lustre/osd-ldiskfs/osd_handler.c

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