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 #include <linux/uidgid.h>
  46
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/types.h>
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/fs.h>
  51 /* XATTR_{REPLACE,CREATE} */
  52 #include <linux/xattr.h>
  53
  54 #include <ldiskfs/ldiskfs.h>
  55 #include <ldiskfs/xattr.h>
  56 #include <ldiskfs/ldiskfs_extents.h>
  57 #undef ENTRY
  58 /*
  59  * struct OBD_{ALLOC,FREE}*()
  60  * OBD_FAIL_CHECK
  61  */
  62 #include <obd_support.h>
  63 /* struct ptlrpc_thread */
  64 #include <lustre_net.h>
  65 #include <lustre_fid.h>
  66 /* process_config */
  67 #include <uapi/linux/lustre/lustre_param.h>
  68
  69 #include "osd_internal.h"
  70 #include "osd_dynlocks.h"
  71
  72 /* llo_* api support */
  73 #include <md_object.h>
  74 #include <lustre_quota.h>
  75
  76 #include <lustre_linkea.h>
  77
  78 /* Maximum EA size is limited by LNET_MTU for remote objects */
  79 #define OSD_MAX_EA_SIZE 1048364
  80
  81 int ldiskfs_pdo = 1;
  82 module_param(ldiskfs_pdo, int, 0644);
  83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  84
  85 int ldiskfs_track_declares_assert;
  86 module_param(ldiskfs_track_declares_assert, int, 0644);
  87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  88
  89 /* Slab to allocate dynlocks */
  90 struct kmem_cache *dynlock_cachep;
  91
  92 /* Slab to allocate osd_it_ea */
  93 struct kmem_cache *osd_itea_cachep;
  94
  95 static struct lu_kmem_descr ldiskfs_caches[] = {
  96         {
  97                 .ckd_cache = &dynlock_cachep,
  98                 .ckd_name  = "dynlock_cache",
  99                 .ckd_size  = sizeof(struct dynlock_handle)
 100         },
 101         {
 102                 .ckd_cache = &osd_itea_cachep,
 103                 .ckd_name  = "osd_itea_cache",
 104                 .ckd_size  = sizeof(struct osd_it_ea)
 105         },
 106         {
 107                 .ckd_cache = NULL
 108         }
 109 };
 110
 111 static const char dot[] = ".";
 112 static const char dotdot[] = "..";
 113
 114 static const struct lu_object_operations      osd_lu_obj_ops;
 115 static const struct dt_object_operations      osd_obj_ops;
 116 static const struct dt_object_operations      osd_obj_otable_it_ops;
 117 static const struct dt_index_operations       osd_index_iam_ops;
 118 static const struct dt_index_operations       osd_index_ea_ops;
 119
 120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 121                           const struct lu_fid *fid);
 122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 123                                                 struct osd_device *osd);
 124
 125 int osd_trans_declare_op2rb[] = {
 126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 137 };
 138
 139 static int osd_has_index(const struct osd_object *obj)
 140 {
 141         return obj->oo_dt.do_index_ops != NULL;
 142 }
 143
 144 static int osd_object_invariant(const struct lu_object *l)
 145 {
 146         return osd_invariant(osd_obj(l));
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 153 {
 154         struct osd_thread_info *oti = osd_oti_get(env);
 155
 156         return oti->oti_w_locks > 0 && o->oo_owner == env;
 157 }
 158
 159 /*
 160  * Concurrency: doesn't access mutable data
 161  */
 162 static int osd_root_get(const struct lu_env *env,
 163                         struct dt_device *dev, struct lu_fid *f)
 164 {
 165         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 166         return 0;
 167 }
 168
 169 /*
 170  * the following set of functions are used to maintain per-thread
 171  * cache of FID->ino mapping. this mechanism is needed to resolve
 172  * FID to inode at dt_insert() which in turn stores ino in the
 173  * directory entries to keep ldiskfs compatible with ext[34].
