result;
3240 }
3241
3242 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3243                               struct lu_attr *,
3244                               struct dt_allocation_hint *hint,
3245                               struct dt_object_format *dof,
3246                               struct thandle *);
3247
3248 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3249 {
3250         osd_obj_type_f result;
3251
3252         switch (type) {
3253         case DFT_DIR:
3254                 result = osd_mkdir;
3255                 break;
3256         case DFT_REGULAR:
3257                 result = osd_mkreg;
3258                 break;
3259         case DFT_SYM:
3260                 result = osd_mksym;
3261                 break;
3262         case DFT_NODE:
3263                 result = osd_mknod;
3264                 break;
3265         case DFT_INDEX:
3266                 result = osd_mk_index;
3267                 break;
3268
3269         default:
3270                 LBUG();
3271                 break;
3272         }
3273         return result;
3274 }
3275
3276
3277 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3278                         struct dt_object *parent, struct dt_object *child,
3279                         umode_t child_mode)
3280 {
3281         LASSERT(ah);
3282
3283         ah->dah_parent = parent;
3284         ah->dah_mode = child_mode;
3285
3286         if (parent != NULL && !dt_object_remote(parent)) {
3287                 /* will help to find FID->ino at dt_insert("..") */
3288                 struct osd_object *pobj = osd_dt_obj(parent);
3289
3290                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3291         }
3292 }
3293
3294 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3295                           struct lu_attr *attr, struct dt_object_format *dof)
3296 {
3297         struct inode *inode = obj->oo_inode;
3298         __u64 valid = attr->la_valid;
3299         int result;
3300
3301         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3302
3303         if (dof->dof_type != DFT_NODE)
3304                 attr->la_valid &= ~LA_RDEV;
3305         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3306                 attr->la_valid &= ~LA_ATIME;
3307         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3308                 attr->la_valid &= ~LA_CTIME;
3309         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3310                 attr->la_valid &= ~LA_MTIME;
3311
3312         result = osd_quota_transfer(inode, attr);
3313         if (result)
3314                 return;
3315
3316         if (attr->la_valid != 0) {
3317                 result = osd_inode_setattr(info->oti_env, inode, attr);
3318                 /*
3319                  * The osd_inode_setattr() should always succeed here.  The
3320                  * only error that could be returned is EDQUOT when we are
3321                  * trying to change the UID or GID of the inode. However, this
3322                  * should not happen since quota enforcement is no longer
3323                  * enabled on ldiskfs (lquota takes care of it).
3324                  */
3325                 LASSERTF(result == 0, "%d\n", result);
3326                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3327         }
3328
3329         attr->la_valid = valid;
3330 }
3331
3332 /**
3333  * Helper function for osd_create()
3334  *
3335  * \retval 0, on success
3336  */
3337 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3338                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3339                         struct dt_object_format *dof, struct thandle *th)
3340 {
3341         int result;
3342         __u32 umask;
3343
3344         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3345
3346         /* we drop umask so that permissions we pass are not affected */
3347         umask = current->fs->umask;
3348         current->fs->umask = 0;
3349
3350         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3351                                                   th);
3352         if (likely(obj->oo_inode != NULL)) {
3353                 LASSERT(obj->oo_inode->i_state & I_NEW);
3354
3355                 /*
3356                  * Unlock the inode before attr initialization to avoid
3357                  * unnecessary dqget operations. LU-6378
3358                  */
3359                 unlock_new_inode(obj->oo_inode);
3360         }
3361
3362         if (likely(result == 0)) {
3363                 osd_attr_init(info, obj, attr, dof);
3364                 osd_object_init0(obj);
3365         }
3366
3367         /* restore previous umask value */
3368         current->fs->umask = umask;
3369
3370         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3371
3372         return result;
3373 }
3374
3375 /**
3376  * Helper function for osd_create()
3377  *
3378  * \retval 0, on success
3379  */
3380 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3381                            const struct lu_fid *fid, struct thandle *th)
3382 {
3383         struct osd_thread_info *info = osd_oti_get(env);
3384         struct osd_inode_id    *id   = &info->oti_id;
3385         struct osd_device      *osd  = osd_obj2dev(obj);
3386         struct osd_thandle     *oh;
3387         int rc;
3388
3389         LASSERT(obj->oo_inode != NULL);
3390
3391         oh = container_of0(th, struct osd_thandle, ot_super);
3392         LASSERT(oh->ot_handle);
3393         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3394
3395         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3396         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3397                            OI_CHECK_FLD, NULL);
3398         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3399
3400         return rc;
3401 }
3402
3403 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3404                    u64 seq, struct lu_seq_range *range)
3405 {
3406         struct seq_server_site *ss = osd_seq_site(osd);
3407
3408         if (fid_seq_is_idif(seq)) {
3409                 fld_range_set_ost(range);
3410                 range->lsr_index = idif_ost_idx(seq);
3411                 return 0;
3412         }
3413
3414         if (!fid_seq_in_fldb(seq)) {
3415                 fld_range_set_mdt(range);
3416                 if (ss != NULL)
3417                         /*
3418                          * FIXME: If ss is NULL, it suppose not get lsr_index
3419                          * at all
3420                          */
3421                         range->lsr_index = ss->ss_node_id;
3422                 return 0;
3423         }
3424
3425         LASSERT(ss != NULL);
3426         fld_range_set_any(range);
3427         /* OSD will only do local fld lookup */
3428         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3429 }
3430
3431 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3432                               struct lu_attr *attr,
3433                               struct dt_allocation_hint *hint,
3434                               struct dt_object_format *dof,
3435                               struct thandle *handle)
3436 {
3437         struct osd_thandle *oh;
3438         int rc;
3439
3440         ENTRY;
3441
3442         LASSERT(handle != NULL);
3443
3444         oh = container_of0(handle, struct osd_thandle, ot_super);
3445         LASSERT(oh->ot_handle == NULL);
3446
3447         /*
3448          * EA object consumes more credits than regular object: osd_mk_index
3449          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3450          * leaf_node, could involves the block, block bitmap, groups, GDT
3451          * change for each block, so add 4 * 2 credits in that case.
3452          */
3453         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3454                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3455                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3456         /*
3457          * Reuse idle OI block may cause additional one OI block
3458          * to be changed.
3459          */
3460         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3461                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3462
3463         /* will help to find FID->ino mapping at dt_insert() */
3464         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3465                                    osd_dt_obj(dt));
3466         if (rc != 0)
3467                 RETURN(rc);
3468
3469         if (!attr)
3470                 RETURN(0);
3471
3472         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3473                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3474                                    NULL, OSD_QID_INODE);
3475         if (rc != 0)
3476                 RETURN(rc);
3477
3478         RETURN(rc);
3479 }
3480
3481 /**
3482  * Called to destroy on-disk representation of the object
3483  *
3484  * Concurrency: must be locked
3485  */
3486 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3487                                struct thandle *th)
3488 {
3489         struct osd_object *obj = osd_dt_obj(dt);
3490         struct inode *inode = obj->oo_inode;
3491         struct osd_thandle *oh;
3492         int rc;
3493
3494         ENTRY;
3495
3496         if (inode == NULL)
3497                 RETURN(-ENOENT);
3498
3499         oh = container_of0(th, struct osd_thandle, ot_super);
3500         LASSERT(oh->ot_handle == NULL);
3501
3502         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3503                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3504
3505         /* For removing agent entry */
3506         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3507                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3508
3509         /*
3510          * Recycle idle OI leaf may cause additional three OI blocks
3511          * to be changed.
3512          */
3513         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3514                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3515                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3516         /* one less inode */
3517         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3518                                    i_projid_read(inode), -1, oh, obj, NULL,
3519                                    OSD_QID_INODE);
3520         if (rc)
3521                 RETURN(rc);
3522         /* data to be truncated */
3523         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3524                                    i_projid_read(inode), 0, oh, obj, NULL,
3525                                    OSD_QID_BLK);
3526         if (rc)
3527                 RETURN(rc);
3528
3529         /*
3530          * will help to find FID->ino when this object is being
3531          * added to PENDING
3532          */
3533         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3534
3535         RETURN(rc);
3536 }
3537
3538 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3539                        struct thandle *th)
3540 {
3541         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3542         struct osd_object *obj = osd_dt_obj(dt);
3543         struct inode *inode = obj->oo_inode;
3544         struct osd_device *osd = osd_obj2dev(obj);
3545         struct osd_thandle *oh;
3546         int result;
3547
3548         ENTRY;
3549
3550         oh = container_of0(th, struct osd_thandle, ot_super);
3551         LASSERT(oh->ot_handle);
3552         LASSERT(inode);
3553         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3554
3555         if (unlikely(fid_is_acct(fid)))
3556                 RETURN(-EPERM);
3557
3558         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3559                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3560                 if (result != 0)
3561                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3562                                osd_name(osd), PFID(fid), result);
3563         }
3564
3565         if (S_ISDIR(inode->i_mode)) {
3566                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3567                         inode->i_nlink == 2);
3568
3569                 spin_lock(&obj->oo_guard);
3570                 clear_nlink(inode);
3571                 spin_unlock(&obj->oo_guard);
3572                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3573         }
3574
3575         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3576
3577         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3578
3579         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3580                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3581                                        oh->ot_handle, OI_CHECK_FLD);
3582
3583         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3584         /* XXX: add to ext3 orphan list */
3585         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3586
3587         /* not needed in the cache anymore */
3588         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3589         obj->oo_destroyed = 1;
3590
3591         RETURN(0);
3592 }
3593
3594 /**
3595  * Put the fid into lustre_mdt_attrs, and then place the structure
3596  * inode's ea. This fid should not be altered during the life time
3597  * of the inode.
3598  *
3599  * \retval +ve, on success
3600  * \retval -ve, on error
3601  *
3602  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3603  */
3604 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3605                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3606 {
3607         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3608         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3609         int rc;
3610
3611         ENTRY;
3612
3613         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3614                 RETURN(0);
3615
3616         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3617                 rc = -ENOMEM;
3618
3619         lustre_loa_init(loa, fid, compat, incompat);
3620         lustre_loa_swab(loa, false);
3621
3622         /*
3623          * For the OST device with 256 bytes inode size by default,
3624          * the PFID EA will be stored together with LMA EA to avoid
3625          * performance trouble. Otherwise the PFID EA can be stored
3626          * independently. LU-8998
3627          */
3628         if ((compat & LMAC_FID_ON_OST) &&
3629             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3630                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3631                                      sizeof(*loa), XATTR_CREATE);
3632         else
3633                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3634                                      sizeof(*lma), XATTR_CREATE);
3635         /*
3636          * LMA may already exist, but we need to check that all the
3637          * desired compat/incompat flags have been added.
3638          */
3639         if (unlikely(rc == -EEXIST)) {
3640                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3641                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3642                 if (rc < 0)
3643                         RETURN(rc);
3644
3645                 if (rc < sizeof(*lma))
3646                         RETURN(-EINVAL);
3647
3648                 lustre_loa_swab(loa, true);
3649                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3650                     ((compat == 0 && incompat == 0) ||
3651                      (!(~lma->lma_compat & compat) &&
3652                       !(~lma->lma_incompat & incompat))))
3653                         RETURN(0);
3654
3655                 lma->lma_self_fid = *fid;
3656                 lma->lma_compat |= compat;
3657                 lma->lma_incompat |= incompat;
3658                 if (rc == sizeof(*lma)) {
3659                         lustre_lma_swab(lma);
3660                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3661                                              sizeof(*lma), XATTR_REPLACE);
3662                 } else {
3663                         lustre_loa_swab(loa, false);
3664                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3665                                              sizeof(*loa), XATTR_REPLACE);
3666                 }
3667         }
3668
3669         RETURN(rc);
3670 }
3671
3672 /**
3673  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3674  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3675  * To have compatilibility with 1.8 ldiskfs driver we need to have
3676  * magic number at start of fid data.
3677  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3678  * its inmemory API.
3679  */
3680 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3681                                          const struct lu_fid *fid)
3682 {
3683         if (!fid_is_namespace_visible(fid) ||
3684             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3685                 param->edp_magic = 0;
3686                 return;
3687         }
3688
3689         param->edp_magic = LDISKFS_LUFID_MAGIC;
3690         param->edp_len =  sizeof(struct lu_fid) + 1;
3691         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3692 }
3693
3694 /**
3695  * Try to read the fid from inode ea into dt_rec.
3696  *
3697  * \param fid object fid.
3698  *
3699  * \retval 0 on success
3700  */
3701 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3702                           __u32 ino, struct lu_fid *fid,
3703                           struct osd_inode_id *id)
3704 {
3705         struct osd_thread_info *info  = osd_oti_get(env);
3706         struct inode *inode;
3707
3708         ENTRY;
3709
3710         osd_id_gen(id, ino, OSD_OII_NOGEN);
3711         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3712         if (IS_ERR(inode))
3713                 RETURN(PTR_ERR(inode));
3714
3715         iput(inode);
3716         RETURN(0);
3717 }
3718
3719 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3720                                         struct inode *dir,
3721                                         struct inode *parent_dir,
3722                                         const struct lu_fid *dot_fid,
3723                                         const struct lu_fid *dot_dot_fid,
3724                                         struct osd_thandle *oth)
3725 {
3726         struct ldiskfs_dentry_param *dot_ldp;
3727         struct ldiskfs_dentry_param *dot_dot_ldp;
3728         __u32 saved_nlink = dir->i_nlink;
3729         int rc;
3730
3731         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3732         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3733
3734         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3735         dot_ldp->edp_magic = 0;
3736
3737         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3738                                     dir, dot_ldp, dot_dot_ldp);
3739         /*
3740          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3741          * the subseqent ref_add() will increase the dir->i_nlink
3742          * as 3. That is incorrect for new created directory.
3743          *
3744          * It looks like hack, because we want to make the OSD API
3745          * to be order-independent for new created directory object
3746          * between dt_insert(..) and ref_add() operations.
3747          *
3748          * Here, we only restore the in-RAM dir-inode's nlink attr,
3749          * becuase if the nlink attr is not 2, then there will be
3750          * ref_add() called following the dt_insert(..), such call
3751          * will make both the in-RAM and on-disk dir-inode's nlink
3752          * attr to be set as 2. LU-7447
3753          */
3754         set_nlink(dir, saved_nlink);
3755         return rc;
3756 }
3757
3758 /**
3759  * Create an local agent inode for remote entry
3760  */
3761 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3762                                                   struct osd_device *osd,
3763                                                   struct osd_object *pobj,
3764                                                   const struct lu_fid *fid,
3765                                                   __u32 type,
3766                                                   struct thandle *th)
3767 {
3768         struct osd_thread_info *info = osd_oti_get(env);
3769         struct inode *local;
3770         struct osd_thandle *oh;
3771         int rc;
3772
3773         ENTRY;
3774
3775         LASSERT(th);
3776         oh = container_of(th, struct osd_thandle, ot_super);
3777         LASSERT(oh->ot_handle->h_transaction != NULL);
3778
3779         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3780                                      NULL);
3781         if (IS_ERR(local)) {
3782                 CERROR("%s: create local error %d\n", osd_name(osd),
3783                        (int)PTR_ERR(local));
3784                 RETURN(local);
3785         }
3786
3787         /*
3788          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3789          * correct gid on remote file, not agent here
3790          */
3791         local->i_gid = current_fsgid();
3792         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3793
3794         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3795          * That gives e2fsck something to look at and be happy, and allows
3796          * debugging if we need to determine where this symlink came from.
3797          */
3798         if (S_ISLNK(type)) {
3799                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3800                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3801                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3802
3803                 i_size_write(local, rc);
3804                 LDISKFS_I(local)->i_disksize = rc;
3805         }
3806         unlock_new_inode(local);
3807
3808         /* Agent inode should not have project ID */
3809 #ifdef  HAVE_PROJECT_QUOTA
3810         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3811             i_projid_read(pobj->oo_inode) != 0) {
3812                 rc = osd_transfer_project(local, 0);
3813                 if (rc) {
3814                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3815                                "quota enforcement enabled on the ldiskfs "
3816                                "filesystem?\n", local->i_sb->s_id, rc);
3817                         RETURN(ERR_PTR(rc));
3818                 }
3819         }
3820 #endif
3821         /* Set special LMA flag for local agent inode */
3822         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3823         if (rc != 0) {
3824                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3825                        osd_name(osd), PFID(fid), rc);
3826                 RETURN(ERR_PTR(rc));
3827         }
3828
3829         if (!S_ISDIR(type))
3830                 RETURN(local);
3831
3832         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3833                                          lu_object_fid(&pobj->oo_dt.do_lu),
3834                                          fid, oh);
3835         if (rc != 0) {
3836                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3837                         osd_name(osd), PFID(fid), rc);
3838                 RETURN(ERR_PTR(rc));
3839         }
3840
3841         RETURN(local);
3842 }
3843
3844 /**
3845  * when direntry is deleted, we have to take care of possible agent inode
3846  * referenced by that. unfortunately we can't do this at that point:
3847  * iget() within a running transaction leads to deadlock and we better do
3848  * not call that every delete declaration to save performance. so we put
3849  * a potention agent inode on a list and process that once the transaction
3850  * is over. Notice it's not any worse in terms of real orphans as regular
3851  * object destroy doesn't put inodes on the on-disk orphan list. this should
3852  * be addressed separately
3853  */
3854 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3855                                             struct osd_thandle *oh,
3856                                             __u32 ino)
3857 {
3858         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3859         struct osd_obj_orphan *oor;
3860
3861         OBD_ALLOC_PTR(oor);
3862         if (oor == NULL)
3863                 return -ENOMEM;
3864
3865         oor->oor_ino = ino;
3866         oor->oor_env = (struct lu_env *)env;
3867         spin_lock(&osd->od_osfs_lock);
3868         list_add(&oor->oor_list, &osd->od_orphan_list);
3869         spin_unlock(&osd->od_osfs_lock);
3870
3871         oh->ot_remove_agents = 1;
3872
3873         return 0;
3874
3875 }
3876
3877 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3878                                                 struct osd_device *osd)
3879 {
3880         struct osd_thread_info *info = osd_oti_get(env);
3881         struct osd_obj_orphan *oor, *tmp;
3882         struct osd_inode_id id;
3883         struct list_head list;
3884         struct inode *inode;
3885         struct lu_fid fid;
3886         handle_t *jh;
3887         __u32 ino;
3888
3889         INIT_LIST_HEAD(&list);
3890
3891         spin_lock(&osd->od_osfs_lock);
3892         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3893                 if (oor->oor_env == env) {
3894                         list_del(&oor->oor_list);
3895                         list_add(&oor->oor_list, &list);
3896                 }
3897         }
3898         spin_unlock(&osd->od_osfs_lock);
3899
3900         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3901
3902                 ino = oor->oor_ino;
3903
3904                 list_del(&oor->oor_list);
3905                 OBD_FREE_PTR(oor);
3906
3907                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3908                 inode = osd_iget_fid(info, osd, &id, &fid);
3909                 if (IS_ERR(inode))
3910                         continue;
3911
3912                 if (!osd_remote_fid(env, osd, &fid)) {
3913                         iput(inode);
3914                         continue;
3915                 }
3916
3917                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3918                 clear_nlink(inode);
3919                 mark_inode_dirty(inode);
3920                 ldiskfs_journal_stop(jh);
3921                 iput(inode);
3922         }
3923
3924         return 0;
3925 }
3926
3927 /**
3928  * OSD layer object create function for OST objects (b=11826).
