lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #ifdef HAVE_UIDGID_HEADER
  46 # include <linux/uidgid.h>
  47 #endif
  48
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/types.h>
  51 /* prerequisite for linux/xattr.h */
  52 #include <linux/fs.h>
  53 /* XATTR_{REPLACE,CREATE} */
  54 #include <linux/xattr.h>
  55
  56 #include <ldiskfs/ldiskfs.h>
  57 #include <ldiskfs/xattr.h>
  58 #include <ldiskfs/ldiskfs_extents.h>
  59 #undef ENTRY
  60 /*
  61  * struct OBD_{ALLOC,FREE}*()
  62  * OBD_FAIL_CHECK
  63  */
  64 #include <obd_support.h>
  65 /* struct ptlrpc_thread */
  66 #include <lustre_net.h>
  67 #include <lustre_fid.h>
  68 /* process_config */
  69 #include <uapi/linux/lustre/lustre_param.h>
  70
  71 #include "osd_internal.h"
  72 #include "osd_dynlocks.h"
  73
  74 /* llo_* api support */
  75 #include <md_object.h>
  76 #include <lustre_quota.h>
  77
  78 #include <lustre_linkea.h>
  79
  80 /* Maximum EA size is limited by LNET_MTU for remote objects */
  81 #define OSD_MAX_EA_SIZE 1048364
  82
  83 int ldiskfs_pdo = 1;
  84 module_param(ldiskfs_pdo, int, 0644);
  85 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  86
  87 int ldiskfs_track_declares_assert;
  88 module_param(ldiskfs_track_declares_assert, int, 0644);
  89 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  90
  91 /* Slab to allocate dynlocks */
  92 struct kmem_cache *dynlock_cachep;
  93
  94 /* Slab to allocate osd_it_ea */
  95 struct kmem_cache *osd_itea_cachep;
  96
  97 static struct lu_kmem_descr ldiskfs_caches[] = {
  98         {
  99                 .ckd_cache = &dynlock_cachep,
 100                 .ckd_name  = "dynlock_cache",
 101                 .ckd_size  = sizeof(struct dynlock_handle)
 102         },
 103         {
 104                 .ckd_cache = &osd_itea_cachep,
 105                 .ckd_name  = "osd_itea_cache",
 106                 .ckd_size  = sizeof(struct osd_it_ea)
 107         },
 108         {
 109                 .ckd_cache = NULL
 110         }
 111 };
 112
 113 static const char dot[] = ".";
 114 static const char dotdot[] = "..";
 115
 116 static const struct lu_object_operations      osd_lu_obj_ops;
 117 static const struct dt_object_operations      osd_obj_ops;
 118 static const struct dt_object_operations      osd_obj_otable_it_ops;
 119 static const struct dt_index_operations       osd_index_iam_ops;
 120 static const struct dt_index_operations       osd_index_ea_ops;
 121
 122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 123                           const struct lu_fid *fid);
 124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 125                                                 struct osd_device *osd);
 126
 127 int osd_trans_declare_op2rb[] = {
 128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 139 };
 140
 141 static int osd_has_index(const struct osd_object *obj)
 142 {
 143         return obj->oo_dt.do_index_ops != NULL;
 144 }
 145
 146 static int osd_object_invariant(const struct lu_object *l)
 147 {
 148         return osd_invariant(osd_obj(l));
 149 }
 150
 151 /*
 152  * Concurrency: doesn't matter
 153  */
 154 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 155 {
 156         struct osd_thread_info *oti = osd_oti_get(env);
 157
 158         return oti->oti_w_locks > 0 && o->oo_owner == env;
 159 }
 160
 161 /*
 162  * Concurrency: doesn't access mutable data
 163  */
 164 static int osd_root_get(const struct lu_env *env,
 165                         struct dt_device *dev, struct lu_fid *f)
 166 {
 167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 168         return 0;
 169 }
 170
 171 /*
 172  * the following set of functions are used to maintain per-thread
 173  * cache of FID->ino mapping. this mechanism is needed to resolve
 174  * FID to inode at dt_insert() which in turn stores ino in the
 175  * directory entries to keep ldiskfs compatible with ext[34].
 176  * due to locking-originated restrictions we can't lookup ino
 177  * using LU cache (deadlock is possible). lookup using OI is quite
 178  * expensive. so instead we maintain this cache and methods like
 179  * dt_create() fill it. so in the majority of cases dt_insert() is
 180  * able to find needed mapping in lockless manner.
 181  */
 182 static struct osd_idmap_cache *
 183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 184              const struct lu_fid *fid)
 185 {
 186         struct osd_thread_info *oti = osd_oti_get(env);
 187         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 188         int i;
 189
 190         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 191                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 192                         continue;
 193                 if (idc[i].oic_dev != osd)
 194                         continue;
 195
 196                 return idc + i;
 197         }
 198
 199         return NULL;
 200 }
 201
 202 static struct osd_idmap_cache *
 203 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 204             const struct lu_fid *fid)
 205 {
 206         struct osd_thread_info *oti   = osd_oti_get(env);
 207         struct osd_idmap_cache *idc;
 208         int i;
 209
 210         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 211                 i = oti->oti_ins_cache_size * 2;
 212                 if (i == 0)
 213                         i = OSD_INS_CACHE_SIZE;
 214                 OBD_ALLOC(idc, sizeof(*idc) * i);
 215                 if (idc == NULL)
 216                         return ERR_PTR(-ENOMEM);
 217                 if (oti->oti_ins_cache != NULL) {
 218                         memcpy(idc, oti->oti_ins_cache,
 219                                oti->oti_ins_cache_used * sizeof(*idc));
 220                         OBD_FREE(oti->oti_ins_cache,
 221                                  oti->oti_ins_cache_used * sizeof(*idc));
 222                 }
 223                 oti->oti_ins_cache = idc;
 224                 oti->oti_ins_cache_size = i;
 225         }
 226
 227         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 228         idc->oic_fid = *fid;
 229         idc->oic_dev = osd;
 230         idc->oic_lid.oii_ino = 0;
 231         idc->oic_lid.oii_gen = 0;
 232         idc->oic_remote = 0;
 233
 234         return idc;
 235 }
 236
 237 /*
 238  * lookup mapping for the given fid in the cache, initialize a
 239  * new one if not found. the initialization checks whether the
 240  * object is local or remote. for local objects, OI is used to
 241  * learn ino/generation. the function is used when the caller
 242  * has no information about the object, e.g. at dt_insert().
 243  */
 244 static struct osd_idmap_cache *
 245 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 246                      const struct lu_fid *fid)
 247 {
 248         struct osd_idmap_cache *idc;
 249         int rc;
 250
 251         idc = osd_idc_find(env, osd, fid);
 252         LASSERT(!IS_ERR(idc));
 253         if (idc != NULL)
 254                 return idc;
 255
 256         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 257                osd->od_svname, PFID(fid));
 258
 259         /* new mapping is needed */
 260         idc = osd_idc_add(env, osd, fid);
 261         if (IS_ERR(idc)) {
 262                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 263                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 264                 return idc;
 265         }
 266
 267         /* initialize it */
 268         rc = osd_remote_fid(env, osd, fid);
 269         if (unlikely(rc < 0))
 270                 return ERR_PTR(rc);
 271
 272         if (rc == 0) {
 273                 /* the object is local, lookup in OI */
 274                 /* XXX: probably cheaper to lookup in LU first? */
 275                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 276                                    &idc->oic_lid, 0);
 277                 if (unlikely(rc < 0)) {
 278                         CERROR("can't lookup: rc = %d\n", rc);
 279                         return ERR_PTR(rc);
 280                 }
 281         } else {
 282                 /* the object is remote */
 283                 idc->oic_remote = 1;
 284         }
 285
 286         return idc;
 287 }
 288
 289 /*
 290  * lookup mapping for given FID and fill it from the given object.
 291  * the object is lolcal by definition.
 292  */
 293 static int osd_idc_find_and_init(const struct lu_env *env,
 294                                  struct osd_device *osd,
 295                                  struct osd_object *obj)
 296 {
 297         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 298         struct osd_idmap_cache *idc;
 299
 300         idc = osd_idc_find(env, osd, fid);
 301         LASSERT(!IS_ERR(idc));
 302         if (idc != NULL) {
 303                 if (obj->oo_inode == NULL)
 304                         return 0;
 305                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 306                         LASSERT(idc->oic_lid.oii_ino == 0);
 307                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 308                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 309                 }
 310                 return 0;
 311         }
 312
 313         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 314                osd->od_svname, PFID(fid));
 315
 316         /* new mapping is needed */
 317         idc = osd_idc_add(env, osd, fid);
 318         if (IS_ERR(idc)) {
 319                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 320                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 321                 return PTR_ERR(idc);
 322         }
 323
 324         if (obj->oo_inode != NULL) {
 325                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 326                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 327         }
 328         return 0;
 329 }
 330
 331 /*
 332  * OSD object methods.
 333  */
 334
 335 /*
 336  * Concurrency: no concurrent access is possible that early in object
 337  * life-cycle.
 338  */
 339 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 340                                           const struct lu_object_header *hdr,
 341                                           struct lu_device *d)
 342 {
 343         struct osd_object *mo;
 344
 345         OBD_ALLOC_PTR(mo);
 346         if (mo != NULL) {
 347                 struct lu_object *l;
 348                 struct lu_object_header *h;
 349                 struct osd_device *o = osd_dev(d);
 350
 351                 l = &mo->oo_dt.do_lu;
 352                 if (unlikely(o->od_in_init)) {
 353                         OBD_ALLOC_PTR(h);
 354                         if (!h) {
 355                                 OBD_FREE_PTR(mo);
 356                                 return NULL;
 357                         }
 358
 359                         lu_object_header_init(h);
 360                         lu_object_init(l, h, d);
 361                         lu_object_add_top(h, l);
 362                         mo->oo_header = h;
 363                 } else {
 364                         dt_object_init(&mo->oo_dt, NULL, d);
 365                         mo->oo_header = NULL;
 366                 }
 367
 368                 mo->oo_dt.do_ops = &osd_obj_ops;
 369                 l->lo_ops = &osd_lu_obj_ops;
 370                 init_rwsem(&mo->oo_sem);
 371                 init_rwsem(&mo->oo_ext_idx_sem);
 372                 spin_lock_init(&mo->oo_guard);
 373                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 374                 return l;
 375         }
 376         return NULL;
 377 }
 378
 379 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 380                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 381 {
 382         int rc;
 383
 384         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 385                              (void *)loa, sizeof(*loa));
 386         if (rc > 0) {
 387                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 388
 389                 if (rc < sizeof(*lma))
 390                         return -EINVAL;
 391
 392                 rc = 0;
 393                 lustre_loa_swab(loa, true);
 394                 /* Check LMA compatibility */
 395                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 396                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 397                               "for fid = "DFID", ino = %lu\n",
 398                               osd_ino2name(inode),
 399                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 400                               PFID(&lma->lma_self_fid), inode->i_ino);
 401                         rc = -EOPNOTSUPP;
 402                 }
 403         } else if (rc == 0) {
 404                 rc = -ENODATA;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /*
 411  * retrieve object from backend ext fs.
 412  **/
 413 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 414                        struct osd_inode_id *id)
 415 {
 416         int rc;
 417         struct inode *inode = NULL;
 418
 419         /*
 420          * if we look for an inode withing a running
 421          * transaction, then we risk to deadlock
 422          * osd_dirent_check_repair() breaks this
 423          */
 424          /* LASSERT(current->journal_info == NULL); */
 425
 426         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 427         if (IS_ERR(inode)) {
 428                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 429                        id->oii_ino, PTR_ERR(inode));
 430         } else if (id->oii_gen != OSD_OII_NOGEN &&
 431                    inode->i_generation != id->oii_gen) {
 432                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 433                        "i_generation = %u\n",
 434                        id->oii_ino, id->oii_gen, inode->i_generation);
 435                 iput(inode);
 436                 inode = ERR_PTR(-ESTALE);
 437         } else if (inode->i_nlink == 0) {
 438                 /*
 439                  * due to parallel readdir and unlink,
 440                  * we can have dead inode here.
 441                  */
 442                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 443                 iput(inode);
 444                 inode = ERR_PTR(-ESTALE);
 445         } else if (is_bad_inode(inode)) {
 446                 CWARN("%s: bad inode: ino = %u\n",
 447                       osd_dev2name(dev), id->oii_ino);
 448                 iput(inode);
 449                 inode = ERR_PTR(-ENOENT);
 450         } else if ((rc = osd_attach_jinode(inode))) {
 451                 iput(inode);
 452                 inode = ERR_PTR(rc);
 453         } else {
 454                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 455                 if (id->oii_gen == OSD_OII_NOGEN)
 456                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 457
 458                 /*
 459                  * Do not update file c/mtime in ldiskfs.
 460                  * NB: we don't have any lock to protect this because we don't
 461                  * have reference on osd_object now, but contention with
 462                  * another lookup + attr_set can't happen in the tiny window
 463                  * between if (...) and set S_NOCMTIME.
 464                  */
 465                 if (!(inode->i_flags & S_NOCMTIME))
 466                         inode->i_flags |= S_NOCMTIME;
 467         }
 468         return inode;
 469 }
 470
 471 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 472                           handle_t *handle, struct dentry *child,
 473                           struct inode *inode, struct htree_lock *hlock)
 474 {
 475         int rc, rc2;
 476
 477         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 478         if (rc == -ENOBUFS || rc == -ENOSPC) {
 479                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 480                 struct inode *parent = child->d_parent->d_inode;
 481                 struct lu_fid *fid = NULL;
 482
 483                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 484                 if (!rc2) {
 485                         fid = &loa->loa_lma.lma_self_fid;
 486                 } else if (rc2 == -ENODATA) {
 487                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 488                                 fid = &info->oti_fid3;
 489                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 490                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 491                                 fid = &info->oti_fid3;
 492                                 lu_igif_build(fid, parent->i_ino,
 493                                               parent->i_generation);
 494                         }
 495                 }
 496
 497                 if (fid != NULL)
 498                         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 = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 563         if (IS_ERR(inode)) {
 564                 rc = PTR_ERR(inode);
 565                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 566                         goto check_oi;
 567
 568                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 569                        PFID(fid), id->oii_ino, rc);
 570                 GOTO(put, rc);
 571         }
 572
 573         if (is_bad_inode(inode)) {
 574                 rc = -ENOENT;
 575                 if (!trusted)
 576                         goto check_oi;
 577
 578                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 579                        PFID(fid), id->oii_ino);
 580                 GOTO(put, rc);
 581         }
 582
 583         if (id->oii_gen != OSD_OII_NOGEN &&
 584             inode->i_generation != id->oii_gen) {
 585                 rc = -ESTALE;
 586                 if (!trusted)
 587                         goto check_oi;
 588
 589                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 590                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 591                        id->oii_ino, id->oii_gen, inode->i_generation);
 592                 GOTO(put, rc);
 593         }
 594
 595         if (inode->i_nlink == 0) {
 596                 rc = -ENOENT;
 597                 if (!trusted)
 598                         goto check_oi;
 599
 600                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 601                        PFID(fid), id->oii_ino);
 602                 GOTO(put, rc);
 603         }
 604
 605         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 606
 607 check_oi:
 608         if (rc != 0) {
 609                 __u32 saved_ino = id->oii_ino;
 610                 __u32 saved_gen = id->oii_gen;
 611
 612                 LASSERT(!trusted);
 613                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 614
 615                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 616                 /*
 617                  * XXX: There are four possible cases:
 618                  *      1. rc = 0.
 619                  *         Backup/restore caused the OI invalid.
 620                  *      2. rc = 0.
 621                  *         Someone unlinked the object but NOT removed
 622                  *         the OI mapping, such as mount target device
 623                  *         as ldiskfs, and modify something directly.
 624                  *      3. rc = -ENOENT.
 625                  *         Someone just removed the object between the
 626                  *         former oi_lookup and the iget. It is normal.
 627                  *      4. Other failure cases.
 628                  *
 629                  *      Generally, when the device is mounted, it will
 630                  *      auto check whether the system is restored from
 631                  *      file-level backup or not. We trust such detect
 632                  *      to distinguish the 1st case from the 2nd case:
 633                  *      if the OI files are consistent but may contain
 634                  *      stale OI mappings because of case 2, if iget()
 635                  *      returns -ENOENT or -ESTALE, then it should be
 636                  *      the case 2.
 637                  */
 638                 if (rc != 0)
 639                         /*
 640                          * If the OI mapping was in OI file before the
 641                          * osd_iget_check(), but now, it is disappear,
 642                          * then it must be removed by race. That is a
 643                          * normal race case.
 644                          */
 645                         GOTO(put, rc);
 646
 647                 /*
 648                  * It is the OI scrub updated the OI mapping by race.
 649                  * The new OI mapping must be valid.
 650                  */
 651                 if (saved_ino != id->oii_ino ||
 652                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 653                         if (!IS_ERR(inode))
 654                                 iput(inode);
 655
 656                         trusted = true;
 657                         goto again;
 658                 }
 659
 660                 if (IS_ERR(inode)) {
 661                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 662                             SF_INCONSISTENT)
 663                                 /*
 664                                  * It still can be the case 2, but we cannot
 665                                  * distinguish it from the case 1. So return
 666                                  * -EREMCHG to block current operation until
 667                                  *  OI scrub rebuilt the OI mappings.
 668                                  */
 669                                 rc = -EREMCHG;
 670                         else
 671                                 rc = -ENOENT;
 672
 673                         GOTO(put, rc);
 674                 }
 675
 676                 if (inode->i_generation == id->oii_gen)
 677                         rc = -ENOENT;
 678                 else
 679                         rc = -EREMCHG;
 680         } else {
 681                 if (id->oii_gen == OSD_OII_NOGEN)
 682                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 683
 684                 /*
 685                  * Do not update file c/mtime in ldiskfs.
 686                  * NB: we don't have any lock to protect this because we don't
 687                  * have reference on osd_object now, but contention with
 688                  * another lookup + attr_set can't happen in the tiny window
 689                  * between if (...) and set S_NOCMTIME.
 690                  */
 691                 if (!(inode->i_flags & S_NOCMTIME))
 692                         inode->i_flags |= S_NOCMTIME;
 693         }
 694
 695         GOTO(put, rc);
 696
 697 put:
 698         if (rc != 0) {
 699                 if (!IS_ERR(inode))
 700                         iput(inode);
 701
 702                 inode = ERR_PTR(rc);
 703         }
 704
 705         return inode;
 706 }
 707
 708 /**
 709  * \retval +v: new filter_fid does not contain self-fid
 710  * \retval 0:  filter_fid_18_23, contains self-fid
 711  * \retval -v: other failure cases
 712  */
 713 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 714                  struct dentry *dentry, struct lu_fid *fid)
 715 {
 716         struct filter_fid *ff = &info->oti_ff;
 717         struct ost_id *ostid = &info->oti_ostid;
 718         int rc;
 719
 720         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 721         if (rc == sizeof(struct filter_fid_18_23)) {
 722                 struct filter_fid_18_23 *ff_old = (void *)ff;
 723
 724                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 725                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 726                 /*
 727                  * XXX: use 0 as the index for compatibility, the caller will
 728                  * handle index related issues when necessary.
 729                  */
 730                 if (!rc)
 731                         ostid_to_fid(fid, ostid, 0);
 732         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 733                 rc = 1;
 734         } else if (rc >= 0) {
 735                 rc = -EINVAL;
 736         }
 737
 738         return rc;
 739 }
 740
 741 static int osd_lma_self_repair(struct osd_thread_info *info,
 742                                struct osd_device *osd, struct inode *inode,
 743                                const struct lu_fid *fid, __u32 compat)
 744 {
 745         handle_t *jh;
 746         int rc;
 747
 748         LASSERT(current->journal_info == NULL);
 749
 750         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 751                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 752         if (IS_ERR(jh)) {
 753                 rc = PTR_ERR(jh);
 754                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 755                       osd_name(osd), rc);
 756                 return rc;
 757         }
 758
 759         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 760         if (rc != 0)
 761                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 762                       osd_name(osd), rc);
 763         ldiskfs_journal_stop(jh);
 764         return rc;
 765 }
 766
 767 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 768 {
 769         struct osd_thread_info *info = osd_oti_get(env);
 770         struct osd_device *osd = osd_obj2dev(obj);
 771         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 772         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 773         struct inode *inode = obj->oo_inode;
 774         struct dentry *dentry = &info->oti_obj_dentry;
 775         struct lu_fid *fid = NULL;
 776         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 777         int rc;
 778
 779         ENTRY;
 780
 781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 782                              (void *)loa, sizeof(*loa));
 783         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 784                 fid = &lma->lma_self_fid;
 785                 rc = osd_get_idif(info, inode, dentry, fid);
 786                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 787                         /*
 788                          * For the given OST-object, if it has neither LMA nor
 789                          * FID in XATTR_NAME_FID, then the given FID (which is
 790                          * contained in the @obj, from client RPC for locating
 791                          * the OST-object) is trusted. We use it to generate
 792                          * the LMA.
 793                          */
 794                         osd_lma_self_repair(info, osd, inode, rfid,
 795                                             LMAC_FID_ON_OST);
 796                         RETURN(0);
 797                 }
 798         }
 799
 800         if (rc < 0)
 801                 RETURN(rc);
 802
 803         if (rc > 0) {
 804                 rc = 0;
 805                 lustre_lma_swab(lma);
 806                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 807                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 808                               S_ISREG(inode->i_mode)))) {
 809                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 810                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 811                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 812                               PFID(rfid), inode->i_ino);
 813                         rc = -EOPNOTSUPP;
 814                 } else {
 815                         fid = &lma->lma_self_fid;
 816                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 817                             osd->od_is_ost)
 818                                 obj->oo_pfid_in_lma = 1;
 819                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 820                             !osd->od_is_ost)
 821                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 822                 }
 823         }
 824
 825         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 826                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 827                         struct ost_id   *oi   = &info->oti_ostid;
 828                         struct lu_fid   *fid1 = &info->oti_fid3;
 829                         __u32            idx  = fid_idif_ost_idx(rfid);
 830
 831                         /*
 832                          * For old IDIF, the OST index is not part of the IDIF,
 833                          * Means that different OSTs may have the same IDIFs.
 834                          * Under such case, we need to make some compatible
 835                          * check to make sure to trigger OI scrub properly.
 836                          */
 837                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 838                                 /* Given @rfid is new, LMA is old. */
 839                                 fid_to_ostid(fid, oi);
 840                                 ostid_to_fid(fid1, oi, idx);
 841                                 if (lu_fid_eq(fid1, rfid)) {
 842                                         if (osd->od_index_in_idif)
 843                                                 osd_lma_self_repair(info, osd,
 844                                                         inode, rfid,
 845                                                         LMAC_FID_ON_OST);
 846                                         RETURN(0);
 847                                 }
 848                         }
 849                 }
 850
 851                 rc = -EREMCHG;
 852         }
 853
 854         RETURN(rc);
 855 }
 856
 857 struct osd_check_lmv_buf {
 858 #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
2202         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2203
2204         if (ldiskfs_has_feature_extents(sb))
2205                 sfs->os_maxbytes = sb->s_maxbytes;
2206         else
2207                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2208
2209         /*
2210          * Reserve some space so to avoid fragmenting the filesystem too much.
2211          * Fragmentation not only impacts performance, but can also increase
2212          * metadata overhead significantly, causing grant calculation to be
2213          * wrong.
2214          *
2215          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2216          */
2217         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2218         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2219         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2220                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2221
2222         sfs->os_blocks -= reserved;
2223         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2224         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2225
2226 out:
2227         if (unlikely(env == NULL))
2228                 OBD_FREE_PTR(ksfs);
2229         return result;
2230 }
2231
2232 /**
2233  * Estimate space needed for file creations. We assume the largest filename
2234  * which is 2^64 - 1, hence a filename of 20 chars.
2235  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2236  */
2237 #ifdef __LDISKFS_DIR_REC_LEN
2238 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2239 #else
2240 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2241 #endif
2242
2243 /*
2244  * Concurrency: doesn't access mutable data.
2245  */
2246 static void osd_conf_get(const struct lu_env *env,
2247                          const struct dt_device *dev,
2248                          struct dt_device_param *param)
2249 {
2250         struct osd_device *d = osd_dt_dev(dev);
2251         struct super_block *sb = osd_sb(d);
2252         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2253         const char *name;
2254         int ea_overhead;
2255
2256         /*
2257          * XXX should be taken from not-yet-existing fs abstraction layer.
2258          */
2259         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2260         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2261         param->ddp_symlink_max  = sb->s_blocksize;
2262         param->ddp_mount_type   = LDD_MT_LDISKFS;
2263         if (ldiskfs_has_feature_extents(sb))
2264                 param->ddp_maxbytes = sb->s_maxbytes;
2265         else
2266                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2267         /*
2268          * inode are statically allocated, so per-inode space consumption
2269          * is the space consumed by the directory entry
2270          */
2271         param->ddp_inodespace     = PER_OBJ_USAGE;
2272         /*
2273          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2274          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2275          * maximum length to not under-count the actual number of extents
2276          * needed for writing a file if there are sub-optimal block allocations.
2277          */
2278         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2279         /* worst-case extent insertion metadata overhead */
2280         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2281         param->ddp_mntopts = 0;
2282         if (test_opt(sb, XATTR_USER))
2283                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2284         if (test_opt(sb, POSIX_ACL))
2285                 param->ddp_mntopts |= MNTOPT_ACL;
2286
2287         /*
2288          * LOD might calculate the max stripe count based on max_ea_size,
2289          * so we need take account in the overhead as well,
2290          * xattr_header + magic + xattr_entry_head
2291          */
2292         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2293                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2294
2295 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2296         if (ldiskfs_has_feature_ea_inode(sb))
2297                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2298                                                                 ea_overhead;
2299         else
2300 #endif
2301                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2302
2303         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE - ea_overhead)
2304                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE - ea_overhead;
2305
2306         /*
2307          * Preferred RPC size for efficient disk IO.  4MB shows good
2308          * all-around performance for ldiskfs, but use bigalloc chunk size
2309          * by default if larger.