 174  * due to locking-originated restrictions we can't lookup ino
 175  * using LU cache (deadlock is possible). lookup using OI is quite
 176  * expensive. so instead we maintain this cache and methods like
 177  * dt_create() fill it. so in the majority of cases dt_insert() is
 178  * able to find needed mapping in lockless manner.
 179  */
 180 static struct osd_idmap_cache *
 181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 182              const struct lu_fid *fid)
 183 {
 184         struct osd_thread_info *oti = osd_oti_get(env);
 185         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 186         int i;
 187
 188         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 189                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 190                         continue;
 191                 if (idc[i].oic_dev != osd)
 192                         continue;
 193
 194                 return idc + i;
 195         }
 196
 197         return NULL;
 198 }
 199
 200 static struct osd_idmap_cache *
 201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 202             const struct lu_fid *fid)
 203 {
 204         struct osd_thread_info *oti   = osd_oti_get(env);
 205         struct osd_idmap_cache *idc;
 206         int i;
 207
 208         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 209                 i = oti->oti_ins_cache_size * 2;
 210                 if (i == 0)
 211                         i = OSD_INS_CACHE_SIZE;
 212                 OBD_ALLOC(idc, sizeof(*idc) * i);
 213                 if (idc == NULL)
 214                         return ERR_PTR(-ENOMEM);
 215                 if (oti->oti_ins_cache != NULL) {
 216                         memcpy(idc, oti->oti_ins_cache,
 217                                oti->oti_ins_cache_used * sizeof(*idc));
 218                         OBD_FREE(oti->oti_ins_cache,
 219                                  oti->oti_ins_cache_used * sizeof(*idc));
 220                 }
 221                 oti->oti_ins_cache = idc;
 222                 oti->oti_ins_cache_size = i;
 223         }
 224
 225         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 226         idc->oic_fid = *fid;
 227         idc->oic_dev = osd;
 228         idc->oic_lid.oii_ino = 0;
 229         idc->oic_lid.oii_gen = 0;
 230         idc->oic_remote = 0;
 231
 232         return idc;
 233 }
 234
 235 /*
 236  * lookup mapping for the given fid in the cache, initialize a
 237  * new one if not found. the initialization checks whether the
 238  * object is local or remote. for local objects, OI is used to
 239  * learn ino/generation. the function is used when the caller
 240  * has no information about the object, e.g. at dt_insert().
 241  */
 242 static struct osd_idmap_cache *
 243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 244                      const struct lu_fid *fid)
 245 {
 246         struct osd_idmap_cache *idc;
 247         int rc;
 248
 249         idc = osd_idc_find(env, osd, fid);
 250         LASSERT(!IS_ERR(idc));
 251         if (idc != NULL)
 252                 return idc;
 253
 254         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 255                osd->od_svname, PFID(fid));
 256
 257         /* new mapping is needed */
 258         idc = osd_idc_add(env, osd, fid);
 259         if (IS_ERR(idc)) {
 260                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 261                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 262                 return idc;
 263         }
 264
 265         /* initialize it */
 266         rc = osd_remote_fid(env, osd, fid);
 267         if (unlikely(rc < 0))
 268                 return ERR_PTR(rc);
 269
 270         if (rc == 0) {
 271                 /* the object is local, lookup in OI */
 272                 /* XXX: probably cheaper to lookup in LU first? */
 273                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 274                                    &idc->oic_lid, 0);
 275                 if (unlikely(rc < 0)) {
 276                         CERROR("can't lookup: rc = %d\n", rc);
 277                         return ERR_PTR(rc);
 278                 }
 279         } else {
 280                 /* the object is remote */
 281                 idc->oic_remote = 1;
 282         }
 283
 284         return idc;
 285 }
 286
 287 /*
 288  * lookup mapping for given FID and fill it from the given object.
 289  * the object is lolcal by definition.