3929  *
3930  * The FID is inserted into inode xattr here.
3931  *
3932  * \retval   0, on success
3933  * \retval -ve, on error
3934  */
3935 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3936                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3937                       struct dt_object_format *dof, struct thandle *th)
3938 {
3939         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3940         struct osd_object *obj = osd_dt_obj(dt);
3941         struct osd_thread_info *info = osd_oti_get(env);
3942         int result, on_ost = 0;
3943
3944         ENTRY;
3945
3946         if (dt_object_exists(dt))
3947                 RETURN(-EEXIST);
3948
3949         LINVRNT(osd_invariant(obj));
3950         LASSERT(!dt_object_remote(dt));
3951         LASSERT(osd_is_write_locked(env, obj));
3952         LASSERT(th != NULL);
3953
3954         if (unlikely(fid_is_acct(fid)))
3955                 /*
3956                  * Quota files can't be created from the kernel any more,
3957                  * 'tune2fs -O quota' will take care of creating them
3958                  */
3959                 RETURN(-EPERM);
3960
3961         result = __osd_create(info, obj, attr, hint, dof, th);
3962         if (result == 0) {
3963                 if (fid_is_idif(fid) &&
3964                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3965                         struct lu_fid *tfid = &info->oti_fid;
3966                         struct ost_id *oi   = &info->oti_ostid;
3967
3968                         fid_to_ostid(fid, oi);
3969                         ostid_to_fid(tfid, oi, 0);
3970                         on_ost = 1;
3971                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3972                                                 LMAC_FID_ON_OST, 0);
3973                 } else {
3974                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3975                                                fid, OI_CHECK_FLD);
3976                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3977                                                 on_ost ? LMAC_FID_ON_OST : 0,
3978                                                 0);
3979                 }
3980                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3981                         obj->oo_dt.do_body_ops = &osd_body_ops;
3982         }
3983
3984         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3985                 result = __osd_oi_insert(env, obj, fid, th);
3986
3987         /*
3988          * a small optimization - dt_insert() isn't usually applied
3989          * to OST objects, so we don't need to cache OI mapping for
3990          * OST objects
3991          */
3992         if (result == 0 && on_ost == 0) {
3993                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3994
3995                 result = osd_idc_find_and_init(env, osd, obj);
3996                 LASSERT(result == 0);
3997         }
3998
3999         LASSERT(ergo(result == 0,
4000                      dt_object_exists(dt) && !dt_object_remote(dt)));
4001         LINVRNT(osd_invariant(obj));
4002         RETURN(result);
4003 }
4004
4005 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4006                                struct thandle *handle)
4007 {
4008         struct osd_thandle *oh;
4009
4010         /* it's possible that object doesn't exist yet */
4011         LASSERT(handle != NULL);
4012
4013         oh = container_of0(handle, struct osd_thandle, ot_super);
4014         LASSERT(oh->ot_handle == NULL);
4015
4016         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4017                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4018
4019         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4020
4021         return 0;
4022 }
4023
4024 /*
4025  * Concurrency: @dt is write locked.
4026  */
4027 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4028                        struct thandle *th)
4029 {
4030         struct osd_object *obj = osd_dt_obj(dt);
4031         struct inode *inode = obj->oo_inode;
4032         struct osd_thandle *oh;
4033         int rc = 0;
4034
4035         if (!dt_object_exists(dt) || obj->oo_destroyed)
4036                 return -ENOENT;
4037
4038         LINVRNT(osd_invariant(obj));
4039         LASSERT(!dt_object_remote(dt));
4040         LASSERT(osd_is_write_locked(env, obj));
4041         LASSERT(th != NULL);
4042
4043         oh = container_of0(th, struct osd_thandle, ot_super);
4044         LASSERT(oh->ot_handle != NULL);
4045
4046         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4047
4048         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4049                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4050         /*
4051          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4052          * (65000) subdirectories by storing "1" in i_nlink if the link count
4053          * would otherwise overflow. Directory tranversal tools understand
4054          * that (st_nlink == 1) indicates that the filesystem dose not track
4055          * hard links count on the directory, and will not abort subdirectory
4056          * scanning early once (st_nlink - 2) subdirs have been found.
4057          *
4058          * This also has to properly handle the case of inodes with nlink == 0
4059          * in case they are being linked into the PENDING directory
4060          */
4061         spin_lock(&obj->oo_guard);
4062         if (unlikely(inode->i_nlink == 0))
4063                 /* inc_nlink from 0 may cause WARN_ON */
4064                 set_nlink(inode, 1);
4065         else {
4066                 ldiskfs_inc_count(oh->ot_handle, inode);
4067                 if (!S_ISDIR(inode->i_mode))
4068                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4069         }
4070         spin_unlock(&obj->oo_guard);
4071
4072         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4073         LINVRNT(osd_invariant(obj));
4074
4075         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4076
4077         return rc;
4078 }
4079
4080 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4081                                struct thandle *handle)
4082 {
4083         struct osd_thandle *oh;
4084
4085         if (!dt_object_exists(dt))
4086                 return -ENOENT;
4087
4088         LASSERT(!dt_object_remote(dt));
4089         LASSERT(handle != NULL);
4090
4091         oh = container_of0(handle, struct osd_thandle, ot_super);
4092         LASSERT(oh->ot_handle == NULL);
4093
4094         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4095                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4096
4097         return 0;
4098 }
4099
4100 /*
4101  * Concurrency: @dt is write locked.
4102  */
4103 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4104                        struct thandle *th)
4105 {
4106         struct osd_object *obj = osd_dt_obj(dt);
4107         struct inode *inode = obj->oo_inode;
4108         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4109         struct osd_thandle *oh;
4110
4111         if (!dt_object_exists(dt))
4112                 return -ENOENT;
4113
4114         LINVRNT(osd_invariant(obj));
4115         LASSERT(!dt_object_remote(dt));
4116         LASSERT(osd_is_write_locked(env, obj));
4117         LASSERT(th != NULL);
4118
4119         oh = container_of0(th, struct osd_thandle, ot_super);
4120         LASSERT(oh->ot_handle != NULL);
4121
4122         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4123
4124         spin_lock(&obj->oo_guard);
4125         /*
4126          * That can be result of upgrade from old Lustre version and
4127          * applied only to local files.  Just skip this ref_del call.
4128          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4129          */
4130         if (inode->i_nlink == 0) {
4131                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4132                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4133                              ", maybe an upgraded file? (LU-3915)\n",
4134                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4135                 spin_unlock(&obj->oo_guard);
4136                 return 0;
4137         }
4138
4139         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4140                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4141
4142         ldiskfs_dec_count(oh->ot_handle, inode);
4143         spin_unlock(&obj->oo_guard);
4144
4145         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4146         LINVRNT(osd_invariant(obj));
4147
4148         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4149
4150         return 0;
4151 }
4152
4153 /*
4154  * Concurrency: @dt is read locked.
4155  */
4156 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4157                          struct lu_buf *buf, const char *name)
4158 {
4159         struct osd_object      *obj    = osd_dt_obj(dt);
4160         struct inode           *inode  = obj->oo_inode;
4161         struct osd_thread_info *info   = osd_oti_get(env);
4162         struct dentry          *dentry = &info->oti_obj_dentry;
4163         bool cache_xattr = false;
4164         int rc;
4165
4166         LASSERT(buf);
4167
4168         /* version get is not real XATTR but uses xattr API */
4169         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4170                 dt_obj_version_t *ver = buf->lb_buf;
4171
4172                 /*
4173                  * for version we are just using xattr API but change inode
4174                  * field instead
4175                  */
4176                 if (buf->lb_len == 0)
4177                         return sizeof(dt_obj_version_t);
4178
4179                 if (buf->lb_len < sizeof(dt_obj_version_t))
4180                         return -ERANGE;
4181
4182                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4183                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4184
4185                 *ver = LDISKFS_I(inode)->i_fs_version;
4186
4187                 return sizeof(dt_obj_version_t);
4188         }
4189
4190         if (!dt_object_exists(dt))
4191                 return -ENOENT;
4192
4193         LASSERT(!dt_object_remote(dt));
4194         LASSERT(inode->i_op != NULL);
4195 #ifdef HAVE_IOP_XATTR
4196         LASSERT(inode->i_op->getxattr != NULL);
4197 #endif
4198
4199         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4200             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4201                 cache_xattr = true;
4202
4203         if (cache_xattr) {
4204                 rc = osd_oxc_get(obj, name, buf);
4205                 if (rc != -ENOENT)
4206                         return rc;
4207         }
4208
4209         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4210                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4211                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4212                 struct filter_fid *ff;
4213                 struct ost_layout *ol;
4214
4215                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4216                 if (rc)
4217                         return rc;
4218
4219                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4220
4221                 rc = sizeof(*ff);
4222                 if (buf->lb_len == 0 || !buf->lb_buf)
4223                         return rc;
4224
4225                 if (buf->lb_len < rc)
4226                         return -ERANGE;
4227
4228                 ff = buf->lb_buf;
4229                 ol = &ff->ff_layout;
4230                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4231                                                   PFID_STRIPE_IDX_BITS);
4232                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4233                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4234                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4235                 if (lma->lma_compat & LMAC_COMP_INFO) {
4236                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4237                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4238                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4239                 } else {
4240                         ol->ol_comp_start = 0;
4241                         ol->ol_comp_end = 0;
4242                         ol->ol_comp_id = 0;
4243                 }
4244         } else {
4245                 rc = __osd_xattr_get(inode, dentry, name,
4246                                      buf->lb_buf, buf->lb_len);
4247         }
4248
4249         if (cache_xattr) {
4250                 if (rc == -ENOENT || rc == -ENODATA)
4251                         osd_oxc_add(obj, name, NULL, 0);
4252                 else if (rc > 0 && buf->lb_buf != NULL)
4253                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4254         }
4255
4256         return rc;
4257 }
4258
4259 static int osd_declare_xattr_set(const struct lu_env *env,
4260                                  struct dt_object *dt,
4261                                  const struct lu_buf *buf, const char *name,
4262                                  int fl, struct thandle *handle)
4263 {
4264         struct osd_thandle *oh;
4265         int credits = 0;
4266         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4267
4268         LASSERT(handle != NULL);
4269
4270         oh = container_of0(handle, struct osd_thandle, ot_super);
4271         LASSERT(oh->ot_handle == NULL);
4272
4273         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4274                 /*
4275                  * For non-upgrading case, the LMA is set first and
4276                  * usually fit inode. But for upgrade case, the LMA
4277                  * may be in another separated EA block.
4278                  */
4279                 if (dt_object_exists(dt)) {
4280                         if (fl == LU_XATTR_REPLACE)
4281                                 credits = 1;
4282                         else
4283                                 goto upgrade;
4284                 }
4285         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4286                 credits = 1;
4287         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4288                 /* We may need to delete the old PFID EA. */
4289                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4290                 if (fl == LU_XATTR_REPLACE)
4291                         credits += 1;
4292                 else
4293                         goto upgrade;
4294         } else {
4295                 /*
4296                  * If some name entry resides on remote MDT, then will create
4297                  * agent entry under remote parent. On the other hand, if the
4298                  * remote entry will be removed, then related agent entry may
4299                  * need to be removed from the remote parent. So there may be
4300                  * kinds of cases, let's declare enough credits. The credits
4301                  * for create agent entry is enough for remove case.
4302                  */
4303                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4304                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4305                         if (dt_object_exists(dt))
4306                                 credits += 1; /* For updating LMA */
4307                 }
4308
4309 upgrade:
4310                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4311
4312                 if (buf != NULL) {
4313                         ssize_t buflen;
4314
4315                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4316                                 /*
4317                                  * learn xattr size from osd_xattr_get if
4318                                  * attribute has not been read yet
4319                                  */
4320                                 buflen = __osd_xattr_get(
4321                                     osd_dt_obj(dt)->oo_inode,
4322                                     &osd_oti_get(env)->oti_obj_dentry,
4323                                     name, NULL, 0);
4324                                 if (buflen < 0)
4325                                         buflen = 0;
4326                         } else {
4327                                 buflen = buf->lb_len;
4328                         }
4329
4330                         if (buflen > sb->s_blocksize) {
4331                                 credits += osd_calc_bkmap_credits(
4332                                     sb, NULL, 0, -1,
4333                                     (buflen + sb->s_blocksize - 1) >>
4334                                     sb->s_blocksize_bits);
4335                         }
4336                 }
4337                 /*
4338                  * xattr set may involve inode quota change, reserve credits for
4339                  * dquot_initialize()
4340                  */
4341                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4342         }
4343
4344         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4345
4346         return 0;
4347 }
4348
4349 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4350                               const struct lu_buf *buf, int fl,
4351                               struct thandle *handle)
4352 {
4353         struct osd_thread_info *info = osd_oti_get(env);
4354         struct dentry *dentry = &info->oti_obj_dentry;
4355         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4356         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4357         struct inode *inode = obj->oo_inode;
4358         struct filter_fid *ff = buf->lb_buf;
4359         struct ost_layout *ol = &ff->ff_layout;
4360         int flags = XATTR_REPLACE;
4361         int rc;
4362
4363         ENTRY;
4364
4365         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4366                 RETURN(-EINVAL);
4367
4368         rc = osd_get_lma(info, inode, dentry, loa);
4369         if (rc == -ENODATA) {
4370                 /* Usually for upgarding from old device */
4371                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4372                                 LMAC_FID_ON_OST, 0);
4373                 flags = XATTR_CREATE;
4374         } else if (rc) {
4375                 RETURN(rc);
4376         }
4377
4378         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4379                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4380                         RETURN(-EEXIST);
4381
4382                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4383                         /* Separate PFID EA from LMA */
4384                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4385                         lustre_lma_swab(lma);
4386                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4387                                              sizeof(*lma), XATTR_REPLACE);
4388                         if (!rc) {
4389                                 obj->oo_pfid_in_lma = 0;
4390                                 rc = LU_XATTR_CREATE;
4391                         }
4392
4393                         RETURN(rc);
4394                 }
4395         } else {
4396                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4397                         RETURN(fl);
4398
4399                 /*
4400                  * Old client does not send stripe information,
4401                  * then store the PFID EA on disk separatedly.
4402                  */
4403                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4404                              ol->ol_stripe_size == 0))
4405                         RETURN(fl);
4406
4407                 /* Remove old PFID EA entry firstly. */
4408                 ll_vfs_dq_init(inode);
4409                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4410                 if (rc == -ENODATA) {
4411                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4412                                 RETURN(rc);
4413                 } else if (rc) {
4414                         RETURN(rc);
4415                 }
4416         }
4417
4418         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4419         if (likely(ol->ol_stripe_size != 0)) {
4420                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4421                                              PFID_STRIPE_IDX_BITS;
4422                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4423                 lma->lma_compat |= LMAC_STRIPE_INFO;
4424                 if (ol->ol_comp_id != 0) {
4425                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4426                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4427                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4428                         lma->lma_compat |= LMAC_COMP_INFO;
4429                 }
4430         }
4431
4432         lustre_loa_swab(loa, false);
4433
4434         /* Store the PFID EA inside LMA. */
4435         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4436                              flags);
4437         if (!rc)
4438                 obj->oo_pfid_in_lma = 1;
4439
4440         RETURN(rc);
4441 }
4442
4443 /*
4444  * In DNE environment, the object (in spite of regular file or directory)
4445  * and its name entry may reside on different MDTs. Under such case, we will
4446  * create an agent entry on the MDT where the object resides. The agent entry
4447  * references the object locally, that makes the object to be visible to the
4448  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4449  * such as 'tar' can handle the object properly.
4450  *
4451  * We handle the agent entry during set linkEA that is the common interface
4452  * for both regular file and directroy, can handle kinds of cases, such as
4453  * create/link/unlink/rename, and so on.
4454  *
4455  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4456  *
4457  * XXX: There are two known issues:
4458  * 1. For one object, we will create at most one agent entry even if there
4459  *    may be more than one cross-MDTs hard links on the object. So the local
4460  *    e2fsck may claim that the object's nlink is larger than the name entries
4461  *    that reference such inode. And in further, the e2fsck will fix the nlink
4462  *    attribute to match the local references. Then it will cause the object's
4463  *    nlink attribute to be inconsistent with the global references. it is bad
4464  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4465  *    other hand, the global namespace LFSCK can repair the object's attribute
4466  *    according to the linkEA.
4467  * 2. There may be too many hard links on the object as to its linkEA overflow,
4468  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4469  *    case happened, then at this point, we do not know whether there are some
4470  *    cross-MDTs reference. But there are local references, it guarantees that
4471  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4472  */
4473 static int osd_xattr_handle_linkea(const struct lu_env *env,
4474                                    struct osd_device *osd,
4475                                    struct osd_object *obj,
4476                                    const struct lu_buf *buf,
4477                                    struct thandle *handle)
4478 {
4479         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4480         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4481         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4482         struct lu_name tmpname;
4483         struct osd_thandle *oh;
4484         int rc;
4485         bool remote = false;
4486
4487         ENTRY;
4488
4489         oh = container_of0(handle, struct osd_thandle, ot_super);
4490         LASSERT(oh->ot_handle != NULL);
4491
4492         rc = linkea_init_with_rec(&ldata);
4493         if (!rc) {
4494                 linkea_first_entry(&ldata);
4495                 while (ldata.ld_lee != NULL && !remote) {
4496                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4497                                             &tmpname, tfid);
4498                         if (osd_remote_fid(env, osd, tfid) > 0)
4499                                 remote = true;
4500                         else
4501                                 linkea_next_entry(&ldata);
4502                 }
4503         } else if (rc == -ENODATA) {
4504                 rc = 0;
4505         } else {
4506                 RETURN(rc);
4507         }
4508
4509         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4510                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4511                 if (rc)
4512                         CERROR("%s: failed to remove agent entry for "DFID
4513                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4514         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4515                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4516                 if (rc)
4517                         CERROR("%s: failed to create agent entry for "DFID
4518                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4519         }
4520
4521         RETURN(rc);
4522 }
4523
4524 /*
4525  * Concurrency: @dt is write locked.