2310          */
2311 #if defined(LDISKFS_CLUSTER_SIZE)
2312         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2313                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2314         else
2315 #endif
2316                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2317
2318         param->ddp_t10_cksum_type = 0;
2319         if (bi) {
2320                 unsigned short interval = blk_integrity_interval(bi);
2321                 name = blk_integrity_name(bi);
2322                 /*
2323                  * Expected values:
2324                  * T10-DIF-TYPE1-CRC
2325                  * T10-DIF-TYPE3-CRC
2326                  * T10-DIF-TYPE1-IP
2327                  * T10-DIF-TYPE3-IP
2328                  */
2329                 if (strncmp(name, "T10-DIF-TYPE",
2330                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2331                         /* also skip "1/3-" at end */
2332                         const int type_off = sizeof("T10-DIF-TYPE.");
2333                         char type_number = name[type_off - 2];
2334
2335                         if (interval != 512 && interval != 4096) {
2336                                 CERROR("%s: unsupported T10PI sector size %u\n",
2337                                        d->od_svname, interval);
2338                         } else if (type_number != '1' && type_number != '3') {
2339                                 CERROR("%s: unsupported T10PI type %s\n",
2340                                        d->od_svname, name);
2341                         } else if (strcmp(name + type_off, "CRC") == 0) {
2342                                 d->od_t10_type = type_number == '1' ?
2343                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2344                                 param->ddp_t10_cksum_type = interval == 512 ?
2345                                         OBD_CKSUM_T10CRC512 :
2346                                         OBD_CKSUM_T10CRC4K;
2347                         } else if (strcmp(name + type_off, "IP") == 0) {
2348                                 d->od_t10_type = type_number == '1' ?
2349                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2350                                 param->ddp_t10_cksum_type = interval == 512 ?
2351                                         OBD_CKSUM_T10IP512 :
2352                                         OBD_CKSUM_T10IP4K;
2353                         } else {
2354                                 CERROR("%s: unsupported checksum type of "
2355                                        "T10PI type '%s'",
2356                                        d->od_svname, name);
2357                         }
2358
2359                 } else {
2360                         CERROR("%s: unsupported T10PI type '%s'",
2361                                d->od_svname, name);
2362                 }
2363         }
2364 }
2365
2366 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2367 {
2368         return osd_sb(osd_dt_dev(d));
2369 }
2370
2371 /*
2372  * Concurrency: shouldn't matter.
2373  */
2374 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2375 {
2376         int rc;
2377         struct super_block *s = osd_sb(osd_dt_dev(d));
2378         ENTRY;
2379
2380         down_read(&s->s_umount);
2381         rc = s->s_op->sync_fs(s, 1);
2382         up_read(&s->s_umount);
2383
2384         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2385                rc);
2386
2387         return rc;
2388 }
2389
2390 /**
2391  * Start commit for OSD device.
2392  *
2393  * An implementation of dt_commit_async method for OSD device.
2394  * Asychronously starts underlayng fs sync and thereby a transaction
2395  * commit.
2396  *
2397  * \param env environment
2398  * \param d dt device
2399  *
2400  * \see dt_device_operations
2401  */
2402 static int osd_commit_async(const struct lu_env *env,
2403                             struct dt_device *d)
2404 {
2405         struct super_block *s = osd_sb(osd_dt_dev(d));
2406         int rc;
2407
2408         ENTRY;
2409
2410         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2411         down_read(&s->s_umount);
2412         rc = s->s_op->sync_fs(s, 0);
2413         up_read(&s->s_umount);
2414
2415         RETURN(rc);
2416 }
2417
2418 /* Our own copy of the set readonly functions if present, or NU if not. */
2419 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2420 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2421 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2422
2423 /*
2424  * Concurrency: shouldn't matter.
2425  */
2426 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2427 {
2428         struct super_block *sb = osd_sb(osd_dt_dev(d));
2429         struct block_device *dev = sb->s_bdev;
2430         int rc = -EOPNOTSUPP;
2431
2432         ENTRY;
2433
2434         if (priv_dev_set_rdonly) {
2435                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2436
2437                 rc = 0;
2438                 CERROR("*** setting %s read-only ***\n",
2439                        osd_dt_dev(d)->od_svname);
2440
2441                 if (sb->s_op->freeze_fs) {
2442                         rc = sb->s_op->freeze_fs(sb);
2443                         if (rc)
2444                                 goto out;
2445                 }
2446
2447                 if (jdev && (jdev != dev)) {
2448                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2449                                (long)jdev);
2450                         priv_dev_set_rdonly(jdev);
2451                 }
2452                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2453                 priv_dev_set_rdonly(dev);
2454
2455                 if (sb->s_op->unfreeze_fs)
2456                         sb->s_op->unfreeze_fs(sb);
2457         }
2458
2459 out:
2460         if (rc)
2461                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2462                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2463
2464         RETURN(rc);
2465 }
2466
2467 /**
2468  * Note: we do not count into QUOTA here.
2469  * If we mount with --data_journal we may need more.
2470  */
2471 const int osd_dto_credits_noquota[DTO_NR] = {
2472         /**
2473          * Insert.
2474          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2475          * SINGLEDATA_TRANS_BLOCKS(8)
2476          * XXX Note: maybe iam need more, since iam have more level than
2477          *           EXT3 htree.
2478          */
2479         [DTO_INDEX_INSERT]  = 16,
2480         /**
2481          * Delete
2482          * just modify a single entry, probably merge few within a block
2483          */
2484         [DTO_INDEX_DELETE]  = 1,
2485         /**
2486          * Used for OI scrub
2487          */
2488         [DTO_INDEX_UPDATE]  = 16,
2489         /**
2490          * 4(inode, inode bits, groups, GDT)
2491          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2492          */
2493         [DTO_OBJECT_CREATE] = 4,
2494         /**
2495          * 4(inode, inode bits, groups, GDT)
2496          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2497          */
2498         [DTO_OBJECT_DELETE] = 4,
2499         /**
2500          * Attr set credits (inode)
2501          */
2502         [DTO_ATTR_SET_BASE] = 1,
2503         /**
2504          * Xattr set. The same as xattr of EXT3.
2505          * DATA_TRANS_BLOCKS(14)
2506          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2507          * are also counted in. Do not know why?
2508          */
2509         [DTO_XATTR_SET]     = 14,
2510         /**
2511          * credits for inode change during write.
2512          */
2513         [DTO_WRITE_BASE]    = 3,
2514         /**
2515          * credits for single block write.
2516          */
2517         [DTO_WRITE_BLOCK]   = 14,
2518         /**
2519          * Attr set credits for chown.
2520          * This is extra credits for setattr, and it is null without quota
2521          */
2522         [DTO_ATTR_SET_CHOWN] = 0
2523 };
2524
2525 static const struct dt_device_operations osd_dt_ops = {
2526         .dt_root_get       = osd_root_get,
2527         .dt_statfs         = osd_statfs,
2528         .dt_trans_create   = osd_trans_create,
2529         .dt_trans_start    = osd_trans_start,
2530         .dt_trans_stop     = osd_trans_stop,
2531         .dt_trans_cb_add   = osd_trans_cb_add,
2532         .dt_conf_get       = osd_conf_get,
2533         .dt_mnt_sb_get     = osd_mnt_sb_get,
2534         .dt_sync           = osd_sync,
2535         .dt_ro             = osd_ro,
2536         .dt_commit_async   = osd_commit_async,
2537 };
2538
2539 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2540                           unsigned int role)
2541 {
2542         struct osd_object *obj = osd_dt_obj(dt);
2543         struct osd_thread_info *oti = osd_oti_get(env);
2544
2545         LINVRNT(osd_invariant(obj));
2546
2547         LASSERT(obj->oo_owner != env);
2548         down_read_nested(&obj->oo_sem, role);
2549
2550         LASSERT(obj->oo_owner == NULL);
2551         oti->oti_r_locks++;
2552 }
2553
2554 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2555                            unsigned int role)
2556 {
2557         struct osd_object *obj = osd_dt_obj(dt);
2558         struct osd_thread_info *oti = osd_oti_get(env);
2559
2560         LINVRNT(osd_invariant(obj));
2561
2562         LASSERT(obj->oo_owner != env);
2563         down_write_nested(&obj->oo_sem, role);
2564
2565         LASSERT(obj->oo_owner == NULL);
2566         obj->oo_owner = env;
2567         oti->oti_w_locks++;
2568 }
2569
2570 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2571 {
2572         struct osd_object *obj = osd_dt_obj(dt);
2573         struct osd_thread_info *oti = osd_oti_get(env);
2574
2575         LINVRNT(osd_invariant(obj));
2576
2577         LASSERT(oti->oti_r_locks > 0);
2578         oti->oti_r_locks--;
2579         up_read(&obj->oo_sem);
2580 }
2581
2582 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2583 {
2584         struct osd_object *obj = osd_dt_obj(dt);
2585         struct osd_thread_info *oti = osd_oti_get(env);
2586
2587         LINVRNT(osd_invariant(obj));
2588
2589         LASSERT(obj->oo_owner == env);
2590         LASSERT(oti->oti_w_locks > 0);
2591         oti->oti_w_locks--;
2592         obj->oo_owner = NULL;
2593         up_write(&obj->oo_sem);
2594 }
2595
2596 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2597 {
2598         struct osd_object *obj = osd_dt_obj(dt);
2599
2600         LINVRNT(osd_invariant(obj));
2601
2602         return obj->oo_owner == env;
2603 }
2604
2605 static void osd_inode_getattr(const struct lu_env *env,
2606                               struct inode *inode, struct lu_attr *attr)
2607 {
2608         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2609                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2610                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2611                            LA_BLKSIZE | LA_TYPE;
2612
2613         attr->la_atime = inode->i_atime.tv_sec;
2614         attr->la_mtime = inode->i_mtime.tv_sec;
2615         attr->la_ctime = inode->i_ctime.tv_sec;
2616         attr->la_mode    = inode->i_mode;
2617         attr->la_size    = i_size_read(inode);
2618         attr->la_blocks  = inode->i_blocks;
2619         attr->la_uid     = i_uid_read(inode);
2620         attr->la_gid     = i_gid_read(inode);
2621         attr->la_projid  = i_projid_read(inode);
2622         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2623         attr->la_nlink   = inode->i_nlink;
2624         attr->la_rdev    = inode->i_rdev;
2625         attr->la_blksize = 1 << inode->i_blkbits;
2626         attr->la_blkbits = inode->i_blkbits;
2627         /*
2628          * Ext4 did not transfer inherit flags from raw inode
2629          * to inode flags, and ext4 internally test raw inode
2630          * @i_flags directly. Instead of patching ext4, we do it here.
2631          */
2632         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2633                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2634 }
2635
2636 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2637                         struct lu_attr *attr)
2638 {
2639         struct osd_object *obj = osd_dt_obj(dt);
2640
2641         if (unlikely(!dt_object_exists(dt)))
2642                 return -ENOENT;
2643         if (unlikely(obj->oo_destroyed))
2644                 return -ENOENT;
2645
2646         LASSERT(!dt_object_remote(dt));
2647         LINVRNT(osd_invariant(obj));
2648
2649         spin_lock(&obj->oo_guard);
2650         osd_inode_getattr(env, obj->oo_inode, attr);
2651         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2652                 attr->la_valid |= LA_FLAGS;
2653                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2654         }
2655         spin_unlock(&obj->oo_guard);
2656
2657         return 0;
2658 }
2659
2660 static int osd_declare_attr_qid(const struct lu_env *env,
2661                                 struct osd_object *obj,
2662                                 struct osd_thandle *oh, long long bspace,
2663                                 qid_t old_id, qid_t new_id, bool enforce,
2664                                 unsigned int type, bool ignore_edquot)
2665 {
2666         int rc;
2667         struct osd_thread_info *info = osd_oti_get(env);
2668         struct lquota_id_info  *qi = &info->oti_qi;
2669
2670         qi->lqi_type = type;
2671         /* inode accounting */
2672         qi->lqi_is_blk = false;
2673
2674         /* one more inode for the new id ... */
2675         qi->lqi_id.qid_uid = new_id;
2676         qi->lqi_space      = 1;
2677         /* Reserve credits for the new id */
2678         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2679         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2680                 rc = 0;
2681         if (rc)
2682                 RETURN(rc);
2683
2684         /* and one less inode for the current id */
2685         qi->lqi_id.qid_uid = old_id;
2686         qi->lqi_space = -1;
2687         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2688         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2689                 rc = 0;
2690         if (rc)
2691                 RETURN(rc);
2692
2693         /* block accounting */
2694         qi->lqi_is_blk = true;
2695
2696         /* more blocks for the new id ... */
2697         qi->lqi_id.qid_uid = new_id;
2698         qi->lqi_space      = bspace;
2699         /*
2700          * Credits for the new uid has been reserved, re-use "obj"
2701          * to save credit reservation.
2702          */
2703         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2704         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2705                 rc = 0;
2706         if (rc)
2707                 RETURN(rc);
2708
2709         /* and finally less blocks for the current uid */
2710         qi->lqi_id.qid_uid = old_id;
2711         qi->lqi_space      = -bspace;
2712         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2713         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2714                 rc = 0;
2715
2716         RETURN(rc);
2717 }
2718
2719 static int osd_declare_attr_set(const struct lu_env *env,
2720                                 struct dt_object *dt,
2721                                 const struct lu_attr *attr,
2722                                 struct thandle *handle)
2723 {
2724         struct osd_thandle *oh;
2725         struct osd_object *obj;
2726         qid_t uid;
2727         qid_t gid;
2728         long long bspace;
2729         int rc = 0;
2730         bool enforce;
2731
2732         ENTRY;
2733
2734         LASSERT(dt != NULL);
2735         LASSERT(handle != NULL);
2736
2737         obj = osd_dt_obj(dt);
2738         LASSERT(osd_invariant(obj));
2739
2740         oh = container_of0(handle, struct osd_thandle, ot_super);
2741         LASSERT(oh->ot_handle == NULL);
2742
2743         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2744                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2745
2746         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2747                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2748
2749         if (attr == NULL || obj->oo_inode == NULL)
2750                 RETURN(rc);
2751
2752         bspace   = obj->oo_inode->i_blocks << 9;
2753         bspace   = toqb(bspace);
2754
2755         /*
2756          * Changing ownership is always preformed by super user, it should not
2757          * fail with EDQUOT unless required explicitly.
2758          *
2759          * We still need to call the osd_declare_qid() to calculate the journal
2760          * credits for updating quota accounting files and to trigger quota
2761          * space adjustment once the operation is completed.
2762          */
2763         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2764                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2765
2766                 if (!ignore_edquot)
2767                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2768                                "(the quota space is %lld)\n",
2769                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2770                                attr->la_gid, bspace);
2771
2772                 /* USERQUOTA */
2773                 uid = i_uid_read(obj->oo_inode);
2774                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2775                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2776                                           attr->la_uid, enforce, USRQUOTA,
2777                                           true);
2778                 if (rc)
2779                         RETURN(rc);
2780
2781                 gid = i_gid_read(obj->oo_inode);
2782                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2783                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2784                                           i_gid_read(obj->oo_inode),
2785                                           attr->la_gid, enforce, GRPQUOTA,
2786                                           ignore_edquot);
2787                 if (rc)
2788                         RETURN(rc);
2789
2790         }
2791 #ifdef HAVE_PROJECT_QUOTA
2792         if (attr->la_valid & LA_PROJID) {
2793                 __u32 projid = i_projid_read(obj->oo_inode);
2794
2795                 enforce = (attr->la_valid & LA_PROJID) &&
2796                                         (attr->la_projid != projid);
2797                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2798                                           (qid_t)projid, (qid_t)attr->la_projid,
2799                                           enforce, PRJQUOTA, true);
2800                 if (rc)
2801                         RETURN(rc);
2802         }
2803 #endif
2804         RETURN(rc);
2805 }
2806
2807 static int osd_inode_setattr(const struct lu_env *env,
2808                              struct inode *inode, const struct lu_attr *attr)
2809 {
2810         __u64 bits = attr->la_valid;
2811
2812         /* Only allow set size for regular file */
2813         if (!S_ISREG(inode->i_mode))
2814                 bits &= ~(LA_SIZE | LA_BLOCKS);
2815
2816         if (bits == 0)
2817                 return 0;
2818
2819         if (bits & LA_ATIME)
2820                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2821         if (bits & LA_CTIME)
2822                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2823         if (bits & LA_MTIME)
2824                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2825         if (bits & LA_SIZE) {
2826                 spin_lock(&inode->i_lock);
2827                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2828                 i_size_write(inode, attr->la_size);
2829                 spin_unlock(&inode->i_lock);
2830         }
2831
2832         /*
2833          * OSD should not change "i_blocks" which is used by quota.
2834          * "i_blocks" should be changed by ldiskfs only.
2835          */
2836         if (bits & LA_MODE)
2837                 inode->i_mode = (inode->i_mode & S_IFMT) |
2838                                 (attr->la_mode & ~S_IFMT);
2839         if (bits & LA_UID)
2840                 i_uid_write(inode, attr->la_uid);
2841         if (bits & LA_GID)
2842                 i_gid_write(inode, attr->la_gid);
2843         if (bits & LA_PROJID)
2844                 i_projid_write(inode, attr->la_projid);
2845         if (bits & LA_NLINK)
2846                 set_nlink(inode, attr->la_nlink);
2847         if (bits & LA_RDEV)
2848                 inode->i_rdev = attr->la_rdev;
2849
2850         if (bits & LA_FLAGS) {
2851                 /* always keep S_NOCMTIME */
2852                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2853                                  S_NOCMTIME;
2854                 /*
2855                  * Ext4 did not transfer inherit flags from
2856                  * @inode->i_flags to raw inode i_flags when writing
2857                  * flags, we do it explictly here.
2858                  */
2859                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2860                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2861                 else
2862                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2863         }
2864         return 0;
2865 }
2866
2867 #ifdef HAVE_PROJECT_QUOTA
2868 static int osd_transfer_project(struct inode *inode, __u32 projid)
2869 {
2870         struct super_block *sb = inode->i_sb;
2871         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2872         int err;
2873         kprojid_t kprojid;
2874         struct ldiskfs_iloc iloc;
2875         struct ldiskfs_inode *raw_inode;
2876         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2877
2878         if (!ldiskfs_has_feature_project(sb)) {
2879                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2880                         == LDISKFS_DEF_PROJID);
2881                 if (projid != LDISKFS_DEF_PROJID)
2882                         return -EOPNOTSUPP;
2883                 else
2884                         return 0;
2885         }
2886
2887         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2888                 return -EOPNOTSUPP;
2889
2890         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2891         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2892                 return 0;
2893
2894         err = ldiskfs_get_inode_loc(inode, &iloc);
2895         if (err)
2896                 return err;
2897
2898         raw_inode = ldiskfs_raw_inode(&iloc);
2899         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2900                 err = -EOVERFLOW;
2901                 brelse(iloc.bh);
2902                 return err;
2903         }
2904         brelse(iloc.bh);
2905
2906         dquot_initialize(inode);
2907         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2908         if (transfer_to[PRJQUOTA]) {
2909                 err = __dquot_transfer(inode, transfer_to);
2910                 dqput(transfer_to[PRJQUOTA]);
2911                 if (err)
2912                         return err;
2913         }
2914
2915         return err;
2916 }
2917 #endif
2918
2919 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2920 {
2921         int rc;
2922
2923         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2924             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2925                 struct iattr iattr;
2926
2927                 ll_vfs_dq_init(inode);
2928                 iattr.ia_valid = 0;
2929                 if (attr->la_valid & LA_UID)
2930                         iattr.ia_valid |= ATTR_UID;
2931                 if (attr->la_valid & LA_GID)
2932                         iattr.ia_valid |= ATTR_GID;
2933                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2934                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2935
2936                 rc = ll_vfs_dq_transfer(inode, &iattr);
2937                 if (rc) {
2938                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2939                                "enforcement enabled on the ldiskfs "
2940                                "filesystem?\n", inode->i_sb->s_id, rc);
2941                         return rc;
2942                 }
2943         }
2944
2945         /* Handle project id transfer here properly */
2946         if (attr->la_valid & LA_PROJID &&
2947             attr->la_projid != i_projid_read(inode)) {
2948 #ifdef HAVE_PROJECT_QUOTA
2949                 rc = osd_transfer_project(inode, attr->la_projid);
2950 #else
2951                 rc = -ENOTSUPP;
2952 #endif
2953                 if (rc) {
2954                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2955                                "enforcement enabled on the ldiskfs "
2956                                "filesystem?\n", inode->i_sb->s_id, rc);
2957                         return rc;
2958                 }
2959         }
2960         return 0;
2961 }
2962
2963 static int osd_attr_set(const struct lu_env *env,
2964                         struct dt_object *dt,
2965                         const struct lu_attr *attr,
2966                         struct thandle *handle)
2967 {
2968         struct osd_object *obj = osd_dt_obj(dt);
2969         struct inode *inode;
2970         int rc;
2971
2972         if (!dt_object_exists(dt))
2973                 return -ENOENT;
2974
2975         LASSERT(handle != NULL);
2976         LASSERT(!dt_object_remote(dt));
2977         LASSERT(osd_invariant(obj));
2978
2979         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2980
2981         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2982             !osd_obj2dev(obj)->od_is_ost) {
2983                 struct osd_thread_info *oti = osd_oti_get(env);
2984                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
2985                 struct lu_fid *fid1 = &oti->oti_fid;
2986                 struct osd_inode_id *id = &oti->oti_id;
2987                 struct iam_path_descr *ipd;
2988                 struct iam_container *bag;
2989                 struct osd_thandle *oh;
2990                 int rc;
2991
2992                 fid_cpu_to_be(fid1, fid0);
2993                 memset(id, 1, sizeof(*id));
2994                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2995                                   fid0)->oi_dir.od_container;
2996                 ipd = osd_idx_ipd_get(env, bag);
2997                 if (unlikely(ipd == NULL))
2998                         RETURN(-ENOMEM);
2999
3000                 oh = container_of0(handle, struct osd_thandle, ot_super);
3001                 rc = iam_update(oh->ot_handle, bag,
3002                                 (const struct iam_key *)fid1,
3003                                 (const struct iam_rec *)id, ipd);
3004                 osd_ipd_put(env, bag, ipd);
3005                 return(rc > 0 ? 0 : rc);
3006         }
3007
3008         inode = obj->oo_inode;
3009
3010         rc = osd_quota_transfer(inode, attr);
3011         if (rc)
3012                 return rc;
3013
3014         spin_lock(&obj->oo_guard);
3015         rc = osd_inode_setattr(env, inode, attr);
3016         spin_unlock(&obj->oo_guard);
3017         if (rc != 0)
3018                 GOTO(out, rc);
3019
3020         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3021
3022         if (!(attr->la_valid & LA_FLAGS))
3023                 GOTO(out, rc);
3024
3025         /* Let's check if there are extra flags need to be set into LMA */
3026         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3027                 struct osd_thread_info *info = osd_oti_get(env);
3028                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3029
3030                 LASSERT(!obj->oo_pfid_in_lma);
3031
3032                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3033                                  &info->oti_ost_attrs);
3034                 if (rc)
3035                         GOTO(out, rc);
3036
3037                 lma->lma_incompat |=
3038                         lustre_to_lma_flags(attr->la_flags);
3039                 lustre_lma_swab(lma);
3040                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3041                                      lma, sizeof(*lma), XATTR_REPLACE);
3042                 if (rc != 0) {
3043                         struct osd_device *osd = osd_obj2dev(obj);
3044
3045                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3046                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3047                               lma->lma_incompat, rc);
3048                 } else {
3049                         obj->oo_lma_flags =
3050                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3051                 }
3052                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3053         }
3054 out:
3055         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3056
3057         return rc;
3058 }
3059
3060 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3061                                            struct osd_object *obj,
3062                                            const char *name, const int namelen)
3063 {
3064         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3065 }
3066
3067 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3068                       umode_t mode, struct dt_allocation_hint *hint,
3069                       struct thandle *th, struct lu_attr *attr)
3070 {
3071         int result;
3072         struct osd_device *osd = osd_obj2dev(obj);
3073         struct osd_thandle *oth;
3074         struct dt_object *parent = NULL;
3075         struct inode *inode;
3076         uid_t owner[2] = {0, 0};
3077
3078         if (attr->la_valid & LA_UID)
3079                 owner[0] = attr->la_uid;
3080         if (attr->la_valid & LA_GID)
3081                 owner[1] = attr->la_gid;
3082
3083         LINVRNT(osd_invariant(obj));
3084         LASSERT(obj->oo_inode == NULL);
3085         LASSERT(obj->oo_hl_head == NULL);
3086
3087         if (S_ISDIR(mode) && ldiskfs_pdo) {
3088                 obj->oo_hl_head =
3089                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3090                 if (obj->oo_hl_head == NULL)
3091                         return -ENOMEM;
3092         }
3093
3094         oth = container_of(th, struct osd_thandle, ot_super);
3095         LASSERT(oth->ot_handle->h_transaction != NULL);
3096
3097         if (hint != NULL && hint->dah_parent != NULL &&
3098             !dt_object_remote(hint->dah_parent))
3099                 parent = hint->dah_parent;
3100
3101         inode = ldiskfs_create_inode(oth->ot_handle,
3102                                      parent ? osd_dt_obj(parent)->oo_inode :
3103                                               osd_sb(osd)->s_root->d_inode,
3104                                      mode, owner);
3105         if (!IS_ERR(inode)) {
3106                 /* Do not update file c/mtime in ldiskfs. */
3107                 inode->i_flags |= S_NOCMTIME;
3108
3109                 /*
3110                  * For new created object, it must be consistent,
3111                  * and it is unnecessary to scrub against it.