 290  */
 291 static int osd_idc_find_and_init(const struct lu_env *env,
 292                                  struct osd_device *osd,
 293                                  struct osd_object *obj)
 294 {
 295         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 296         struct osd_idmap_cache *idc;
 297
 298         idc = osd_idc_find(env, osd, fid);
 299         LASSERT(!IS_ERR(idc));
 300         if (idc != NULL) {
 301                 if (obj->oo_inode == NULL)
 302                         return 0;
 303                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 304                         LASSERT(idc->oic_lid.oii_ino == 0);
 305                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 306                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 307                 }
 308                 return 0;
 309         }
 310
 311         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 312                osd->od_svname, PFID(fid));
 313
 314         /* new mapping is needed */
 315         idc = osd_idc_add(env, osd, fid);
 316         if (IS_ERR(idc)) {
 317                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 318                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 319                 return PTR_ERR(idc);
 320         }
 321
 322         if (obj->oo_inode != NULL) {
 323                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 324                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 325         }
 326         return 0;
 327 }
 328
 329 /*
 330  * OSD object methods.
 331  */
 332
 333 /*
 334  * Concurrency: no concurrent access is possible that early in object
 335  * life-cycle.
 336  */
 337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 338                                           const struct lu_object_header *hdr,
 339                                           struct lu_device *d)
 340 {
 341         struct osd_object *mo;
 342
 343         OBD_ALLOC_PTR(mo);
 344         if (mo != NULL) {
 345                 struct lu_object *l;
 346                 struct lu_object_header *h;
 347                 struct osd_device *o = osd_dev(d);
 348
 349                 l = &mo->oo_dt.do_lu;
 350                 if (unlikely(o->od_in_init)) {
 351                         OBD_ALLOC_PTR(h);
 352                         if (!h) {
 353                                 OBD_FREE_PTR(mo);
 354                                 return NULL;
 355                         }
 356
 357                         lu_object_header_init(h);
 358                         lu_object_init(l, h, d);
 359                         lu_object_add_top(h, l);
 360                         mo->oo_header = h;
 361                 } else {
 362                         dt_object_init(&mo->oo_dt, NULL, d);
 363                         mo->oo_header = NULL;
 364                 }
 365
 366                 mo->oo_dt.do_ops = &osd_obj_ops;
 367                 l->lo_ops = &osd_lu_obj_ops;
 368                 init_rwsem(&mo->oo_sem);
 369                 init_rwsem(&mo->oo_ext_idx_sem);
 370                 spin_lock_init(&mo->oo_guard);
 371                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 372                 return l;
 373         }
 374         return NULL;
 375 }
 376
 377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 378                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 379 {
 380         int rc;
 381
 382         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 383                              (void *)loa, sizeof(*loa));
 384         if (rc > 0) {
 385                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 386
 387                 if (rc < sizeof(*lma))
 388                         return -EINVAL;
 389
 390                 rc = 0;
 391                 lustre_loa_swab(loa, true);
 392                 /* Check LMA compatibility */
 393                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 394                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 395                               "for fid = "DFID", ino = %lu\n",
 396                               osd_ino2name(inode),
 397                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 398                               PFID(&lma->lma_self_fid), inode->i_ino);
 399                         rc = -EOPNOTSUPP;
 400                 }
 401         } else if (rc == 0) {
 402                 rc = -ENODATA;
 403         }
 404
 405         return rc;
 406 }
 407
 408 /*
 409  * retrieve object from backend ext fs.
 410  **/
 411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 412                        struct osd_inode_id *id)
 413 {
 414         int rc;
 415         struct inode *inode = NULL;
 416
 417         /*
 418          * if we look for an inode withing a running
 419          * transaction, then we risk to deadlock
 420          * osd_dirent_check_repair() breaks this
 421          */
 422          /* LASSERT(current->journal_info == NULL); */
 423
 424         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 425         if (IS_ERR(inode)) {
 426                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 427                        id->oii_ino, PTR_ERR(inode));
 428         } else if (id->oii_gen != OSD_OII_NOGEN &&
 429                    inode->i_generation != id->oii_gen) {
 430                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 431                        "i_generation = %u\n",
 432                        id->oii_ino, id->oii_gen, inode->i_generation);
 433                 iput(inode);
 434                 inode = ERR_PTR(-ESTALE);
 435         } else if (inode->i_nlink == 0) {
 436                 /*
 437                  * due to parallel readdir and unlink,
 438                  * we can have dead inode here.