4526  */
4527 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4528                          const struct lu_buf *buf, const char *name, int fl,
4529                          struct thandle *handle)
4530 {
4531         struct osd_object *obj = osd_dt_obj(dt);
4532         struct osd_device *osd = osd_obj2dev(obj);
4533         struct inode *inode = obj->oo_inode;
4534         struct osd_thread_info *info = osd_oti_get(env);
4535         int fs_flags = 0;
4536         int len;
4537         int rc;
4538
4539         ENTRY;
4540
4541         LASSERT(handle);
4542         LASSERT(buf);
4543
4544         /* version set is not real XATTR */
4545         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4546                 dt_obj_version_t *version = buf->lb_buf;
4547
4548                 /*
4549                  * for version we are just using xattr API but change inode
4550                  * field instead
4551                  */
4552                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4553
4554                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4555                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4556
4557                 LDISKFS_I(inode)->i_fs_version = *version;
4558                 /*
4559                  * Version is set after all inode operations are finished,
4560                  * so we should mark it dirty here
4561                  */
4562                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4563
4564                 RETURN(0);
4565         }
4566
4567         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4568                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4569
4570         len = buf->lb_len;
4571         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4572
4573         /*
4574          * For the OST device with 256 bytes inode size by default,
4575          * the PFID EA will be stored together with LMA EA to avoid
4576          * performance trouble. Otherwise the PFID EA can be stored
4577          * independently. LU-8998
4578          */
4579         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4580             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4581                 LASSERT(buf->lb_buf);
4582
4583                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4584                 if (fl <= 0)
4585                         RETURN(fl);
4586         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4587                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4588                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4589
4590                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4591                 if (rc)
4592                         RETURN(rc);
4593
4594                 lma->lma_incompat |= LMAI_STRIPED;
4595                 lustre_lma_swab(lma);
4596                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4597                                      sizeof(*lma), XATTR_REPLACE);
4598                 if (rc != 0)
4599                         RETURN(rc);
4600         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4601                 LASSERT(!osd->od_is_ost);
4602
4603                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4604                 if (rc)
4605                         RETURN(rc);
4606         }
4607
4608         if (fl & LU_XATTR_REPLACE)
4609                 fs_flags |= XATTR_REPLACE;
4610
4611         if (fl & LU_XATTR_CREATE)
4612                 fs_flags |= XATTR_CREATE;
4613
4614         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4615         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4616
4617         if (rc == 0 &&
4618             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4619              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4620                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4621
4622         return rc;
4623 }
4624
4625 /*
4626  * Concurrency: @dt is read locked.
4627  */
4628 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4629                           const struct lu_buf *buf)
4630 {
4631         struct osd_object *obj = osd_dt_obj(dt);
4632         struct inode *inode = obj->oo_inode;
4633         struct osd_thread_info *info = osd_oti_get(env);
4634         struct dentry *dentry = &info->oti_obj_dentry;
4635
4636         if (!dt_object_exists(dt))
4637                 return -ENOENT;
4638
4639         LASSERT(!dt_object_remote(dt));
4640         LASSERT(inode->i_op != NULL);
4641         LASSERT(inode->i_op->listxattr != NULL);
4642
4643         dentry->d_inode = inode;
4644         dentry->d_sb = inode->i_sb;
4645         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4646 }
4647
4648 static int osd_declare_xattr_del(const struct lu_env *env,
4649                                  struct dt_object *dt, const char *name,
4650                                  struct thandle *handle)
4651 {
4652         struct osd_thandle *oh;
4653         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4654
4655         LASSERT(!dt_object_remote(dt));
4656         LASSERT(handle != NULL);
4657
4658         oh = container_of0(handle, struct osd_thandle, ot_super);
4659         LASSERT(oh->ot_handle == NULL);
4660
4661         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4662                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4663         /*
4664          * xattr del may involve inode quota change, reserve credits for
4665          * dquot_initialize()
4666          */
4667         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4668
4669         return 0;
4670 }
4671
4672 /*
4673  * Concurrency: @dt is write locked.
4674  */
4675 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4676                          const char *name, struct thandle *handle)
4677 {
4678         struct osd_object *obj = osd_dt_obj(dt);
4679         struct inode *inode = obj->oo_inode;
4680         struct osd_thread_info *info = osd_oti_get(env);
4681         struct dentry *dentry = &info->oti_obj_dentry;
4682         int rc;
4683
4684         if (!dt_object_exists(dt))
4685                 return -ENOENT;
4686
4687         LASSERT(!dt_object_remote(dt));
4688         LASSERT(inode->i_op != NULL);
4689         LASSERT(handle != NULL);
4690 #ifdef HAVE_IOP_XATTR
4691         LASSERT(inode->i_op->removexattr != NULL);
4692 #endif
4693
4694         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4695
4696         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4697                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4698
4699                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4700                                  &info->oti_ost_attrs);
4701                 if (!rc) {
4702                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4703
4704                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4705                         lustre_lma_swab(lma);
4706                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4707                                              sizeof(*lma), XATTR_REPLACE);
4708                         if (!rc)
4709                                 obj->oo_pfid_in_lma = 0;
4710                 }
4711         } else {
4712                 ll_vfs_dq_init(inode);
4713                 dentry->d_inode = inode;
4714                 dentry->d_sb = inode->i_sb;
4715                 rc = osd_removexattr(dentry, inode, name);
4716         }
4717
4718         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4719
4720         if (rc == 0 &&
4721             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4722              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4723                 osd_oxc_del(obj, name);
4724
4725         return rc;
4726 }
4727
4728 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4729                            __u64 start, __u64 end)
4730 {
4731         struct osd_object *obj = osd_dt_obj(dt);
4732         struct inode *inode = obj->oo_inode;
4733         struct osd_thread_info *info = osd_oti_get(env);
4734         struct dentry *dentry = &info->oti_obj_dentry;
4735         struct file *file = &info->oti_file;
4736         int rc;
4737
4738         ENTRY;
4739
4740         dentry->d_inode = inode;
4741         dentry->d_sb = inode->i_sb;
4742         file->f_path.dentry = dentry;
4743         file->f_mapping = inode->i_mapping;
4744         file->f_op = inode->i_fop;
4745         set_file_inode(file, inode);
4746
4747         rc = ll_vfs_fsync_range(file, start, end, 0);
4748
4749         RETURN(rc);
4750 }
4751
4752 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4753 {
4754         return 0;
4755 }
4756
4757 /*
4758  * Index operations.
4759  */
4760
4761 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4762                                const struct dt_index_features *feat)
4763 {
4764         struct iam_descr *descr;
4765
4766         if (osd_object_is_root(o))
4767                 return feat == &dt_directory_features;
4768
4769         LASSERT(o->oo_dir != NULL);
4770
4771         descr = o->oo_dir->od_container.ic_descr;
4772         if (feat == &dt_directory_features) {
4773                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4774                         return 1;
4775                 else
4776                         return 0;
4777         } else {
4778                 return feat->dif_keysize_min <= descr->id_key_size &&
4779                        descr->id_key_size <= feat->dif_keysize_max &&
4780                        feat->dif_recsize_min <= descr->id_rec_size &&
4781                        descr->id_rec_size <= feat->dif_recsize_max &&
4782                        !(feat->dif_flags & (DT_IND_VARKEY |
4783                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4784                        ergo(feat->dif_flags & DT_IND_UPDATE,
4785                             1 /* XXX check that object (and fs) is writable */);
4786         }
4787 }
4788
4789 static int osd_iam_container_init(const struct lu_env *env,
4790                                   struct osd_object *obj,
4791                                   struct osd_directory *dir)
4792 {
4793         struct iam_container *bag = &dir->od_container;
4794         int result;
4795
4796         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4797         if (result != 0)
4798                 return result;
4799
4800         result = iam_container_setup(bag);
4801         if (result == 0)
4802                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4803         else
4804                 iam_container_fini(bag);
4805
4806         return result;
4807 }
4808
4809
4810 /*
4811  * Concurrency: no external locking is necessary.
4812  */
4813 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4814                          const struct dt_index_features *feat)
4815 {
4816         int result;
4817         int skip_iam = 0;
4818         struct osd_object *obj = osd_dt_obj(dt);
4819
4820         LINVRNT(osd_invariant(obj));
4821
4822         if (osd_object_is_root(obj)) {
4823                 dt->do_index_ops = &osd_index_ea_ops;
4824                 result = 0;
4825         } else if (feat == &dt_directory_features) {
4826                 dt->do_index_ops = &osd_index_ea_ops;
4827                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4828                         result = 0;
4829                 else
4830                         result = -ENOTDIR;
4831                 skip_iam = 1;
4832         } else if (unlikely(feat == &dt_otable_features)) {
4833                 dt->do_index_ops = &osd_otable_ops;
4834                 return 0;
4835         } else if (unlikely(feat == &dt_acct_features)) {
4836                 dt->do_index_ops = &osd_acct_index_ops;
4837                 result = 0;
4838                 skip_iam = 1;
4839         } else if (!osd_has_index(obj)) {
4840                 struct osd_directory *dir;
4841                 struct osd_device *osd = osd_obj2dev(obj);
4842                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4843
4844                 OBD_ALLOC_PTR(dir);
4845                 if (dir) {
4846
4847                         spin_lock(&obj->oo_guard);
4848                         if (obj->oo_dir == NULL)
4849                                 obj->oo_dir = dir;
4850                         else
4851                                 /*
4852                                  * Concurrent thread allocated container data.
4853                                  */
4854                                 OBD_FREE_PTR(dir);
4855                         spin_unlock(&obj->oo_guard);
4856                         /*
4857                          * Now, that we have container data, serialize its
4858                          * initialization.
4859                          */
4860                         down_write(&obj->oo_ext_idx_sem);
4861                         /*
4862                          * recheck under lock.
4863                          */
4864
4865                         if (osd_has_index(obj)) {
4866                                 result = 0;
4867                                 goto unlock;
4868                         }
4869
4870                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4871                         if (result || feat == &dt_lfsck_namespace_features ||
4872                             feat == &dt_lfsck_layout_orphan_features ||
4873                             feat == &dt_lfsck_layout_dangling_features)
4874                                 goto unlock;
4875
4876                         result = osd_index_register(osd, fid,
4877                                                     feat->dif_keysize_max,
4878                                                     feat->dif_recsize_max);
4879                         if (result < 0)
4880                                 CWARN("%s: failed to register index "
4881                                       DFID": rc = %d\n",
4882                                       osd_name(osd), PFID(fid), result);
4883                         else if (result > 0)
4884                                 result = 0;
4885                         else
4886                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4887                                        " (%d/%d) registered\n",
4888                                        osd_name(osd), PFID(fid),
4889                                        (int)feat->dif_keysize_max,
4890                                        (int)feat->dif_recsize_max);
4891
4892 unlock:
4893                         up_write(&obj->oo_ext_idx_sem);
4894                 } else {
4895                         result = -ENOMEM;
4896                 }
4897         } else {
4898                 result = 0;
4899         }
4900
4901         if (result == 0 && skip_iam == 0) {
4902                 if (!osd_iam_index_probe(env, obj, feat))
4903                         result = -ENOTDIR;
4904         }
4905         LINVRNT(osd_invariant(obj));
4906
4907         return result;
4908 }
4909
4910 static int osd_otable_it_attr_get(const struct lu_env *env,
4911                                  struct dt_object *dt,
4912                                  struct lu_attr *attr)
4913 {
4914         attr->la_valid = 0;
4915         return 0;
4916 }
4917
4918 static const struct dt_object_operations osd_obj_ops = {
4919         .do_read_lock           = osd_read_lock,
4920         .do_write_lock          = osd_write_lock,
4921         .do_read_unlock         = osd_read_unlock,
4922         .do_write_unlock        = osd_write_unlock,
4923         .do_write_locked        = osd_write_locked,
4924         .do_attr_get            = osd_attr_get,
4925         .do_declare_attr_set    = osd_declare_attr_set,
4926         .do_attr_set            = osd_attr_set,
4927         .do_ah_init             = osd_ah_init,
4928         .do_declare_create      = osd_declare_create,
4929         .do_create              = osd_create,
4930         .do_declare_destroy     = osd_declare_destroy,
4931         .do_destroy             = osd_destroy,
4932         .do_index_try           = osd_index_try,
4933         .do_declare_ref_add     = osd_declare_ref_add,
4934         .do_ref_add             = osd_ref_add,
4935         .do_declare_ref_del     = osd_declare_ref_del,
4936         .do_ref_del             = osd_ref_del,
4937         .do_xattr_get           = osd_xattr_get,
4938         .do_declare_xattr_set   = osd_declare_xattr_set,
4939         .do_xattr_set           = osd_xattr_set,
4940         .do_declare_xattr_del   = osd_declare_xattr_del,
4941         .do_xattr_del           = osd_xattr_del,
4942         .do_xattr_list          = osd_xattr_list,
4943         .do_object_sync         = osd_object_sync,
4944         .do_invalidate          = osd_invalidate,
4945 };
4946
4947 static const struct dt_object_operations osd_obj_otable_it_ops = {
4948         .do_attr_get    = osd_otable_it_attr_get,
4949         .do_index_try   = osd_index_try,
4950 };
4951
4952 static int osd_index_declare_iam_delete(const struct lu_env *env,
4953                                         struct dt_object *dt,
4954                                         const struct dt_key *key,
4955                                         struct thandle *handle)
4956 {
4957         struct osd_thandle *oh;
4958
4959         oh = container_of0(handle, struct osd_thandle, ot_super);
4960         LASSERT(oh->ot_handle == NULL);
4961
4962         /* Recycle  may cause additional three blocks to be changed. */
4963         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4964                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4965
4966         return 0;
4967 }
4968
4969 /**
4970  *      delete a (key, value) pair from index \a dt specified by \a key
4971  *
4972  *      \param  dt      osd index object
4973  *      \param  key     key for index
4974  *      \param  rec     record reference
4975  *      \param  handle  transaction handler
4976  *
4977  *      \retval  0  success
4978  *      \retval -ve   failure
4979  */
4980 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4981                                 const struct dt_key *key,
4982                                 struct thandle *handle)
4983 {
4984         struct osd_thread_info *oti = osd_oti_get(env);
4985         struct osd_object *obj = osd_dt_obj(dt);
4986         struct osd_thandle *oh;
4987         struct iam_path_descr *ipd;
4988         struct iam_container *bag = &obj->oo_dir->od_container;
4989         int rc;
4990
4991         ENTRY;
4992
4993         if (!dt_object_exists(dt))
4994                 RETURN(-ENOENT);
4995
4996         LINVRNT(osd_invariant(obj));
4997         LASSERT(!dt_object_remote(dt));
4998         LASSERT(bag->ic_object == obj->oo_inode);
4999         LASSERT(handle != NULL);
5000
5001         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5002
5003         ipd = osd_idx_ipd_get(env, bag);
5004         if (unlikely(ipd == NULL))
5005                 RETURN(-ENOMEM);
5006
5007         oh = container_of0(handle, struct osd_thandle, ot_super);
5008         LASSERT(oh->ot_handle != NULL);
5009         LASSERT(oh->ot_handle->h_transaction != NULL);
5010
5011         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5012                 /* swab quota uid/gid provided by caller */
5013                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5014                 key = (const struct dt_key *)&oti->oti_quota_id;
5015         }
5016
5017         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5018         osd_ipd_put(env, bag, ipd);
5019         LINVRNT(osd_invariant(obj));
5020         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5021         RETURN(rc);
5022 }
5023
5024 static int osd_index_declare_ea_delete(const struct lu_env *env,
5025                                        struct dt_object *dt,
5026                                        const struct dt_key *key,
5027                                        struct thandle *handle)
5028 {
5029         struct osd_thandle *oh;
5030         struct inode *inode;
5031         int rc, credits;
5032
5033         ENTRY;
5034
5035         LASSERT(!dt_object_remote(dt));
5036         LASSERT(handle != NULL);
5037
5038         oh = container_of0(handle, struct osd_thandle, ot_super);
5039         LASSERT(oh->ot_handle == NULL);
5040
5041         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5042         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5043
5044         inode = osd_dt_obj(dt)->oo_inode;
5045         if (inode == NULL)
5046                 RETURN(-ENOENT);
5047
5048         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5049                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5050                                    NULL, OSD_QID_BLK);
5051         RETURN(rc);
5052 }
5053
5054 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5055                                           struct dt_rec *fid)
5056 {
5057         struct osd_fid_pack *rec;
5058         int rc = -ENODATA;
5059
5060         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5061                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5062                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5063                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5064                         rc = -EINVAL;
5065         }
5066         return rc;
5067 }
5068
5069 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5070                           const struct lu_fid *fid)
5071 {
5072         struct seq_server_site *ss = osd_seq_site(osd);
5073
5074         ENTRY;
5075
5076         /* FID seqs not in FLDB, must be local seq */
5077         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5078                 RETURN(0);
5079
5080         /*
5081          * If FLD is not being initialized yet, it only happens during the
5082          * initialization, likely during mgs initialization, and we assume
5083          * this is local FID.
5084          */
5085         if (ss == NULL || ss->ss_server_fld == NULL)
5086                 RETURN(0);
5087
5088         /* Only check the local FLDB here */
5089         if (osd_seq_exists(env, osd, fid_seq(fid)))
5090                 RETURN(0);
5091
5092         RETURN(1);
5093 }
5094
5095 static void osd_take_care_of_agent(const struct lu_env *env,
5096                                    struct osd_device *osd,
5097                                    struct osd_thandle *oh,
5098                                    struct ldiskfs_dir_entry_2 *de)
5099 {
5100         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5101         struct osd_idmap_cache *idc;
5102         int rc, schedule = 0;
5103
5104         LASSERT(de != NULL);
5105
5106         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5107         if (likely(rc == 0)) {
5108                 idc = osd_idc_find_or_init(env, osd, fid);
5109                 if (IS_ERR(idc) || idc->oic_remote)
5110                         schedule = 1;
5111         } else if (rc == -ENODATA) {
5112                 /*
5113                  * can't get FID, postpone to the end of the
5114                  * transaction when iget() is safe
5115                  */
5116                 schedule = 1;
5117         } else {
5118                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5119         }
5120         if (schedule)
5121                 osd_schedule_agent_inode_removal(env, oh,
5122                                                  le32_to_cpu(de->inode));
5123 }
5124
5125 /**
5126  * Index delete function for interoperability mode (b11826).
5127  * It will remove the directory entry added by osd_index_ea_insert().
5128  * This entry is needed to maintain name->fid mapping.
5129  *
5130  * \param key,  key i.e. file entry to be deleted
5131  *
5132  * \retval   0, on success
5133  * \retval -ve, on error
5134  */
5135 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5136                                const struct dt_key *key, struct thandle *handle)
5137 {
5138         struct osd_object *obj = osd_dt_obj(dt);
5139         struct inode *dir = obj->oo_inode;
5140         struct dentry *dentry;
5141         struct osd_thandle *oh;
5142         struct ldiskfs_dir_entry_2 *de = NULL;
5143         struct buffer_head *bh;
5144         struct htree_lock *hlock = NULL;
5145         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5146         int rc;
5147
5148         ENTRY;
5149
5150         if (!dt_object_exists(dt))
5151                 RETURN(-ENOENT);
5152
5153         LINVRNT(osd_invariant(obj));
5154         LASSERT(!dt_object_remote(dt));
5155         LASSERT(handle != NULL);
5156
5157         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5158
5159         oh = container_of(handle, struct osd_thandle, ot_super);
5160         LASSERT(oh->ot_handle != NULL);
5161         LASSERT(oh->ot_handle->h_transaction != NULL);
5162
5163         ll_vfs_dq_init(dir);
5164         dentry = osd_child_dentry_get(env, obj,
5165                                       (char *)key, strlen((char *)key));
5166
5167         if (obj->oo_hl_head != NULL) {
5168                 hlock = osd_oti_get(env)->oti_hlock;
5169                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5170                                    dir, LDISKFS_HLOCK_DEL);
5171         } else {
5172                 down_write(&obj->oo_ext_idx_sem);
5173         }
5174
5175         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5176         if (!IS_ERR(bh)) {
5177                 /*
5178                  * If this is not the ".." entry, it might be a remote DNE
5179                  * entry and  we need to check if the FID is for a remote
5180                  * MDT.  If the FID is  not in the directory entry (e.g.