3112                  */
3113                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3114
3115                 obj->oo_inode = inode;
3116                 result = 0;
3117         } else {
3118                 if (obj->oo_hl_head != NULL) {
3119                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3120                         obj->oo_hl_head = NULL;
3121                 }
3122                 result = PTR_ERR(inode);
3123         }
3124         LINVRNT(osd_invariant(obj));
3125         return result;
3126 }
3127
3128 enum {
3129         OSD_NAME_LEN = 255
3130 };
3131
3132 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3133                      struct lu_attr *attr,
3134                      struct dt_allocation_hint *hint,
3135                      struct dt_object_format *dof,
3136                      struct thandle *th)
3137 {
3138         int result;
3139         struct osd_thandle *oth;
3140         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3141
3142         LASSERT(S_ISDIR(attr->la_mode));
3143
3144         oth = container_of(th, struct osd_thandle, ot_super);
3145         LASSERT(oth->ot_handle->h_transaction != NULL);
3146         result = osd_mkfile(info, obj, mode, hint, th, attr);
3147
3148         return result;
3149 }
3150
3151 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3152                         struct lu_attr *attr,
3153                         struct dt_allocation_hint *hint,
3154                         struct dt_object_format *dof,
3155                         struct thandle *th)
3156 {
3157         int result;
3158         struct osd_thandle *oth;
3159         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3160
3161         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3162
3163         LASSERT(S_ISREG(attr->la_mode));
3164
3165         oth = container_of(th, struct osd_thandle, ot_super);
3166         LASSERT(oth->ot_handle->h_transaction != NULL);
3167
3168         result = osd_mkfile(info, obj, mode, hint, th, attr);
3169         if (result == 0) {
3170                 LASSERT(obj->oo_inode != NULL);
3171                 if (feat->dif_flags & DT_IND_VARKEY)
3172                         result = iam_lvar_create(obj->oo_inode,
3173                                                  feat->dif_keysize_max,
3174                                                  feat->dif_ptrsize,
3175                                                  feat->dif_recsize_max,
3176                                                  oth->ot_handle);
3177                 else
3178                         result = iam_lfix_create(obj->oo_inode,
3179                                                  feat->dif_keysize_max,
3180                                                  feat->dif_ptrsize,
3181                                                  feat->dif_recsize_max,
3182                                                  oth->ot_handle);
3183         }
3184         return result;
3185 }
3186
3187 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3188                      struct lu_attr *attr,
3189                      struct dt_allocation_hint *hint,
3190                      struct dt_object_format *dof,
3191                      struct thandle *th)
3192 {
3193         LASSERT(S_ISREG(attr->la_mode));
3194         return osd_mkfile(info, obj, (attr->la_mode &
3195                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3196                           attr);
3197 }
3198
3199 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3200                      struct lu_attr *attr,
3201                      struct dt_allocation_hint *hint,
3202                      struct dt_object_format *dof,
3203                      struct thandle *th)
3204 {
3205         LASSERT(S_ISLNK(attr->la_mode));
3206         return osd_mkfile(info, obj, (attr->la_mode &
3207                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3208                           attr);
3209 }
3210
3211 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3212                      struct lu_attr *attr,
3213                      struct dt_allocation_hint *hint,
3214                      struct dt_object_format *dof,
3215                      struct thandle *th)
3216 {
3217         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3218         int result;
3219
3220         LINVRNT(osd_invariant(obj));
3221         LASSERT(obj->oo_inode == NULL);
3222         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3223                 S_ISFIFO(mode) || S_ISSOCK(mode));
3224
3225         result = osd_mkfile(info, obj, mode, hint, th, attr);
3226         if (result == 0) {
3227                 LASSERT(obj->oo_inode != NULL);
3228                 /*
3229                  * This inode should be marked dirty for i_rdev.  Currently
3230                  * that is done in the osd_attr_init().
3231                  */
3232                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3233                                    attr->la_rdev);
3234         }
3235         LINVRNT(osd_invariant(obj));
3236         return result;
3237 }
3238
3239 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3240                               struct lu_attr *,
3241                               struct dt_allocation_hint *hint,
3242                               struct dt_object_format *dof,
3243                               struct thandle *);
3244
3245 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3246 {
3247         osd_obj_type_f result;
3248
3249         switch (type) {
3250         case DFT_DIR:
3251                 result = osd_mkdir;
3252                 break;
3253         case DFT_REGULAR:
3254                 result = osd_mkreg;
3255                 break;
3256         case DFT_SYM:
3257                 result = osd_mksym;
3258                 break;
3259         case DFT_NODE:
3260                 result = osd_mknod;
3261                 break;
3262         case DFT_INDEX:
3263                 result = osd_mk_index;
3264                 break;
3265
3266         default:
3267                 LBUG();
3268                 break;
3269         }
3270         return result;
3271 }
3272
3273
3274 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3275                         struct dt_object *parent, struct dt_object *child,
3276                         umode_t child_mode)
3277 {
3278         LASSERT(ah);
3279
3280         ah->dah_parent = parent;
3281         ah->dah_mode = child_mode;
3282
3283         if (parent != NULL && !dt_object_remote(parent)) {
3284                 /* will help to find FID->ino at dt_insert("..") */
3285                 struct osd_object *pobj = osd_dt_obj(parent);
3286
3287                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3288         }
3289 }
3290
3291 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3292                           struct lu_attr *attr, struct dt_object_format *dof)
3293 {
3294         struct inode *inode = obj->oo_inode;
3295         __u64 valid = attr->la_valid;
3296         int result;
3297
3298         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3299
3300         if (dof->dof_type != DFT_NODE)
3301                 attr->la_valid &= ~LA_RDEV;
3302         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3303                 attr->la_valid &= ~LA_ATIME;
3304         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3305                 attr->la_valid &= ~LA_CTIME;
3306         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3307                 attr->la_valid &= ~LA_MTIME;
3308
3309         result = osd_quota_transfer(inode, attr);
3310         if (result)
3311                 return;
3312
3313         if (attr->la_valid != 0) {
3314                 result = osd_inode_setattr(info->oti_env, inode, attr);
3315                 /*
3316                  * The osd_inode_setattr() should always succeed here.  The
3317                  * only error that could be returned is EDQUOT when we are
3318                  * trying to change the UID or GID of the inode. However, this
3319                  * should not happen since quota enforcement is no longer
3320                  * enabled on ldiskfs (lquota takes care of it).
3321                  */
3322                 LASSERTF(result == 0, "%d\n", result);
3323                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3324         }
3325
3326         attr->la_valid = valid;
3327 }
3328
3329 /**
3330  * Helper function for osd_create()
3331  *
3332  * \retval 0, on success
3333  */
3334 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3335                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3336                         struct dt_object_format *dof, struct thandle *th)
3337 {
3338         int result;
3339         __u32 umask;
3340
3341         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3342
3343         /* we drop umask so that permissions we pass are not affected */
3344         umask = current->fs->umask;
3345         current->fs->umask = 0;
3346
3347         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3348                                                   th);
3349         if (likely(obj->oo_inode != NULL)) {
3350                 LASSERT(obj->oo_inode->i_state & I_NEW);
3351
3352                 /*
3353                  * Unlock the inode before attr initialization to avoid
3354                  * unnecessary dqget operations. LU-6378
3355                  */
3356                 unlock_new_inode(obj->oo_inode);
3357         }
3358
3359         if (likely(result == 0)) {
3360                 osd_attr_init(info, obj, attr, dof);
3361                 osd_object_init0(obj);
3362         }
3363
3364         /* restore previous umask value */
3365         current->fs->umask = umask;
3366
3367         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3368
3369         return result;
3370 }
3371
3372 /**
3373  * Helper function for osd_create()
3374  *
3375  * \retval 0, on success
3376  */
3377 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3378                            const struct lu_fid *fid, struct thandle *th)
3379 {
3380         struct osd_thread_info *info = osd_oti_get(env);
3381         struct osd_inode_id    *id   = &info->oti_id;
3382         struct osd_device      *osd  = osd_obj2dev(obj);
3383         struct osd_thandle     *oh;
3384         int rc;
3385
3386         LASSERT(obj->oo_inode != NULL);
3387
3388         oh = container_of0(th, struct osd_thandle, ot_super);
3389         LASSERT(oh->ot_handle);
3390         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3391
3392         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3393         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3394                            OI_CHECK_FLD, NULL);
3395         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3396
3397         return rc;
3398 }
3399
3400 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3401                    u64 seq, struct lu_seq_range *range)
3402 {
3403         struct seq_server_site *ss = osd_seq_site(osd);
3404
3405         if (fid_seq_is_idif(seq)) {
3406                 fld_range_set_ost(range);
3407                 range->lsr_index = idif_ost_idx(seq);
3408                 return 0;
3409         }
3410
3411         if (!fid_seq_in_fldb(seq)) {
3412                 fld_range_set_mdt(range);
3413                 if (ss != NULL)
3414                         /*
3415                          * FIXME: If ss is NULL, it suppose not get lsr_index
3416                          * at all
3417                          */
3418                         range->lsr_index = ss->ss_node_id;
3419                 return 0;
3420         }
3421
3422         LASSERT(ss != NULL);
3423         fld_range_set_any(range);
3424         /* OSD will only do local fld lookup */
3425         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3426 }
3427
3428 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3429                               struct lu_attr *attr,
3430                               struct dt_allocation_hint *hint,
3431                               struct dt_object_format *dof,
3432                               struct thandle *handle)
3433 {
3434         struct osd_thandle *oh;
3435         int rc;
3436
3437         ENTRY;
3438
3439         LASSERT(handle != NULL);
3440
3441         oh = container_of0(handle, struct osd_thandle, ot_super);
3442         LASSERT(oh->ot_handle == NULL);
3443
3444         /*
3445          * EA object consumes more credits than regular object: osd_mk_index
3446          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3447          * leaf_node, could involves the block, block bitmap, groups, GDT
3448          * change for each block, so add 4 * 2 credits in that case.
3449          */
3450         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3451                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3452                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3453         /*
3454          * Reuse idle OI block may cause additional one OI block
3455          * to be changed.
3456          */
3457         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3458                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3459
3460         /* will help to find FID->ino mapping at dt_insert() */
3461         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3462                                    osd_dt_obj(dt));
3463         if (rc != 0)
3464                 RETURN(rc);
3465
3466         if (!attr)
3467                 RETURN(0);
3468
3469         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3470                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3471                                    NULL, OSD_QID_INODE);
3472         if (rc != 0)
3473                 RETURN(rc);
3474
3475         RETURN(rc);
3476 }
3477
3478 /**
3479  * Called to destroy on-disk representation of the object
3480  *
3481  * Concurrency: must be locked
3482  */
3483 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3484                                struct thandle *th)
3485 {
3486         struct osd_object *obj = osd_dt_obj(dt);
3487         struct inode *inode = obj->oo_inode;
3488         struct osd_thandle *oh;
3489         int rc;
3490
3491         ENTRY;
3492
3493         if (inode == NULL)
3494                 RETURN(-ENOENT);
3495
3496         oh = container_of0(th, struct osd_thandle, ot_super);
3497         LASSERT(oh->ot_handle == NULL);
3498
3499         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3500                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3501
3502         /* For removing agent entry */
3503         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3504                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3505
3506         /*
3507          * Recycle idle OI leaf may cause additional three OI blocks
3508          * to be changed.
3509          */
3510         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3511                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3512                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3513         /* one less inode */
3514         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3515                                    i_projid_read(inode), -1, oh, obj, NULL,
3516                                    OSD_QID_INODE);
3517         if (rc)
3518                 RETURN(rc);
3519         /* data to be truncated */
3520         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3521                                    i_projid_read(inode), 0, oh, obj, NULL,
3522                                    OSD_QID_BLK);
3523         if (rc)
3524                 RETURN(rc);
3525
3526         /*
3527          * will help to find FID->ino when this object is being
3528          * added to PENDING
3529          */
3530         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3531
3532         RETURN(rc);
3533 }
3534
3535 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3536                        struct thandle *th)
3537 {
3538         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3539         struct osd_object *obj = osd_dt_obj(dt);
3540         struct inode *inode = obj->oo_inode;
3541         struct osd_device *osd = osd_obj2dev(obj);
3542         struct osd_thandle *oh;
3543         int result;
3544
3545         ENTRY;
3546
3547         oh = container_of0(th, struct osd_thandle, ot_super);
3548         LASSERT(oh->ot_handle);
3549         LASSERT(inode);
3550         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3551
3552         if (unlikely(fid_is_acct(fid)))
3553                 RETURN(-EPERM);
3554
3555         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3556                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3557                 if (result != 0)
3558                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3559                                osd_name(osd), PFID(fid), result);
3560         }
3561
3562         if (S_ISDIR(inode->i_mode)) {
3563                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3564                         inode->i_nlink == 2);
3565
3566                 spin_lock(&obj->oo_guard);
3567                 clear_nlink(inode);
3568                 spin_unlock(&obj->oo_guard);
3569                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
3570         }
3571
3572         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3573
3574         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3575
3576         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3577                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3578                                        oh->ot_handle, OI_CHECK_FLD);
3579
3580         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3581         /* XXX: add to ext3 orphan list */
3582         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3583
3584         /* not needed in the cache anymore */
3585         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3586         obj->oo_destroyed = 1;
3587
3588         RETURN(0);
3589 }
3590
3591 /**
3592  * Put the fid into lustre_mdt_attrs, and then place the structure
3593  * inode's ea. This fid should not be altered during the life time
3594  * of the inode.
3595  *
3596  * \retval +ve, on success
3597  * \retval -ve, on error
3598  *
3599  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3600  */
3601 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3602                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3603 {
3604         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3605         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3606         int rc;
3607
3608         ENTRY;
3609
3610         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3611                 RETURN(0);
3612
3613         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3614                 rc = -ENOMEM;
3615
3616         lustre_loa_init(loa, fid, compat, incompat);
3617         lustre_loa_swab(loa, false);
3618
3619         /*
3620          * For the OST device with 256 bytes inode size by default,
3621          * the PFID EA will be stored together with LMA EA to avoid
3622          * performance trouble. Otherwise the PFID EA can be stored
3623          * independently. LU-8998
3624          */
3625         if ((compat & LMAC_FID_ON_OST) &&
3626             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3627                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3628                                      sizeof(*loa), XATTR_CREATE);
3629         else
3630                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3631                                      sizeof(*lma), XATTR_CREATE);
3632         /*
3633          * LMA may already exist, but we need to check that all the
3634          * desired compat/incompat flags have been added.
3635          */
3636         if (unlikely(rc == -EEXIST)) {
3637                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3638                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3639                 if (rc < 0)
3640                         RETURN(rc);
3641
3642                 if (rc < sizeof(*lma))
3643                         RETURN(-EINVAL);
3644
3645                 lustre_loa_swab(loa, true);
3646                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3647                     ((compat == 0 && incompat == 0) ||
3648                      (!(~lma->lma_compat & compat) &&
3649                       !(~lma->lma_incompat & incompat))))
3650                         RETURN(0);
3651
3652                 lma->lma_self_fid = *fid;
3653                 lma->lma_compat |= compat;
3654                 lma->lma_incompat |= incompat;
3655                 if (rc == sizeof(*lma)) {
3656                         lustre_lma_swab(lma);
3657                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3658                                              sizeof(*lma), XATTR_REPLACE);
3659                 } else {
3660                         lustre_loa_swab(loa, false);
3661                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3662                                              sizeof(*loa), XATTR_REPLACE);
3663                 }
3664         }
3665
3666         RETURN(rc);
3667 }
3668
3669 /**
3670  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3671  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3672  * To have compatilibility with 1.8 ldiskfs driver we need to have
3673  * magic number at start of fid data.
3674  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3675  * its inmemory API.
3676  */
3677 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3678                                          const struct lu_fid *fid)
3679 {
3680         if (!fid_is_namespace_visible(fid) ||
3681             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3682                 param->edp_magic = 0;
3683                 return;
3684         }
3685
3686         param->edp_magic = LDISKFS_LUFID_MAGIC;
3687         param->edp_len =  sizeof(struct lu_fid) + 1;
3688         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3689 }
3690
3691 /**
3692  * Try to read the fid from inode ea into dt_rec.
3693  *
3694  * \param fid object fid.
3695  *
3696  * \retval 0 on success
3697  */
3698 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3699                           __u32 ino, struct lu_fid *fid,
3700                           struct osd_inode_id *id)
3701 {
3702         struct osd_thread_info *info  = osd_oti_get(env);
3703         struct inode *inode;
3704
3705         ENTRY;
3706
3707         osd_id_gen(id, ino, OSD_OII_NOGEN);
3708         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3709         if (IS_ERR(inode))
3710                 RETURN(PTR_ERR(inode));
3711
3712         iput(inode);
3713         RETURN(0);
3714 }
3715
3716 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3717                                         struct inode *dir,
3718                                         struct inode *parent_dir,
3719                                         const struct lu_fid *dot_fid,
3720                                         const struct lu_fid *dot_dot_fid,
3721                                         struct osd_thandle *oth)
3722 {
3723         struct ldiskfs_dentry_param *dot_ldp;
3724         struct ldiskfs_dentry_param *dot_dot_ldp;
3725         __u32 saved_nlink = dir->i_nlink;
3726         int rc;
3727
3728         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3729         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3730
3731         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3732         dot_ldp->edp_magic = 0;
3733
3734         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3735                                     dir, dot_ldp, dot_dot_ldp);
3736         /*
3737          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3738          * the subseqent ref_add() will increase the dir->i_nlink
3739          * as 3. That is incorrect for new created directory.
3740          *
3741          * It looks like hack, because we want to make the OSD API
3742          * to be order-independent for new created directory object
3743          * between dt_insert(..) and ref_add() operations.
3744          *
3745          * Here, we only restore the in-RAM dir-inode's nlink attr,
3746          * becuase if the nlink attr is not 2, then there will be
3747          * ref_add() called following the dt_insert(..), such call
3748          * will make both the in-RAM and on-disk dir-inode's nlink
3749          * attr to be set as 2. LU-7447
3750          */
3751         set_nlink(dir, saved_nlink);
3752         return rc;
3753 }
3754
3755 /**
3756  * Create an local agent inode for remote entry
3757  */
3758 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3759                                                   struct osd_device *osd,
3760                                                   struct osd_object *pobj,
3761                                                   const struct lu_fid *fid,
3762                                                   __u32 type,
3763                                                   struct thandle *th)
3764 {
3765         struct osd_thread_info *info = osd_oti_get(env);
3766         struct inode *local;
3767         struct osd_thandle *oh;
3768         int rc;
3769
3770         ENTRY;
3771
3772         LASSERT(th);
3773         oh = container_of(th, struct osd_thandle, ot_super);
3774         LASSERT(oh->ot_handle->h_transaction != NULL);
3775
3776         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3777                                      NULL);
3778         if (IS_ERR(local)) {
3779                 CERROR("%s: create local error %d\n", osd_name(osd),
3780                        (int)PTR_ERR(local));
3781                 RETURN(local);
3782         }
3783
3784         /*
3785          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3786          * correct gid on remote file, not agent here
3787          */
3788         local->i_gid = current_fsgid();
3789         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3790
3791         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3792          * That gives e2fsck something to look at and be happy, and allows
3793          * debugging if we need to determine where this symlink came from.
3794          */
3795         if (S_ISLNK(type)) {
3796                 CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
3797                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3798                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3799
3800                 i_size_write(local, rc);
3801                 LDISKFS_I(local)->i_disksize = rc;
3802         }
3803         unlock_new_inode(local);
3804
3805         /* Agent inode should not have project ID */
3806 #ifdef  HAVE_PROJECT_QUOTA
3807         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3808             i_projid_read(pobj->oo_inode) != 0) {
3809                 rc = osd_transfer_project(local, 0);
3810                 if (rc) {
3811                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3812                                "quota enforcement enabled on the ldiskfs "
3813                                "filesystem?\n", local->i_sb->s_id, rc);
3814                         RETURN(ERR_PTR(rc));
3815                 }
3816         }
3817 #endif
3818         /* Set special LMA flag for local agent inode */
3819         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3820         if (rc != 0) {
3821                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3822                        osd_name(osd), PFID(fid), rc);
3823                 RETURN(ERR_PTR(rc));
3824         }
3825
3826         if (!S_ISDIR(type))
3827                 RETURN(local);
3828
3829         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3830                                          lu_object_fid(&pobj->oo_dt.do_lu),
3831                                          fid, oh);
3832         if (rc != 0) {
3833                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3834                         osd_name(osd), PFID(fid), rc);
3835                 RETURN(ERR_PTR(rc));
3836         }
3837
3838         RETURN(local);
3839 }
3840
3841 /**
3842  * when direntry is deleted, we have to take care of possible agent inode
3843  * referenced by that. unfortunately we can't do this at that point:
3844  * iget() within a running transaction leads to deadlock and we better do
3845  * not call that every delete declaration to save performance. so we put
3846  * a potention agent inode on a list and process that once the transaction
3847  * is over. Notice it's not any worse in terms of real orphans as regular
3848  * object destroy doesn't put inodes on the on-disk orphan list. this should
3849  * be addressed separately
3850  */
3851 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3852                                             struct osd_thandle *oh,
3853                                             __u32 ino)
3854 {
3855         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3856         struct osd_obj_orphan *oor;
3857
3858         OBD_ALLOC_PTR(oor);
3859         if (oor == NULL)
3860                 return -ENOMEM;
3861
3862         oor->oor_ino = ino;
3863         oor->oor_env = (struct lu_env *)env;
3864         spin_lock(&osd->od_osfs_lock);
3865         list_add(&oor->oor_list, &osd->od_orphan_list);
3866         spin_unlock(&osd->od_osfs_lock);
3867
3868         oh->ot_remove_agents = 1;
3869
3870         return 0;
3871
3872 }
3873
3874 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3875                                                 struct osd_device *osd)
3876 {
3877         struct osd_thread_info *info = osd_oti_get(env);
3878         struct osd_obj_orphan *oor, *tmp;
3879         struct osd_inode_id id;
3880         struct list_head list;
3881         struct inode *inode;
3882         struct lu_fid fid;
3883         handle_t *jh;
3884         __u32 ino;
3885
3886         INIT_LIST_HEAD(&list);
3887
3888         spin_lock(&osd->od_osfs_lock);
3889         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3890                 if (oor->oor_env == env) {
3891                         list_del(&oor->oor_list);
3892                         list_add(&oor->oor_list, &list);
3893                 }
3894         }
3895         spin_unlock(&osd->od_osfs_lock);
3896
3897         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3898
3899                 ino = oor->oor_ino;
3900
3901                 list_del(&oor->oor_list);
3902                 OBD_FREE_PTR(oor);
3903
3904                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3905                 inode = osd_iget_fid(info, osd, &id, &fid);
3906                 if (IS_ERR(inode))
3907                         continue;
3908
3909                 if (!osd_remote_fid(env, osd, &fid)) {
3910                         iput(inode);
3911                         continue;
3912                 }
3913
3914                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3915                 clear_nlink(inode);
3916                 mark_inode_dirty(inode);
3917                 ldiskfs_journal_stop(jh);
3918                 iput(inode);
3919         }
3920
3921         return 0;
3922 }
3923
3924 /**
3925  * OSD layer object create function for OST objects (b=11826).
3926  *
3927  * The FID is inserted into inode xattr here.
3928  *
3929  * \retval   0, on success
3930  * \retval -ve, on error
3931  */
3932 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3933                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3934                       struct dt_object_format *dof, struct thandle *th)
3935 {
3936         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3937         struct osd_object *obj = osd_dt_obj(dt);
3938         struct osd_thread_info *info = osd_oti_get(env);
3939         int result, on_ost = 0;
3940
3941         ENTRY;
3942
3943         if (dt_object_exists(dt))
3944                 RETURN(-EEXIST);
3945
3946         LINVRNT(osd_invariant(obj));
3947         LASSERT(!dt_object_remote(dt));
3948         LASSERT(osd_is_write_locked(env, obj));
3949         LASSERT(th != NULL);
3950
3951         if (unlikely(fid_is_acct(fid)))
3952                 /*
3953                  * Quota files can't be created from the kernel any more,
3954                  * 'tune2fs -O quota' will take care of creating them
3955                  */
3956                 RETURN(-EPERM);
3957
3958         result = __osd_create(info, obj, attr, hint, dof, th);
3959         if (result == 0) {
3960                 if (fid_is_idif(fid) &&
3961                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3962                         struct lu_fid *tfid = &info->oti_fid;
3963                         struct ost_id *oi   = &info->oti_ostid;
3964
3965                         fid_to_ostid(fid, oi);
3966                         ostid_to_fid(tfid, oi, 0);
3967                         on_ost = 1;
3968                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
3969                                                 LMAC_FID_ON_OST, 0);
3970                 } else {
3971                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
3972                                                fid, OI_CHECK_FLD);
3973                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
3974                                                 on_ost ? LMAC_FID_ON_OST : 0,
3975                                                 0);
3976                 }
3977                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3978                         obj->oo_dt.do_body_ops = &osd_body_ops;
3979         }
3980
3981         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
3982                 result = __osd_oi_insert(env, obj, fid, th);
3983
3984         /*
3985          * a small optimization - dt_insert() isn't usually applied
3986          * to OST objects, so we don't need to cache OI mapping for
3987          * OST objects
3988          */
3989         if (result == 0 && on_ost == 0) {
3990                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
3991
3992                 result = osd_idc_find_and_init(env, osd, obj);
3993                 LASSERT(result == 0);
3994         }
3995
3996         LASSERT(ergo(result == 0,
3997                      dt_object_exists(dt) && !dt_object_remote(dt)));
3998         LINVRNT(osd_invariant(obj));
3999         RETURN(result);
4000 }
4001
4002 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4003                                struct thandle *handle)
4004 {
4005         struct osd_thandle *oh;
4006
4007         /* it's possible that object doesn't exist yet */
4008         LASSERT(handle != NULL);
4009
4010         oh = container_of0(handle, struct osd_thandle, ot_super);
4011         LASSERT(oh->ot_handle == NULL);
4012
4013         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4014                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4015
4016         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4017
4018         return 0;
4019 }
4020
4021 /*
4022  * Concurrency: @dt is write locked.