 439                  */
 440                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 441                 iput(inode);
 442                 inode = ERR_PTR(-ESTALE);
 443         } else if (is_bad_inode(inode)) {
 444                 CWARN("%s: bad inode: ino = %u\n",
 445                       osd_dev2name(dev), id->oii_ino);
 446                 iput(inode);
 447                 inode = ERR_PTR(-ENOENT);
 448         } else if ((rc = osd_attach_jinode(inode))) {
 449                 iput(inode);
 450                 inode = ERR_PTR(rc);
 451         } else {
 452                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 453                 if (id->oii_gen == OSD_OII_NOGEN)
 454                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 455
 456                 /*
 457                  * Do not update file c/mtime in ldiskfs.
 458                  * NB: we don't have any lock to protect this because we don't
 459                  * have reference on osd_object now, but contention with
 460                  * another lookup + attr_set can't happen in the tiny window
 461                  * between if (...) and set S_NOCMTIME.
 462                  */
 463                 if (!(inode->i_flags & S_NOCMTIME))
 464                         inode->i_flags |= S_NOCMTIME;
 465         }
 466         return inode;
 467 }
 468
 469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 470                           handle_t *handle, struct dentry *child,
 471                           struct inode *inode, struct htree_lock *hlock)
 472 {
 473         int rc, rc2;
 474
 475         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 476         if (rc == -ENOBUFS || rc == -ENOSPC) {
 477                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 478                 struct inode *parent = child->d_parent->d_inode;
 479                 struct lu_fid *fid = NULL;
 480
 481                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 482                 if (!rc2) {
 483                         fid = &loa->loa_lma.lma_self_fid;
 484                 } else if (rc2 == -ENODATA) {
 485                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 486                                 fid = &info->oti_fid3;
 487                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 488                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 489                                 fid = &info->oti_fid3;
 490                                 lu_igif_build(fid, parent->i_ino,
 491                                               parent->i_generation);
 492                         }
 493                 }
 494
 495                 if (fid != NULL)
 496                         /* below message is checked in sanity.sh test_129 */
 497                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
 498                               osd_name(osd), parent->i_ino, PFID(fid),
 499                               rc == -ENOSPC ? "has reached" : "is approaching");
 500                 else
 501                         /* below message is checked in sanity.sh test_129 */
 502                         CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
 503                               osd_name(osd), parent->i_ino,
 504                               rc == -ENOSPC ? "has reached" : "is approaching");
 505
 506                 /* ignore such error now */
 507                 if (rc == -ENOBUFS)
 508                         rc = 0;
 509         }
 510
 511         return rc;
 512 }
 513
 514
 515 struct inode *
 516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 517              struct osd_inode_id *id, struct lu_fid *fid)
 518 {
 519         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 520         struct inode *inode;
 521         int rc;
 522
 523         inode = osd_iget(info, dev, id);
 524         if (IS_ERR(inode))
 525                 return inode;
 526
 527         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 528         if (!rc) {
 529                 *fid = loa->loa_lma.lma_self_fid;
 530         } else if (rc == -ENODATA) {
 531                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 532                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 533                 else
 534                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 535         } else {
 536                 iput(inode);
 537                 inode = ERR_PTR(rc);
 538         }
 539         return inode;
 540 }
 541
 542 static struct inode *osd_iget_check(struct osd_thread_info *info,
 543                                     struct osd_device *dev,
 544                                     const struct lu_fid *fid,
 545                                     struct osd_inode_id *id,
 546                                     bool trusted)
 547 {
 548         struct inode *inode;
 549         int rc = 0;
 550
 551         ENTRY;
 552
 553         /*
 554          * The cached OI mapping is trustable. If we cannot locate the inode
 555          * via the cached OI mapping, then return the failure to the caller
 556          * directly without further OI checking.