5181                  * upgraded 1.8 filesystem without dirdata enabled) then
5182                  * we need to get the FID from the LMA. For a remote directory
5183                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5184                  * case.
5185                  *
5186                  * Delete the entry before the agent inode in order to
5187                  * simplify error handling.  At worst an error after deleting
5188                  * the entry first might leak the agent inode afterward. The
5189                  * reverse would need filesystem abort in case of error deleting
5190                  * the entry after the agent had been removed, or leave a
5191                  * dangling entry pointing at a random inode.
5192                  */
5193                 if (strcmp((char *)key, dotdot) != 0)
5194                         osd_take_care_of_agent(env, osd, oh, de);
5195                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5196                 brelse(bh);
5197         } else {
5198                 rc = PTR_ERR(bh);
5199         }
5200         if (hlock != NULL)
5201                 ldiskfs_htree_unlock(hlock);
5202         else
5203                 up_write(&obj->oo_ext_idx_sem);
5204
5205         GOTO(out, rc);
5206 out:
5207         LASSERT(osd_invariant(obj));
5208         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5209         RETURN(rc);
5210 }
5211
5212 /**
5213  *      Lookup index for \a key and copy record to \a rec.
5214  *
5215  *      \param  dt      osd index object
5216  *      \param  key     key for index
5217  *      \param  rec     record reference
5218  *
5219  *      \retval  +ve  success : exact mach
5220  *      \retval  0    return record with key not greater than \a key
5221  *      \retval -ve   failure
5222  */
5223 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5224                                 struct dt_rec *rec, const struct dt_key *key)
5225 {
5226         struct osd_object *obj = osd_dt_obj(dt);
5227         struct iam_path_descr *ipd;
5228         struct iam_container *bag = &obj->oo_dir->od_container;
5229         struct osd_thread_info *oti = osd_oti_get(env);
5230         struct iam_iterator *it = &oti->oti_idx_it;
5231         struct iam_rec *iam_rec;
5232         int rc;
5233
5234         ENTRY;
5235
5236         if (!dt_object_exists(dt))
5237                 RETURN(-ENOENT);
5238
5239         LASSERT(osd_invariant(obj));
5240         LASSERT(!dt_object_remote(dt));
5241         LASSERT(bag->ic_object == obj->oo_inode);
5242
5243         ipd = osd_idx_ipd_get(env, bag);
5244         if (IS_ERR(ipd))
5245                 RETURN(-ENOMEM);
5246
5247         /* got ipd now we can start iterator. */
5248         iam_it_init(it, bag, 0, ipd);
5249
5250         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5251                 /* swab quota uid/gid provided by caller */
5252                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5253                 key = (const struct dt_key *)&oti->oti_quota_id;
5254         }
5255
5256         rc = iam_it_get(it, (struct iam_key *)key);
5257         if (rc >= 0) {
5258                 if (S_ISDIR(obj->oo_inode->i_mode))
5259                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5260                 else
5261                         iam_rec = (struct iam_rec *)rec;
5262
5263                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5264
5265                 if (S_ISDIR(obj->oo_inode->i_mode))
5266                         osd_fid_unpack((struct lu_fid *)rec,
5267                                        (struct osd_fid_pack *)iam_rec);
5268                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5269                         osd_quota_unpack(obj, rec);
5270         }
5271
5272         iam_it_put(it);
5273         iam_it_fini(it);
5274         osd_ipd_put(env, bag, ipd);
5275
5276         LINVRNT(osd_invariant(obj));
5277
5278         RETURN(rc);
5279 }
5280
5281 static int osd_index_declare_iam_insert(const struct lu_env *env,
5282                                         struct dt_object *dt,
5283                                         const struct dt_rec *rec,
5284                                         const struct dt_key *key,
5285                                         struct thandle *handle)
5286 {
5287         struct osd_thandle *oh;
5288
5289         LASSERT(handle != NULL);
5290
5291         oh = container_of0(handle, struct osd_thandle, ot_super);
5292         LASSERT(oh->ot_handle == NULL);
5293
5294         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5295                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5296
5297         return 0;
5298 }
5299
5300 /**
5301  *      Inserts (key, value) pair in \a dt index object.
5302  *
5303  *      \param  dt      osd index object
5304  *      \param  key     key for index
5305  *      \param  rec     record reference
5306  *      \param  th      transaction handler
5307  *
5308  *      \retval  0  success
5309  *      \retval -ve failure
5310  */
5311 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5312                                 const struct dt_rec *rec,
5313                                 const struct dt_key *key, struct thandle *th)
5314 {
5315         struct osd_object *obj = osd_dt_obj(dt);
5316         struct iam_path_descr *ipd;
5317         struct osd_thandle *oh;
5318         struct iam_container *bag;
5319         struct osd_thread_info *oti = osd_oti_get(env);
5320         struct iam_rec *iam_rec;
5321         int rc;
5322
5323         ENTRY;
5324
5325         if (!dt_object_exists(dt))
5326                 RETURN(-ENOENT);
5327
5328         LINVRNT(osd_invariant(obj));
5329         LASSERT(!dt_object_remote(dt));
5330
5331         bag = &obj->oo_dir->od_container;
5332         LASSERT(bag->ic_object == obj->oo_inode);
5333         LASSERT(th != NULL);
5334
5335         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5336
5337         ipd = osd_idx_ipd_get(env, bag);
5338         if (unlikely(ipd == NULL))
5339                 RETURN(-ENOMEM);
5340
5341         oh = container_of0(th, struct osd_thandle, ot_super);
5342         LASSERT(oh->ot_handle != NULL);
5343         LASSERT(oh->ot_handle->h_transaction != NULL);
5344         if (S_ISDIR(obj->oo_inode->i_mode)) {
5345                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5346                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5347                              &oti->oti_fid);
5348         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5349                 /* pack quota uid/gid */
5350                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5351                 key = (const struct dt_key *)&oti->oti_quota_id;
5352                 /* pack quota record */
5353                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5354                 iam_rec = (struct iam_rec *)rec;
5355         } else {
5356                 iam_rec = (struct iam_rec *)rec;
5357         }
5358
5359         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5360                         iam_rec, ipd);
5361         osd_ipd_put(env, bag, ipd);
5362         LINVRNT(osd_invariant(obj));
5363         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5364         RETURN(rc);
5365 }
5366
5367 /**
5368  * Calls ldiskfs_add_entry() to add directory entry
5369  * into the directory. This is required for
5370  * interoperability mode (b11826)
5371  *
5372  * \retval   0, on success
5373  * \retval -ve, on error
5374  */
5375 static int __osd_ea_add_rec(struct osd_thread_info *info,
5376                             struct osd_object *pobj, struct inode  *cinode,
5377                             const char *name, const struct lu_fid *fid,
5378                             struct htree_lock *hlock, struct thandle *th)
5379 {
5380         struct ldiskfs_dentry_param *ldp;
5381         struct dentry *child;
5382         struct osd_thandle *oth;
5383         int rc;
5384
5385         oth = container_of(th, struct osd_thandle, ot_super);
5386         LASSERT(oth->ot_handle != NULL);
5387         LASSERT(oth->ot_handle->h_transaction != NULL);
5388         LASSERT(pobj->oo_inode);
5389
5390         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5391         if (unlikely(pobj->oo_inode ==
5392                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5393                 ldp->edp_magic = 0;
5394         else
5395                 osd_get_ldiskfs_dirent_param(ldp, fid);
5396         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5397         child->d_fsdata = (void *)ldp;
5398         ll_vfs_dq_init(pobj->oo_inode);
5399         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5400                                    child, cinode, hlock);
5401         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5402                 struct ldiskfs_dir_entry_2      *de;
5403                 struct buffer_head              *bh;
5404                 int                              rc1;
5405
5406                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5407                                             NULL, hlock);
5408                 if (!IS_ERR(bh)) {
5409                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5410                                                                bh);
5411                         if (rc1 == 0) {
5412                                 if (S_ISDIR(cinode->i_mode))
5413                                         de->file_type = LDISKFS_DIRENT_LUFID |
5414                                                         LDISKFS_FT_REG_FILE;
5415                                 else
5416                                         de->file_type = LDISKFS_DIRENT_LUFID |
5417                                                         LDISKFS_FT_DIR;
5418                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5419                                                               NULL, bh);
5420                         }
5421                         brelse(bh);
5422                 }
5423         }
5424
5425         RETURN(rc);
5426 }
5427
5428 /**
5429  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5430  * into the directory.Also sets flags into osd object to
5431  * indicate dot and dotdot are created. This is required for
5432  * interoperability mode (b11826)
5433  *
5434  * \param dir   directory for dot and dotdot fixup.
5435  * \param obj   child object for linking
5436  *
5437  * \retval   0, on success
5438  * \retval -ve, on error
5439  */
5440 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5441                               struct osd_object *dir,
5442                               struct inode *parent_dir, const char *name,
5443                               const struct lu_fid *dot_fid,
5444                               const struct lu_fid *dot_dot_fid,
5445                               struct thandle *th)
5446 {
5447         struct inode *inode = dir->oo_inode;
5448         struct osd_thandle *oth;
5449         int result = 0;
5450
5451         oth = container_of(th, struct osd_thandle, ot_super);
5452         LASSERT(oth->ot_handle->h_transaction != NULL);
5453         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5454
5455         if (strcmp(name, dot) == 0) {
5456                 if (dir->oo_compat_dot_created) {
5457                         result = -EEXIST;
5458                 } else {
5459                         LASSERT(inode->i_ino == parent_dir->i_ino);
5460                         dir->oo_compat_dot_created = 1;
5461                         result = 0;
5462                 }
5463         } else if (strcmp(name, dotdot) == 0) {
5464                 if (!dir->oo_compat_dot_created)
5465                         return -EINVAL;
5466                 /* in case of rename, dotdot is already created */
5467                 if (dir->oo_compat_dotdot_created) {
5468                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5469                                                 dot_dot_fid, NULL, th);
5470                 }
5471
5472                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5473                         struct lu_fid tfid = *dot_dot_fid;
5474
5475                         tfid.f_oid--;
5476                         result = osd_add_dot_dotdot_internal(info,
5477                                         dir->oo_inode, parent_dir, dot_fid,
5478                                         &tfid, oth);
5479                 } else {
5480                         result = osd_add_dot_dotdot_internal(info,
5481                                         dir->oo_inode, parent_dir, dot_fid,
5482                                         dot_dot_fid, oth);
5483                 }
5484
5485                 if (result == 0)
5486                         dir->oo_compat_dotdot_created = 1;
5487         }
5488
5489         return result;
5490 }
5491
5492
5493 /**
5494  * It will call the appropriate osd_add* function and return the
5495  * value, return by respective functions.
5496  */
5497 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5498                           struct inode *cinode, const char *name,
5499                           const struct lu_fid *fid, struct thandle *th)
5500 {
5501         struct osd_thread_info *info = osd_oti_get(env);
5502         struct htree_lock *hlock;
5503         int rc;
5504
5505         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5506
5507         if (name[0] == '.' && (name[1] == '\0' ||
5508                                (name[1] == '.' && name[2] == '\0'))) {
5509                 if (hlock != NULL) {
5510                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5511                                            pobj->oo_inode, 0);
5512                 } else {
5513                         down_write(&pobj->oo_ext_idx_sem);
5514                 }
5515
5516                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5517                                         lu_object_fid(&pobj->oo_dt.do_lu),
5518                                         fid, th);
5519         } else {
5520                 if (hlock != NULL) {
5521                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5522                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5523                 } else {
5524                         down_write(&pobj->oo_ext_idx_sem);
5525                 }
5526
5527                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5528                         struct lu_fid *tfid = &info->oti_fid;
5529
5530                         *tfid = *fid;
5531                         tfid->f_ver = ~0;
5532                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5533                                               tfid, hlock, th);
5534                 } else {
5535                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5536                                               hlock, th);
5537                 }
5538         }
5539         if (hlock != NULL)
5540                 ldiskfs_htree_unlock(hlock);
5541         else
5542                 up_write(&pobj->oo_ext_idx_sem);
5543
5544         return rc;
5545 }
5546
5547 static int
5548 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5549                       struct osd_idmap_cache *oic)
5550 {
5551         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5552         struct lu_fid *fid = &oic->oic_fid;
5553         struct osd_inode_id *id = &oic->oic_lid;
5554         struct inode *inode = NULL;
5555         int once = 0;
5556         bool insert;
5557         int rc;
5558
5559         ENTRY;
5560
5561         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5562                 RETURN(0);
5563
5564         if (thread_is_running(&scrub->os_thread) &&
5565             scrub->os_pos_current > id->oii_ino)
5566                 RETURN(0);
5567
5568         if (dev->od_auto_scrub_interval == AS_NEVER ||
5569             ktime_get_real_seconds() <
5570             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5571                 RETURN(0);
5572
5573 again:
5574         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5575         if (rc == -ENOENT) {
5576                 __u32 gen = id->oii_gen;
5577
5578                 insert = true;
5579                 if (inode != NULL)
5580                         goto trigger;
5581
5582                 inode = osd_iget(oti, dev, id);
5583                 /* The inode has been removed (by race maybe). */
5584                 if (IS_ERR(inode)) {
5585                         rc = PTR_ERR(inode);
5586
5587                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5588                 }
5589
5590                 /* The OI mapping is lost. */
5591                 if (gen != OSD_OII_NOGEN)
5592                         goto trigger;
5593
5594                 /*
5595                  * The inode may has been reused by others, we do not know,
5596                  * leave it to be handled by subsequent osd_fid_lookup().
5597                  */
5598                 GOTO(out, rc = 0);
5599         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5600                 GOTO(out, rc);
5601         }
5602
5603         insert = false;
5604
5605 trigger:
5606         if (thread_is_running(&scrub->os_thread)) {
5607                 if (inode == NULL) {
5608                         inode = osd_iget(oti, dev, id);
5609                         /* The inode has been removed (by race maybe). */
5610                         if (IS_ERR(inode)) {
5611                                 rc = PTR_ERR(inode);
5612
5613                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5614                         }
5615                 }
5616
5617                 rc = osd_oii_insert(dev, oic, insert);
5618                 /*
5619                  * There is race condition between osd_oi_lookup and OI scrub.
5620                  * The OI scrub finished just after osd_oi_lookup() failure.
5621                  * Under such case, it is unnecessary to trigger OI scrub again,
5622                  * but try to call osd_oi_lookup() again.
5623                  */
5624                 if (unlikely(rc == -EAGAIN))
5625                         goto again;
5626
5627                 if (!S_ISDIR(inode->i_mode))
5628                         rc = 0;
5629                 else
5630                         rc = osd_check_lmv(oti, dev, inode, oic);
5631
5632                 GOTO(out, rc);
5633         }
5634
5635         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5636                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5637                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5638                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5639                        "%s: trigger partial OI scrub for RPC inconsistency "
5640                        "checking FID "DFID": rc = %d\n",
5641                        osd_dev2name(dev), PFID(fid), rc);
5642                 if (rc == 0 || rc == -EALREADY)
5643                         goto again;
5644         }
5645
5646         GOTO(out, rc);
5647
5648 out:
5649         if (inode)
5650                 iput(inode);
5651
5652         RETURN(rc);
5653 }
5654
5655 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5656                                struct osd_device *dev,
5657                                struct osd_idmap_cache *oic,
5658                                struct lu_fid *fid, __u32 ino)
5659 {
5660         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5661         struct inode *inode;
5662         int rc;
5663
5664         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5665         inode = osd_iget(oti, dev, &oic->oic_lid);
5666         if (IS_ERR(inode)) {
5667                 fid_zero(&oic->oic_fid);
5668                 return PTR_ERR(inode);
5669         }
5670
5671         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5672         iput(inode);
5673         if (rc != 0)
5674                 fid_zero(&oic->oic_fid);
5675         else
5676                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5677         return rc;
5678 }
5679
5680 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5681                       struct osd_inode_id *id, const struct lu_fid *fid)
5682 {
5683         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5684                id->oii_ino, id->oii_gen, info);
5685         info->oti_cache.oic_lid = *id;
5686         info->oti_cache.oic_fid = *fid;
5687         info->oti_cache.oic_dev = osd;
5688 }
5689
5690 /**
5691  * Get parent FID from the linkEA.
5692  *
5693  * For a directory which parent resides on remote MDT, to satisfy the
5694  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5695  * the other hand, to make the lookup(..) on the directory can return
5696  * the real parent FID, we append the real parent FID after its ".."
5697  * name entry in the /REMOTE_PARENT_DIR.
5698  *
5699  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5700  * backup. So after the restore, we cannot get the right parent FID from
5701  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5702  * we have stored the real parent FID in the directory object's linkEA,
5703  * we can parse the linkEA for the real parent FID.