4023  */
4024 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4025                        struct thandle *th)
4026 {
4027         struct osd_object *obj = osd_dt_obj(dt);
4028         struct inode *inode = obj->oo_inode;
4029         struct osd_thandle *oh;
4030         int rc = 0;
4031
4032         if (!dt_object_exists(dt) || obj->oo_destroyed)
4033                 return -ENOENT;
4034
4035         LINVRNT(osd_invariant(obj));
4036         LASSERT(!dt_object_remote(dt));
4037         LASSERT(osd_is_write_locked(env, obj));
4038         LASSERT(th != NULL);
4039
4040         oh = container_of0(th, struct osd_thandle, ot_super);
4041         LASSERT(oh->ot_handle != NULL);
4042
4043         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4044
4045         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4046                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4047         /*
4048          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4049          * (65000) subdirectories by storing "1" in i_nlink if the link count
4050          * would otherwise overflow. Directory tranversal tools understand
4051          * that (st_nlink == 1) indicates that the filesystem dose not track
4052          * hard links count on the directory, and will not abort subdirectory
4053          * scanning early once (st_nlink - 2) subdirs have been found.
4054          *
4055          * This also has to properly handle the case of inodes with nlink == 0
4056          * in case they are being linked into the PENDING directory
4057          */
4058         spin_lock(&obj->oo_guard);
4059         if (unlikely(inode->i_nlink == 0))
4060                 /* inc_nlink from 0 may cause WARN_ON */
4061                 set_nlink(inode, 1);
4062         else {
4063                 ldiskfs_inc_count(oh->ot_handle, inode);
4064                 if (!S_ISDIR(inode->i_mode))
4065                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4066         }
4067         spin_unlock(&obj->oo_guard);
4068
4069         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4070         LINVRNT(osd_invariant(obj));
4071
4072         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4073
4074         return rc;
4075 }
4076
4077 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4078                                struct thandle *handle)
4079 {
4080         struct osd_thandle *oh;
4081
4082         if (!dt_object_exists(dt))
4083                 return -ENOENT;
4084
4085         LASSERT(!dt_object_remote(dt));
4086         LASSERT(handle != NULL);
4087
4088         oh = container_of0(handle, struct osd_thandle, ot_super);
4089         LASSERT(oh->ot_handle == NULL);
4090
4091         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4092                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4093
4094         return 0;
4095 }
4096
4097 /*
4098  * Concurrency: @dt is write locked.
4099  */
4100 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4101                        struct thandle *th)
4102 {
4103         struct osd_object *obj = osd_dt_obj(dt);
4104         struct inode *inode = obj->oo_inode;
4105         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4106         struct osd_thandle *oh;
4107
4108         if (!dt_object_exists(dt))
4109                 return -ENOENT;
4110
4111         LINVRNT(osd_invariant(obj));
4112         LASSERT(!dt_object_remote(dt));
4113         LASSERT(osd_is_write_locked(env, obj));
4114         LASSERT(th != NULL);
4115
4116         oh = container_of0(th, struct osd_thandle, ot_super);
4117         LASSERT(oh->ot_handle != NULL);
4118
4119         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4120
4121         spin_lock(&obj->oo_guard);
4122         /*
4123          * That can be result of upgrade from old Lustre version and
4124          * applied only to local files.  Just skip this ref_del call.
4125          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4126          */
4127         if (inode->i_nlink == 0) {
4128                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4129                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4130                              ", maybe an upgraded file? (LU-3915)\n",
4131                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4132                 spin_unlock(&obj->oo_guard);
4133                 return 0;
4134         }
4135
4136         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4137                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4138
4139         ldiskfs_dec_count(oh->ot_handle, inode);
4140         spin_unlock(&obj->oo_guard);
4141
4142         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4143         LINVRNT(osd_invariant(obj));
4144
4145         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4146
4147         return 0;
4148 }
4149
4150 /*
4151  * Concurrency: @dt is read locked.
4152  */
4153 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4154                          struct lu_buf *buf, const char *name)
4155 {
4156         struct osd_object      *obj    = osd_dt_obj(dt);
4157         struct inode           *inode  = obj->oo_inode;
4158         struct osd_thread_info *info   = osd_oti_get(env);
4159         struct dentry          *dentry = &info->oti_obj_dentry;
4160         bool cache_xattr = false;
4161         int rc;
4162
4163         LASSERT(buf);
4164
4165         /* version get is not real XATTR but uses xattr API */
4166         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4167                 dt_obj_version_t *ver = buf->lb_buf;
4168
4169                 /*
4170                  * for version we are just using xattr API but change inode
4171                  * field instead
4172                  */
4173                 if (buf->lb_len == 0)
4174                         return sizeof(dt_obj_version_t);
4175
4176                 if (buf->lb_len < sizeof(dt_obj_version_t))
4177                         return -ERANGE;
4178
4179                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4180                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4181
4182                 *ver = LDISKFS_I(inode)->i_fs_version;
4183
4184                 return sizeof(dt_obj_version_t);
4185         }
4186
4187         if (!dt_object_exists(dt))
4188                 return -ENOENT;
4189
4190         LASSERT(!dt_object_remote(dt));
4191         LASSERT(inode->i_op != NULL);
4192 #ifdef HAVE_IOP_XATTR
4193         LASSERT(inode->i_op->getxattr != NULL);
4194 #endif
4195
4196         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4197             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4198                 cache_xattr = true;
4199
4200         if (cache_xattr) {
4201                 rc = osd_oxc_get(obj, name, buf);
4202                 if (rc != -ENOENT)
4203                         return rc;
4204         }
4205
4206         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4207                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4208                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4209                 struct filter_fid *ff;
4210                 struct ost_layout *ol;
4211
4212                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4213                 if (rc)
4214                         return rc;
4215
4216                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4217
4218                 rc = sizeof(*ff);
4219                 if (buf->lb_len == 0 || !buf->lb_buf)
4220                         return rc;
4221
4222                 if (buf->lb_len < rc)
4223                         return -ERANGE;
4224
4225                 ff = buf->lb_buf;
4226                 ol = &ff->ff_layout;
4227                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4228                                                   PFID_STRIPE_IDX_BITS);
4229                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4230                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4231                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4232                 if (lma->lma_compat & LMAC_COMP_INFO) {
4233                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4234                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4235                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4236                 } else {
4237                         ol->ol_comp_start = 0;
4238                         ol->ol_comp_end = 0;
4239                         ol->ol_comp_id = 0;
4240                 }
4241         } else {
4242                 rc = __osd_xattr_get(inode, dentry, name,
4243                                      buf->lb_buf, buf->lb_len);
4244         }
4245
4246         if (cache_xattr) {
4247                 if (rc == -ENOENT || rc == -ENODATA)
4248                         osd_oxc_add(obj, name, NULL, 0);
4249                 else if (rc > 0 && buf->lb_buf != NULL)
4250                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4251         }
4252
4253         return rc;
4254 }
4255
4256 static int osd_declare_xattr_set(const struct lu_env *env,
4257                                  struct dt_object *dt,
4258                                  const struct lu_buf *buf, const char *name,
4259                                  int fl, struct thandle *handle)
4260 {
4261         struct osd_thandle *oh;
4262         int credits = 0;
4263         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4264
4265         LASSERT(handle != NULL);
4266
4267         oh = container_of0(handle, struct osd_thandle, ot_super);
4268         LASSERT(oh->ot_handle == NULL);
4269
4270         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4271                 /*
4272                  * For non-upgrading case, the LMA is set first and
4273                  * usually fit inode. But for upgrade case, the LMA
4274                  * may be in another separated EA block.
4275                  */
4276                 if (dt_object_exists(dt)) {
4277                         if (fl == LU_XATTR_REPLACE)
4278                                 credits = 1;
4279                         else
4280                                 goto upgrade;
4281                 }
4282         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4283                 credits = 1;
4284         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4285                 /* We may need to delete the old PFID EA. */
4286                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4287                 if (fl == LU_XATTR_REPLACE)
4288                         credits += 1;
4289                 else
4290                         goto upgrade;
4291         } else {
4292                 /*
4293                  * If some name entry resides on remote MDT, then will create
4294                  * agent entry under remote parent. On the other hand, if the
4295                  * remote entry will be removed, then related agent entry may
4296                  * need to be removed from the remote parent. So there may be
4297                  * kinds of cases, let's declare enough credits. The credits
4298                  * for create agent entry is enough for remove case.
4299                  */
4300                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4301                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4302                         if (dt_object_exists(dt))
4303                                 credits += 1; /* For updating LMA */
4304                 }
4305
4306 upgrade:
4307                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4308
4309                 if (buf != NULL) {
4310                         ssize_t buflen;
4311
4312                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4313                                 /*
4314                                  * learn xattr size from osd_xattr_get if
4315                                  * attribute has not been read yet
4316                                  */
4317                                 buflen = __osd_xattr_get(
4318                                     osd_dt_obj(dt)->oo_inode,
4319                                     &osd_oti_get(env)->oti_obj_dentry,
4320                                     name, NULL, 0);
4321                                 if (buflen < 0)
4322                                         buflen = 0;
4323                         } else {
4324                                 buflen = buf->lb_len;
4325                         }
4326
4327                         if (buflen > sb->s_blocksize) {
4328                                 credits += osd_calc_bkmap_credits(
4329                                     sb, NULL, 0, -1,
4330                                     (buflen + sb->s_blocksize - 1) >>
4331                                     sb->s_blocksize_bits);
4332                         }
4333                 }
4334                 /*
4335                  * xattr set may involve inode quota change, reserve credits for
4336                  * dquot_initialize()
4337                  */
4338                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4339         }
4340
4341         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4342
4343         return 0;
4344 }
4345
4346 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4347                               const struct lu_buf *buf, int fl,
4348                               struct thandle *handle)
4349 {
4350         struct osd_thread_info *info = osd_oti_get(env);
4351         struct dentry *dentry = &info->oti_obj_dentry;
4352         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4353         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4354         struct inode *inode = obj->oo_inode;
4355         struct filter_fid *ff = buf->lb_buf;
4356         struct ost_layout *ol = &ff->ff_layout;
4357         int flags = XATTR_REPLACE;
4358         int rc;
4359
4360         ENTRY;
4361
4362         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4363                 RETURN(-EINVAL);
4364
4365         rc = osd_get_lma(info, inode, dentry, loa);
4366         if (rc == -ENODATA) {
4367                 /* Usually for upgarding from old device */
4368                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4369                                 LMAC_FID_ON_OST, 0);
4370                 flags = XATTR_CREATE;
4371         } else if (rc) {
4372                 RETURN(rc);
4373         }
4374
4375         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4376                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4377                         RETURN(-EEXIST);
4378
4379                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4380                         /* Separate PFID EA from LMA */
4381                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4382                         lustre_lma_swab(lma);
4383                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4384                                              sizeof(*lma), XATTR_REPLACE);
4385                         if (!rc) {
4386                                 obj->oo_pfid_in_lma = 0;
4387                                 rc = LU_XATTR_CREATE;
4388                         }
4389
4390                         RETURN(rc);
4391                 }
4392         } else {
4393                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4394                         RETURN(fl);
4395
4396                 /*
4397                  * Old client does not send stripe information,
4398                  * then store the PFID EA on disk separatedly.
4399                  */
4400                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4401                              ol->ol_stripe_size == 0))
4402                         RETURN(fl);
4403
4404                 /* Remove old PFID EA entry firstly. */
4405                 ll_vfs_dq_init(inode);
4406                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4407                 if (rc == -ENODATA) {
4408                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4409                                 RETURN(rc);
4410                 } else if (rc) {
4411                         RETURN(rc);
4412                 }
4413         }
4414
4415         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4416         if (likely(ol->ol_stripe_size != 0)) {
4417                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4418                                              PFID_STRIPE_IDX_BITS;
4419                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4420                 lma->lma_compat |= LMAC_STRIPE_INFO;
4421                 if (ol->ol_comp_id != 0) {
4422                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4423                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4424                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4425                         lma->lma_compat |= LMAC_COMP_INFO;
4426                 }
4427         }
4428
4429         lustre_loa_swab(loa, false);
4430
4431         /* Store the PFID EA inside LMA. */
4432         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4433                              flags);
4434         if (!rc)
4435                 obj->oo_pfid_in_lma = 1;
4436
4437         RETURN(rc);
4438 }
4439
4440 /*
4441  * In DNE environment, the object (in spite of regular file or directory)
4442  * and its name entry may reside on different MDTs. Under such case, we will
4443  * create an agent entry on the MDT where the object resides. The agent entry
4444  * references the object locally, that makes the object to be visible to the
4445  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4446  * such as 'tar' can handle the object properly.
4447  *
4448  * We handle the agent entry during set linkEA that is the common interface
4449  * for both regular file and directroy, can handle kinds of cases, such as
4450  * create/link/unlink/rename, and so on.
4451  *
4452  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4453  *
4454  * XXX: There are two known issues:
4455  * 1. For one object, we will create at most one agent entry even if there
4456  *    may be more than one cross-MDTs hard links on the object. So the local
4457  *    e2fsck may claim that the object's nlink is larger than the name entries
4458  *    that reference such inode. And in further, the e2fsck will fix the nlink
4459  *    attribute to match the local references. Then it will cause the object's
4460  *    nlink attribute to be inconsistent with the global references. it is bad
4461  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4462  *    other hand, the global namespace LFSCK can repair the object's attribute
4463  *    according to the linkEA.
4464  * 2. There may be too many hard links on the object as to its linkEA overflow,
4465  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4466  *    case happened, then at this point, we do not know whether there are some
4467  *    cross-MDTs reference. But there are local references, it guarantees that
4468  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4469  */
4470 static int osd_xattr_handle_linkea(const struct lu_env *env,
4471                                    struct osd_device *osd,
4472                                    struct osd_object *obj,
4473                                    const struct lu_buf *buf,
4474                                    struct thandle *handle)
4475 {
4476         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4477         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4478         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4479         struct lu_name tmpname;
4480         struct osd_thandle *oh;
4481         int rc;
4482         bool remote = false;
4483
4484         ENTRY;
4485
4486         oh = container_of0(handle, struct osd_thandle, ot_super);
4487         LASSERT(oh->ot_handle != NULL);
4488
4489         rc = linkea_init_with_rec(&ldata);
4490         if (!rc) {
4491                 linkea_first_entry(&ldata);
4492                 while (ldata.ld_lee != NULL && !remote) {
4493                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4494                                             &tmpname, tfid);
4495                         if (osd_remote_fid(env, osd, tfid) > 0)
4496                                 remote = true;
4497                         else
4498                                 linkea_next_entry(&ldata);
4499                 }
4500         } else if (rc == -ENODATA) {
4501                 rc = 0;
4502         } else {
4503                 RETURN(rc);
4504         }
4505
4506         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4507                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4508                 if (rc)
4509                         CERROR("%s: failed to remove agent entry for "DFID
4510                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4511         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4512                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4513                 if (rc)
4514                         CERROR("%s: failed to create agent entry for "DFID
4515                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4516         }
4517
4518         RETURN(rc);
4519 }
4520
4521 /*
4522  * Concurrency: @dt is write locked.
4523  */
4524 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4525                          const struct lu_buf *buf, const char *name, int fl,
4526                          struct thandle *handle)
4527 {
4528         struct osd_object *obj = osd_dt_obj(dt);
4529         struct osd_device *osd = osd_obj2dev(obj);
4530         struct inode *inode = obj->oo_inode;
4531         struct osd_thread_info *info = osd_oti_get(env);
4532         int fs_flags = 0;
4533         int len;
4534         int rc;
4535
4536         ENTRY;
4537
4538         LASSERT(handle);
4539         LASSERT(buf);
4540
4541         /* version set is not real XATTR */
4542         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4543                 dt_obj_version_t *version = buf->lb_buf;
4544
4545                 /*
4546                  * for version we are just using xattr API but change inode
4547                  * field instead
4548                  */
4549                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4550
4551                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4552                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4553
4554                 LDISKFS_I(inode)->i_fs_version = *version;
4555                 /*
4556                  * Version is set after all inode operations are finished,
4557                  * so we should mark it dirty here
4558                  */
4559                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
4560
4561                 RETURN(0);
4562         }
4563
4564         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4565                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4566
4567         len = buf->lb_len;
4568         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4569
4570         /*
4571          * For the OST device with 256 bytes inode size by default,
4572          * the PFID EA will be stored together with LMA EA to avoid
4573          * performance trouble. Otherwise the PFID EA can be stored
4574          * independently. LU-8998
4575          */
4576         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4577             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4578                 LASSERT(buf->lb_buf);
4579
4580                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4581                 if (fl <= 0)
4582                         RETURN(fl);
4583         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4584                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4585                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4586
4587                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4588                 if (rc)
4589                         RETURN(rc);
4590
4591                 lma->lma_incompat |= LMAI_STRIPED;
4592                 lustre_lma_swab(lma);
4593                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4594                                      sizeof(*lma), XATTR_REPLACE);
4595                 if (rc != 0)
4596                         RETURN(rc);
4597         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4598                 LASSERT(!osd->od_is_ost);
4599
4600                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4601                 if (rc)
4602                         RETURN(rc);
4603         }
4604
4605         if (fl & LU_XATTR_REPLACE)
4606                 fs_flags |= XATTR_REPLACE;
4607
4608         if (fl & LU_XATTR_CREATE)
4609                 fs_flags |= XATTR_CREATE;
4610
4611         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4612         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4613
4614         if (rc == 0 &&
4615             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4616              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4617                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4618
4619         return rc;
4620 }
4621
4622 /*
4623  * Concurrency: @dt is read locked.
4624  */
4625 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4626                           const struct lu_buf *buf)
4627 {
4628         struct osd_object *obj = osd_dt_obj(dt);
4629         struct inode *inode = obj->oo_inode;
4630         struct osd_thread_info *info = osd_oti_get(env);
4631         struct dentry *dentry = &info->oti_obj_dentry;
4632
4633         if (!dt_object_exists(dt))
4634                 return -ENOENT;
4635
4636         LASSERT(!dt_object_remote(dt));
4637         LASSERT(inode->i_op != NULL);
4638         LASSERT(inode->i_op->listxattr != NULL);
4639
4640         dentry->d_inode = inode;
4641         dentry->d_sb = inode->i_sb;
4642         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4643 }
4644
4645 static int osd_declare_xattr_del(const struct lu_env *env,
4646                                  struct dt_object *dt, const char *name,
4647                                  struct thandle *handle)
4648 {
4649         struct osd_thandle *oh;
4650         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4651
4652         LASSERT(!dt_object_remote(dt));
4653         LASSERT(handle != NULL);
4654
4655         oh = container_of0(handle, struct osd_thandle, ot_super);
4656         LASSERT(oh->ot_handle == NULL);
4657
4658         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4659                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4660         /*
4661          * xattr del may involve inode quota change, reserve credits for
4662          * dquot_initialize()
4663          */
4664         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4665
4666         return 0;
4667 }
4668
4669 /*
4670  * Concurrency: @dt is write locked.
4671  */
4672 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4673                          const char *name, struct thandle *handle)
4674 {
4675         struct osd_object *obj = osd_dt_obj(dt);
4676         struct inode *inode = obj->oo_inode;
4677         struct osd_thread_info *info = osd_oti_get(env);
4678         struct dentry *dentry = &info->oti_obj_dentry;
4679         int rc;
4680
4681         if (!dt_object_exists(dt))
4682                 return -ENOENT;
4683
4684         LASSERT(!dt_object_remote(dt));
4685         LASSERT(inode->i_op != NULL);
4686         LASSERT(handle != NULL);
4687 #ifdef HAVE_IOP_XATTR
4688         LASSERT(inode->i_op->removexattr != NULL);
4689 #endif
4690
4691         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4692
4693         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4694                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4695
4696                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4697                                  &info->oti_ost_attrs);
4698                 if (!rc) {
4699                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4700
4701                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4702                         lustre_lma_swab(lma);
4703                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4704                                              sizeof(*lma), XATTR_REPLACE);
4705                         if (!rc)
4706                                 obj->oo_pfid_in_lma = 0;
4707                 }
4708         } else {
4709                 ll_vfs_dq_init(inode);
4710                 dentry->d_inode = inode;
4711                 dentry->d_sb = inode->i_sb;
4712                 rc = osd_removexattr(dentry, inode, name);
4713         }
4714
4715         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4716
4717         if (rc == 0 &&
4718             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4719              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4720                 osd_oxc_del(obj, name);
4721
4722         return rc;
4723 }
4724
4725 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4726                            __u64 start, __u64 end)
4727 {
4728         struct osd_object *obj = osd_dt_obj(dt);
4729         struct inode *inode = obj->oo_inode;
4730         struct osd_thread_info *info = osd_oti_get(env);
4731         struct dentry *dentry = &info->oti_obj_dentry;
4732         struct file *file = &info->oti_file;
4733         int rc;
4734
4735         ENTRY;
4736
4737         dentry->d_inode = inode;
4738         dentry->d_sb = inode->i_sb;
4739         file->f_path.dentry = dentry;
4740         file->f_mapping = inode->i_mapping;
4741         file->f_op = inode->i_fop;
4742         set_file_inode(file, inode);
4743
4744         rc = ll_vfs_fsync_range(file, start, end, 0);
4745
4746         RETURN(rc);
4747 }
4748
4749 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4750 {
4751         return 0;
4752 }
4753
4754 /*
4755  * Index operations.
4756  */
4757
4758 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4759                                const struct dt_index_features *feat)
4760 {
4761         struct iam_descr *descr;
4762
4763         if (osd_object_is_root(o))
4764                 return feat == &dt_directory_features;
4765
4766         LASSERT(o->oo_dir != NULL);
4767
4768         descr = o->oo_dir->od_container.ic_descr;
4769         if (feat == &dt_directory_features) {
4770                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4771                         return 1;
4772                 else
4773                         return 0;
4774         } else {
4775                 return feat->dif_keysize_min <= descr->id_key_size &&
4776                        descr->id_key_size <= feat->dif_keysize_max &&
4777                        feat->dif_recsize_min <= descr->id_rec_size &&
4778                        descr->id_rec_size <= feat->dif_recsize_max &&
4779                        !(feat->dif_flags & (DT_IND_VARKEY |
4780                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4781                        ergo(feat->dif_flags & DT_IND_UPDATE,
4782                             1 /* XXX check that object (and fs) is writable */);
4783         }
4784 }
4785
4786 static int osd_iam_container_init(const struct lu_env *env,
4787                                   struct osd_object *obj,
4788                                   struct osd_directory *dir)
4789 {
4790         struct iam_container *bag = &dir->od_container;
4791         int result;
4792
4793         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4794         if (result != 0)
4795                 return result;
4796
4797         result = iam_container_setup(bag);
4798         if (result == 0)
4799                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4800         else
4801                 iam_container_fini(bag);
4802
4803         return result;
4804 }
4805
4806
4807 /*
4808  * Concurrency: no external locking is necessary.
4809  */
4810 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4811                          const struct dt_index_features *feat)
4812 {
4813         int result;
4814         int skip_iam = 0;
4815         struct osd_object *obj = osd_dt_obj(dt);
4816
4817         LINVRNT(osd_invariant(obj));
4818
4819         if (osd_object_is_root(obj)) {
4820                 dt->do_index_ops = &osd_index_ea_ops;
4821                 result = 0;
4822         } else if (feat == &dt_directory_features) {
4823                 dt->do_index_ops = &osd_index_ea_ops;
4824                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4825                         result = 0;
4826                 else
4827                         result = -ENOTDIR;
4828                 skip_iam = 1;
4829         } else if (unlikely(feat == &dt_otable_features)) {
4830                 dt->do_index_ops = &osd_otable_ops;
4831                 return 0;
4832         } else if (unlikely(feat == &dt_acct_features)) {
4833                 dt->do_index_ops = &osd_acct_index_ops;
4834                 result = 0;
4835                 skip_iam = 1;
4836         } else if (!osd_has_index(obj)) {
4837                 struct osd_directory *dir;
4838                 struct osd_device *osd = osd_obj2dev(obj);
4839                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4840
4841                 OBD_ALLOC_PTR(dir);
4842                 if (dir) {
4843
4844                         spin_lock(&obj->oo_guard);
4845                         if (obj->oo_dir == NULL)
4846                                 obj->oo_dir = dir;
4847                         else
4848                                 /*
4849                                  * Concurrent thread allocated container data.
4850                                  */
4851                                 OBD_FREE_PTR(dir);
4852                         spin_unlock(&obj->oo_guard);
4853                         /*
4854                          * Now, that we have container data, serialize its
4855                          * initialization.
4856                          */
4857                         down_write(&obj->oo_ext_idx_sem);
4858                         /*
4859                          * recheck under lock.