 557          */
 558
 559 again:
 560         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 561         if (IS_ERR(inode)) {
 562                 rc = PTR_ERR(inode);
 563                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 564                         goto check_oi;
 565
 566                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 567                        PFID(fid), id->oii_ino, rc);
 568                 GOTO(put, rc);
 569         }
 570
 571         if (is_bad_inode(inode)) {
 572                 rc = -ENOENT;
 573                 if (!trusted)
 574                         goto check_oi;
 575
 576                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 577                        PFID(fid), id->oii_ino);
 578                 GOTO(put, rc);
 579         }
 580
 581         if (id->oii_gen != OSD_OII_NOGEN &&
 582             inode->i_generation != id->oii_gen) {
 583                 rc = -ESTALE;
 584                 if (!trusted)
 585                         goto check_oi;
 586
 587                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 588                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 589                        id->oii_ino, id->oii_gen, inode->i_generation);
 590                 GOTO(put, rc);
 591         }
 592
 593         if (inode->i_nlink == 0) {
 594                 rc = -ENOENT;
 595                 if (!trusted)
 596                         goto check_oi;
 597
 598                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 599                        PFID(fid), id->oii_ino);
 600                 GOTO(put, rc);
 601         }
 602
 603         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 604
 605 check_oi:
 606         if (rc != 0) {
 607                 __u32 saved_ino = id->oii_ino;
 608                 __u32 saved_gen = id->oii_gen;
 609
 610                 LASSERT(!trusted);
 611                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 612
 613                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 614                 /*
 615                  * XXX: There are four possible cases:
 616                  *      1. rc = 0.
 617                  *         Backup/restore caused the OI invalid.
 618                  *      2. rc = 0.
 619                  *         Someone unlinked the object but NOT removed
 620                  *         the OI mapping, such as mount target device
 621                  *         as ldiskfs, and modify something directly.
 622                  *      3. rc = -ENOENT.
 623                  *         Someone just removed the object between the
 624                  *         former oi_lookup and the iget. It is normal.
 625                  *      4. Other failure cases.
 626                  *
 627                  *      Generally, when the device is mounted, it will
 628                  *      auto check whether the system is restored from
 629                  *      file-level backup or not. We trust such detect
 630                  *      to distinguish the 1st case from the 2nd case:
 631                  *      if the OI files are consistent but may contain
 632                  *      stale OI mappings because of case 2, if iget()
 633                  *      returns -ENOENT or -ESTALE, then it should be
 634                  *      the case 2.
 635                  */
 636                 if (rc != 0)
 637                         /*
 638                          * If the OI mapping was in OI file before the
 639                          * osd_iget_check(), but now, it is disappear,
 640                          * then it must be removed by race. That is a
 641                          * normal race case.
 642                          */
 643                         GOTO(put, rc);
 644
 645                 /*
 646                  * It is the OI scrub updated the OI mapping by race.
 647                  * The new OI mapping must be valid.
 648                  */
 649                 if (saved_ino != id->oii_ino ||
 650                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 651                         if (!IS_ERR(inode))
 652                                 iput(inode);
 653
 654                         trusted = true;
 655                         goto again;
 656                 }
 657
 658                 if (IS_ERR(inode)) {
 659                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 660                             SF_INCONSISTENT)
 661                                 /*
 662                                  * It still can be the case 2, but we cannot
 663                                  * distinguish it from the case 1. So return
 664                                  * -EREMCHG to block current operation until
 665                                  *  OI scrub rebuilt the OI mappings.