5704  *
5705  * \param[in] env       pointer to the thread context
5706  * \param[in] obj       pointer to the object to be handled
5707  * \param[out]fid       pointer to the buffer to hold the parent FID
5708  *
5709  * \retval              0 for getting the real parent FID successfully
5710  * \retval              negative error number on failure
5711  */
5712 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5713                                     struct osd_object *obj,
5714                                     struct lu_fid *fid)
5715 {
5716         struct osd_thread_info *oti = osd_oti_get(env);
5717         struct lu_buf *buf = &oti->oti_big_buf;
5718         struct dentry *dentry = &oti->oti_obj_dentry;
5719         struct inode *inode = obj->oo_inode;
5720         struct linkea_data ldata = { NULL };
5721         int rc;
5722
5723         ENTRY;
5724
5725         fid_zero(fid);
5726         if (!S_ISDIR(inode->i_mode))
5727                 RETURN(-EIO);
5728
5729 again:
5730         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5731                              buf->lb_buf, buf->lb_len);
5732         if (rc == -ERANGE) {
5733                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5734                                      NULL, 0);
5735                 if (rc > 0) {
5736                         lu_buf_realloc(buf, rc);
5737                         if (buf->lb_buf == NULL)
5738                                 RETURN(-ENOMEM);
5739
5740                         goto again;
5741                 }
5742         }
5743
5744         if (unlikely(rc == 0))
5745                 RETURN(-ENODATA);
5746
5747         if (rc < 0)
5748                 RETURN(rc);
5749
5750         if (unlikely(buf->lb_buf == NULL)) {
5751                 lu_buf_realloc(buf, rc);
5752                 if (buf->lb_buf == NULL)
5753                         RETURN(-ENOMEM);
5754
5755                 goto again;
5756         }
5757
5758         ldata.ld_buf = buf;
5759         rc = linkea_init_with_rec(&ldata);
5760         if (!rc) {
5761                 linkea_first_entry(&ldata);
5762                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5763         }
5764
5765         RETURN(rc);
5766 }
5767
5768 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5769                                     struct inode *inode,
5770                                     const struct lu_fid *pfid,
5771                                     const char *name, const int namelen)
5772 {
5773         struct osd_thread_info *oti = osd_oti_get(env);
5774         struct lu_buf *buf = &oti->oti_big_buf;
5775         struct dentry *dentry = &oti->oti_obj_dentry;
5776         struct linkea_data ldata = { NULL };
5777         struct lu_name cname = { .ln_name = name,
5778                                  .ln_namelen = namelen };
5779         int rc;
5780
5781         ENTRY;
5782
5783 again:
5784         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5785                              buf->lb_buf, buf->lb_len);
5786         if (rc == -ERANGE)
5787                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5788
5789         if (rc < 0)
5790                 RETURN(rc);
5791
5792         if (unlikely(rc == 0))
5793                 RETURN(-ENODATA);
5794
5795         if (buf->lb_len < rc) {
5796                 lu_buf_realloc(buf, rc);
5797                 if (buf->lb_buf == NULL)
5798                         RETURN(-ENOMEM);
5799
5800                 goto again;
5801         }
5802
5803         ldata.ld_buf = buf;
5804         rc = linkea_init_with_rec(&ldata);
5805         if (!rc)
5806                 rc = linkea_links_find(&ldata, &cname, pfid);
5807
5808         RETURN(rc);
5809 }
5810
5811 /**
5812  * Calls ->lookup() to find dentry. From dentry get inode and
5813  * read inode's ea to get fid. This is required for  interoperability
5814  * mode (b11826)
5815  *
5816  * \retval   0, on success
5817  * \retval -ve, on error
5818  */
5819 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5820                              struct dt_rec *rec, const struct dt_key *key)
5821 {
5822         struct inode *dir = obj->oo_inode;
5823         struct dentry *dentry;
5824         struct ldiskfs_dir_entry_2 *de;
5825         struct buffer_head *bh;
5826         struct lu_fid *fid = (struct lu_fid *)rec;
5827         struct htree_lock *hlock = NULL;
5828         int ino;
5829         int rc;
5830
5831         ENTRY;
5832
5833         LASSERT(dir->i_op != NULL);
5834         LASSERT(dir->i_op->lookup != NULL);
5835
5836         dentry = osd_child_dentry_get(env, obj,
5837                                       (char *)key, strlen((char *)key));
5838
5839         if (obj->oo_hl_head != NULL) {
5840                 hlock = osd_oti_get(env)->oti_hlock;
5841                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5842                                    dir, LDISKFS_HLOCK_LOOKUP);
5843         } else {
5844                 down_read(&obj->oo_ext_idx_sem);
5845         }
5846
5847         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5848         if (!IS_ERR(bh)) {
5849                 struct osd_thread_info *oti = osd_oti_get(env);
5850                 struct osd_inode_id *id = &oti->oti_id;
5851                 struct osd_idmap_cache *oic = &oti->oti_cache;
5852                 struct osd_device *dev = osd_obj2dev(obj);
5853
5854                 ino = le32_to_cpu(de->inode);
5855                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5856                         brelse(bh);
5857                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5858                         GOTO(out, rc);
5859                 }
5860
5861                 rc = osd_get_fid_from_dentry(de, rec);
5862
5863                 /* done with de, release bh */
5864                 brelse(bh);
5865                 if (rc != 0) {
5866                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5867                                 const char *name = (const char *)key;
5868
5869                                 /*
5870                                  * If the parent is on remote MDT, and there
5871                                  * is no FID-in-dirent, then we have to get
5872                                  * the parent FID from the linkEA.
5873                                  */
5874                                 if (likely(strlen(name) == 2 &&
5875                                            name[0] == '.' && name[1] == '.'))
5876                                         rc = osd_get_pfid_from_linkea(env, obj,
5877                                                                       fid);
5878                         } else {
5879                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5880                         }
5881                 } else {
5882                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5883                 }
5884
5885                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5886                         fid_zero(&oic->oic_fid);
5887
5888                         GOTO(out, rc);
5889                 }
5890
5891                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5892                 rc = osd_consistency_check(oti, dev, oic);
5893                 if (rc == -ENOENT)
5894                         fid_zero(&oic->oic_fid);
5895                 else
5896                         /* Other error should not affect lookup result. */
5897                         rc = 0;
5898         } else {
5899                 rc = PTR_ERR(bh);
5900         }
5901
5902         GOTO(out, rc);
5903
5904 out:
5905         if (hlock != NULL)
5906                 ldiskfs_htree_unlock(hlock);
5907         else
5908                 up_read(&obj->oo_ext_idx_sem);
5909         return rc;
5910 }
5911
5912 static int osd_index_declare_ea_insert(const struct lu_env *env,
5913                                        struct dt_object *dt,
5914                                        const struct dt_rec *rec,
5915                                        const struct dt_key *key,
5916                                        struct thandle *handle)
5917 {
5918         struct osd_thandle *oh;
5919         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5920         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5921         const struct lu_fid *fid = rec1->rec_fid;
5922         int credits, rc = 0;
5923         struct osd_idmap_cache *idc;
5924
5925         ENTRY;
5926
5927         LASSERT(!dt_object_remote(dt));
5928         LASSERT(handle != NULL);
5929         LASSERT(fid != NULL);
5930         LASSERT(rec1->rec_type != 0);
5931
5932         oh = container_of0(handle, struct osd_thandle, ot_super);
5933         LASSERT(oh->ot_handle == NULL);
5934
5935         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5936
5937         /*
5938          * we can't call iget() while a transactions is running
5939          * (this can lead to a deadlock), but we need to know
5940          * inum and object type. so we find this information at
5941          * declaration and cache in per-thread info
5942          */
5943         idc = osd_idc_find_or_init(env, osd, fid);
5944         if (IS_ERR(idc))
5945                 RETURN(PTR_ERR(idc));
5946         if (idc->oic_remote) {
5947                 /*
5948                  * a reference to remote inode is represented by an
5949                  * agent inode which we have to create
5950                  */
5951                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5952                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5953         }
5954
5955         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5956
5957         if (osd_dt_obj(dt)->oo_inode != NULL) {
5958                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5959
5960                 /*
5961                  * We ignore block quota on meta pool (MDTs), so needn't
5962                  * calculate how many blocks will be consumed by this index
5963                  * insert
5964                  */
5965                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5966                                            i_gid_read(inode),
5967                                            i_projid_read(inode), 0,
5968                                            oh, osd_dt_obj(dt), NULL,
5969                                            OSD_QID_BLK);
5970                 if (rc)
5971                         RETURN(rc);
5972
5973 #ifdef HAVE_PROJECT_QUOTA
5974                 /*
5975                  * Reserve credits for local agent inode to transfer
5976                  * to 0, quota enforcement is ignored in this case.
5977                  */
5978                 if (idc->oic_remote &&
5979                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5980                     i_projid_read(inode) != 0)
5981                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5982                                                   0, i_projid_read(inode),
5983                                                   0, false, PRJQUOTA, true);
5984 #endif
5985         }
5986
5987         RETURN(rc);
5988 }
5989
5990 /**
5991  * Index add function for interoperability mode (b11826).
5992  * It will add the directory entry.This entry is needed to
5993  * maintain name->fid mapping.
5994  *
5995  * \param key it is key i.e. file entry to be inserted
5996  * \param rec it is value of given key i.e. fid
5997  *
5998  * \retval   0, on success
5999  * \retval -ve, on error
6000  */
6001 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6002                                const struct dt_rec *rec,
6003                                const struct dt_key *key, struct thandle *th)
6004 {
6005         struct osd_object *obj = osd_dt_obj(dt);
6006         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6007         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6008         const struct lu_fid *fid = rec1->rec_fid;
6009         const char *name = (const char *)key;
6010         struct osd_thread_info *oti = osd_oti_get(env);
6011         struct inode *child_inode = NULL;
6012         struct osd_idmap_cache *idc;
6013         int rc;
6014
6015         ENTRY;
6016
6017         if (!dt_object_exists(dt))
6018                 RETURN(-ENOENT);
6019
6020         LASSERT(osd_invariant(obj));
6021         LASSERT(!dt_object_remote(dt));
6022         LASSERT(th != NULL);
6023
6024         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6025
6026         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6027
6028         idc = osd_idc_find(env, osd, fid);
6029         if (unlikely(idc == NULL)) {
6030                 idc = osd_idc_find_or_init(env, osd, fid);
6031                 if (IS_ERR(idc)) {
6032                         /*
6033                          * this dt_insert() wasn't declared properly, so
6034                          * FID is missing in OI cache. we better do not
6035                          * lookup FID in FLDB/OI and don't risk to deadlock,
6036                          * but in some special cases (lfsck testing, etc)
6037                          * it's much simpler than fixing a caller.
6038                          *
6039                          * normally this error should be placed after the first
6040                          * find, but migrate may attach source stripes to
6041                          * target, which doesn't create stripes.
6042                          */
6043                         CERROR("%s: "DFID" wasn't declared for insert\n",
6044                                osd_name(osd), PFID(fid));
6045                         dump_stack();
6046                         RETURN(PTR_ERR(idc));
6047                 }
6048         }
6049
6050         if (idc->oic_remote) {
6051                 /* Insert remote entry */
6052                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6053                         child_inode =
6054                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6055                 } else {
6056                         child_inode = osd_create_local_agent_inode(env, osd,
6057                                         obj, fid, rec1->rec_type & S_IFMT, th);
6058                         if (IS_ERR(child_inode))
6059                                 RETURN(PTR_ERR(child_inode));
6060                 }
6061         } else {
6062                 /* Insert local entry */
6063                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6064                         /* for a reason OI cache wasn't filled properly */
6065                         CERROR("%s: OIC for "DFID" isn't filled\n",
6066                                osd_name(osd), PFID(fid));
6067                         RETURN(-EINVAL);
6068                 }
6069                 child_inode = oti->oti_inode;
6070                 if (unlikely(child_inode == NULL)) {
6071                         struct ldiskfs_inode_info *lii;
6072
6073                         OBD_ALLOC_PTR(lii);
6074                         if (lii == NULL)
6075                                 RETURN(-ENOMEM);
6076                         child_inode = oti->oti_inode = &lii->vfs_inode;
6077                 }
6078                 child_inode->i_sb = osd_sb(osd);
6079                 child_inode->i_ino = idc->oic_lid.oii_ino;
6080                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6081         }
6082
6083         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6084
6085         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6086                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6087
6088         if (child_inode && child_inode != oti->oti_inode)
6089                 iput(child_inode);
6090         LASSERT(osd_invariant(obj));
6091         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6092         RETURN(rc);
6093 }
6094
6095 /**
6096  *  Initialize osd Iterator for given osd index object.
6097  *
6098  *  \param  dt      osd index object
6099  */
6100
6101 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6102                                      struct dt_object *dt,
6103                                      __u32 unused)
6104 {
6105         struct osd_it_iam *it;
6106         struct osd_object *obj = osd_dt_obj(dt);
6107         struct lu_object *lo = &dt->do_lu;
6108         struct iam_path_descr *ipd;
6109         struct iam_container *bag = &obj->oo_dir->od_container;
6110
6111         if (!dt_object_exists(dt))
6112                 return ERR_PTR(-ENOENT);
6113
6114         OBD_ALLOC_PTR(it);
6115         if (it == NULL)
6116                 return ERR_PTR(-ENOMEM);
6117
6118         ipd = osd_it_ipd_get(env, bag);
6119         if (likely(ipd != NULL)) {
6120                 it->oi_obj = obj;
6121                 it->oi_ipd = ipd;
6122                 lu_object_get(lo);
6123                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6124                 return (struct dt_it *)it;
6125         } else {
6126                 OBD_FREE_PTR(it);
6127                 return ERR_PTR(-ENOMEM);
6128         }
6129 }
6130
6131 /**
6132  * free given Iterator.
6133  */
6134 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6135 {
6136         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6137         struct osd_object *obj = it->oi_obj;
6138
6139         iam_it_fini(&it->oi_it);
6140         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6141         osd_object_put(env, obj);
6142         OBD_FREE_PTR(it);
6143 }
6144
6145 /**
6146  *  Move Iterator to record specified by \a key
6147  *
6148  *  \param  di      osd iterator
6149  *  \param  key     key for index
6150  *
6151  *  \retval +ve  di points to record with least key not larger than key
6152  *  \retval  0   di points to exact matched key
6153  *  \retval -ve  failure
6154  */
6155
6156 static int osd_it_iam_get(const struct lu_env *env,
6157                           struct dt_it *di, const struct dt_key *key)
6158 {
6159         struct osd_thread_info *oti = osd_oti_get(env);
6160         struct osd_it_iam *it = (struct osd_it_iam *)di;
6161
6162         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6163                 /* swab quota uid/gid */
6164                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6165                 key = (struct dt_key *)&oti->oti_quota_id;
6166         }
6167
6168         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6169 }
6170
6171 /**
6172  *  Release Iterator
6173  *
6174  *  \param  di      osd iterator
6175  */
6176 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6177 {
6178         struct osd_it_iam *it = (struct osd_it_iam *)di;
6179
6180         iam_it_put(&it->oi_it);
6181 }
6182
6183 /**
6184  *  Move iterator by one record
6185  *
6186  *  \param  di      osd iterator
6187  *
6188  *  \retval +1   end of container reached
6189  *  \retval  0   success
6190  *  \retval -ve  failure
6191  */
6192
6193 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6194 {
6195         struct osd_it_iam *it = (struct osd_it_iam *)di;
6196
6197         return iam_it_next(&it->oi_it);
6198 }
6199
6200 /**
6201  * Return pointer to the key under iterator.
6202  */
6203
6204 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6205                                      const struct dt_it *di)
6206 {
6207         struct osd_thread_info *oti = osd_oti_get(env);
6208         struct osd_it_iam *it = (struct osd_it_iam *)di;
6209         struct osd_object *obj = it->oi_obj;
6210         struct dt_key *key;
6211
6212         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6213
6214         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6215                 /* swab quota uid/gid */
6216                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6217                 key = (struct dt_key *)&oti->oti_quota_id;
6218         }
6219
6220         return key;
6221 }
6222
6223 /**
6224  * Return size of key under iterator (in bytes)
6225  */
6226
6227 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6228 {
6229         struct osd_it_iam *it = (struct osd_it_iam *)di;
6230
6231         return iam_it_key_size(&it->oi_it);
6232 }
6233
6234 static inline void
6235 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6236 {
6237         /* check if file type is required */
6238         if (ent->lde_attrs & LUDA_TYPE) {
6239                 struct luda_type *lt;
6240                 int align = sizeof(*lt) - 1;
6241
6242                 len = (len + align) & ~align;
6243                 lt = (struct luda_type *)(ent->lde_name + len);
6244                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6245         }
6246
6247         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6248 }
6249
6250 /**
6251  * build lu direct from backend fs dirent.
6252  */
6253
6254 static inline void
6255 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6256                    char *name, __u16 namelen, __u16 type, __u32 attr)
6257 {
6258         ent->lde_attrs = attr | LUDA_FID;
6259         fid_cpu_to_le(&ent->lde_fid, fid);
6260
6261         ent->lde_hash = cpu_to_le64(offset);
6262         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6263
6264         strncpy(ent->lde_name, name, namelen);
6265         ent->lde_name[namelen] = '\0';
6266         ent->lde_namelen = cpu_to_le16(namelen);
6267
6268         /* append lustre attributes */
6269         osd_it_append_attrs(ent, namelen, type);
6270 }
6271
6272 /**
6273  * Return pointer to the record under iterator.
6274  */
6275 static int osd_it_iam_rec(const struct lu_env *env,
6276                           const struct dt_it *di,
6277                           struct dt_rec *dtrec, __u32 attr)
6278 {
6279         struct osd_it_iam *it = (struct osd_it_iam *)di;
6280         struct osd_thread_info *info = osd_oti_get(env);
6281
6282         ENTRY;
6283
6284         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6285                 const struct osd_fid_pack *rec;
6286                 struct lu_fid *fid = &info->oti_fid;
6287                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6288                 char *name;
6289                 int namelen;
6290                 __u64 hash;
6291                 int rc;
6292
6293                 name = (char *)iam_it_key_get(&it->oi_it);
6294                 if (IS_ERR(name))
6295                         RETURN(PTR_ERR(name));
6296
6297                 namelen = iam_it_key_size(&it->oi_it);
6298
6299                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6300                 if (IS_ERR(rec))
6301                         RETURN(PTR_ERR(rec));
6302
6303                 rc = osd_fid_unpack(fid, rec);
6304                 if (rc)
6305                         RETURN(rc);
6306
6307                 hash = iam_it_store(&it->oi_it);
6308
6309                 /* IAM does not store object type in IAM index (dir) */
6310                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6311                                    0, LUDA_FID);
6312         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6313                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6314                            (struct iam_rec *)dtrec);
6315                 osd_quota_unpack(it->oi_obj, dtrec);
6316         } else {
6317                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6318                            (struct iam_rec *)dtrec);
6319         }
6320
6321         RETURN(0);
6322 }
6323
6324 /**
6325  * Returns cookie for current Iterator position.
6326  */
6327 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6328 {
6329         struct osd_it_iam *it = (struct osd_it_iam *)di;
6330
6331         return iam_it_store(&it->oi_it);
6332 }
6333
6334 /**
6335  * Restore iterator from cookie.
6336  *
6337  * \param  di      osd iterator
6338  * \param  hash    Iterator location cookie
6339  *
6340  * \retval +ve  di points to record with least key not larger than key.
6341  * \retval  0   di points to exact matched key
6342  * \retval -ve  failure
6343  */
6344
6345 static int osd_it_iam_load(const struct lu_env *env,
6346                            const struct dt_it *di, __u64 hash)
6347 {
6348         struct osd_it_iam *it = (struct osd_it_iam *)di;
6349
6350         return iam_it_load(&it->oi_it, hash);
6351 }
6352
6353 static const struct dt_index_operations osd_index_iam_ops = {
6354         .dio_lookup         = osd_index_iam_lookup,
6355         .dio_declare_insert = osd_index_declare_iam_insert,
6356         .dio_insert         = osd_index_iam_insert,
6357         .dio_declare_delete = osd_index_declare_iam_delete,
6358         .dio_delete         = osd_index_iam_delete,
6359         .dio_it     = {
6360                 .init     = osd_it_iam_init,
6361                 .fini     = osd_it_iam_fini,
6362                 .get      = osd_it_iam_get,
6363                 .put      = osd_it_iam_put,
6364                 .next     = osd_it_iam_next,
6365                 .key      = osd_it_iam_key,
6366                 .key_size = osd_it_iam_key_size,
6367                 .rec      = osd_it_iam_rec,
6368                 .store    = osd_it_iam_store,
6369                 .load     = osd_it_iam_load
6370         }
6371 };
6372
6373
6374 /**
6375  * Creates or initializes iterator context.