4860                          */
4861
4862                         if (osd_has_index(obj)) {
4863                                 result = 0;
4864                                 goto unlock;
4865                         }
4866
4867                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4868                         if (result || feat == &dt_lfsck_namespace_features ||
4869                             feat == &dt_lfsck_layout_orphan_features ||
4870                             feat == &dt_lfsck_layout_dangling_features)
4871                                 goto unlock;
4872
4873                         result = osd_index_register(osd, fid,
4874                                                     feat->dif_keysize_max,
4875                                                     feat->dif_recsize_max);
4876                         if (result < 0)
4877                                 CWARN("%s: failed to register index "
4878                                       DFID": rc = %d\n",
4879                                       osd_name(osd), PFID(fid), result);
4880                         else if (result > 0)
4881                                 result = 0;
4882                         else
4883                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4884                                        " (%d/%d) registered\n",
4885                                        osd_name(osd), PFID(fid),
4886                                        (int)feat->dif_keysize_max,
4887                                        (int)feat->dif_recsize_max);
4888
4889 unlock:
4890                         up_write(&obj->oo_ext_idx_sem);
4891                 } else {
4892                         result = -ENOMEM;
4893                 }
4894         } else {
4895                 result = 0;
4896         }
4897
4898         if (result == 0 && skip_iam == 0) {
4899                 if (!osd_iam_index_probe(env, obj, feat))
4900                         result = -ENOTDIR;
4901         }
4902         LINVRNT(osd_invariant(obj));
4903
4904         return result;
4905 }
4906
4907 static int osd_otable_it_attr_get(const struct lu_env *env,
4908                                  struct dt_object *dt,
4909                                  struct lu_attr *attr)
4910 {
4911         attr->la_valid = 0;
4912         return 0;
4913 }
4914
4915 static const struct dt_object_operations osd_obj_ops = {
4916         .do_read_lock           = osd_read_lock,
4917         .do_write_lock          = osd_write_lock,
4918         .do_read_unlock         = osd_read_unlock,
4919         .do_write_unlock        = osd_write_unlock,
4920         .do_write_locked        = osd_write_locked,
4921         .do_attr_get            = osd_attr_get,
4922         .do_declare_attr_set    = osd_declare_attr_set,
4923         .do_attr_set            = osd_attr_set,
4924         .do_ah_init             = osd_ah_init,
4925         .do_declare_create      = osd_declare_create,
4926         .do_create              = osd_create,
4927         .do_declare_destroy     = osd_declare_destroy,
4928         .do_destroy             = osd_destroy,
4929         .do_index_try           = osd_index_try,
4930         .do_declare_ref_add     = osd_declare_ref_add,
4931         .do_ref_add             = osd_ref_add,
4932         .do_declare_ref_del     = osd_declare_ref_del,
4933         .do_ref_del             = osd_ref_del,
4934         .do_xattr_get           = osd_xattr_get,
4935         .do_declare_xattr_set   = osd_declare_xattr_set,
4936         .do_xattr_set           = osd_xattr_set,
4937         .do_declare_xattr_del   = osd_declare_xattr_del,
4938         .do_xattr_del           = osd_xattr_del,
4939         .do_xattr_list          = osd_xattr_list,
4940         .do_object_sync         = osd_object_sync,
4941         .do_invalidate          = osd_invalidate,
4942 };
4943
4944 static const struct dt_object_operations osd_obj_otable_it_ops = {
4945         .do_attr_get    = osd_otable_it_attr_get,
4946         .do_index_try   = osd_index_try,
4947 };
4948
4949 static int osd_index_declare_iam_delete(const struct lu_env *env,
4950                                         struct dt_object *dt,
4951                                         const struct dt_key *key,
4952                                         struct thandle *handle)
4953 {
4954         struct osd_thandle *oh;
4955
4956         oh = container_of0(handle, struct osd_thandle, ot_super);
4957         LASSERT(oh->ot_handle == NULL);
4958
4959         /* Recycle  may cause additional three blocks to be changed. */
4960         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4961                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4962
4963         return 0;
4964 }
4965
4966 /**
4967  *      delete a (key, value) pair from index \a dt specified by \a key
4968  *
4969  *      \param  dt      osd index object
4970  *      \param  key     key for index
4971  *      \param  rec     record reference
4972  *      \param  handle  transaction handler
4973  *
4974  *      \retval  0  success
4975  *      \retval -ve   failure
4976  */
4977 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
4978                                 const struct dt_key *key,
4979                                 struct thandle *handle)
4980 {
4981         struct osd_thread_info *oti = osd_oti_get(env);
4982         struct osd_object *obj = osd_dt_obj(dt);
4983         struct osd_thandle *oh;
4984         struct iam_path_descr *ipd;
4985         struct iam_container *bag = &obj->oo_dir->od_container;
4986         int rc;
4987
4988         ENTRY;
4989
4990         if (!dt_object_exists(dt))
4991                 RETURN(-ENOENT);
4992
4993         LINVRNT(osd_invariant(obj));
4994         LASSERT(!dt_object_remote(dt));
4995         LASSERT(bag->ic_object == obj->oo_inode);
4996         LASSERT(handle != NULL);
4997
4998         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
4999
5000         ipd = osd_idx_ipd_get(env, bag);
5001         if (unlikely(ipd == NULL))
5002                 RETURN(-ENOMEM);
5003
5004         oh = container_of0(handle, struct osd_thandle, ot_super);
5005         LASSERT(oh->ot_handle != NULL);
5006         LASSERT(oh->ot_handle->h_transaction != NULL);
5007
5008         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5009                 /* swab quota uid/gid provided by caller */
5010                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5011                 key = (const struct dt_key *)&oti->oti_quota_id;
5012         }
5013
5014         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5015         osd_ipd_put(env, bag, ipd);
5016         LINVRNT(osd_invariant(obj));
5017         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5018         RETURN(rc);
5019 }
5020
5021 static int osd_index_declare_ea_delete(const struct lu_env *env,
5022                                        struct dt_object *dt,
5023                                        const struct dt_key *key,
5024                                        struct thandle *handle)
5025 {
5026         struct osd_thandle *oh;
5027         struct inode *inode;
5028         int rc, credits;
5029
5030         ENTRY;
5031
5032         LASSERT(!dt_object_remote(dt));
5033         LASSERT(handle != NULL);
5034
5035         oh = container_of0(handle, struct osd_thandle, ot_super);
5036         LASSERT(oh->ot_handle == NULL);
5037
5038         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5039         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5040
5041         inode = osd_dt_obj(dt)->oo_inode;
5042         if (inode == NULL)
5043                 RETURN(-ENOENT);
5044
5045         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5046                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5047                                    NULL, OSD_QID_BLK);
5048         RETURN(rc);
5049 }
5050
5051 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5052                                           struct dt_rec *fid)
5053 {
5054         struct osd_fid_pack *rec;
5055         int rc = -ENODATA;
5056
5057         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5058                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5059                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5060                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5061                         rc = -EINVAL;
5062         }
5063         return rc;
5064 }
5065
5066 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5067                           const struct lu_fid *fid)
5068 {
5069         struct seq_server_site *ss = osd_seq_site(osd);
5070
5071         ENTRY;
5072
5073         /* FID seqs not in FLDB, must be local seq */
5074         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5075                 RETURN(0);
5076
5077         /*
5078          * If FLD is not being initialized yet, it only happens during the
5079          * initialization, likely during mgs initialization, and we assume
5080          * this is local FID.
5081          */
5082         if (ss == NULL || ss->ss_server_fld == NULL)
5083                 RETURN(0);
5084
5085         /* Only check the local FLDB here */
5086         if (osd_seq_exists(env, osd, fid_seq(fid)))
5087                 RETURN(0);
5088
5089         RETURN(1);
5090 }
5091
5092 static void osd_take_care_of_agent(const struct lu_env *env,
5093                                    struct osd_device *osd,
5094                                    struct osd_thandle *oh,
5095                                    struct ldiskfs_dir_entry_2 *de)
5096 {
5097         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5098         struct osd_idmap_cache *idc;
5099         int rc, schedule = 0;
5100
5101         LASSERT(de != NULL);
5102
5103         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5104         if (likely(rc == 0)) {
5105                 idc = osd_idc_find_or_init(env, osd, fid);
5106                 if (IS_ERR(idc) || idc->oic_remote)
5107                         schedule = 1;
5108         } else if (rc == -ENODATA) {
5109                 /*
5110                  * can't get FID, postpone to the end of the
5111                  * transaction when iget() is safe
5112                  */
5113                 schedule = 1;
5114         } else {
5115                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5116         }
5117         if (schedule)
5118                 osd_schedule_agent_inode_removal(env, oh,
5119                                                  le32_to_cpu(de->inode));
5120 }
5121
5122 /**
5123  * Index delete function for interoperability mode (b11826).
5124  * It will remove the directory entry added by osd_index_ea_insert().
5125  * This entry is needed to maintain name->fid mapping.
5126  *
5127  * \param key,  key i.e. file entry to be deleted
5128  *
5129  * \retval   0, on success
5130  * \retval -ve, on error
5131  */
5132 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5133                                const struct dt_key *key, struct thandle *handle)
5134 {
5135         struct osd_object *obj = osd_dt_obj(dt);
5136         struct inode *dir = obj->oo_inode;
5137         struct dentry *dentry;
5138         struct osd_thandle *oh;
5139         struct ldiskfs_dir_entry_2 *de = NULL;
5140         struct buffer_head *bh;
5141         struct htree_lock *hlock = NULL;
5142         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5143         int rc;
5144
5145         ENTRY;
5146
5147         if (!dt_object_exists(dt))
5148                 RETURN(-ENOENT);
5149
5150         LINVRNT(osd_invariant(obj));
5151         LASSERT(!dt_object_remote(dt));
5152         LASSERT(handle != NULL);
5153
5154         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5155
5156         oh = container_of(handle, struct osd_thandle, ot_super);
5157         LASSERT(oh->ot_handle != NULL);
5158         LASSERT(oh->ot_handle->h_transaction != NULL);
5159
5160         ll_vfs_dq_init(dir);
5161         dentry = osd_child_dentry_get(env, obj,
5162                                       (char *)key, strlen((char *)key));
5163
5164         if (obj->oo_hl_head != NULL) {
5165                 hlock = osd_oti_get(env)->oti_hlock;
5166                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5167                                    dir, LDISKFS_HLOCK_DEL);
5168         } else {
5169                 down_write(&obj->oo_ext_idx_sem);
5170         }
5171
5172         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5173         if (!IS_ERR(bh)) {
5174                 /*
5175                  * If this is not the ".." entry, it might be a remote DNE
5176                  * entry and  we need to check if the FID is for a remote
5177                  * MDT.  If the FID is  not in the directory entry (e.g.
5178                  * upgraded 1.8 filesystem without dirdata enabled) then
5179                  * we need to get the FID from the LMA. For a remote directory
5180                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5181                  * case.
5182                  *
5183                  * Delete the entry before the agent inode in order to
5184                  * simplify error handling.  At worst an error after deleting
5185                  * the entry first might leak the agent inode afterward. The
5186                  * reverse would need filesystem abort in case of error deleting
5187                  * the entry after the agent had been removed, or leave a
5188                  * dangling entry pointing at a random inode.
5189                  */
5190                 if (strcmp((char *)key, dotdot) != 0)
5191                         osd_take_care_of_agent(env, osd, oh, de);
5192                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5193                 brelse(bh);
5194         } else {
5195                 rc = PTR_ERR(bh);
5196         }
5197         if (hlock != NULL)
5198                 ldiskfs_htree_unlock(hlock);
5199         else
5200                 up_write(&obj->oo_ext_idx_sem);
5201
5202         GOTO(out, rc);
5203 out:
5204         LASSERT(osd_invariant(obj));
5205         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5206         RETURN(rc);
5207 }
5208
5209 /**
5210  *      Lookup index for \a key and copy record to \a rec.
5211  *
5212  *      \param  dt      osd index object
5213  *      \param  key     key for index
5214  *      \param  rec     record reference
5215  *
5216  *      \retval  +ve  success : exact mach
5217  *      \retval  0    return record with key not greater than \a key
5218  *      \retval -ve   failure
5219  */
5220 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5221                                 struct dt_rec *rec, const struct dt_key *key)
5222 {
5223         struct osd_object *obj = osd_dt_obj(dt);
5224         struct iam_path_descr *ipd;
5225         struct iam_container *bag = &obj->oo_dir->od_container;
5226         struct osd_thread_info *oti = osd_oti_get(env);
5227         struct iam_iterator *it = &oti->oti_idx_it;
5228         struct iam_rec *iam_rec;
5229         int rc;
5230
5231         ENTRY;
5232
5233         if (!dt_object_exists(dt))
5234                 RETURN(-ENOENT);
5235
5236         LASSERT(osd_invariant(obj));
5237         LASSERT(!dt_object_remote(dt));
5238         LASSERT(bag->ic_object == obj->oo_inode);
5239
5240         ipd = osd_idx_ipd_get(env, bag);
5241         if (IS_ERR(ipd))
5242                 RETURN(-ENOMEM);
5243
5244         /* got ipd now we can start iterator. */
5245         iam_it_init(it, bag, 0, ipd);
5246
5247         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5248                 /* swab quota uid/gid provided by caller */
5249                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5250                 key = (const struct dt_key *)&oti->oti_quota_id;
5251         }
5252
5253         rc = iam_it_get(it, (struct iam_key *)key);
5254         if (rc >= 0) {
5255                 if (S_ISDIR(obj->oo_inode->i_mode))
5256                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5257                 else
5258                         iam_rec = (struct iam_rec *)rec;
5259
5260                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5261
5262                 if (S_ISDIR(obj->oo_inode->i_mode))
5263                         osd_fid_unpack((struct lu_fid *)rec,
5264                                        (struct osd_fid_pack *)iam_rec);
5265                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5266                         osd_quota_unpack(obj, rec);
5267         }
5268
5269         iam_it_put(it);
5270         iam_it_fini(it);
5271         osd_ipd_put(env, bag, ipd);
5272
5273         LINVRNT(osd_invariant(obj));
5274
5275         RETURN(rc);
5276 }
5277
5278 static int osd_index_declare_iam_insert(const struct lu_env *env,
5279                                         struct dt_object *dt,
5280                                         const struct dt_rec *rec,
5281                                         const struct dt_key *key,
5282                                         struct thandle *handle)
5283 {
5284         struct osd_thandle *oh;
5285
5286         LASSERT(handle != NULL);
5287
5288         oh = container_of0(handle, struct osd_thandle, ot_super);
5289         LASSERT(oh->ot_handle == NULL);
5290
5291         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5292                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5293
5294         return 0;
5295 }
5296
5297 /**
5298  *      Inserts (key, value) pair in \a dt index object.
5299  *
5300  *      \param  dt      osd index object
5301  *      \param  key     key for index
5302  *      \param  rec     record reference
5303  *      \param  th      transaction handler
5304  *
5305  *      \retval  0  success
5306  *      \retval -ve failure
5307  */
5308 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5309                                 const struct dt_rec *rec,
5310                                 const struct dt_key *key, struct thandle *th)
5311 {
5312         struct osd_object *obj = osd_dt_obj(dt);
5313         struct iam_path_descr *ipd;
5314         struct osd_thandle *oh;
5315         struct iam_container *bag;
5316         struct osd_thread_info *oti = osd_oti_get(env);
5317         struct iam_rec *iam_rec;
5318         int rc;
5319
5320         ENTRY;
5321
5322         if (!dt_object_exists(dt))
5323                 RETURN(-ENOENT);
5324
5325         LINVRNT(osd_invariant(obj));
5326         LASSERT(!dt_object_remote(dt));
5327
5328         bag = &obj->oo_dir->od_container;
5329         LASSERT(bag->ic_object == obj->oo_inode);
5330         LASSERT(th != NULL);
5331
5332         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5333
5334         ipd = osd_idx_ipd_get(env, bag);
5335         if (unlikely(ipd == NULL))
5336                 RETURN(-ENOMEM);
5337
5338         oh = container_of0(th, struct osd_thandle, ot_super);
5339         LASSERT(oh->ot_handle != NULL);
5340         LASSERT(oh->ot_handle->h_transaction != NULL);
5341         if (S_ISDIR(obj->oo_inode->i_mode)) {
5342                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5343                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5344                              &oti->oti_fid);
5345         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5346                 /* pack quota uid/gid */
5347                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5348                 key = (const struct dt_key *)&oti->oti_quota_id;
5349                 /* pack quota record */
5350                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5351                 iam_rec = (struct iam_rec *)rec;
5352         } else {
5353                 iam_rec = (struct iam_rec *)rec;
5354         }
5355
5356         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5357                         iam_rec, ipd);
5358         osd_ipd_put(env, bag, ipd);
5359         LINVRNT(osd_invariant(obj));
5360         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5361         RETURN(rc);
5362 }
5363
5364 /**
5365  * Calls ldiskfs_add_entry() to add directory entry
5366  * into the directory. This is required for
5367  * interoperability mode (b11826)
5368  *
5369  * \retval   0, on success
5370  * \retval -ve, on error
5371  */
5372 static int __osd_ea_add_rec(struct osd_thread_info *info,
5373                             struct osd_object *pobj, struct inode  *cinode,
5374                             const char *name, const struct lu_fid *fid,
5375                             struct htree_lock *hlock, struct thandle *th)
5376 {
5377         struct ldiskfs_dentry_param *ldp;
5378         struct dentry *child;
5379         struct osd_thandle *oth;
5380         int rc;
5381
5382         oth = container_of(th, struct osd_thandle, ot_super);
5383         LASSERT(oth->ot_handle != NULL);
5384         LASSERT(oth->ot_handle->h_transaction != NULL);
5385         LASSERT(pobj->oo_inode);
5386
5387         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5388         if (unlikely(pobj->oo_inode ==
5389                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5390                 ldp->edp_magic = 0;
5391         else
5392                 osd_get_ldiskfs_dirent_param(ldp, fid);
5393         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5394         child->d_fsdata = (void *)ldp;
5395         ll_vfs_dq_init(pobj->oo_inode);
5396         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5397                                    child, cinode, hlock);
5398         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5399                 struct ldiskfs_dir_entry_2      *de;
5400                 struct buffer_head              *bh;
5401                 int                              rc1;
5402
5403                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5404                                             NULL, hlock);
5405                 if (!IS_ERR(bh)) {
5406                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5407                                                                bh);
5408                         if (rc1 == 0) {
5409                                 if (S_ISDIR(cinode->i_mode))
5410                                         de->file_type = LDISKFS_DIRENT_LUFID |
5411                                                         LDISKFS_FT_REG_FILE;
5412                                 else
5413                                         de->file_type = LDISKFS_DIRENT_LUFID |
5414                                                         LDISKFS_FT_DIR;
5415                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5416                                                               NULL, bh);
5417                         }
5418                         brelse(bh);
5419                 }
5420         }
5421
5422         RETURN(rc);
5423 }
5424
5425 /**
5426  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5427  * into the directory.Also sets flags into osd object to
5428  * indicate dot and dotdot are created. This is required for
5429  * interoperability mode (b11826)
5430  *
5431  * \param dir   directory for dot and dotdot fixup.
5432  * \param obj   child object for linking
5433  *
5434  * \retval   0, on success
5435  * \retval -ve, on error
5436  */
5437 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5438                               struct osd_object *dir,
5439                               struct inode *parent_dir, const char *name,
5440                               const struct lu_fid *dot_fid,
5441                               const struct lu_fid *dot_dot_fid,
5442                               struct thandle *th)
5443 {
5444         struct inode *inode = dir->oo_inode;
5445         struct osd_thandle *oth;
5446         int result = 0;
5447
5448         oth = container_of(th, struct osd_thandle, ot_super);
5449         LASSERT(oth->ot_handle->h_transaction != NULL);
5450         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5451
5452         if (strcmp(name, dot) == 0) {
5453                 if (dir->oo_compat_dot_created) {
5454                         result = -EEXIST;
5455                 } else {
5456                         LASSERT(inode->i_ino == parent_dir->i_ino);
5457                         dir->oo_compat_dot_created = 1;
5458                         result = 0;
5459                 }
5460         } else if (strcmp(name, dotdot) == 0) {
5461                 if (!dir->oo_compat_dot_created)
5462                         return -EINVAL;
5463                 /* in case of rename, dotdot is already created */
5464                 if (dir->oo_compat_dotdot_created) {
5465                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5466                                                 dot_dot_fid, NULL, th);
5467                 }
5468
5469                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5470                         struct lu_fid tfid = *dot_dot_fid;
5471
5472                         tfid.f_oid--;
5473                         result = osd_add_dot_dotdot_internal(info,
5474                                         dir->oo_inode, parent_dir, dot_fid,
5475                                         &tfid, oth);
5476                 } else {
5477                         result = osd_add_dot_dotdot_internal(info,
5478                                         dir->oo_inode, parent_dir, dot_fid,
5479                                         dot_dot_fid, oth);
5480                 }
5481
5482                 if (result == 0)
5483                         dir->oo_compat_dotdot_created = 1;
5484         }
5485
5486         return result;
5487 }
5488
5489
5490 /**
5491  * It will call the appropriate osd_add* function and return the
5492  * value, return by respective functions.
5493  */
5494 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5495                           struct inode *cinode, const char *name,
5496                           const struct lu_fid *fid, struct thandle *th)
5497 {
5498         struct osd_thread_info *info = osd_oti_get(env);
5499         struct htree_lock *hlock;
5500         int rc;
5501
5502         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5503
5504         if (name[0] == '.' && (name[1] == '\0' ||
5505                                (name[1] == '.' && name[2] == '\0'))) {
5506                 if (hlock != NULL) {
5507                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5508                                            pobj->oo_inode, 0);
5509                 } else {
5510                         down_write(&pobj->oo_ext_idx_sem);
5511                 }
5512
5513                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5514                                         lu_object_fid(&pobj->oo_dt.do_lu),
5515                                         fid, th);
5516         } else {
5517                 if (hlock != NULL) {
5518                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5519                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5520                 } else {
5521                         down_write(&pobj->oo_ext_idx_sem);
5522                 }
5523
5524                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5525                         struct lu_fid *tfid = &info->oti_fid;
5526
5527                         *tfid = *fid;
5528                         tfid->f_ver = ~0;
5529                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5530                                               tfid, hlock, th);
5531                 } else {
5532                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5533                                               hlock, th);
5534                 }
5535         }
5536         if (hlock != NULL)
5537                 ldiskfs_htree_unlock(hlock);
5538         else
5539                 up_write(&pobj->oo_ext_idx_sem);
5540
5541         return rc;
5542 }
5543
5544 static int
5545 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5546                       struct osd_idmap_cache *oic)
5547 {
5548         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5549         struct lu_fid *fid = &oic->oic_fid;
5550         struct osd_inode_id *id = &oic->oic_lid;
5551         struct inode *inode = NULL;
5552         int once = 0;
5553         bool insert;
5554         int rc;
5555
5556         ENTRY;
5557
5558         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5559                 RETURN(0);
5560
5561         if (thread_is_running(&scrub->os_thread) &&
5562             scrub->os_pos_current > id->oii_ino)
5563                 RETURN(0);
5564
5565         if (dev->od_auto_scrub_interval == AS_NEVER ||
5566             ktime_get_real_seconds() <
5567             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5568                 RETURN(0);
5569
5570 again:
5571         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5572         if (rc == -ENOENT) {
5573                 __u32 gen = id->oii_gen;
5574
5575                 insert = true;
5576                 if (inode != NULL)
5577                         goto trigger;
5578
5579                 inode = osd_iget(oti, dev, id);
5580                 /* The inode has been removed (by race maybe). */
5581                 if (IS_ERR(inode)) {
5582                         rc = PTR_ERR(inode);
5583
5584                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5585                 }
5586
5587                 /* The OI mapping is lost. */
5588                 if (gen != OSD_OII_NOGEN)
5589                         goto trigger;
5590
5591                 /*
5592                  * The inode may has been reused by others, we do not know,
5593                  * leave it to be handled by subsequent osd_fid_lookup().
5594                  */
5595                 GOTO(out, rc = 0);
5596         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5597                 GOTO(out, rc);
5598         }
5599
5600         insert = false;
5601
5602 trigger:
5603         if (thread_is_running(&scrub->os_thread)) {
5604                 if (inode == NULL) {
5605                         inode = osd_iget(oti, dev, id);
5606                         /* The inode has been removed (by race maybe). */
5607                         if (IS_ERR(inode)) {
5608                                 rc = PTR_ERR(inode);
5609
5610                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5611                         }
5612                 }
5613
5614                 rc = osd_oii_insert(dev, oic, insert);
5615                 /*
5616                  * There is race condition between osd_oi_lookup and OI scrub.
5617                  * The OI scrub finished just after osd_oi_lookup() failure.
5618                  * Under such case, it is unnecessary to trigger OI scrub again,
5619                  * but try to call osd_oi_lookup() again.
5620                  */
5621                 if (unlikely(rc == -EAGAIN))
5622                         goto again;
5623
5624                 if (!S_ISDIR(inode->i_mode))
5625                         rc = 0;
5626                 else
5627                         rc = osd_check_lmv(oti, dev, inode, oic);
5628
5629                 GOTO(out, rc);
5630         }
5631
5632         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5633                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5634                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5635                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5636                        "%s: trigger partial OI scrub for RPC inconsistency "
5637                        "checking FID "DFID": rc = %d\n",
5638                        osd_dev2name(dev), PFID(fid), rc);
5639                 if (rc == 0 || rc == -EALREADY)
5640                         goto again;
5641         }
5642
5643         GOTO(out, rc);
5644
5645 out:
5646         if (inode)
5647                 iput(inode);
5648
5649         RETURN(rc);
5650 }
5651
5652 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5653                                struct osd_device *dev,
5654                                struct osd_idmap_cache *oic,
5655                                struct lu_fid *fid, __u32 ino)
5656 {
5657         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5658         struct inode *inode;
5659         int rc;
5660
5661         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5662         inode = osd_iget(oti, dev, &oic->oic_lid);
5663         if (IS_ERR(inode)) {
5664                 fid_zero(&oic->oic_fid);
5665                 return PTR_ERR(inode);
5666         }
5667
5668         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5669         iput(inode);
5670         if (rc != 0)
5671                 fid_zero(&oic->oic_fid);
5672         else
5673                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5674         return rc;
5675 }
5676
5677 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5678                       struct osd_inode_id *id, const struct lu_fid *fid)
5679 {
5680         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5681                id->oii_ino, id->oii_gen, info);
5682         info->oti_cache.oic_lid = *id;
5683         info->oti_cache.oic_fid = *fid;
5684         info->oti_cache.oic_dev = osd;
5685 }
5686
5687 /**
5688  * Get parent FID from the linkEA.
5689  *
5690  * For a directory which parent resides on remote MDT, to satisfy the
5691  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5692  * the other hand, to make the lookup(..) on the directory can return
5693  * the real parent FID, we append the real parent FID after its ".."
5694  * name entry in the /REMOTE_PARENT_DIR.
5695  *
5696  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5697  * backup. So after the restore, we cannot get the right parent FID from
5698  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5699  * we have stored the real parent FID in the directory object's linkEA,
5700  * we can parse the linkEA for the real parent FID.
5701  *
5702  * \param[in] env       pointer to the thread context
5703  * \param[in] obj       pointer to the object to be handled
5704  * \param[out]fid       pointer to the buffer to hold the parent FID
5705  *
5706  * \retval              0 for getting the real parent FID successfully
5707  * \retval              negative error number on failure
5708  */
5709 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5710                                     struct osd_object *obj,
5711                                     struct lu_fid *fid)
5712 {
5713         struct osd_thread_info *oti = osd_oti_get(env);
5714         struct lu_buf *buf = &oti->oti_big_buf;
5715         struct dentry *dentry = &oti->oti_obj_dentry;
5716         struct inode *inode = obj->oo_inode;
5717         struct linkea_data ldata = { NULL };
5718         int rc;
5719
5720         ENTRY;
5721
5722         fid_zero(fid);
5723         if (!S_ISDIR(inode->i_mode))
5724                 RETURN(-EIO);
5725
5726 again:
5727         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5728                              buf->lb_buf, buf->lb_len);
5729         if (rc == -ERANGE) {
5730                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5731                                      NULL, 0);
5732                 if (rc > 0) {
5733                         lu_buf_realloc(buf, rc);
5734                         if (buf->lb_buf == NULL)
5735                                 RETURN(-ENOMEM);
5736
5737                         goto again;
5738                 }
5739         }
5740
5741         if (unlikely(rc == 0))
5742                 RETURN(-ENODATA);
5743
5744         if (rc < 0)
5745                 RETURN(rc);
5746
5747         if (unlikely(buf->lb_buf == NULL)) {
5748                 lu_buf_realloc(buf, rc);
5749                 if (buf->lb_buf == NULL)
5750                         RETURN(-ENOMEM);
5751
5752                 goto again;
5753         }
5754
5755         ldata.ld_buf = buf;
5756         rc = linkea_init_with_rec(&ldata);
5757         if (!rc) {
5758                 linkea_first_entry(&ldata);
5759                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5760         }
5761
5762         RETURN(rc);
5763 }
5764
5765 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5766                                     struct inode *inode,
5767                                     const struct lu_fid *pfid,
5768                                     const char *name, const int namelen)
5769 {
5770         struct osd_thread_info *oti = osd_oti_get(env);
5771         struct lu_buf *buf = &oti->oti_big_buf;
5772         struct dentry *dentry = &oti->oti_obj_dentry;
5773         struct linkea_data ldata = { NULL };
5774         struct lu_name cname = { .ln_name = name,
5775                                  .ln_namelen = namelen };
5776         int rc;
5777
5778         ENTRY;
5779
5780 again:
5781         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5782                              buf->lb_buf, buf->lb_len);
5783         if (rc == -ERANGE)
5784                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5785
5786         if (rc < 0)
5787                 RETURN(rc);
5788
5789         if (unlikely(rc == 0))
5790                 RETURN(-ENODATA);
5791
5792         if (buf->lb_len < rc) {
5793                 lu_buf_realloc(buf, rc);
5794                 if (buf->lb_buf == NULL)
5795                         RETURN(-ENOMEM);
5796
5797                 goto again;
5798         }
5799
5800         ldata.ld_buf = buf;
5801         rc = linkea_init_with_rec(&ldata);
5802         if (!rc)
5803                 rc = linkea_links_find(&ldata, &cname, pfid);
5804
5805         RETURN(rc);
5806 }
5807
5808 /**
5809  * Calls ->lookup() to find dentry. From dentry get inode and
5810  * read inode's ea to get fid. This is required for  interoperability
5811  * mode (b11826)
5812  *
5813  * \retval   0, on success
5814  * \retval -ve, on error
5815  */
5816 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5817                              struct dt_rec *rec, const struct dt_key *key)
5818 {
5819         struct inode *dir = obj->oo_inode;
5820         struct dentry *dentry;
5821         struct ldiskfs_dir_entry_2 *de;
5822         struct buffer_head *bh;
5823         struct lu_fid *fid = (struct lu_fid *)rec;
5824         struct htree_lock *hlock = NULL;
5825         int ino;
5826         int rc;
5827
5828         ENTRY;
5829
5830         LASSERT(dir->i_op != NULL);
5831         LASSERT(dir->i_op->lookup != NULL);
5832
5833         dentry = osd_child_dentry_get(env, obj,
5834                                       (char *)key, strlen((char *)key));
5835
5836         if (obj->oo_hl_head != NULL) {
5837                 hlock = osd_oti_get(env)->oti_hlock;
5838                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5839                                    dir, LDISKFS_HLOCK_LOOKUP);
5840         } else {
5841                 down_read(&obj->oo_ext_idx_sem);
5842         }
5843
5844         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5845         if (!IS_ERR(bh)) {
5846                 struct osd_thread_info *oti = osd_oti_get(env);
5847                 struct osd_inode_id *id = &oti->oti_id;
5848                 struct osd_idmap_cache *oic = &oti->oti_cache;
5849                 struct osd_device *dev = osd_obj2dev(obj);
5850
5851                 ino = le32_to_cpu(de->inode);
5852                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5853                         brelse(bh);
5854                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5855                         GOTO(out, rc);
5856                 }
5857
5858                 rc = osd_get_fid_from_dentry(de, rec);
5859
5860                 /* done with de, release bh */
5861                 brelse(bh);
5862                 if (rc != 0) {
5863                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5864                                 const char *name = (const char *)key;
5865
5866                                 /*
5867                                  * If the parent is on remote MDT, and there
5868                                  * is no FID-in-dirent, then we have to get
5869                                  * the parent FID from the linkEA.
5870                                  */
5871                                 if (likely(strlen(name) == 2 &&
5872                                            name[0] == '.' && name[1] == '.'))
5873                                         rc = osd_get_pfid_from_linkea(env, obj,
5874                                                                       fid);
5875                         } else {
5876                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5877                         }
5878                 } else {
5879                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5880                 }
5881
5882                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5883                         fid_zero(&oic->oic_fid);
5884
5885                         GOTO(out, rc);
5886                 }
5887
5888                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5889                 rc = osd_consistency_check(oti, dev, oic);
5890                 if (rc == -ENOENT)
5891                         fid_zero(&oic->oic_fid);
5892                 else
5893                         /* Other error should not affect lookup result. */
5894                         rc = 0;
5895         } else {
5896                 rc = PTR_ERR(bh);
5897         }
5898
5899         GOTO(out, rc);
5900
5901 out:
5902         if (hlock != NULL)
5903                 ldiskfs_htree_unlock(hlock);
5904         else
5905                 up_read(&obj->oo_ext_idx_sem);
5906         return rc;
5907 }
5908
5909 static int osd_index_declare_ea_insert(const struct lu_env *env,
5910                                        struct dt_object *dt,
5911                                        const struct dt_rec *rec,
5912                                        const struct dt_key *key,
5913                                        struct thandle *handle)
5914 {
5915         struct osd_thandle *oh;
5916         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5917         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5918         const struct lu_fid *fid = rec1->rec_fid;
5919         int credits, rc = 0;
5920         struct osd_idmap_cache *idc;
5921
5922         ENTRY;
5923
5924         LASSERT(!dt_object_remote(dt));
5925         LASSERT(handle != NULL);
5926         LASSERT(fid != NULL);
5927         LASSERT(rec1->rec_type != 0);
5928
5929         oh = container_of0(handle, struct osd_thandle, ot_super);
5930         LASSERT(oh->ot_handle == NULL);
5931
5932         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5933
5934         /*
5935          * we can't call iget() while a transactions is running
5936          * (this can lead to a deadlock), but we need to know
5937          * inum and object type. so we find this information at
5938          * declaration and cache in per-thread info
5939          */
5940         idc = osd_idc_find_or_init(env, osd, fid);
5941         if (IS_ERR(idc))
5942                 RETURN(PTR_ERR(idc));
5943         if (idc->oic_remote) {
5944                 /*
5945                  * a reference to remote inode is represented by an
5946                  * agent inode which we have to create
5947                  */
5948                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5949                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5950         }
5951
5952         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5953
5954         if (osd_dt_obj(dt)->oo_inode != NULL) {
5955                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5956
5957                 /*
5958                  * We ignore block quota on meta pool (MDTs), so needn't
5959                  * calculate how many blocks will be consumed by this index
5960                  * insert
5961                  */
5962                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5963                                            i_gid_read(inode),
5964                                            i_projid_read(inode), 0,
5965                                            oh, osd_dt_obj(dt), NULL,
5966                                            OSD_QID_BLK);
5967                 if (rc)
5968                         RETURN(rc);
5969
5970 #ifdef HAVE_PROJECT_QUOTA
5971                 /*
5972                  * Reserve credits for local agent inode to transfer
5973                  * to 0, quota enforcement is ignored in this case.
5974                  */
5975                 if (idc->oic_remote &&
5976                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
5977                     i_projid_read(inode) != 0)
5978                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
5979                                                   0, i_projid_read(inode),
5980                                                   0, false, PRJQUOTA, true);
5981 #endif
5982         }
5983
5984         RETURN(rc);
5985 }
5986
5987 /**
5988  * Index add function for interoperability mode (b11826).
5989  * It will add the directory entry.This entry is needed to
5990  * maintain name->fid mapping.
5991  *
5992  * \param key it is key i.e. file entry to be inserted
5993  * \param rec it is value of given key i.e. fid
5994  *
5995  * \retval   0, on success
5996  * \retval -ve, on error
5997  */
5998 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
5999                                const struct dt_rec *rec,
6000                                const struct dt_key *key, struct thandle *th)
6001 {
6002         struct osd_object *obj = osd_dt_obj(dt);
6003         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6004         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6005         const struct lu_fid *fid = rec1->rec_fid;
6006         const char *name = (const char *)key;
6007         struct osd_thread_info *oti = osd_oti_get(env);
6008         struct inode *child_inode = NULL;
6009         struct osd_idmap_cache *idc;
6010         int rc;
6011
6012         ENTRY;
6013
6014         if (!dt_object_exists(dt))
6015                 RETURN(-ENOENT);
6016
6017         LASSERT(osd_invariant(obj));
6018         LASSERT(!dt_object_remote(dt));
6019         LASSERT(th != NULL);
6020
6021         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6022
6023         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6024
6025         idc = osd_idc_find(env, osd, fid);
6026         if (unlikely(idc == NULL)) {
6027                 idc = osd_idc_find_or_init(env, osd, fid);
6028                 if (IS_ERR(idc)) {
6029                         /*
6030                          * this dt_insert() wasn't declared properly, so
6031                          * FID is missing in OI cache. we better do not
6032                          * lookup FID in FLDB/OI and don't risk to deadlock,
6033                          * but in some special cases (lfsck testing, etc)
6034                          * it's much simpler than fixing a caller.
6035                          *
6036                          * normally this error should be placed after the first
6037                          * find, but migrate may attach source stripes to
6038                          * target, which doesn't create stripes.
6039                          */
6040                         CERROR("%s: "DFID" wasn't declared for insert\n",
6041                                osd_name(osd), PFID(fid));
6042                         dump_stack();
6043                         RETURN(PTR_ERR(idc));
6044                 }
6045         }
6046
6047         if (idc->oic_remote) {
6048                 /* Insert remote entry */
6049                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6050                         child_inode =
6051                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6052                 } else {
6053                         child_inode = osd_create_local_agent_inode(env, osd,
6054                                         obj, fid, rec1->rec_type & S_IFMT, th);
6055                         if (IS_ERR(child_inode))
6056                                 RETURN(PTR_ERR(child_inode));
6057                 }
6058         } else {
6059                 /* Insert local entry */
6060                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6061                         /* for a reason OI cache wasn't filled properly */
6062                         CERROR("%s: OIC for "DFID" isn't filled\n",
6063                                osd_name(osd), PFID(fid));
6064                         RETURN(-EINVAL);
6065                 }
6066                 child_inode = oti->oti_inode;
6067                 if (unlikely(child_inode == NULL)) {
6068                         struct ldiskfs_inode_info *lii;
6069
6070                         OBD_ALLOC_PTR(lii);
6071                         if (lii == NULL)
6072                                 RETURN(-ENOMEM);
6073                         child_inode = oti->oti_inode = &lii->vfs_inode;
6074                 }
6075                 child_inode->i_sb = osd_sb(osd);
6076                 child_inode->i_ino = idc->oic_lid.oii_ino;
6077                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6078         }
6079
6080         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6081
6082         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6083                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6084
6085         if (child_inode && child_inode != oti->oti_inode)
6086                 iput(child_inode);
6087         LASSERT(osd_invariant(obj));
6088         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6089         RETURN(rc);
6090 }
6091
6092 /**
6093  *  Initialize osd Iterator for given osd index object.
6094  *
6095  *  \param  dt      osd index object
6096  */
6097
6098 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6099                                      struct dt_object *dt,
6100                                      __u32 unused)
6101 {
6102         struct osd_it_iam *it;
6103         struct osd_object *obj = osd_dt_obj(dt);
6104         struct lu_object *lo = &dt->do_lu;
6105         struct iam_path_descr *ipd;
6106         struct iam_container *bag = &obj->oo_dir->od_container;
6107
6108         if (!dt_object_exists(dt))
6109                 return ERR_PTR(-ENOENT);
6110
6111         OBD_ALLOC_PTR(it);
6112         if (it == NULL)
6113                 return ERR_PTR(-ENOMEM);
6114
6115         ipd = osd_it_ipd_get(env, bag);
6116         if (likely(ipd != NULL)) {
6117                 it->oi_obj = obj;
6118                 it->oi_ipd = ipd;
6119                 lu_object_get(lo);
6120                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6121                 return (struct dt_it *)it;
6122         } else {
6123                 OBD_FREE_PTR(it);
6124                 return ERR_PTR(-ENOMEM);
6125         }
6126 }
6127
6128 /**
6129  * free given Iterator.
6130  */
6131 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6132 {
6133         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6134         struct osd_object *obj = it->oi_obj;
6135
6136         iam_it_fini(&it->oi_it);
6137         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6138         osd_object_put(env, obj);
6139         OBD_FREE_PTR(it);
6140 }
6141
6142 /**
6143  *  Move Iterator to record specified by \a key
6144  *
6145  *  \param  di      osd iterator
6146  *  \param  key     key for index
6147  *
6148  *  \retval +ve  di points to record with least key not larger than key
6149  *  \retval  0   di points to exact matched key
6150  *  \retval -ve  failure
6151  */
6152
6153 static int osd_it_iam_get(const struct lu_env *env,
6154                           struct dt_it *di, const struct dt_key *key)
6155 {
6156         struct osd_thread_info *oti = osd_oti_get(env);
6157         struct osd_it_iam *it = (struct osd_it_iam *)di;
6158
6159         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6160                 /* swab quota uid/gid */
6161                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6162                 key = (struct dt_key *)&oti->oti_quota_id;
6163         }
6164
6165         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6166 }
6167
6168 /**
6169  *  Release Iterator
6170  *
6171  *  \param  di      osd iterator
6172  */
6173 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6174 {
6175         struct osd_it_iam *it = (struct osd_it_iam *)di;
6176
6177         iam_it_put(&it->oi_it);
6178 }
6179
6180 /**
6181  *  Move iterator by one record
6182  *
6183  *  \param  di      osd iterator
6184  *
6185  *  \retval +1   end of container reached
6186  *  \retval  0   success
6187  *  \retval -ve  failure
6188  */
6189
6190 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6191 {
6192         struct osd_it_iam *it = (struct osd_it_iam *)di;
6193
6194         return iam_it_next(&it->oi_it);
6195 }
6196
6197 /**
6198  * Return pointer to the key under iterator.
6199  */
6200
6201 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6202                                      const struct dt_it *di)
6203 {
6204         struct osd_thread_info *oti = osd_oti_get(env);
6205         struct osd_it_iam *it = (struct osd_it_iam *)di;
6206         struct osd_object *obj = it->oi_obj;
6207         struct dt_key *key;
6208
6209         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6210
6211         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6212                 /* swab quota uid/gid */
6213                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6214                 key = (struct dt_key *)&oti->oti_quota_id;
6215         }
6216
6217         return key;
6218 }
6219
6220 /**
6221  * Return size of key under iterator (in bytes)
6222  */
6223
6224 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6225 {
6226         struct osd_it_iam *it = (struct osd_it_iam *)di;
6227
6228         return iam_it_key_size(&it->oi_it);
6229 }
6230
6231 static inline void
6232 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6233 {
6234         /* check if file type is required */
6235         if (ent->lde_attrs & LUDA_TYPE) {
6236                 struct luda_type *lt;
6237                 int align = sizeof(*lt) - 1;
6238
6239                 len = (len + align) & ~align;
6240                 lt = (struct luda_type *)(ent->lde_name + len);
6241                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6242         }
6243
6244         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6245 }
6246
6247 /**
6248  * build lu direct from backend fs dirent.
6249  */
6250
6251 static inline void
6252 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6253                    char *name, __u16 namelen, __u16 type, __u32 attr)
6254 {
6255         ent->lde_attrs = attr | LUDA_FID;
6256         fid_cpu_to_le(&ent->lde_fid, fid);
6257
6258         ent->lde_hash = cpu_to_le64(offset);
6259         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6260
6261         strncpy(ent->lde_name, name, namelen);
6262         ent->lde_name[namelen] = '\0';
6263         ent->lde_namelen = cpu_to_le16(namelen);
6264
6265         /* append lustre attributes */
6266         osd_it_append_attrs(ent, namelen, type);
6267 }
6268
6269 /**
6270  * Return pointer to the record under iterator.
6271  */
6272 static int osd_it_iam_rec(const struct lu_env *env,
6273                           const struct dt_it *di,
6274                           struct dt_rec *dtrec, __u32 attr)
6275 {
6276         struct osd_it_iam *it = (struct osd_it_iam *)di;
6277         struct osd_thread_info *info = osd_oti_get(env);
6278
6279         ENTRY;
6280
6281         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6282                 const struct osd_fid_pack *rec;
6283                 struct lu_fid *fid = &info->oti_fid;
6284                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6285                 char *name;
6286                 int namelen;
6287                 __u64 hash;
6288                 int rc;
6289
6290                 name = (char *)iam_it_key_get(&it->oi_it);
6291                 if (IS_ERR(name))
6292                         RETURN(PTR_ERR(name));
6293
6294                 namelen = iam_it_key_size(&it->oi_it);
6295
6296                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6297                 if (IS_ERR(rec))
6298                         RETURN(PTR_ERR(rec));
6299
6300                 rc = osd_fid_unpack(fid, rec);
6301                 if (rc)
6302                         RETURN(rc);
6303
6304                 hash = iam_it_store(&it->oi_it);
6305
6306                 /* IAM does not store object type in IAM index (dir) */
6307                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6308                                    0, LUDA_FID);
6309         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6310                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6311                            (struct iam_rec *)dtrec);
6312                 osd_quota_unpack(it->oi_obj, dtrec);
6313         } else {
6314                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6315                            (struct iam_rec *)dtrec);
6316         }
6317
6318         RETURN(0);
6319 }
6320
6321 /**
6322  * Returns cookie for current Iterator position.
6323  */
6324 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6325 {
6326         struct osd_it_iam *it = (struct osd_it_iam *)di;
6327
6328         return iam_it_store(&it->oi_it);
6329 }
6330
6331 /**
6332  * Restore iterator from cookie.
6333  *
6334  * \param  di      osd iterator
6335  * \param  hash    Iterator location cookie
6336  *
6337  * \retval +ve  di points to record with least key not larger than key.
6338  * \retval  0   di points to exact matched key
6339  * \retval -ve  failure
6340  */
6341
6342 static int osd_it_iam_load(const struct lu_env *env,
6343                            const struct dt_it *di, __u64 hash)
6344 {
6345         struct osd_it_iam *it = (struct osd_it_iam *)di;
6346
6347         return iam_it_load(&it->oi_it, hash);
6348 }
6349
6350 static const struct dt_index_operations osd_index_iam_ops = {
6351         .dio_lookup         = osd_index_iam_lookup,
6352         .dio_declare_insert = osd_index_declare_iam_insert,
6353         .dio_insert         = osd_index_iam_insert,
6354         .dio_declare_delete = osd_index_declare_iam_delete,
6355         .dio_delete         = osd_index_iam_delete,
6356         .dio_it     = {
6357                 .init     = osd_it_iam_init,
6358                 .fini     = osd_it_iam_fini,
6359                 .get      = osd_it_iam_get,
6360                 .put      = osd_it_iam_put,
6361                 .next     = osd_it_iam_next,
6362                 .key      = osd_it_iam_key,
6363                 .key_size = osd_it_iam_key_size,
6364                 .rec      = osd_it_iam_rec,
6365                 .store    = osd_it_iam_store,
6366                 .load     = osd_it_iam_load
6367         }
6368 };
6369
6370
6371 /**
6372  * Creates or initializes iterator context.
6373  *
6374  * \retval struct osd_it_ea, iterator structure on success
6375  *
6376  */
6377 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6378                                     struct dt_object *dt,
6379                                     __u32 attr)
6380 {
6381         struct osd_object *obj = osd_dt_obj(dt);
6382         struct osd_thread_info *info = osd_oti_get(env);
6383         struct osd_it_ea *oie;
6384         struct file *file;
6385         struct lu_object *lo = &dt->do_lu;
6386         struct dentry *obj_dentry;
6387
6388         ENTRY;
6389
6390         if (!dt_object_exists(dt) || obj->oo_destroyed)
6391                 RETURN(ERR_PTR(-ENOENT));
6392
6393         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6394         if (oie == NULL)
6395                 RETURN(ERR_PTR(-ENOMEM));
6396         obj_dentry = &oie->oie_dentry;
6397
6398         obj_dentry->d_inode = obj->oo_inode;
6399         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6400         obj_dentry->d_name.hash = 0;
6401
6402         oie->oie_rd_dirent       = 0;
6403         oie->oie_it_dirent       = 0;
6404         oie->oie_dirent          = NULL;
6405         if (unlikely(!info->oti_it_ea_buf_used)) {
6406                 oie->oie_buf = info->oti_it_ea_buf;
6407                 info->oti_it_ea_buf_used = 1;
6408         } else {
6409                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6410                 if (oie->oie_buf == NULL)
6411                         RETURN(ERR_PTR(-ENOMEM));
6412         }
6413         oie->oie_obj = obj;
6414
6415         file = &oie->oie_file;
6416
6417         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6418         if (attr & LUDA_64BITHASH)
6419                 file->f_mode    = FMODE_64BITHASH;
6420         else
6421                 file->f_mode    = FMODE_32BITHASH;
6422         file->f_path.dentry     = obj_dentry;
6423         file->f_mapping         = obj->oo_inode->i_mapping;
6424         file->f_op              = obj->oo_inode->i_fop;
6425         set_file_inode(file, obj->oo_inode);
6426
6427         lu_object_get(lo);
6428         RETURN((struct dt_it *)oie);
6429 }
6430
6431 /**
6432  * Destroy or finishes iterator context.
6433  *
6434  * \param di iterator structure to be destroyed
6435  */
6436 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6437 {
6438         struct osd_thread_info *info = osd_oti_get(env);
6439         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6440         struct osd_object *obj = oie->oie_obj;
6441         struct inode *inode = obj->oo_inode;
6442
6443         ENTRY;
6444         oie->oie_file.f_op->release(inode, &oie->oie_file);
6445         osd_object_put(env, obj);
6446         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6447                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6448         else
6449                 info->oti_it_ea_buf_used = 0;
6450         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6451         EXIT;
6452 }
6453
6454 /**
6455  * It position the iterator at given key, so that next lookup continues from
6456  * that key Or it is similar to dio_it->load() but based on a key,
6457  * rather than file position.
6458  *
6459  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6460  * to the beginning.
6461  *
6462  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6463  */
6464 static int osd_it_ea_get(const struct lu_env *env,
6465                          struct dt_it *di, const struct dt_key *key)
6466 {
6467         struct osd_it_ea *it = (struct osd_it_ea *)di;
6468
6469         ENTRY;
6470         LASSERT(((const char *)key)[0] == '\0');
6471         it->oie_file.f_pos = 0;
6472         it->oie_rd_dirent = 0;
6473         it->oie_it_dirent = 0;
6474         it->oie_dirent = NULL;
6475
6476         RETURN(+1);
6477 }
6478
6479 /**
6480  * Does nothing
6481  */
6482 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6483 {
6484 }
6485
6486 struct osd_filldir_cbs {
6487 #ifdef HAVE_DIR_CONTEXT
6488         struct dir_context ctx;
6489 #endif
6490         struct osd_it_ea  *it;
6491 };
6492 /**
6493  * It is called internally by ->readdir(). It fills the
6494  * iterator's in-memory data structure with required
6495  * information i.e. name, namelen, rec_size etc.
6496  *
6497  * \param buf in which information to be filled in.
6498  * \param name name of the file in given dir
6499  *
6500  * \retval 0 on success
6501  * \retval 1 on buffer full
6502  */
6503 #ifdef HAVE_FILLDIR_USE_CTX
6504 static int osd_ldiskfs_filldir(struct dir_context *buf,
6505 #else
6506 static int osd_ldiskfs_filldir(void *buf,
6507 #endif
6508                                const char *name, int namelen,
6509                                loff_t offset, __u64 ino, unsigned int d_type)
6510 {
6511         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6512         struct osd_object *obj = it->oie_obj;
6513         struct osd_it_ea_dirent *ent = it->oie_dirent;
6514         struct lu_fid *fid = &ent->oied_fid;
6515         struct osd_fid_pack *rec;
6516
6517         ENTRY;
6518
6519 /* this should never happen */
6520         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6521                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6522                 RETURN(-EIO);
6523         }
6524
6525         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6526             OSD_IT_EA_BUFSIZE)
6527                 RETURN(1);
6528
6529         /* "." is just the object itself. */
6530         if (namelen == 1 && name[0] == '.') {
6531                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6532         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6533                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6534                 if (osd_fid_unpack(fid, rec) != 0)
6535                         fid_zero(fid);
6536         } else {
6537                 fid_zero(fid);
6538         }
6539         d_type &= ~LDISKFS_DIRENT_LUFID;
6540
6541         /* NOT export local root. */
6542         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6543                 ino = obj->oo_inode->i_ino;
6544                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6545         }
6546
6547         ent->oied_ino     = ino;
6548         ent->oied_off     = offset;
6549         ent->oied_namelen = namelen;
6550         ent->oied_type    = d_type;
6551
6552         memcpy(ent->oied_name, name, namelen);
6553
6554         it->oie_rd_dirent++;
6555         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6556         RETURN(0);
6557 }
6558
6559 /**
6560  * Calls ->readdir() to load a directory entry at a time
6561  * and stored it in iterator's in-memory data structure.