6376  *
6377  * \retval struct osd_it_ea, iterator structure on success
6378  *
6379  */
6380 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6381                                     struct dt_object *dt,
6382                                     __u32 attr)
6383 {
6384         struct osd_object *obj = osd_dt_obj(dt);
6385         struct osd_thread_info *info = osd_oti_get(env);
6386         struct osd_it_ea *oie;
6387         struct file *file;
6388         struct lu_object *lo = &dt->do_lu;
6389         struct dentry *obj_dentry;
6390
6391         ENTRY;
6392
6393         if (!dt_object_exists(dt) || obj->oo_destroyed)
6394                 RETURN(ERR_PTR(-ENOENT));
6395
6396         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6397         if (oie == NULL)
6398                 RETURN(ERR_PTR(-ENOMEM));
6399         obj_dentry = &oie->oie_dentry;
6400
6401         obj_dentry->d_inode = obj->oo_inode;
6402         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6403         obj_dentry->d_name.hash = 0;
6404
6405         oie->oie_rd_dirent       = 0;
6406         oie->oie_it_dirent       = 0;
6407         oie->oie_dirent          = NULL;
6408         if (unlikely(!info->oti_it_ea_buf_used)) {
6409                 oie->oie_buf = info->oti_it_ea_buf;
6410                 info->oti_it_ea_buf_used = 1;
6411         } else {
6412                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6413                 if (oie->oie_buf == NULL)
6414                         RETURN(ERR_PTR(-ENOMEM));
6415         }
6416         oie->oie_obj = obj;
6417
6418         file = &oie->oie_file;
6419
6420         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6421         if (attr & LUDA_64BITHASH)
6422                 file->f_mode    = FMODE_64BITHASH;
6423         else
6424                 file->f_mode    = FMODE_32BITHASH;
6425         file->f_path.dentry     = obj_dentry;
6426         file->f_mapping         = obj->oo_inode->i_mapping;
6427         file->f_op              = obj->oo_inode->i_fop;
6428         set_file_inode(file, obj->oo_inode);
6429
6430         lu_object_get(lo);
6431         RETURN((struct dt_it *)oie);
6432 }
6433
6434 /**
6435  * Destroy or finishes iterator context.
6436  *
6437  * \param di iterator structure to be destroyed
6438  */
6439 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6440 {
6441         struct osd_thread_info *info = osd_oti_get(env);
6442         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6443         struct osd_object *obj = oie->oie_obj;
6444         struct inode *inode = obj->oo_inode;
6445
6446         ENTRY;
6447         oie->oie_file.f_op->release(inode, &oie->oie_file);
6448         osd_object_put(env, obj);
6449         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6450                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6451         else
6452                 info->oti_it_ea_buf_used = 0;
6453         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6454         EXIT;
6455 }
6456
6457 /**
6458  * It position the iterator at given key, so that next lookup continues from
6459  * that key Or it is similar to dio_it->load() but based on a key,
6460  * rather than file position.
6461  *
6462  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6463  * to the beginning.
6464  *
6465  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6466  */
6467 static int osd_it_ea_get(const struct lu_env *env,
6468                          struct dt_it *di, const struct dt_key *key)
6469 {
6470         struct osd_it_ea *it = (struct osd_it_ea *)di;
6471
6472         ENTRY;
6473         LASSERT(((const char *)key)[0] == '\0');
6474         it->oie_file.f_pos = 0;
6475         it->oie_rd_dirent = 0;
6476         it->oie_it_dirent = 0;
6477         it->oie_dirent = NULL;
6478
6479         RETURN(+1);
6480 }
6481
6482 /**
6483  * Does nothing
6484  */
6485 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6486 {
6487 }
6488
6489 struct osd_filldir_cbs {
6490 #ifdef HAVE_DIR_CONTEXT
6491         struct dir_context ctx;
6492 #endif
6493         struct osd_it_ea  *it;
6494 };
6495 /**
6496  * It is called internally by ->readdir(). It fills the
6497  * iterator's in-memory data structure with required
6498  * information i.e. name, namelen, rec_size etc.
6499  *
6500  * \param buf in which information to be filled in.
6501  * \param name name of the file in given dir
6502  *
6503  * \retval 0 on success
6504  * \retval 1 on buffer full
6505  */
6506 #ifdef HAVE_FILLDIR_USE_CTX
6507 static int osd_ldiskfs_filldir(struct dir_context *buf,
6508 #else
6509 static int osd_ldiskfs_filldir(void *buf,
6510 #endif
6511                                const char *name, int namelen,
6512                                loff_t offset, __u64 ino, unsigned int d_type)
6513 {
6514         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6515         struct osd_object *obj = it->oie_obj;
6516         struct osd_it_ea_dirent *ent = it->oie_dirent;
6517         struct lu_fid *fid = &ent->oied_fid;
6518         struct osd_fid_pack *rec;
6519
6520         ENTRY;
6521
6522 /* this should never happen */
6523         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6524                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6525                 RETURN(-EIO);
6526         }
6527
6528         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6529             OSD_IT_EA_BUFSIZE)
6530                 RETURN(1);
6531
6532         /* "." is just the object itself. */
6533         if (namelen == 1 && name[0] == '.') {
6534                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6535         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6536                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6537                 if (osd_fid_unpack(fid, rec) != 0)
6538                         fid_zero(fid);
6539         } else {
6540                 fid_zero(fid);
6541         }
6542         d_type &= ~LDISKFS_DIRENT_LUFID;
6543
6544         /* NOT export local root. */
6545         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6546                 ino = obj->oo_inode->i_ino;
6547                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6548         }
6549
6550         ent->oied_ino     = ino;
6551         ent->oied_off     = offset;
6552         ent->oied_namelen = namelen;
6553         ent->oied_type    = d_type;
6554
6555         memcpy(ent->oied_name, name, namelen);
6556
6557         it->oie_rd_dirent++;
6558         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6559         RETURN(0);
6560 }
6561
6562 /**
6563  * Calls ->readdir() to load a directory entry at a time
6564  * and stored it in iterator's in-memory data structure.
6565  *
6566  * \param di iterator's in memory structure
6567  *
6568  * \retval   0 on success
6569  * \retval -ve on error
6570  * \retval +1 reach the end of entry
6571  */
6572 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6573                                const struct dt_it *di)
6574 {
6575         struct osd_it_ea *it = (struct osd_it_ea *)di;
6576         struct osd_object *obj = it->oie_obj;
6577         struct inode *inode = obj->oo_inode;
6578         struct htree_lock *hlock = NULL;
6579         struct file *filp = &it->oie_file;
6580         int rc = 0;
6581         struct osd_filldir_cbs buf = {
6582 #ifdef HAVE_DIR_CONTEXT
6583                 .ctx.actor = osd_ldiskfs_filldir,
6584 #endif
6585                 .it = it
6586         };
6587
6588         ENTRY;
6589         it->oie_dirent = it->oie_buf;
6590         it->oie_rd_dirent = 0;
6591
6592         if (obj->oo_hl_head != NULL) {
6593                 hlock = osd_oti_get(env)->oti_hlock;
6594                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6595                                    inode, LDISKFS_HLOCK_READDIR);
6596         } else {
6597                 down_read(&obj->oo_ext_idx_sem);
6598         }
6599
6600 #ifdef HAVE_DIR_CONTEXT
6601         buf.ctx.pos = filp->f_pos;
6602 #ifdef HAVE_ITERATE_SHARED
6603         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6604 #else
6605         rc = inode->i_fop->iterate(filp, &buf.ctx);
6606 #endif
6607         filp->f_pos = buf.ctx.pos;
6608 #else
6609         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6610 #endif
6611
6612         if (hlock != NULL)
6613                 ldiskfs_htree_unlock(hlock);
6614         else
6615                 up_read(&obj->oo_ext_idx_sem);
6616
6617         if (it->oie_rd_dirent == 0) {
6618                 /*
6619                  * If it does not get any dirent, it means it has been reached
6620                  * to the end of the dir
6621                  */
6622                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6623                 if (rc == 0)
6624                         rc = 1;
6625         } else {
6626                 it->oie_dirent = it->oie_buf;
6627                 it->oie_it_dirent = 1;
6628         }
6629
6630         RETURN(rc);
6631 }
6632
6633 /**
6634  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6635  * to load a directory entry at a time and stored it in
6636  * iterator's in-memory data structure.
6637  *
6638  * \param di iterator's in memory structure
6639  *
6640  * \retval +ve iterator reached to end
6641  * \retval   0 iterator not reached to end
6642  * \retval -ve on error
6643  */
6644 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6645 {
6646         struct osd_it_ea *it = (struct osd_it_ea *)di;
6647         int rc;
6648
6649         ENTRY;
6650
6651         if (it->oie_it_dirent < it->oie_rd_dirent) {
6652                 it->oie_dirent =
6653                         (void *)it->oie_dirent +
6654                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6655                                        it->oie_dirent->oied_namelen);
6656                 it->oie_it_dirent++;
6657                 rc = 0;
6658         } else {
6659                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6660                         rc = 1;
6661                 else
6662                         rc = osd_ldiskfs_it_fill(env, di);
6663         }
6664
6665         RETURN(rc);
6666 }
6667
6668 /**
6669  * Returns the key at current position from iterator's in memory structure.
6670  *
6671  * \param di iterator's in memory structure
6672  *
6673  * \retval key i.e. struct dt_key on success
6674  */
6675 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6676                                     const struct dt_it *di)
6677 {
6678         struct osd_it_ea *it = (struct osd_it_ea *)di;
6679
6680         return (struct dt_key *)it->oie_dirent->oied_name;
6681 }
6682
6683 /**
6684  * Returns key's size at current position from iterator's in memory structure.
6685  *
6686  * \param di iterator's in memory structure
6687  *
6688  * \retval key_size i.e. struct dt_key on success
6689  */
6690 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6691 {
6692         struct osd_it_ea *it = (struct osd_it_ea *)di;
6693
6694         return it->oie_dirent->oied_namelen;
6695 }
6696
6697 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6698 {
6699         if (LDISKFS_DIR_REC_LEN(de) >=
6700             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6701                 return true;
6702
6703         return false;
6704 }
6705
6706 static inline bool
6707 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6708                      unsigned int blocksize, bool dotdot)
6709 {
6710         if (dotdot)
6711                 return osd_dotdot_has_space(de);
6712
6713         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6714             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6715                 return true;
6716
6717         return false;
6718 }
6719
6720 static int
6721 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6722                     handle_t *jh, struct dentry *dentry,
6723                     const struct lu_fid *fid, struct buffer_head *bh,
6724                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6725                     bool dotdot)
6726 {
6727         struct inode *dir = dentry->d_parent->d_inode;
6728         struct inode *inode = dentry->d_inode;
6729         struct osd_fid_pack *rec;
6730         struct ldiskfs_dentry_param *ldp;
6731         int namelen = dentry->d_name.len;
6732         int rc;
6733         struct osd_thread_info *info = osd_oti_get(env);
6734
6735         ENTRY;
6736
6737         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6738                 RETURN(0);
6739
6740         /* There is enough space to hold the FID-in-dirent. */
6741         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6742                 rc = ldiskfs_journal_get_write_access(jh, bh);
6743                 if (rc != 0)
6744                         RETURN(rc);
6745
6746                 de->name[namelen] = 0;
6747                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6748                 rec->fp_len = sizeof(struct lu_fid) + 1;
6749                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6750                 de->file_type |= LDISKFS_DIRENT_LUFID;
6751                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6752
6753                 RETURN(rc);
6754         }
6755
6756         LASSERT(!dotdot);
6757
6758         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6759         if (rc != 0)
6760                 RETURN(rc);
6761
6762         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6763         osd_get_ldiskfs_dirent_param(ldp, fid);
6764         dentry->d_fsdata = (void *)ldp;
6765         ll_vfs_dq_init(dir);
6766         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6767         /*
6768          * It is too bad, we cannot reinsert the name entry back.
6769          * That means we lose it!
6770          */
6771         if (rc != 0)
6772                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6773                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6774                        osd_ino2name(inode),
6775                        dir->i_ino, dir->i_generation, namelen,
6776                        dentry->d_name.name, PFID(fid), rc);
6777
6778         RETURN(rc);
6779 }
6780
6781 static int
6782 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6783                         struct osd_it_ea *it, struct lu_fid *fid,
6784                         struct osd_inode_id *id, __u32 *attr)
6785 {
6786         struct osd_thread_info *info = osd_oti_get(env);
6787         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6788         struct osd_device *dev = osd_obj2dev(obj);
6789         struct super_block *sb = osd_sb(dev);
6790         const char *devname = osd_name(dev);
6791         struct osd_it_ea_dirent *ent = it->oie_dirent;
6792         struct inode *dir = obj->oo_inode;
6793         struct htree_lock *hlock = NULL;
6794         struct buffer_head *bh = NULL;
6795         handle_t *jh = NULL;
6796         struct ldiskfs_dir_entry_2 *de;
6797         struct dentry *dentry;
6798         struct inode *inode;
6799         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6800         int credits;
6801         int rc;
6802         bool dotdot = false;
6803         bool dirty = false;
6804
6805         ENTRY;
6806
6807         if (ent->oied_name[0] == '.') {
6808                 if (ent->oied_namelen == 1)
6809                         RETURN(0);
6810
6811                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6812                         dotdot = true;
6813         }
6814
6815         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6816         inode = osd_iget(info, dev, id);
6817         if (IS_ERR(inode)) {
6818                 rc = PTR_ERR(inode);
6819                 if (rc == -ENOENT || rc == -ESTALE) {
6820                         /*
6821                          * Maybe dangling name entry, or
6822                          * corrupted directory entry.
6823                          */
6824                         *attr |= LUDA_UNKNOWN;
6825                         rc = 0;
6826                 } else {
6827                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6828                                "check_repair, dir = %lu/%u, name = %.*s, "
6829                                "ino = %llu, rc = %d\n",
6830                                devname, dir->i_ino, dir->i_generation,
6831                                ent->oied_namelen, ent->oied_name,
6832                                ent->oied_ino, rc);
6833                 }
6834
6835                 RETURN(rc);
6836         }
6837
6838         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6839                                            ent->oied_namelen);
6840         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6841         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6842                 lma = NULL;
6843         else if (rc != 0)
6844                 GOTO(out, rc);
6845
6846         /*
6847          * We need to ensure that the name entry is still valid.
6848          * Because it may be removed or renamed by other already.
6849          *
6850          * The unlink or rename operation will start journal before PDO lock,
6851          * so to avoid deadlock, here we need to start journal handle before
6852          * related PDO lock also. But because we do not know whether there
6853          * will be something to be repaired before PDO lock, we just start
6854          * journal without conditions.
6855          *
6856          * We may need to remove the name entry firstly, then insert back.
6857          * One credit is for user quota file update.
6858          * One credit is for group quota file update.
6859          * Two credits are for dirty inode.
6860          */
6861         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6862                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6863
6864         if (dev->od_dirent_journal != 0) {
6865
6866 again:
6867                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6868                 if (IS_ERR(jh)) {
6869                         rc = PTR_ERR(jh);
6870                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6871                                "check_repair, dir = %lu/%u, credits = %d, "
6872                                "name = %.*s, ino = %llu: rc = %d\n",
6873                                devname, dir->i_ino, dir->i_generation, credits,
6874                                ent->oied_namelen, ent->oied_name,
6875                                ent->oied_ino, rc);
6876
6877                         GOTO(out_inode, rc);
6878                 }
6879
6880                 if (obj->oo_hl_head != NULL) {
6881                         hlock = osd_oti_get(env)->oti_hlock;
6882                         /*
6883                          * "0" means exclusive lock for the whole directory.
6884                          * We need to prevent others access such name entry
6885                          * during the delete + insert. Neither HLOCK_ADD nor
6886                          * HLOCK_DEL cannot guarantee the atomicity.
6887                          */
6888                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6889                 } else {
6890                         down_write(&obj->oo_ext_idx_sem);
6891                 }
6892         } else {
6893                 if (obj->oo_hl_head != NULL) {
6894                         hlock = osd_oti_get(env)->oti_hlock;
6895                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6896                                            LDISKFS_HLOCK_LOOKUP);
6897                 } else {
6898                         down_read(&obj->oo_ext_idx_sem);
6899                 }
6900         }
6901
6902         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6903         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6904                 *attr |= LUDA_IGNORE;
6905
6906                 GOTO(out, rc = 0);
6907         }
6908
6909         /*
6910          * For dotdot entry, if there is not enough space to hold the
6911          * FID-in-dirent, just keep them there. It only happens when the
6912          * device upgraded from 1.8 or restored from MDT file-level backup.
6913          * For the whole directory, only dotdot entry have no FID-in-dirent
6914          * and needs to get FID from LMA when readdir, it will not affect the
6915          * performance much.
6916          */
6917         if (dotdot && !osd_dotdot_has_space(de)) {
6918                 *attr |= LUDA_UNKNOWN;
6919
6920                 GOTO(out, rc = 0);
6921         }
6922
6923         if (lma != NULL) {
6924                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6925                         GOTO(out, rc = 0);
6926
6927                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6928                         struct lu_fid *tfid = &lma->lma_self_fid;
6929
6930                         if (likely(dotdot &&
6931                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6932                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6933                                 /*
6934                                  * It must be REMOTE_PARENT_DIR and as the
6935                                  * 'dotdot' entry of remote directory
6936                                  */
6937                                 *attr |= LUDA_IGNORE;
6938                         } else {
6939                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6940                                        "parent directory, but got %.*s under "
6941                                        "dir = %lu/%u with the FID "DFID"\n",
6942                                        devname, ent->oied_namelen,
6943                                        ent->oied_name, dir->i_ino,
6944                                        dir->i_generation, PFID(tfid));
6945
6946                                 *attr |= LUDA_UNKNOWN;
6947                         }
6948
6949                         GOTO(out, rc = 0);
6950                 }
6951         }
6952
6953         if (!fid_is_zero(fid)) {
6954                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6955                                               ent->oied_namelen);
6956                 if (rc == -ENOENT ||
6957                     (rc == -ENODATA &&
6958                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6959                         /*
6960                          * linkEA does not recognize the dirent entry,
6961                          * it may because the dirent entry corruption
6962                          * and points to other's inode.