6562  *
6563  * \param di iterator's in memory structure
6564  *
6565  * \retval   0 on success
6566  * \retval -ve on error
6567  * \retval +1 reach the end of entry
6568  */
6569 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6570                                const struct dt_it *di)
6571 {
6572         struct osd_it_ea *it = (struct osd_it_ea *)di;
6573         struct osd_object *obj = it->oie_obj;
6574         struct inode *inode = obj->oo_inode;
6575         struct htree_lock *hlock = NULL;
6576         struct file *filp = &it->oie_file;
6577         int rc = 0;
6578         struct osd_filldir_cbs buf = {
6579 #ifdef HAVE_DIR_CONTEXT
6580                 .ctx.actor = osd_ldiskfs_filldir,
6581 #endif
6582                 .it = it
6583         };
6584
6585         ENTRY;
6586         it->oie_dirent = it->oie_buf;
6587         it->oie_rd_dirent = 0;
6588
6589         if (obj->oo_hl_head != NULL) {
6590                 hlock = osd_oti_get(env)->oti_hlock;
6591                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6592                                    inode, LDISKFS_HLOCK_READDIR);
6593         } else {
6594                 down_read(&obj->oo_ext_idx_sem);
6595         }
6596
6597 #ifdef HAVE_DIR_CONTEXT
6598         buf.ctx.pos = filp->f_pos;
6599 #ifdef HAVE_ITERATE_SHARED
6600         rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
6601 #else
6602         rc = inode->i_fop->iterate(filp, &buf.ctx);
6603 #endif
6604         filp->f_pos = buf.ctx.pos;
6605 #else
6606         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
6607 #endif
6608
6609         if (hlock != NULL)
6610                 ldiskfs_htree_unlock(hlock);
6611         else
6612                 up_read(&obj->oo_ext_idx_sem);
6613
6614         if (it->oie_rd_dirent == 0) {
6615                 /*
6616                  * If it does not get any dirent, it means it has been reached
6617                  * to the end of the dir
6618                  */
6619                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6620                 if (rc == 0)
6621                         rc = 1;
6622         } else {
6623                 it->oie_dirent = it->oie_buf;
6624                 it->oie_it_dirent = 1;
6625         }
6626
6627         RETURN(rc);
6628 }
6629
6630 /**
6631  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
6632  * to load a directory entry at a time and stored it in
6633  * iterator's in-memory data structure.
6634  *
6635  * \param di iterator's in memory structure
6636  *
6637  * \retval +ve iterator reached to end
6638  * \retval   0 iterator not reached to end
6639  * \retval -ve on error
6640  */
6641 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6642 {
6643         struct osd_it_ea *it = (struct osd_it_ea *)di;
6644         int rc;
6645
6646         ENTRY;
6647
6648         if (it->oie_it_dirent < it->oie_rd_dirent) {
6649                 it->oie_dirent =
6650                         (void *)it->oie_dirent +
6651                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6652                                        it->oie_dirent->oied_namelen);
6653                 it->oie_it_dirent++;
6654                 rc = 0;
6655         } else {
6656                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6657                         rc = 1;
6658                 else
6659                         rc = osd_ldiskfs_it_fill(env, di);
6660         }
6661
6662         RETURN(rc);
6663 }
6664
6665 /**
6666  * Returns the key at current position from iterator's in memory structure.
6667  *
6668  * \param di iterator's in memory structure
6669  *
6670  * \retval key i.e. struct dt_key on success
6671  */
6672 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6673                                     const struct dt_it *di)
6674 {
6675         struct osd_it_ea *it = (struct osd_it_ea *)di;
6676
6677         return (struct dt_key *)it->oie_dirent->oied_name;
6678 }
6679
6680 /**
6681  * Returns key's size at current position from iterator's in memory structure.
6682  *
6683  * \param di iterator's in memory structure
6684  *
6685  * \retval key_size i.e. struct dt_key on success
6686  */
6687 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6688 {
6689         struct osd_it_ea *it = (struct osd_it_ea *)di;
6690
6691         return it->oie_dirent->oied_namelen;
6692 }
6693
6694 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6695 {
6696         if (LDISKFS_DIR_REC_LEN(de) >=
6697             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6698                 return true;
6699
6700         return false;
6701 }
6702
6703 static inline bool
6704 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6705                      unsigned int blocksize, bool dotdot)
6706 {
6707         if (dotdot)
6708                 return osd_dotdot_has_space(de);
6709
6710         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6711             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6712                 return true;
6713
6714         return false;
6715 }
6716
6717 static int
6718 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6719                     handle_t *jh, struct dentry *dentry,
6720                     const struct lu_fid *fid, struct buffer_head *bh,
6721                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6722                     bool dotdot)
6723 {
6724         struct inode *dir = dentry->d_parent->d_inode;
6725         struct inode *inode = dentry->d_inode;
6726         struct osd_fid_pack *rec;
6727         struct ldiskfs_dentry_param *ldp;
6728         int namelen = dentry->d_name.len;
6729         int rc;
6730         struct osd_thread_info *info = osd_oti_get(env);
6731
6732         ENTRY;
6733
6734         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6735                 RETURN(0);
6736
6737         /* There is enough space to hold the FID-in-dirent. */
6738         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6739                 rc = ldiskfs_journal_get_write_access(jh, bh);
6740                 if (rc != 0)
6741                         RETURN(rc);
6742
6743                 de->name[namelen] = 0;
6744                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6745                 rec->fp_len = sizeof(struct lu_fid) + 1;
6746                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6747                 de->file_type |= LDISKFS_DIRENT_LUFID;
6748                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6749
6750                 RETURN(rc);
6751         }
6752
6753         LASSERT(!dotdot);
6754
6755         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6756         if (rc != 0)
6757                 RETURN(rc);
6758
6759         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6760         osd_get_ldiskfs_dirent_param(ldp, fid);
6761         dentry->d_fsdata = (void *)ldp;
6762         ll_vfs_dq_init(dir);
6763         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6764         /*
6765          * It is too bad, we cannot reinsert the name entry back.
6766          * That means we lose it!
6767          */
6768         if (rc != 0)
6769                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6770                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6771                        osd_ino2name(inode),
6772                        dir->i_ino, dir->i_generation, namelen,
6773                        dentry->d_name.name, PFID(fid), rc);
6774
6775         RETURN(rc);
6776 }
6777
6778 static int
6779 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6780                         struct osd_it_ea *it, struct lu_fid *fid,
6781                         struct osd_inode_id *id, __u32 *attr)
6782 {
6783         struct osd_thread_info *info = osd_oti_get(env);
6784         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6785         struct osd_device *dev = osd_obj2dev(obj);
6786         struct super_block *sb = osd_sb(dev);
6787         const char *devname = osd_name(dev);
6788         struct osd_it_ea_dirent *ent = it->oie_dirent;
6789         struct inode *dir = obj->oo_inode;
6790         struct htree_lock *hlock = NULL;
6791         struct buffer_head *bh = NULL;
6792         handle_t *jh = NULL;
6793         struct ldiskfs_dir_entry_2 *de;
6794         struct dentry *dentry;
6795         struct inode *inode;
6796         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6797         int credits;
6798         int rc;
6799         bool dotdot = false;
6800         bool dirty = false;
6801
6802         ENTRY;
6803
6804         if (ent->oied_name[0] == '.') {
6805                 if (ent->oied_namelen == 1)
6806                         RETURN(0);
6807
6808                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6809                         dotdot = true;
6810         }
6811
6812         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6813         inode = osd_iget(info, dev, id);
6814         if (IS_ERR(inode)) {
6815                 rc = PTR_ERR(inode);
6816                 if (rc == -ENOENT || rc == -ESTALE) {
6817                         /*
6818                          * Maybe dangling name entry, or
6819                          * corrupted directory entry.
6820                          */
6821                         *attr |= LUDA_UNKNOWN;
6822                         rc = 0;
6823                 } else {
6824                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6825                                "check_repair, dir = %lu/%u, name = %.*s, "
6826                                "ino = %llu, rc = %d\n",
6827                                devname, dir->i_ino, dir->i_generation,
6828                                ent->oied_namelen, ent->oied_name,
6829                                ent->oied_ino, rc);
6830                 }
6831
6832                 RETURN(rc);
6833         }
6834
6835         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6836                                            ent->oied_namelen);
6837         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6838         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6839                 lma = NULL;
6840         else if (rc != 0)
6841                 GOTO(out, rc);
6842
6843         /*
6844          * We need to ensure that the name entry is still valid.
6845          * Because it may be removed or renamed by other already.
6846          *
6847          * The unlink or rename operation will start journal before PDO lock,
6848          * so to avoid deadlock, here we need to start journal handle before
6849          * related PDO lock also. But because we do not know whether there
6850          * will be something to be repaired before PDO lock, we just start
6851          * journal without conditions.
6852          *
6853          * We may need to remove the name entry firstly, then insert back.
6854          * One credit is for user quota file update.
6855          * One credit is for group quota file update.
6856          * Two credits are for dirty inode.
6857          */
6858         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6859                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6860
6861         if (dev->od_dirent_journal != 0) {
6862
6863 again:
6864                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6865                 if (IS_ERR(jh)) {
6866                         rc = PTR_ERR(jh);
6867                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6868                                "check_repair, dir = %lu/%u, credits = %d, "
6869                                "name = %.*s, ino = %llu: rc = %d\n",
6870                                devname, dir->i_ino, dir->i_generation, credits,
6871                                ent->oied_namelen, ent->oied_name,
6872                                ent->oied_ino, rc);
6873
6874                         GOTO(out_inode, rc);
6875                 }
6876
6877                 if (obj->oo_hl_head != NULL) {
6878                         hlock = osd_oti_get(env)->oti_hlock;
6879                         /*
6880                          * "0" means exclusive lock for the whole directory.
6881                          * We need to prevent others access such name entry
6882                          * during the delete + insert. Neither HLOCK_ADD nor
6883                          * HLOCK_DEL cannot guarantee the atomicity.
6884                          */
6885                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6886                 } else {
6887                         down_write(&obj->oo_ext_idx_sem);
6888                 }
6889         } else {
6890                 if (obj->oo_hl_head != NULL) {
6891                         hlock = osd_oti_get(env)->oti_hlock;
6892                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6893                                            LDISKFS_HLOCK_LOOKUP);
6894                 } else {
6895                         down_read(&obj->oo_ext_idx_sem);
6896                 }
6897         }
6898
6899         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6900         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6901                 *attr |= LUDA_IGNORE;
6902
6903                 GOTO(out, rc = 0);
6904         }
6905
6906         /*
6907          * For dotdot entry, if there is not enough space to hold the
6908          * FID-in-dirent, just keep them there. It only happens when the
6909          * device upgraded from 1.8 or restored from MDT file-level backup.
6910          * For the whole directory, only dotdot entry have no FID-in-dirent
6911          * and needs to get FID from LMA when readdir, it will not affect the
6912          * performance much.
6913          */
6914         if (dotdot && !osd_dotdot_has_space(de)) {
6915                 *attr |= LUDA_UNKNOWN;
6916
6917                 GOTO(out, rc = 0);
6918         }
6919
6920         if (lma != NULL) {
6921                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6922                         GOTO(out, rc = 0);
6923
6924                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6925                         struct lu_fid *tfid = &lma->lma_self_fid;
6926
6927                         if (likely(dotdot &&
6928                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6929                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6930                                 /*
6931                                  * It must be REMOTE_PARENT_DIR and as the
6932                                  * 'dotdot' entry of remote directory
6933                                  */
6934                                 *attr |= LUDA_IGNORE;
6935                         } else {
6936                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6937                                        "parent directory, but got %.*s under "
6938                                        "dir = %lu/%u with the FID "DFID"\n",
6939                                        devname, ent->oied_namelen,
6940                                        ent->oied_name, dir->i_ino,
6941                                        dir->i_generation, PFID(tfid));
6942
6943                                 *attr |= LUDA_UNKNOWN;
6944                         }
6945
6946                         GOTO(out, rc = 0);
6947                 }
6948         }
6949
6950         if (!fid_is_zero(fid)) {
6951                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6952                                               ent->oied_namelen);
6953                 if (rc == -ENOENT ||
6954                     (rc == -ENODATA &&
6955                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6956                         /*
6957                          * linkEA does not recognize the dirent entry,
6958                          * it may because the dirent entry corruption
6959                          * and points to other's inode.
6960                          */
6961                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6962                                "recognize the dirent, dir = %lu/%u, "
6963                                " name = %.*s, ino = %llu, "
6964                                DFID": rc = %d\n", devname, dir->i_ino,
6965                                dir->i_generation, ent->oied_namelen,
6966                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6967                         *attr |= LUDA_UNKNOWN;
6968
6969                         GOTO(out, rc = 0);
6970                 }
6971
6972                 if (rc && rc != -ENODATA) {
6973                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
6974                                "dir = %lu/%u, name = %.*s, ino = %llu, "
6975                                DFID": rc = %d\n", devname, dir->i_ino,
6976                                dir->i_generation, ent->oied_namelen,
6977                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6978                         *attr |= LUDA_UNKNOWN;
6979
6980                         GOTO(out, rc = 0);
6981                 }
6982         }
6983
6984         if (lma != NULL) {
6985                 /*
6986                  * linkEA recognizes the dirent entry, the FID-in-LMA is
6987                  * valid, trusted, in spite of fid_is_sane(fid) or not.
6988                  */
6989                 if (*attr & LUDA_VERIFY_DRYRUN) {
6990                         *fid = lma->lma_self_fid;
6991                         *attr |= LUDA_REPAIR;
6992
6993                         GOTO(out, rc = 0);
6994                 }
6995
6996                 if (jh == NULL) {
6997                         brelse(bh);
6998                         dev->od_dirent_journal = 1;
6999                         if (hlock != NULL) {
7000                                 ldiskfs_htree_unlock(hlock);
7001                                 hlock = NULL;
7002                         } else {
7003                                 up_read(&obj->oo_ext_idx_sem);
7004                         }
7005
7006                         goto again;
7007                 }
7008
7009                 *fid = lma->lma_self_fid;
7010                 dirty = true;
7011                 /* Update or append the FID-in-dirent. */
7012                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7013                                          bh, de, hlock, dotdot);
7014                 if (rc == 0)
7015                         *attr |= LUDA_REPAIR;
7016                 else
7017                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7018                                "the dirent, dir = %lu/%u, name = %.*s, "
7019                                "ino = %llu, "DFID": rc = %d\n",
7020                                devname, dir->i_ino, dir->i_generation,
7021                                ent->oied_namelen, ent->oied_name,
7022                                ent->oied_ino, PFID(fid), rc);
7023         } else {
7024                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7025                 if (*attr & LUDA_VERIFY_DRYRUN) {
7026                         if (fid_is_sane(fid)) {
7027                                 *attr |= LUDA_REPAIR;
7028                         } else if (dev->od_index == 0) {
7029                                 lu_igif_build(fid, inode->i_ino,
7030                                               inode->i_generation);
7031                                 *attr |= LUDA_UPGRADE;
7032                         }
7033
7034                         GOTO(out, rc = 0);
7035                 }
7036
7037                 if (jh == NULL) {
7038                         brelse(bh);
7039                         dev->od_dirent_journal = 1;
7040                         if (hlock != NULL) {
7041                                 ldiskfs_htree_unlock(hlock);
7042                                 hlock = NULL;
7043                         } else {
7044                                 up_read(&obj->oo_ext_idx_sem);
7045                         }
7046
7047                         goto again;
7048                 }
7049
7050                 dirty = true;
7051                 if (unlikely(fid_is_sane(fid))) {
7052                         /*
7053                          * FID-in-dirent exists, but FID-in-LMA is lost.
7054                          * Trust the FID-in-dirent, and add FID-in-LMA.
7055                          */
7056                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7057                         if (rc == 0)
7058                                 *attr |= LUDA_REPAIR;
7059                         else
7060                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7061                                        "update dirent, dir = %lu/%u, "
7062                                        "name = %.*s, ino = %llu, "
7063                                        DFID": rc = %d\n",
7064                                        devname, dir->i_ino, dir->i_generation,
7065                                        ent->oied_namelen, ent->oied_name,
7066                                        ent->oied_ino, PFID(fid), rc);
7067                 } else if (dev->od_index == 0) {
7068                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7069                         /*
7070                          * It is probably IGIF object. Only aappend the
7071                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7072                          */
7073                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7074                                                  bh, de, hlock, dotdot);
7075                         if (rc == 0)
7076                                 *attr |= LUDA_UPGRADE;
7077                         else
7078                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7079                                        "after the dirent, dir = %lu/%u, "
7080                                        "name = %.*s, ino = %llu, "
7081                                        DFID": rc = %d\n",
7082                                        devname, dir->i_ino, dir->i_generation,
7083                                        ent->oied_namelen, ent->oied_name,
7084                                        ent->oied_ino, PFID(fid), rc);
7085                 }
7086         }
7087
7088         GOTO(out, rc);
7089
7090 out:
7091         if (!IS_ERR(bh))
7092                 brelse(bh);
7093         if (hlock != NULL) {
7094                 ldiskfs_htree_unlock(hlock);
7095         } else {
7096                 if (dev->od_dirent_journal != 0)
7097                         up_write(&obj->oo_ext_idx_sem);
7098                 else
7099                         up_read(&obj->oo_ext_idx_sem);
7100         }
7101
7102         if (jh != NULL)
7103                 ldiskfs_journal_stop(jh);
7104
7105 out_inode:
7106         iput(inode);
7107         if (rc >= 0 && !dirty)
7108                 dev->od_dirent_journal = 0;
7109
7110         return rc;
7111 }
7112
7113 /**
7114  * Returns the value at current position from iterator's in memory structure.
7115  *
7116  * \param di struct osd_it_ea, iterator's in memory structure
7117  * \param attr attr requested for dirent.
7118  * \param lde lustre dirent
7119  *
7120  * \retval   0 no error and \param lde has correct lustre dirent.
7121  * \retval -ve on error
7122  */
7123 static inline int osd_it_ea_rec(const struct lu_env *env,
7124                                 const struct dt_it *di,
7125                                 struct dt_rec *dtrec, __u32 attr)
7126 {
7127         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7128         struct osd_object      *obj   = it->oie_obj;
7129         struct osd_device      *dev   = osd_obj2dev(obj);
7130         struct osd_thread_info *oti   = osd_oti_get(env);
7131         struct osd_inode_id    *id    = &oti->oti_id;
7132         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7133         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7134         __u32 ino = it->oie_dirent->oied_ino;
7135         int rc = 0;
7136
7137         ENTRY;
7138
7139         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7140
7141         if (attr & LUDA_VERIFY) {
7142                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7143                         attr |= LUDA_IGNORE;
7144                         /*
7145                          * If the parent is on remote MDT, and there
7146                          * is no FID-in-dirent, then we have to get
7147                          * the parent FID from the linkEA.
7148                          */
7149                         if (!fid_is_sane(fid) &&
7150                             it->oie_dirent->oied_namelen == 2 &&
7151                             it->oie_dirent->oied_name[0] == '.' &&
7152                             it->oie_dirent->oied_name[1] == '.')
7153                                 osd_get_pfid_from_linkea(env, obj, fid);
7154                 } else {
7155                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7156                                                      &attr);
7157                 }
7158
7159                 if (!fid_is_sane(fid))
7160                         attr |= LUDA_UNKNOWN;
7161         } else {
7162                 attr &= ~LU_DIRENT_ATTRS_MASK;
7163                 if (!fid_is_sane(fid)) {
7164                         bool is_dotdot = false;
7165
7166                         if (it->oie_dirent->oied_namelen == 2 &&
7167                             it->oie_dirent->oied_name[0] == '.' &&
7168                             it->oie_dirent->oied_name[1] == '.')
7169                                 is_dotdot = true;
7170                         /*
7171                          * If the parent is on remote MDT, and there
7172                          * is no FID-in-dirent, then we have to get
7173                          * the parent FID from the linkEA.
7174                          */
7175                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7176                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7177                         } else {
7178                                 if (is_dotdot == false &&
7179                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7180                                         RETURN(-ENOENT);
7181
7182                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7183                         }
7184                 } else {
7185                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7186                 }
7187         }
7188
7189         /* Pack the entry anyway, at least the offset is right. */
7190         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7191                            it->oie_dirent->oied_name,
7192                            it->oie_dirent->oied_namelen,
7193                            it->oie_dirent->oied_type, attr);
7194
7195         if (rc < 0)
7196                 RETURN(rc);
7197
7198         if (osd_remote_fid(env, dev, fid))
7199                 RETURN(0);
7200
7201         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7202                 osd_add_oi_cache(oti, dev, id, fid);
7203
7204         RETURN(rc > 0 ? 0 : rc);
7205 }
7206
7207 /**
7208  * Returns the record size size at current position.
7209  *
7210  * This function will return record(lu_dirent) size in bytes.
7211  *
7212  * \param[in] env       execution environment
7213  * \param[in] di        iterator's in memory structure
7214  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7215  *                      calculate the lu_dirent size.
7216  *
7217  * \retval      record size(in bytes & in memory) of the current lu_dirent
7218  *              entry.
7219  */
7220 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7221                               __u32 attr)
7222 {
7223         struct osd_it_ea *it = (struct osd_it_ea *)di;
7224
7225         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7226 }
7227
7228 /**
7229  * Returns a cookie for current position of the iterator head, so that
7230  * user can use this cookie to load/start the iterator next time.
7231  *
7232  * \param di iterator's in memory structure
7233  *
7234  * \retval cookie for current position, on success
7235  */
7236 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7237 {
7238         struct osd_it_ea *it = (struct osd_it_ea *)di;
7239
7240         return it->oie_dirent->oied_off;
7241 }
7242
7243 /**
7244  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
7245  * to load a directory entry at a time and stored it i inn,
7246  * in iterator's in-memory data structure.
7247  *
7248  * \param di struct osd_it_ea, iterator's in memory structure
7249  *
7250  * \retval +ve on success
7251  * \retval -ve on error
7252  */
7253 static int osd_it_ea_load(const struct lu_env *env,
7254                           const struct dt_it *di, __u64 hash)
7255 {
7256         struct osd_it_ea *it = (struct osd_it_ea *)di;
7257         int rc;
7258
7259         ENTRY;
7260         it->oie_file.f_pos = hash;
7261
7262         rc =  osd_ldiskfs_it_fill(env, di);
7263         if (rc > 0)
7264                 rc = -ENODATA;
7265
7266         if (rc == 0)
7267                 rc = 1;
7268
7269         RETURN(rc);
7270 }
7271
7272 /**
7273  * Index lookup function for interoperability mode (b11826).
7274  *
7275  * \param key,  key i.e. file name to be searched
7276  *
7277  * \retval +ve, on success
7278  * \retval -ve, on error
7279  */
7280 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7281                                struct dt_rec *rec, const struct dt_key *key)
7282 {
7283         struct osd_object *obj = osd_dt_obj(dt);
7284         int rc = 0;
7285
7286         ENTRY;
7287
7288         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7289         LINVRNT(osd_invariant(obj));
7290
7291         rc = osd_ea_lookup_rec(env, obj, rec, key);
7292         if (rc == 0)
7293                 rc = 1;
7294         RETURN(rc);
7295 }
7296
7297 /**
7298  * Index and Iterator operations for interoperability
7299  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7300  */
7301 static const struct dt_index_operations osd_index_ea_ops = {
7302         .dio_lookup         = osd_index_ea_lookup,
7303         .dio_declare_insert = osd_index_declare_ea_insert,
7304         .dio_insert         = osd_index_ea_insert,
7305         .dio_declare_delete = osd_index_declare_ea_delete,
7306         .dio_delete         = osd_index_ea_delete,
7307         .dio_it     = {
7308                 .init     = osd_it_ea_init,
7309                 .fini     = osd_it_ea_fini,
7310                 .get      = osd_it_ea_get,
7311                 .put      = osd_it_ea_put,
7312                 .next     = osd_it_ea_next,
7313                 .key      = osd_it_ea_key,
7314                 .key_size = osd_it_ea_key_size,
7315                 .rec      = osd_it_ea_rec,
7316                 .rec_size = osd_it_ea_rec_size,
7317                 .store    = osd_it_ea_store,
7318                 .load     = osd_it_ea_load
7319         }
7320 };
7321
7322 static void *osd_key_init(const struct lu_context *ctx,
7323                           struct lu_context_key *key)
7324 {
7325         struct osd_thread_info *info;
7326
7327         OBD_ALLOC_PTR(info);
7328         if (info == NULL)
7329                 return ERR_PTR(-ENOMEM);
7330
7331         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7332         if (info->oti_it_ea_buf == NULL)
7333                 goto out_free_info;
7334
7335         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7336
7337         info->oti_hlock = ldiskfs_htree_lock_alloc();
7338         if (info->oti_hlock == NULL)
7339                 goto out_free_ea;
7340
7341         return info;
7342
7343 out_free_ea:
7344         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7345 out_free_info:
7346         OBD_FREE_PTR(info);
7347         return ERR_PTR(-ENOMEM);
7348 }
7349
7350 static void osd_key_fini(const struct lu_context *ctx,
7351                          struct lu_context_key *key, void *data)
7352 {
7353         struct osd_thread_info *info = data;
7354         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7355         struct osd_idmap_cache *idc = info->oti_ins_cache;
7356
7357         if (info->oti_dio_pages) {
7358                 int i;
7359                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7360                         if (info->oti_dio_pages[i])
7361                                 __free_page(info->oti_dio_pages[i]);
7362                 }
7363                 OBD_FREE(info->oti_dio_pages,
7364                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7365         }
7366
7367         if (info->oti_inode != NULL)
7368                 OBD_FREE_PTR(lli);
7369         if (info->oti_hlock != NULL)
7370                 ldiskfs_htree_lock_free(info->oti_hlock);
7371         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7372         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7373         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7374         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7375         lu_buf_free(&info->oti_big_buf);
7376         if (idc != NULL) {
7377                 LASSERT(info->oti_ins_cache_size > 0);
7378                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7379                 info->oti_ins_cache = NULL;
7380                 info->oti_ins_cache_size = 0;
7381         }
7382         OBD_FREE_PTR(info);
7383 }
7384
7385 static void osd_key_exit(const struct lu_context *ctx,
7386                          struct lu_context_key *key, void *data)
7387 {
7388         struct osd_thread_info *info = data;
7389
7390         LASSERT(info->oti_r_locks == 0);
7391         LASSERT(info->oti_w_locks == 0);
7392         LASSERT(info->oti_txns    == 0);
7393 }
7394
7395 /* type constructor/destructor: osd_type_init, osd_type_fini */
7396 LU_TYPE_INIT_FINI(osd, &osd_key);
7397
7398 struct lu_context_key osd_key = {
7399         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7400         .lct_init = osd_key_init,
7401         .lct_fini = osd_key_fini,
7402         .lct_exit = osd_key_exit
7403 };
7404
7405
7406 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7407                            const char *name, struct lu_device *next)
7408 {
7409         struct osd_device *osd = osd_dev(d);
7410
7411         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7412             sizeof(osd->od_svname))
7413                 return -E2BIG;
7414         return osd_procfs_init(osd, name);
7415 }
7416
7417 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7418 {
7419         struct seq_server_site *ss = osd_seq_site(osd);
7420         int rc;
7421
7422         ENTRY;
7423
7424         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7425                 RETURN(0);
7426
7427         if (unlikely(ss == NULL))
7428                 RETURN(-ENODEV);
7429
7430         OBD_ALLOC_PTR(osd->od_cl_seq);
7431         if (osd->od_cl_seq == NULL)
7432                 RETURN(-ENOMEM);
7433
7434         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7435                              osd->od_svname, ss->ss_server_seq);
7436         if (rc != 0) {
7437                 OBD_FREE_PTR(osd->od_cl_seq);
7438                 osd->od_cl_seq = NULL;
7439                 RETURN(rc);
7440         }
7441
7442         if (ss->ss_node_id == 0) {
7443                 /*
7444                  * If the OSD on the sequence controller(MDT0), then allocate
7445                  * sequence here, otherwise allocate sequence after connected
7446                  * to MDT0 (see mdt_register_lwp_callback()).
7447                  */
7448                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7449                                    &osd->od_cl_seq->lcs_space, env);
7450         }
7451
7452         RETURN(rc);
7453 }
7454
7455 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7456 {
7457         if (osd->od_cl_seq == NULL)
7458                 return;
7459
7460         seq_client_fini(osd->od_cl_seq);
7461         OBD_FREE_PTR(osd->od_cl_seq);
7462         osd->od_cl_seq = NULL;
7463 }
7464
7465 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7466 {
7467         ENTRY;
7468
7469         /* shutdown quota slave instance associated with the device */
7470         if (o->od_quota_slave_md != NULL) {
7471                 struct qsd_instance *qsd = o->od_quota_slave_md;
7472
7473                 o->od_quota_slave_md = NULL;
7474                 qsd_fini(env, qsd);
7475         }
7476
7477         if (o->od_quota_slave_dt != NULL) {
7478                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7479
7480                 o->od_quota_slave_dt = NULL;
7481                 qsd_fini(env, qsd);
7482         }
7483
7484         osd_fid_fini(env, o);
7485         osd_scrub_cleanup(env, o);
7486
7487         RETURN(0);
7488 }
7489
7490 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7491 {
7492         ENTRY;
7493
7494         if (o->od_mnt != NULL) {
7495                 shrink_dcache_sb(osd_sb(o));
7496                 osd_sync(env, &o->od_dt_dev);
7497
7498                 mntput(o->od_mnt);
7499                 o->od_mnt = NULL;
7500         }
7501
7502         EXIT;
7503 }
7504
7505 static int osd_mount(const struct lu_env *env,
7506                      struct osd_device *o, struct lustre_cfg *cfg)
7507 {
7508         const char *name = lustre_cfg_string(cfg, 0);
7509         const char *dev = lustre_cfg_string(cfg, 1);
7510         const char *opts;
7511         unsigned long page, s_flags, lmd_flags = 0;
7512         struct page *__page;
7513         struct file_system_type *type;
7514         char *options = NULL;
7515         char *str;
7516         struct osd_thread_info *info = osd_oti_get(env);
7517         struct lu_fid *fid = &info->oti_fid;
7518         struct inode *inode;
7519         int rc = 0, force_over_512tb = 0;
7520
7521         ENTRY;
7522
7523         if (o->od_mnt != NULL)
7524                 RETURN(0);
7525
7526         if (strlen(dev) >= sizeof(o->od_mntdev))
7527                 RETURN(-E2BIG);
7528         strcpy(o->od_mntdev, dev);
7529
7530         str = lustre_cfg_string(cfg, 2);
7531         s_flags = simple_strtoul(str, NULL, 0);
7532         str = strstr(str, ":");
7533         if (str)
7534                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7535         opts = lustre_cfg_string(cfg, 3);
7536 #ifdef __BIG_ENDIAN
7537         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7538                 CERROR("%s: device %s extents feature is not guaranteed to "
7539                        "work on big-endian systems. Use \"bigendian_extents\" "
7540                        "mount option to override.\n", name, dev);
7541                 RETURN(-EINVAL);
7542         }
7543 #endif
7544 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7545         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7546                 CWARN("force_over_128tb option is deprecated. "
7547                       "Filesystems less than 512TB can be created without any "
7548                       "force options. Use force_over_512tb option for "
7549                       "filesystems greater than 512TB.\n");
7550         }
7551 #endif
7552 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7553         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7554                 CWARN("force_over_256tb option is deprecated. "
7555                       "Filesystems less than 512TB can be created without any "
7556                       "force options. Use force_over_512tb option for "
7557                       "filesystems greater than 512TB.\n");
7558         }
7559 #endif
7560
7561         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7562                 force_over_512tb = 1;
7563
7564         __page = alloc_page(GFP_KERNEL);
7565         if (__page == NULL)
7566                 GOTO(out, rc = -ENOMEM);
7567         page = (unsigned long)page_address(__page);
7568         options = (char *)page;
7569         *options = '\0';
7570         if (opts != NULL) {
7571                 /* strip out the options for back compatiblity */
7572                 static char *sout[] = {
7573                         "mballoc",
7574                         "iopen",
7575                         "noiopen",
7576                         "iopen_nopriv",
7577                         "extents",
7578                         "noextents",
7579                         /* strip out option we processed in osd */
7580                         "bigendian_extents",
7581                         "force_over_128tb",
7582                         "force_over_256tb",
7583                         "force_over_512tb",
7584                         NULL
7585                 };
7586                 strncat(options, opts, PAGE_SIZE);
7587                 for (rc = 0, str = options; sout[rc]; ) {
7588                         char *op = strstr(str, sout[rc]);
7589
7590                         if (op == NULL) {
7591                                 rc++;
7592                                 str = options;
7593                                 continue;
7594                         }
7595                         if (op == options || *(op - 1) == ',') {
7596                                 str = op + strlen(sout[rc]);
7597                                 if (*str == ',' || *str == '\0') {
7598                                         *str == ',' ? str++ : str;
7599                                         memmove(op, str, strlen(str) + 1);
7600                                 }
7601                         }
7602                         for (str = op; *str != ',' && *str != '\0'; str++)
7603                                 ;
7604                 }
7605         } else {
7606                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7607         }
7608
7609         /* Glom up mount options */
7610         if (*options != '\0')
7611                 strncat(options, ",", PAGE_SIZE);
7612         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7613
7614         type = get_fs_type("ldiskfs");
7615         if (!type) {
7616                 CERROR("%s: cannot find ldiskfs module\n", name);
7617                 GOTO(out, rc = -ENODEV);
7618         }
7619
7620         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7621         module_put(type->owner);
7622
7623         if (IS_ERR(o->od_mnt)) {
7624                 rc = PTR_ERR(o->od_mnt);
7625                 o->od_mnt = NULL;
7626                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7627                 GOTO(out, rc);
7628         }
7629
7630         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7631                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7632                                  force_over_512tb == 0) {
7633                 CERROR("%s: device %s LDISKFS does not support filesystems "
7634                        "greater than 512TB and can cause data corruption. "
7635                        "Use \"force_over_512tb\" mount option to override.\n",
7636                        name, dev);
7637                 GOTO(out_mnt, rc = -EINVAL);
7638         }
7639
7640         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7641                 if (priv_dev_set_rdonly) {
7642                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7643                         o->od_dt_dev.dd_rdonly = 1;
7644                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7645                                       name);
7646                 } else {
7647                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7648                                       name);
7649
7650                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7651                 }
7652         } else if (priv_dev_check_rdonly &&
7653                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7654                 CERROR("%s: underlying device %s is marked as "
7655                        "read-only. Setup failed\n", name, dev);
7656
7657                 GOTO(out_mnt, rc = -EROFS);
7658         }
7659
7660         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7661                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7662                 GOTO(out_mnt, rc = -EINVAL);
7663         }
7664
7665 #ifdef LDISKFS_MOUNT_DIRDATA
7666         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7667                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7668         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7669                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7670                       "enabling the dirdata feature. If you do not want to "
7671                       "downgrade to Lustre-1.x again, you can enable it via "
7672                       "'tune2fs -O dirdata device'\n", name, dev);
7673 #endif
7674         inode = osd_sb(o)->s_root->d_inode;
7675         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7676         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7677         if (rc != 0) {
7678                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7679                 GOTO(out_mnt, rc);
7680         }
7681
7682         if (lmd_flags & LMD_FLG_NOSCRUB)
7683                 o->od_auto_scrub_interval = AS_NEVER;
7684
7685         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7686                 /* do not use pagecache with flash-backed storage */
7687                 o->od_writethrough_cache = 0;
7688                 o->od_read_cache = 0;
7689         }
7690
7691         GOTO(out, rc = 0);
7692
7693 out_mnt:
7694         mntput(o->od_mnt);
7695         o->od_mnt = NULL;
7696
7697 out:
7698         if (__page)
7699                 __free_page(__page);
7700
7701         return rc;
7702 }
7703
7704 static struct lu_device *osd_device_fini(const struct lu_env *env,
7705                                          struct lu_device *d)
7706 {
7707         struct osd_device *o = osd_dev(d);
7708
7709         ENTRY;
7710
7711         osd_index_backup(env, o, false);
7712         osd_shutdown(env, o);
7713         osd_procfs_fini(o);
7714         osd_obj_map_fini(o);
7715         osd_umount(env, o);
7716
7717         RETURN(NULL);
7718 }
7719
7720 static int osd_device_init0(const struct lu_env *env,
7721                             struct osd_device *o,
7722                             struct lustre_cfg *cfg)
7723 {
7724         struct lu_device *l = osd2lu_dev(o);
7725         struct osd_thread_info *info;
7726         int rc;
7727         int cplen = 0;
7728
7729         /* if the module was re-loaded, env can loose its keys */
7730         rc = lu_env_refill((struct lu_env *)env);
7731         if (rc)
7732                 GOTO(out, rc);
7733         info = osd_oti_get(env);
7734         LASSERT(info);
7735
7736         l->ld_ops = &osd_lu_ops;
7737         o->od_dt_dev.dd_ops = &osd_dt_ops;
7738
7739         spin_lock_init(&o->od_osfs_lock);
7740         mutex_init(&o->od_otable_mutex);
7741         INIT_LIST_HEAD(&o->od_orphan_list);
7742         INIT_LIST_HEAD(&o->od_index_backup_list);
7743         INIT_LIST_HEAD(&o->od_index_restore_list);
7744         spin_lock_init(&o->od_lock);
7745         o->od_index_backup_policy = LIBP_NONE;
7746         o->od_t10_type = 0;
7747
7748         o->od_read_cache = 1;
7749         o->od_writethrough_cache = 1;
7750         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7751
7752         o->od_auto_scrub_interval = AS_DEFAULT;
7753
7754         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7755                         sizeof(o->od_svname));
7756         if (cplen >= sizeof(o->od_svname)) {
7757                 rc = -E2BIG;
7758                 GOTO(out, rc);
7759         }
7760
7761         o->od_index_backup_stop = 0;
7762         o->od_index = -1; /* -1 means index is invalid */
7763         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7764         if (rc == LDD_F_SV_TYPE_OST)
7765                 o->od_is_ost = 1;
7766
7767         o->od_full_scrub_ratio = OFSR_DEFAULT;
7768         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7769         rc = osd_mount(env, o, cfg);
7770         if (rc != 0)
7771                 GOTO(out, rc);
7772
7773         /* Can only check block device after mount */
7774         o->od_nonrotational =
7775                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7776
7777         rc = osd_obj_map_init(env, o);
7778         if (rc != 0)
7779                 GOTO(out_mnt, rc);
7780
7781         rc = lu_site_init(&o->od_site, l);
7782         if (rc != 0)
7783                 GOTO(out_compat, rc);
7784         o->od_site.ls_bottom_dev = l;
7785
7786         rc = lu_site_init_finish(&o->od_site);
7787         if (rc != 0)
7788                 GOTO(out_site, rc);
7789
7790         INIT_LIST_HEAD(&o->od_ios_list);
7791         /* setup scrub, including OI files initialization */
7792         o->od_in_init = 1;
7793         rc = osd_scrub_setup(env, o);
7794         o->od_in_init = 0;
7795         if (rc < 0)
7796                 GOTO(out_site, rc);
7797
7798         rc = osd_procfs_init(o, o->od_svname);
7799         if (rc != 0) {
7800                 CERROR("%s: can't initialize procfs: rc = %d\n",
7801                        o->od_svname, rc);
7802                 GOTO(out_scrub, rc);
7803         }
7804
7805         LASSERT(l->ld_site->ls_linkage.next != NULL);
7806         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7807
7808         /* initialize quota slave instance */
7809         /* currently it's no need to prepare qsd_instance_md for OST */
7810         if (!o->od_is_ost) {
7811                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7812                                                 &o->od_dt_dev,
7813                                                 o->od_proc_entry, true);
7814                 if (IS_ERR(o->od_quota_slave_md)) {
7815                         rc = PTR_ERR(o->od_quota_slave_md);
7816                         o->od_quota_slave_md = NULL;
7817                         GOTO(out_procfs, rc);
7818                 }
7819         }
7820
7821         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7822                                         o->od_proc_entry, false);
7823
7824         if (IS_ERR(o->od_quota_slave_dt)) {
7825                 if (o->od_quota_slave_md != NULL) {
7826                         qsd_fini(env, o->od_quota_slave_md);
7827                         o->od_quota_slave_md = NULL;
7828                 }
7829
7830                 rc = PTR_ERR(o->od_quota_slave_dt);
7831                 o->od_quota_slave_dt = NULL;
7832                 GOTO(out_procfs, rc);
7833         }
7834
7835         RETURN(0);
7836
7837 out_procfs:
7838         osd_procfs_fini(o);
7839 out_scrub:
7840         osd_scrub_cleanup(env, o);
7841 out_site:
7842         lu_site_fini(&o->od_site);
7843 out_compat:
7844         osd_obj_map_fini(o);
7845 out_mnt:
7846         osd_umount(env, o);
7847 out:
7848         return rc;
7849 }
7850
7851 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7852                                           struct lu_device_type *t,
7853                                           struct lustre_cfg *cfg)
7854 {
7855         struct osd_device *o;
7856         int rc;
7857
7858         OBD_ALLOC_PTR(o);
7859         if (o == NULL)
7860                 return ERR_PTR(-ENOMEM);
7861
7862         rc = dt_device_init(&o->od_dt_dev, t);
7863         if (rc == 0) {
7864                 /*
7865                  * Because the ctx might be revived in dt_device_init,
7866                  * refill the env here
7867                  */
7868                 lu_env_refill((struct lu_env *)env);
7869                 rc = osd_device_init0(env, o, cfg);
7870                 if (rc)
7871                         dt_device_fini(&o->od_dt_dev);
7872         }
7873
7874         if (unlikely(rc != 0))
7875                 OBD_FREE_PTR(o);
7876
7877         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7878 }
7879
7880 static struct lu_device *osd_device_free(const struct lu_env *env,
7881                                          struct lu_device *d)
7882 {
7883         struct osd_device *o = osd_dev(d);
7884
7885         ENTRY;
7886
7887         /* XXX: make osd top device in order to release reference */
7888         d->ld_site->ls_top_dev = d;
7889         lu_site_purge(env, d->ld_site, -1);
7890         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7891                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7892                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7893         }
7894         lu_site_fini(&o->od_site);
7895         dt_device_fini(&o->od_dt_dev);
7896         OBD_FREE_PTR(o);
7897         RETURN(NULL);
7898 }
7899
7900 static int osd_process_config(const struct lu_env *env,
7901                               struct lu_device *d, struct lustre_cfg *cfg)
7902 {
7903         struct osd_device *o = osd_dev(d);
7904         ssize_t count;
7905         int rc;
7906
7907         ENTRY;
7908
7909         switch (cfg->lcfg_command) {
7910         case LCFG_SETUP:
7911                 rc = osd_mount(env, o, cfg);
7912                 break;
7913         case LCFG_CLEANUP:
7914                 /*
7915                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7916                  * that may happend if hit failure during mount process.
7917                  */
7918                 osd_index_backup(env, o, false);
7919                 lu_dev_del_linkage(d->ld_site, d);
7920                 rc = osd_shutdown(env, o);
7921                 break;
7922         case LCFG_PARAM:
7923                 LASSERT(&o->od_dt_dev);
7924                 count  = class_modify_config(cfg, PARAM_OSD,
7925                                              &o->od_dt_dev.dd_kobj);
7926                 if (count < 0)
7927                         count = class_modify_config(cfg, PARAM_OST,
7928                                                     &o->od_dt_dev.dd_kobj);
7929                 rc = count > 0 ? 0 : count;
7930                 break;
7931         case LCFG_PRE_CLEANUP:
7932                 osd_scrub_stop(o);
7933                 osd_index_backup(env, o,
7934                                  o->od_index_backup_policy != LIBP_NONE);
7935                 rc = 0;
7936                 break;
7937         default:
7938                 rc = -ENOSYS;
7939         }
7940
7941         RETURN(rc);
7942 }
7943
7944 static int osd_recovery_complete(const struct lu_env *env,
7945                                  struct lu_device *d)
7946 {
7947         struct osd_device *osd = osd_dev(d);
7948         int rc = 0;
7949
7950         ENTRY;
7951
7952         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7953                 RETURN(0);
7954
7955         /*
7956          * start qsd instance on recovery completion, this notifies the quota
7957          * slave code that we are about to process new requests now
7958          */
7959         rc = qsd_start(env, osd->od_quota_slave_dt);
7960         if (rc == 0 && osd->od_quota_slave_md != NULL)
7961                 rc = qsd_start(env, osd->od_quota_slave_md);
7962
7963         RETURN(rc);
7964 }
7965
7966 /*
7967  * we use exports to track all osd users
7968  */
7969 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
7970                            struct obd_device *obd, struct obd_uuid *cluuid,
7971                            struct obd_connect_data *data, void *localdata)
7972 {
7973         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
7974         struct lustre_handle conn;
7975         int rc;
7976
7977         ENTRY;
7978
7979         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
7980
7981         rc = class_connect(&conn, obd, cluuid);
7982         if (rc)
7983                 RETURN(rc);
7984
7985         *exp = class_conn2export(&conn);
7986
7987         spin_lock(&osd->od_osfs_lock);
7988         osd->od_connects++;
7989         spin_unlock(&osd->od_osfs_lock);
7990
7991         RETURN(0);
7992 }
7993
7994 /*
7995  * once last export (we don't count self-export) disappeared
7996  * osd can be released
7997  */
7998 static int osd_obd_disconnect(struct obd_export *exp)
7999 {
8000         struct obd_device *obd = exp->exp_obd;
8001         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8002         int rc, release = 0;
8003
8004         ENTRY;
8005
8006         /* Only disconnect the underlying layers on the final disconnect. */
8007         spin_lock(&osd->od_osfs_lock);
8008         osd->od_connects--;
8009         if (osd->od_connects == 0)
8010                 release = 1;
8011         spin_unlock(&osd->od_osfs_lock);
8012
8013         rc = class_disconnect(exp); /* bz 9811 */
8014
8015         if (rc == 0 && release)
8016                 class_manual_cleanup(obd);
8017         RETURN(rc);
8018 }
8019
8020 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8021                        struct lu_device *dev)
8022 {
8023         struct osd_device *osd = osd_dev(dev);
8024         struct lr_server_data *lsd =
8025                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8026         int result = 0;
8027
8028         ENTRY;
8029
8030         if (osd->od_quota_slave_md != NULL) {
8031                 /* set up quota slave objects for inode */
8032                 result = qsd_prepare(env, osd->od_quota_slave_md);
8033                 if (result != 0)
8034                         RETURN(result);
8035         }
8036
8037         if (osd->od_quota_slave_dt != NULL) {
8038                 /* set up quota slave objects for block */
8039                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8040                 if (result != 0)
8041                         RETURN(result);
8042         }
8043
8044
8045         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8046 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8047                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8048                         osd->od_index_in_idif = 1;
8049                 } else {
8050                         osd->od_index_in_idif = 0;
8051                         result = osd_register_proc_index_in_idif(osd);
8052                         if (result != 0)
8053                                 RETURN(result);
8054                 }
8055 #else
8056                 osd->od_index_in_idif = 1;
8057 #endif
8058         }
8059
8060         result = osd_fid_init(env, osd);
8061
8062         RETURN(result);
8063 }
8064
8065 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8066                          struct lu_fid *fid, struct md_op_data *op_data)
8067 {
8068         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8069
8070         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8071 }
8072
8073 static const struct lu_object_operations osd_lu_obj_ops = {
8074         .loo_object_init      = osd_object_init,
8075         .loo_object_delete    = osd_object_delete,
8076         .loo_object_release   = osd_object_release,
8077         .loo_object_free      = osd_object_free,
8078         .loo_object_print     = osd_object_print,
8079         .loo_object_invariant = osd_object_invariant
8080 };
8081
8082 const struct lu_device_operations osd_lu_ops = {
8083         .ldo_object_alloc      = osd_object_alloc,
8084         .ldo_process_config    = osd_process_config,
8085         .ldo_recovery_complete = osd_recovery_complete,
8086         .ldo_prepare           = osd_prepare,
8087 };
8088
8089 static const struct lu_device_type_operations osd_device_type_ops = {
8090         .ldto_init = osd_type_init,
8091         .ldto_fini = osd_type_fini,
8092
8093         .ldto_start = osd_type_start,
8094         .ldto_stop  = osd_type_stop,
8095
8096         .ldto_device_alloc = osd_device_alloc,
8097         .ldto_device_free  = osd_device_free,
8098
8099         .ldto_device_init = osd_device_init,
8100         .ldto_device_fini = osd_device_fini
8101 };
8102
8103 static struct lu_device_type osd_device_type = {
8104         .ldt_tags     = LU_DEVICE_DT,
8105         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8106         .ldt_ops      = &osd_device_type_ops,
8107         .ldt_ctx_tags = LCT_LOCAL,
8108 };
8109
8110 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8111 {
8112         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8113         struct super_block *sb = osd_sb(osd);
8114
8115         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
8116 }
8117
8118 /*
8119  * lprocfs legacy support.
8120  */
8121 static struct obd_ops osd_obd_device_ops = {
8122         .o_owner = THIS_MODULE,
8123         .o_connect      = osd_obd_connect,
8124         .o_disconnect   = osd_obd_disconnect,
8125         .o_fid_alloc    = osd_fid_alloc,
8126         .o_health_check = osd_health_check,
8127 };
8128
8129 static ssize_t track_declares_assert_show(struct kobject *kobj,
8130                                    struct attribute *attr,
8131                                    char *buf)
8132 {
8133         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8134 }
8135
8136 static ssize_t track_declares_assert_store(struct kobject *kobj,
8137                                            struct attribute *attr,
8138                                            const char *buffer, size_t count)
8139 {
8140         bool track_declares_assert;
8141         int rc;
8142
8143         rc = kstrtobool(buffer, &track_declares_assert);
8144         if (rc)
8145                 return rc;
8146
8147         ldiskfs_track_declares_assert = track_declares_assert;
8148
8149         return count;
8150 }
8151 LUSTRE_RW_ATTR(track_declares_assert);
8152
8153 static int __init osd_init(void)
8154 {
8155         struct kobject *kobj;
8156         int rc;
8157
8158         CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
8159 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8160         /* please, try to keep osd_thread_info smaller than a page */
8161         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
8162 #endif
8163
8164         osd_oi_mod_init();
8165
8166         rc = lu_kmem_init(ldiskfs_caches);
8167         if (rc)
8168                 return rc;
8169
8170 #ifdef CONFIG_KALLSYMS
8171         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8172         priv_dev_check_rdonly =
8173                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8174 #endif
8175
8176         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8177                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8178         if (rc) {
8179                 lu_kmem_fini(ldiskfs_caches);
8180                 return rc;
8181         }
8182
8183         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8184         if (kobj) {
8185                 rc = sysfs_create_file(kobj,
8186                                        &lustre_attr_track_declares_assert.attr);
8187                 kobject_put(kobj);
8188                 if (rc) {
8189                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8190                         rc = 0;
8191                 }
8192         }
8193         return rc;
8194 }
8195
8196 static void __exit osd_exit(void)
8197 {
8198         struct kobject *kobj;
8199
8200         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8201         if (kobj) {
8202                 sysfs_remove_file(kobj,
8203                                   &lustre_attr_track_declares_assert.attr);
8204                 kobject_put(kobj);
8205         }
8206         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8207         lu_kmem_fini(ldiskfs_caches);
8208 }
8209
8210 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8211 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8212 MODULE_VERSION(LUSTRE_VERSION_STRING);
8213 MODULE_LICENSE("GPL");
8214
8215 module_init(osd_init);
8216 module_exit(osd_exit);