6963                          */
6964                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6965                                "recognize the dirent, dir = %lu/%u, "
6966                                " name = %.*s, ino = %llu, "
6967                                DFID": rc = %d\n", devname, dir->i_ino,
6968                                dir->i_generation, ent->oied_namelen,
6969                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6970                         *attr |= LUDA_UNKNOWN;
6971
6972                         GOTO(out, rc = 0);
6973                 }
6974
6975                 if (rc && rc != -ENODATA) {
6976                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6977                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6978                                DFID": rc = %d\n", devname, dir->i_ino,
6979                                dir->i_generation, ent->oied_namelen,
6980                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6981                         *attr |= LUDA_UNKNOWN;
6982
6983                         GOTO(out, rc = 0);
6984                 }
6985         }
6986
6987         if (lma != NULL) {
6988                 /*
6989                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6990                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6991                  */
6992                 if (*attr & LUDA_VERIFY_DRYRUN) {
6993                         *fid = lma->lma_self_fid;
6994                         *attr |= LUDA_REPAIR;
6995
6996                         GOTO(out, rc = 0);
6997                 }
6998
6999                 if (jh == NULL) {
7000                         brelse(bh);
7001                         dev->od_dirent_journal = 1;
7002                         if (hlock != NULL) {
7003                                 ldiskfs_htree_unlock(hlock);
7004                                 hlock = NULL;
7005                         } else {
7006                                 up_read(&obj->oo_ext_idx_sem);
7007                         }
7008
7009                         goto again;
7010                 }
7011
7012                 *fid = lma->lma_self_fid;
7013                 dirty = true;
7014                 /* Update or append the FID-in-dirent. */
7015                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7016                                          bh, de, hlock, dotdot);
7017                 if (rc == 0)
7018                         *attr |= LUDA_REPAIR;
7019                 else
7020                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7021                                "the dirent, dir = %lu/%u, name = %.*s, "
7022                                "ino = %llu, "DFID": rc = %d\n",
7023                                devname, dir->i_ino, dir->i_generation,
7024                                ent->oied_namelen, ent->oied_name,
7025                                ent->oied_ino, PFID(fid), rc);
7026         } else {
7027                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7028                 if (*attr & LUDA_VERIFY_DRYRUN) {
7029                         if (fid_is_sane(fid)) {
7030                                 *attr |= LUDA_REPAIR;
7031                         } else if (dev->od_index == 0) {
7032                                 lu_igif_build(fid, inode->i_ino,
7033                                               inode->i_generation);
7034                                 *attr |= LUDA_UPGRADE;
7035                         }
7036
7037                         GOTO(out, rc = 0);
7038                 }
7039
7040                 if (jh == NULL) {
7041                         brelse(bh);
7042                         dev->od_dirent_journal = 1;
7043                         if (hlock != NULL) {
7044                                 ldiskfs_htree_unlock(hlock);
7045                                 hlock = NULL;
7046                         } else {
7047                                 up_read(&obj->oo_ext_idx_sem);
7048                         }
7049
7050                         goto again;
7051                 }
7052
7053                 dirty = true;
7054                 if (unlikely(fid_is_sane(fid))) {
7055                         /*
7056                          * FID-in-dirent exists, but FID-in-LMA is lost.
7057                          * Trust the FID-in-dirent, and add FID-in-LMA.
7058                          */
7059                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7060                         if (rc == 0)
7061                                 *attr |= LUDA_REPAIR;
7062                         else
7063                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7064                                        "update dirent, dir = %lu/%u, "
7065                                        "name = %.*s, ino = %llu, "
7066                                        DFID": rc = %d\n",
7067                                        devname, dir->i_ino, dir->i_generation,
7068                                        ent->oied_namelen, ent->oied_name,
7069                                        ent->oied_ino, PFID(fid), rc);
7070                 } else if (dev->od_index == 0) {
7071                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7072                         /*
7073                          * It is probably IGIF object. Only aappend the
7074                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7075                          */
7076                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7077                                                  bh, de, hlock, dotdot);
7078                         if (rc == 0)
7079                                 *attr |= LUDA_UPGRADE;
7080                         else
7081                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7082                                        "after the dirent, dir = %lu/%u, "
7083                                        "name = %.*s, ino = %llu, "
7084                                        DFID": rc = %d\n",
7085                                        devname, dir->i_ino, dir->i_generation,
7086                                        ent->oied_namelen, ent->oied_name,
7087                                        ent->oied_ino, PFID(fid), rc);
7088                 }
7089         }
7090
7091         GOTO(out, rc);
7092
7093 out:
7094         if (!IS_ERR(bh))
7095                 brelse(bh);
7096         if (hlock != NULL) {
7097                 ldiskfs_htree_unlock(hlock);
7098         } else {
7099                 if (dev->od_dirent_journal != 0)
7100                         up_write(&obj->oo_ext_idx_sem);
7101                 else
7102                         up_read(&obj->oo_ext_idx_sem);
7103         }
7104
7105         if (jh != NULL)
7106                 ldiskfs_journal_stop(jh);
7107
7108 out_inode:
7109         iput(inode);
7110         if (rc >= 0 && !dirty)
7111                 dev->od_dirent_journal = 0;
7112
7113         return rc;
7114 }
7115
7116 /**
7117  * Returns the value at current position from iterator's in memory structure.
7118  *
7119  * \param di struct osd_it_ea, iterator's in memory structure
7120  * \param attr attr requested for dirent.
7121  * \param lde lustre dirent
7122  *
7123  * \retval   0 no error and \param lde has correct lustre dirent.
7124  * \retval -ve on error
7125  */
7126 static inline int osd_it_ea_rec(const struct lu_env *env,
7127                                 const struct dt_it *di,
7128                                 struct dt_rec *dtrec, __u32 attr)
7129 {
7130         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7131         struct osd_object      *obj   = it->oie_obj;
7132         struct osd_device      *dev   = osd_obj2dev(obj);
7133         struct osd_thread_info *oti   = osd_oti_get(env);
7134         struct osd_inode_id    *id    = &oti->oti_id;
7135         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7136         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7137         __u32 ino = it->oie_dirent->oied_ino;
7138         int rc = 0;
7139
7140         ENTRY;
7141
7142         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7143
7144         if (attr & LUDA_VERIFY) {
7145                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7146                         attr |= LUDA_IGNORE;
7147                         /*
7148                          * If the parent is on remote MDT, and there
7149                          * is no FID-in-dirent, then we have to get
7150                          * the parent FID from the linkEA.
7151                          */
7152                         if (!fid_is_sane(fid) &&
7153                             it->oie_dirent->oied_namelen == 2 &&
7154                             it->oie_dirent->oied_name[0] == '.' &&
7155                             it->oie_dirent->oied_name[1] == '.')
7156                                 osd_get_pfid_from_linkea(env, obj, fid);
7157                 } else {
7158                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7159                                                      &attr);
7160                 }
7161
7162                 if (!fid_is_sane(fid))
7163                         attr |= LUDA_UNKNOWN;
7164         } else {
7165                 attr &= ~LU_DIRENT_ATTRS_MASK;
7166                 if (!fid_is_sane(fid)) {
7167                         bool is_dotdot = false;
7168
7169                         if (it->oie_dirent->oied_namelen == 2 &&
7170                             it->oie_dirent->oied_name[0] == '.' &&
7171                             it->oie_dirent->oied_name[1] == '.')
7172                                 is_dotdot = true;
7173                         /*
7174                          * If the parent is on remote MDT, and there
7175                          * is no FID-in-dirent, then we have to get
7176                          * the parent FID from the linkEA.
7177                          */
7178                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7179                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7180                         } else {
7181                                 if (is_dotdot == false &&
7182                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7183                                         RETURN(-ENOENT);
7184
7185                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7186                         }
7187                 } else {
7188                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7189                 }
7190         }
7191
7192         /* Pack the entry anyway, at least the offset is right. */
7193         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7194                            it->oie_dirent->oied_name,
7195                            it->oie_dirent->oied_namelen,
7196                            it->oie_dirent->oied_type, attr);
7197
7198         if (rc < 0)
7199                 RETURN(rc);
7200
7201         if (osd_remote_fid(env, dev, fid))
7202                 RETURN(0);
7203
7204         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7205                 osd_add_oi_cache(oti, dev, id, fid);
7206
7207         RETURN(rc > 0 ? 0 : rc);
7208 }
7209
7210 /**
7211  * Returns the record size size at current position.
7212  *
7213  * This function will return record(lu_dirent) size in bytes.
7214  *
7215  * \param[in] env       execution environment
7216  * \param[in] di        iterator's in memory structure
7217  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7218  *                      calculate the lu_dirent size.
7219  *
7220  * \retval      record size(in bytes & in memory) of the current lu_dirent
7221  *              entry.
7222  */
7223 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7224                               __u32 attr)
7225 {
7226         struct osd_it_ea *it = (struct osd_it_ea *)di;
7227
7228         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7229 }
7230
7231 /**
7232  * Returns a cookie for current position of the iterator head, so that
7233  * user can use this cookie to load/start the iterator next time.
7234  *
7235  * \param di iterator's in memory structure
7236  *
7237  * \retval cookie for current position, on success
7238  */
7239 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7240 {
7241         struct osd_it_ea *it = (struct osd_it_ea *)di;
7242
7243         return it->oie_dirent->oied_off;
7244 }
7245
7246 /**
7247  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7248  * to load a directory entry at a time and stored it i inn,
7249  * in iterator's in-memory data structure.
7250  *
7251  * \param di struct osd_it_ea, iterator's in memory structure
7252  *
7253  * \retval +ve on success
7254  * \retval -ve on error
7255  */
7256 static int osd_it_ea_load(const struct lu_env *env,
7257                           const struct dt_it *di, __u64 hash)
7258 {
7259         struct osd_it_ea *it = (struct osd_it_ea *)di;
7260         int rc;
7261
7262         ENTRY;
7263         it->oie_file.f_pos = hash;
7264
7265         rc =  osd_ldiskfs_it_fill(env, di);
7266         if (rc > 0)
7267                 rc = -ENODATA;
7268
7269         if (rc == 0)
7270                 rc = 1;
7271
7272         RETURN(rc);
7273 }
7274
7275 /**
7276  * Index lookup function for interoperability mode (b11826).
7277  *
7278  * \param key,  key i.e. file name to be searched
7279  *
7280  * \retval +ve, on success
7281  * \retval -ve, on error
7282  */
7283 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7284                                struct dt_rec *rec, const struct dt_key *key)
7285 {
7286         struct osd_object *obj = osd_dt_obj(dt);
7287         int rc = 0;
7288
7289         ENTRY;
7290
7291         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7292         LINVRNT(osd_invariant(obj));
7293
7294         rc = osd_ea_lookup_rec(env, obj, rec, key);
7295         if (rc == 0)
7296                 rc = 1;
7297         RETURN(rc);
7298 }
7299
7300 /**
7301  * Index and Iterator operations for interoperability
7302  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7303  */
7304 static const struct dt_index_operations osd_index_ea_ops = {
7305         .dio_lookup         = osd_index_ea_lookup,
7306         .dio_declare_insert = osd_index_declare_ea_insert,
7307         .dio_insert         = osd_index_ea_insert,
7308         .dio_declare_delete = osd_index_declare_ea_delete,
7309         .dio_delete         = osd_index_ea_delete,
7310         .dio_it     = {
7311                 .init     = osd_it_ea_init,
7312                 .fini     = osd_it_ea_fini,
7313                 .get      = osd_it_ea_get,
7314                 .put      = osd_it_ea_put,
7315                 .next     = osd_it_ea_next,
7316                 .key      = osd_it_ea_key,
7317                 .key_size = osd_it_ea_key_size,
7318                 .rec      = osd_it_ea_rec,
7319                 .rec_size = osd_it_ea_rec_size,
7320                 .store    = osd_it_ea_store,
7321                 .load     = osd_it_ea_load
7322         }
7323 };
7324
7325 static void *osd_key_init(const struct lu_context *ctx,
7326                           struct lu_context_key *key)
7327 {
7328         struct osd_thread_info *info;
7329
7330         OBD_ALLOC_PTR(info);
7331         if (info == NULL)
7332                 return ERR_PTR(-ENOMEM);
7333
7334         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7335         if (info->oti_it_ea_buf == NULL)
7336                 goto out_free_info;
7337
7338         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7339
7340         info->oti_hlock = ldiskfs_htree_lock_alloc();
7341         if (info->oti_hlock == NULL)
7342                 goto out_free_ea;
7343
7344         return info;
7345
7346 out_free_ea:
7347         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7348 out_free_info:
7349         OBD_FREE_PTR(info);
7350         return ERR_PTR(-ENOMEM);
7351 }
7352
7353 static void osd_key_fini(const struct lu_context *ctx,
7354                          struct lu_context_key *key, void *data)
7355 {
7356         struct osd_thread_info *info = data;
7357         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7358         struct osd_idmap_cache *idc = info->oti_ins_cache;
7359
7360         if (info->oti_dio_pages) {
7361                 int i;
7362                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7363                         if (info->oti_dio_pages[i])
7364                                 __free_page(info->oti_dio_pages[i]);
7365                 }
7366                 OBD_FREE(info->oti_dio_pages,
7367                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7368         }
7369
7370         if (info->oti_inode != NULL)
7371                 OBD_FREE_PTR(lli);
7372         if (info->oti_hlock != NULL)
7373                 ldiskfs_htree_lock_free(info->oti_hlock);
7374         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7375         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7376         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7377         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7378         lu_buf_free(&info->oti_big_buf);
7379         if (idc != NULL) {
7380                 LASSERT(info->oti_ins_cache_size > 0);
7381                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7382                 info->oti_ins_cache = NULL;
7383                 info->oti_ins_cache_size = 0;
7384         }
7385         OBD_FREE_PTR(info);
7386 }
7387
7388 static void osd_key_exit(const struct lu_context *ctx,
7389                          struct lu_context_key *key, void *data)
7390 {
7391         struct osd_thread_info *info = data;
7392
7393         LASSERT(info->oti_r_locks == 0);
7394         LASSERT(info->oti_w_locks == 0);
7395         LASSERT(info->oti_txns    == 0);
7396 }
7397
7398 /* type constructor/destructor: osd_type_init, osd_type_fini */
7399 LU_TYPE_INIT_FINI(osd, &osd_key);
7400
7401 struct lu_context_key osd_key = {
7402         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7403         .lct_init = osd_key_init,
7404         .lct_fini = osd_key_fini,
7405         .lct_exit = osd_key_exit
7406 };
7407
7408
7409 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7410                            const char *name, struct lu_device *next)
7411 {
7412         struct osd_device *osd = osd_dev(d);
7413
7414         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7415             sizeof(osd->od_svname))
7416                 return -E2BIG;
7417         return osd_procfs_init(osd, name);
7418 }
7419
7420 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7421 {
7422         struct seq_server_site *ss = osd_seq_site(osd);
7423         int rc;
7424
7425         ENTRY;
7426
7427         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7428                 RETURN(0);
7429
7430         if (unlikely(ss == NULL))
7431                 RETURN(-ENODEV);
7432
7433         OBD_ALLOC_PTR(osd->od_cl_seq);
7434         if (osd->od_cl_seq == NULL)
7435                 RETURN(-ENOMEM);
7436
7437         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7438                              osd->od_svname, ss->ss_server_seq);
7439         if (rc != 0) {
7440                 OBD_FREE_PTR(osd->od_cl_seq);
7441                 osd->od_cl_seq = NULL;
7442                 RETURN(rc);
7443         }
7444
7445         if (ss->ss_node_id == 0) {
7446                 /*
7447                  * If the OSD on the sequence controller(MDT0), then allocate
7448                  * sequence here, otherwise allocate sequence after connected
7449                  * to MDT0 (see mdt_register_lwp_callback()).
7450                  */
7451                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7452                                    &osd->od_cl_seq->lcs_space, env);
7453         }
7454
7455         RETURN(rc);
7456 }
7457
7458 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7459 {
7460         if (osd->od_cl_seq == NULL)
7461                 return;
7462
7463         seq_client_fini(osd->od_cl_seq);
7464         OBD_FREE_PTR(osd->od_cl_seq);
7465         osd->od_cl_seq = NULL;
7466 }
7467
7468 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7469 {
7470         ENTRY;
7471
7472         /* shutdown quota slave instance associated with the device */
7473         if (o->od_quota_slave_md != NULL) {
7474                 struct qsd_instance *qsd = o->od_quota_slave_md;
7475
7476                 o->od_quota_slave_md = NULL;
7477                 qsd_fini(env, qsd);
7478         }
7479
7480         if (o->od_quota_slave_dt != NULL) {
7481                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7482
7483                 o->od_quota_slave_dt = NULL;
7484                 qsd_fini(env, qsd);
7485         }
7486
7487         osd_fid_fini(env, o);
7488         osd_scrub_cleanup(env, o);
7489
7490         RETURN(0);
7491 }
7492
7493 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7494 {
7495         ENTRY;
7496
7497         if (o->od_mnt != NULL) {
7498                 shrink_dcache_sb(osd_sb(o));
7499                 osd_sync(env, &o->od_dt_dev);
7500
7501                 mntput(o->od_mnt);
7502                 o->od_mnt = NULL;
7503         }
7504
7505         EXIT;
7506 }
7507
7508 static int osd_mount(const struct lu_env *env,
7509                      struct osd_device *o, struct lustre_cfg *cfg)
7510 {
7511         const char *name = lustre_cfg_string(cfg, 0);
7512         const char *dev = lustre_cfg_string(cfg, 1);
7513         const char *opts;
7514         unsigned long page, s_flags, lmd_flags = 0;
7515         struct page *__page;
7516         struct file_system_type *type;
7517         char *options = NULL;
7518         char *str;
7519         struct osd_thread_info *info = osd_oti_get(env);
7520         struct lu_fid *fid = &info->oti_fid;
7521         struct inode *inode;
7522         int rc = 0, force_over_512tb = 0;
7523
7524         ENTRY;
7525
7526         if (o->od_mnt != NULL)
7527                 RETURN(0);
7528
7529         if (strlen(dev) >= sizeof(o->od_mntdev))
7530                 RETURN(-E2BIG);
7531         strcpy(o->od_mntdev, dev);
7532
7533         str = lustre_cfg_string(cfg, 2);
7534         s_flags = simple_strtoul(str, NULL, 0);
7535         str = strstr(str, ":");
7536         if (str)
7537                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7538         opts = lustre_cfg_string(cfg, 3);
7539 #ifdef __BIG_ENDIAN
7540         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7541                 CERROR("%s: device %s extents feature is not guaranteed to "
7542                        "work on big-endian systems. Use \"bigendian_extents\" "
7543                        "mount option to override.\n", name, dev);
7544                 RETURN(-EINVAL);
7545         }
7546 #endif
7547 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7548         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7549                 CWARN("force_over_128tb option is deprecated. "
7550                       "Filesystems less than 512TB can be created without any "
7551                       "force options. Use force_over_512tb option for "
7552                       "filesystems greater than 512TB.\n");
7553         }
7554 #endif
7555 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7556         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7557                 CWARN("force_over_256tb option is deprecated. "
7558                       "Filesystems less than 512TB can be created without any "
7559                       "force options. Use force_over_512tb option for "
7560                       "filesystems greater than 512TB.\n");
7561         }
7562 #endif
7563
7564         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7565                 force_over_512tb = 1;
7566
7567         __page = alloc_page(GFP_KERNEL);
7568         if (__page == NULL)
7569                 GOTO(out, rc = -ENOMEM);
7570         page = (unsigned long)page_address(__page);
7571         options = (char *)page;
7572         *options = '\0';
7573         if (opts != NULL) {
7574                 /* strip out the options for back compatiblity */
7575                 static char *sout[] = {
7576                         "mballoc",
7577                         "iopen",
7578                         "noiopen",
7579                         "iopen_nopriv",
7580                         "extents",
7581                         "noextents",
7582                         /* strip out option we processed in osd */
7583                         "bigendian_extents",
7584                         "force_over_128tb",
7585                         "force_over_256tb",
7586                         "force_over_512tb",
7587                         NULL
7588                 };
7589                 strcat(options, opts);
7590                 for (rc = 0, str = options; sout[rc]; ) {
7591                         char *op = strstr(str, sout[rc]);
7592
7593                         if (op == NULL) {
7594                                 rc++;
7595                                 str = options;
7596                                 continue;
7597                         }
7598                         if (op == options || *(op - 1) == ',') {
7599                                 str = op + strlen(sout[rc]);
7600                                 if (*str == ',' || *str == '\0') {
7601                                         *str == ',' ? str++ : str;
7602                                         memmove(op, str, strlen(str) + 1);
7603                                 }
7604                         }
7605                         for (str = op; *str != ',' && *str != '\0'; str++)
7606                                 ;
7607                 }
7608         } else {
7609                 strncat(options, "user_xattr,acl", 14);
7610         }
7611
7612         /* Glom up mount options */
7613         if (*options != '\0')
7614                 strcat(options, ",");
7615         strlcat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7616
7617         type = get_fs_type("ldiskfs");
7618         if (!type) {
7619                 CERROR("%s: cannot find ldiskfs module\n", name);
7620                 GOTO(out, rc = -ENODEV);
7621         }
7622
7623         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7624         module_put(type->owner);
7625
7626         if (IS_ERR(o->od_mnt)) {
7627                 rc = PTR_ERR(o->od_mnt);
7628                 o->od_mnt = NULL;
7629                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7630                 GOTO(out, rc);
7631         }
7632
7633         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7634                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7635                                  force_over_512tb == 0) {
7636                 CERROR("%s: device %s LDISKFS does not support filesystems "
7637                        "greater than 512TB and can cause data corruption. "
7638                        "Use \"force_over_512tb\" mount option to override.\n",
7639                        name, dev);
7640                 GOTO(out_mnt, rc = -EINVAL);
7641         }
7642
7643         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7644                 if (priv_dev_set_rdonly) {
7645                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7646                         o->od_dt_dev.dd_rdonly = 1;
7647                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7648                                       name);
7649                 } else {
7650                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7651                                       name);
7652
7653                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7654                 }
7655         } else if (priv_dev_check_rdonly &&
7656                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7657                 CERROR("%s: underlying device %s is marked as "
7658                        "read-only. Setup failed\n", name, dev);
7659
7660                 GOTO(out_mnt, rc = -EROFS);
7661         }
7662
7663         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7664                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7665                 GOTO(out_mnt, rc = -EINVAL);
7666         }
7667
7668 #ifdef LDISKFS_MOUNT_DIRDATA
7669         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7670                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7671         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7672                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7673                       "enabling the dirdata feature. If you do not want to "
7674                       "downgrade to Lustre-1.x again, you can enable it via "
7675                       "'tune2fs -O dirdata device'\n", name, dev);
7676 #endif
7677         inode = osd_sb(o)->s_root->d_inode;
7678         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7679         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7680         if (rc != 0) {
7681                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7682                 GOTO(out_mnt, rc);
7683         }
7684
7685         if (lmd_flags & LMD_FLG_NOSCRUB)
7686                 o->od_auto_scrub_interval = AS_NEVER;
7687
7688         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7689                 /* do not use pagecache with flash-backed storage */
7690                 o->od_writethrough_cache = 0;
7691                 o->od_read_cache = 0;
7692         }
7693
7694         GOTO(out, rc = 0);
7695
7696 out_mnt:
7697         mntput(o->od_mnt);
7698         o->od_mnt = NULL;
7699
7700 out:
7701         if (__page)
7702                 __free_page(__page);
7703
7704         return rc;
7705 }
7706
7707 static struct lu_device *osd_device_fini(const struct lu_env *env,
7708                                          struct lu_device *d)
7709 {
7710         struct osd_device *o = osd_dev(d);
7711
7712         ENTRY;
7713
7714         osd_index_backup(env, o, false);
7715         osd_shutdown(env, o);
7716         osd_procfs_fini(o);
7717         osd_obj_map_fini(o);
7718         osd_umount(env, o);
7719
7720         RETURN(NULL);
7721 }
7722
7723 static int osd_device_init0(const struct lu_env *env,
7724                             struct osd_device *o,
7725                             struct lustre_cfg *cfg)
7726 {
7727         struct lu_device *l = osd2lu_dev(o);
7728         struct osd_thread_info *info;
7729         int rc;
7730         int cplen = 0;
7731
7732         /* if the module was re-loaded, env can loose its keys */
7733         rc = lu_env_refill((struct lu_env *)env);
7734         if (rc)
7735                 GOTO(out, rc);
7736         info = osd_oti_get(env);
7737         LASSERT(info);
7738
7739         l->ld_ops = &osd_lu_ops;
7740         o->od_dt_dev.dd_ops = &osd_dt_ops;
7741
7742         spin_lock_init(&o->od_osfs_lock);
7743         mutex_init(&o->od_otable_mutex);
7744         INIT_LIST_HEAD(&o->od_orphan_list);
7745         INIT_LIST_HEAD(&o->od_index_backup_list);
7746         INIT_LIST_HEAD(&o->od_index_restore_list);
7747         spin_lock_init(&o->od_lock);
7748         o->od_index_backup_policy = LIBP_NONE;
7749         o->od_t10_type = 0;
7750
7751         o->od_read_cache = 1;
7752         o->od_writethrough_cache = 1;
7753         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7754         o->od_auto_scrub_interval = AS_DEFAULT;
7755
7756         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7757                         sizeof(o->od_svname));
7758         if (cplen >= sizeof(o->od_svname)) {
7759                 rc = -E2BIG;
7760                 GOTO(out, rc);
7761         }
7762
7763         o->od_index_backup_stop = 0;
7764         o->od_index = -1; /* -1 means index is invalid */
7765         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7766         if (rc == LDD_F_SV_TYPE_OST)
7767                 o->od_is_ost = 1;
7768
7769         o->od_full_scrub_ratio = OFSR_DEFAULT;
7770         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7771         rc = osd_mount(env, o, cfg);
7772         if (rc != 0)
7773                 GOTO(out, rc);
7774
7775         rc = osd_obj_map_init(env, o);
7776         if (rc != 0)
7777                 GOTO(out_mnt, rc);
7778
7779         rc = lu_site_init(&o->od_site, l);
7780         if (rc != 0)
7781                 GOTO(out_compat, rc);
7782         o->od_site.ls_bottom_dev = l;
7783
7784         rc = lu_site_init_finish(&o->od_site);
7785         if (rc != 0)
7786                 GOTO(out_site, rc);
7787
7788         INIT_LIST_HEAD(&o->od_ios_list);
7789         /* setup scrub, including OI files initialization */
7790         o->od_in_init = 1;
7791         rc = osd_scrub_setup(env, o);
7792         o->od_in_init = 0;
7793         if (rc < 0)
7794                 GOTO(out_site, rc);
7795
7796         rc = osd_procfs_init(o, o->od_svname);
7797         if (rc != 0) {
7798                 CERROR("%s: can't initialize procfs: rc = %d\n",
7799                        o->od_svname, rc);
7800                 GOTO(out_scrub, rc);
7801         }
7802
7803         LASSERT(l->ld_site->ls_linkage.next != NULL);
7804         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7805
7806         /* initialize quota slave instance */
7807         /* currently it's no need to prepare qsd_instance_md for OST */
7808         if (!o->od_is_ost) {
7809                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7810                                                 &o->od_dt_dev,
7811                                                 o->od_proc_entry, true);
7812                 if (IS_ERR(o->od_quota_slave_md)) {
7813                         rc = PTR_ERR(o->od_quota_slave_md);
7814                         o->od_quota_slave_md = NULL;
7815                         GOTO(out_procfs, rc);
7816                 }
7817         }
7818
7819         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7820                                         o->od_proc_entry, false);
7821
7822         if (IS_ERR(o->od_quota_slave_dt)) {
7823                 if (o->od_quota_slave_md != NULL) {
7824                         qsd_fini(env, o->od_quota_slave_md);
7825                         o->od_quota_slave_md = NULL;
7826                 }
7827
7828                 rc = PTR_ERR(o->od_quota_slave_dt);
7829                 o->od_quota_slave_dt = NULL;
7830                 GOTO(out_procfs, rc);
7831         }
7832
7833         RETURN(0);
7834
7835 out_procfs:
7836         osd_procfs_fini(o);
7837 out_scrub:
7838         osd_scrub_cleanup(env, o);
7839 out_site:
7840         lu_site_fini(&o->od_site);
7841 out_compat:
7842         osd_obj_map_fini(o);
7843 out_mnt:
7844         osd_umount(env, o);
7845 out:
7846         return rc;
7847 }
7848
7849 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7850                                           struct lu_device_type *t,
7851                                           struct lustre_cfg *cfg)
7852 {
7853         struct osd_device *o;
7854         int rc;
7855
7856         OBD_ALLOC_PTR(o);
7857         if (o == NULL)
7858                 return ERR_PTR(-ENOMEM);
7859
7860         rc = dt_device_init(&o->od_dt_dev, t);
7861         if (rc == 0) {
7862                 /*
7863                  * Because the ctx might be revived in dt_device_init,
7864                  * refill the env here
7865                  */
7866                 lu_env_refill((struct lu_env *)env);
7867                 rc = osd_device_init0(env, o, cfg);
7868                 if (rc)
7869                         dt_device_fini(&o->od_dt_dev);
7870         }
7871
7872         if (unlikely(rc != 0))
7873                 OBD_FREE_PTR(o);
7874
7875         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7876 }
7877
7878 static struct lu_device *osd_device_free(const struct lu_env *env,
7879                                          struct lu_device *d)
7880 {
7881         struct osd_device *o = osd_dev(d);
7882
7883         ENTRY;
7884
7885         /* XXX: make osd top device in order to release reference */
7886         d->ld_site->ls_top_dev = d;
7887         lu_site_purge(env, d->ld_site, -1);
7888         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7889                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7890                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7891         }
7892         lu_site_fini(&o->od_site);
7893         dt_device_fini(&o->od_dt_dev);
7894         OBD_FREE_PTR(o);
7895         RETURN(NULL);
7896 }
7897
7898 static int osd_process_config(const struct lu_env *env,
7899                               struct lu_device *d, struct lustre_cfg *cfg)
7900 {
7901         struct osd_device *o = osd_dev(d);
7902         ssize_t count;
7903         int rc;
7904
7905         ENTRY;
7906
7907         switch (cfg->lcfg_command) {
7908         case LCFG_SETUP:
7909                 rc = osd_mount(env, o, cfg);
7910                 break;
7911         case LCFG_CLEANUP:
7912                 /*
7913                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7914                  * that may happend if hit failure during mount process.
7915                  */
7916                 osd_index_backup(env, o, false);
7917                 lu_dev_del_linkage(d->ld_site, d);
7918                 rc = osd_shutdown(env, o);
7919                 break;
7920         case LCFG_PARAM:
7921                 LASSERT(&o->od_dt_dev);
7922                 count  = class_modify_config(cfg, PARAM_OSD,
7923                                              &o->od_dt_dev.dd_kobj);
7924                 if (count < 0)
7925                         count = class_modify_config(cfg, PARAM_OST,
7926                                                     &o->od_dt_dev.dd_kobj);
7927                 rc = count > 0 ? 0 : count;
7928                 break;
7929         case LCFG_PRE_CLEANUP:
7930                 osd_scrub_stop(o);
7931                 osd_index_backup(env, o,
7932                                  o->od_index_backup_policy != LIBP_NONE);
7933                 rc = 0;
7934                 break;
7935         default:
7936                 rc = -ENOSYS;
7937         }
7938
7939         RETURN(rc);
7940 }
7941
7942 static int osd_recovery_complete(const struct lu_env *env,
7943                                  struct lu_device *d)
7944 {
7945         struct osd_device *osd = osd_dev(d);
7946         int rc = 0;
7947
7948         ENTRY;
7949
7950         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7951                 RETURN(0);
7952
7953         /*
7954          * start qsd instance on recovery completion, this notifies the quota
7955          * slave code that we are about to process new requests now
7956          */
7957         rc = qsd_start(env, osd->od_quota_slave_dt);
7958         if (rc == 0 && osd->od_quota_slave_md != NULL)
7959                 rc = qsd_start(env, osd->od_quota_slave_md);
7960
7961         RETURN(rc);
7962 }
7963
7964 /*
7965  * we use exports to track all osd users
7966  */
7967 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7968                            struct obd_device *obd, struct obd_uuid *cluuid,
7969                            struct obd_connect_data *data, void *localdata)
7970 {
7971         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7972         struct lustre_handle conn;
7973         int rc;
7974
7975         ENTRY;
7976
7977         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7978
7979         rc = class_connect(&conn, obd, cluuid);
7980         if (rc)
7981                 RETURN(rc);
7982
7983         *exp = class_conn2export(&conn);
7984
7985         spin_lock(&osd->od_osfs_lock);
7986         osd->od_connects++;
7987         spin_unlock(&osd->od_osfs_lock);
7988
7989         RETURN(0);
7990 }
7991
7992 /*
7993  * once last export (we don't count self-export) disappeared
7994  * osd can be released
7995  */
7996 static int osd_obd_disconnect(struct obd_export *exp)
7997 {
7998         struct obd_device *obd = exp->exp_obd;
7999         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8000         int rc, release = 0;
8001
8002         ENTRY;
8003
8004         /* Only disconnect the underlying layers on the final disconnect. */
8005         spin_lock(&osd->od_osfs_lock);
8006         osd->od_connects--;
8007         if (osd->od_connects == 0)
8008                 release = 1;
8009         spin_unlock(&osd->od_osfs_lock);
8010
8011         rc = class_disconnect(exp); /* bz 9811 */
8012
8013         if (rc == 0 && release)
8014                 class_manual_cleanup(obd);
8015         RETURN(rc);
8016 }
8017
8018 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8019                        struct lu_device *dev)
8020 {
8021         struct osd_device *osd = osd_dev(dev);
8022         struct lr_server_data *lsd =
8023                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8024         int result = 0;
8025
8026         ENTRY;
8027
8028         if (osd->od_quota_slave_md != NULL) {
8029                 /* set up quota slave objects for inode */
8030                 result = qsd_prepare(env, osd->od_quota_slave_md);
8031                 if (result != 0)
8032                         RETURN(result);
8033         }
8034
8035         if (osd->od_quota_slave_dt != NULL) {
8036                 /* set up quota slave objects for block */
8037                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8038                 if (result != 0)
8039                         RETURN(result);
8040         }
8041
8042
8043         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8044 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8045                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8046                         osd->od_index_in_idif = 1;
8047                 } else {
8048                         osd->od_index_in_idif = 0;
8049                         result = osd_register_proc_index_in_idif(osd);
8050                         if (result != 0)
8051                                 RETURN(result);
8052                 }
8053 #else
8054                 osd->od_index_in_idif = 1;
8055 #endif
8056         }
8057
8058         result = osd_fid_init(env, osd);
8059
8060         RETURN(result);
8061 }
8062
8063 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8064                          struct lu_fid *fid, struct md_op_data *op_data)
8065 {
8066         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8067
8068         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8069 }
8070
8071 static const struct lu_object_operations osd_lu_obj_ops = {
8072         .loo_object_init      = osd_object_init,
8073         .loo_object_delete    = osd_object_delete,
8074         .loo_object_release   = osd_object_release,
8075         .loo_object_free      = osd_object_free,
8076         .loo_object_print     = osd_object_print,
8077         .loo_object_invariant = osd_object_invariant
8078 };
8079
8080 const struct lu_device_operations osd_lu_ops = {
8081         .ldo_object_alloc      = osd_object_alloc,
8082         .ldo_process_config    = osd_process_config,
8083         .ldo_recovery_complete = osd_recovery_complete,
8084         .ldo_prepare           = osd_prepare,
8085 };
8086
8087 static const struct lu_device_type_operations osd_device_type_ops = {
8088         .ldto_init = osd_type_init,
8089         .ldto_fini = osd_type_fini,
8090
8091         .ldto_start = osd_type_start,
8092         .ldto_stop  = osd_type_stop,
8093
8094         .ldto_device_alloc = osd_device_alloc,
8095         .ldto_device_free  = osd_device_free,
8096
8097         .ldto_device_init = osd_device_init,
8098         .ldto_device_fini = osd_device_fini
8099 };
8100
8101 static struct lu_device_type osd_device_type = {
8102         .ldt_tags     = LU_DEVICE_DT,
8103         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8104         .ldt_ops      = &osd_device_type_ops,
8105         .ldt_ctx_tags = LCT_LOCAL,
8106 };
8107
8108 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8109 {
8110         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8111         struct super_block *sb = osd_sb(osd);
8112
8113         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8114 }
8115
8116 /*
8117  * lprocfs legacy support.
8118  */
8119 static struct obd_ops osd_obd_device_ops = {
8120         .o_owner = THIS_MODULE,
8121         .o_connect      = osd_obd_connect,
8122         .o_disconnect   = osd_obd_disconnect,
8123         .o_fid_alloc    = osd_fid_alloc,
8124         .o_health_check = osd_health_check,
8125 };
8126
8127 static ssize_t track_declares_assert_show(struct kobject *kobj,
8128                                    struct attribute *attr,
8129                                    char *buf)
8130 {
8131         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8132 }
8133
8134 static ssize_t track_declares_assert_store(struct kobject *kobj,
8135                                            struct attribute *attr,
8136                                            const char *buffer, size_t count)
8137 {
8138         bool track_declares_assert;
8139         int rc;
8140
8141         rc = kstrtobool(buffer, &track_declares_assert);
8142         if (rc)
8143                 return rc;
8144
8145         ldiskfs_track_declares_assert = track_declares_assert;
8146
8147         return count;
8148 }
8149 LUSTRE_RW_ATTR(track_declares_assert);
8150
8151 static int __init osd_init(void)
8152 {
8153         struct kobject *kobj;
8154         int rc;
8155
8156         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8157 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8158         /* please, try to keep osd_thread_info smaller than a page */
8159         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8160 #endif
8161
8162         osd_oi_mod_init();
8163
8164         rc = lu_kmem_init(ldiskfs_caches);
8165         if (rc)
8166                 return rc;
8167
8168 #ifdef CONFIG_KALLSYMS
8169         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8170         priv_dev_check_rdonly =
8171                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8172 #endif
8173
8174         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8175                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8176         if (rc) {
8177                 lu_kmem_fini(ldiskfs_caches);
8178                 return rc;
8179         }
8180
8181         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8182         if (kobj) {
8183                 rc = sysfs_create_file(kobj,
8184                                        &lustre_attr_track_declares_assert.attr);
8185                 kobject_put(kobj);
8186                 if (rc) {
8187                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8188                         rc = 0;
8189                 }
8190         }
8191         return rc;
8192 }
8193
8194 static void __exit osd_exit(void)
8195 {
8196         struct kobject *kobj;
8197
8198         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8199         if (kobj) {
8200                 sysfs_remove_file(kobj,
8201                                   &lustre_attr_track_declares_assert.attr);
8202                 kobject_put(kobj);
8203         }
8204         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8205         lu_kmem_fini(ldiskfs_caches);
8206 }
8207
8208 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8209 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8210 MODULE_VERSION(LUSTRE_VERSION_STRING);
8211 MODULE_LICENSE("GPL");
8212
8213 module_init(osd_init);
8214 module_exit(osd_exit);