 666                                  */
 667                                 rc = -EREMCHG;
 668                         else
 669                                 rc = -ENOENT;
 670
 671                         GOTO(put, rc);
 672                 }
 673
 674                 if (inode->i_generation == id->oii_gen)
 675                         rc = -ENOENT;
 676                 else
 677                         rc = -EREMCHG;
 678         } else {
 679                 if (id->oii_gen == OSD_OII_NOGEN)
 680                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 681
 682                 /*
 683                  * Do not update file c/mtime in ldiskfs.
 684                  * NB: we don't have any lock to protect this because we don't
 685                  * have reference on osd_object now, but contention with
 686                  * another lookup + attr_set can't happen in the tiny window
 687                  * between if (...) and set S_NOCMTIME.
 688                  */
 689                 if (!(inode->i_flags & S_NOCMTIME))
 690                         inode->i_flags |= S_NOCMTIME;
 691         }
 692
 693         GOTO(put, rc);
 694
 695 put:
 696         if (rc != 0) {
 697                 if (!IS_ERR(inode))
 698                         iput(inode);
 699
 700                 inode = ERR_PTR(rc);
 701         }
 702
 703         return inode;
 704 }
 705
 706 /**
 707  * \retval +v: new filter_fid does not contain self-fid
 708  * \retval 0:  filter_fid_18_23, contains self-fid
 709  * \retval -v: other failure cases
 710  */
 711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 712                  struct dentry *dentry, struct lu_fid *fid)
 713 {
 714         struct filter_fid *ff = &info->oti_ff;
 715         struct ost_id *ostid = &info->oti_ostid;
 716         int rc;
 717
 718         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 719         if (rc == sizeof(struct filter_fid_18_23)) {
 720                 struct filter_fid_18_23 *ff_old = (void *)ff;
 721
 722                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 723                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 724                 /*
 725                  * XXX: use 0 as the index for compatibility, the caller will
 726                  * handle index related issues when necessary.
 727                  */
 728                 if (!rc)
 729                         ostid_to_fid(fid, ostid, 0);
 730         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 731                 rc = 1;
 732         } else if (rc >= 0) {
 733                 rc = -EINVAL;
 734         }
 735
 736         return rc;
 737 }
 738
 739 static int osd_lma_self_repair(struct osd_thread_info *info,
 740                                struct osd_device *osd, struct inode *inode,
 741                                const struct lu_fid *fid, __u32 compat)
 742 {
 743         handle_t *jh;
 744         int rc;
 745
 746         LASSERT(current->journal_info == NULL);
 747
 748         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 749                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 750         if (IS_ERR(jh)) {
 751                 rc = PTR_ERR(jh);
 752                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 753                       osd_name(osd), rc);
 754                 return rc;
 755         }
 756
 757         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 758         if (rc != 0)
 759                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 760                       osd_name(osd), rc);
 761         ldiskfs_journal_stop(jh);
 762         return rc;
 763 }
 764
 765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 766 {
 767         struct osd_thread_info *info = osd_oti_get(env);
 768         struct osd_device *osd = osd_obj2dev(obj);
 769         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 770         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 771         struct inode *inode = obj->oo_inode;
 772         struct dentry *dentry = &info->oti_obj_dentry;
 773         struct lu_fid *fid = NULL;
 774         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 775         int rc;
 776
 777         ENTRY;
 778
 779         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 780                              (void *)loa, sizeof(*loa));
 781         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 782                 fid = &lma->lma_self_fid;
 783                 rc = osd_get_idif(info, inode, dentry, fid);
 784                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 785                         /*
 786                          * For the given OST-object, if it has neither LMA nor
 787                          * FID in XATTR_NAME_FID, then the given FID (which is
 788                          * contained in the @obj, from client RPC for locating
 789                          * the OST-object) is trusted. We use it to generate
 790                          * the LMA.
 791                          */
 792                         osd_lma_self_repair(info, osd, inode, rfid,
 793                                             LMAC_FID_ON_OST);
 794                         RETURN(0);
 795                 }
 796         }
 797
 798         if (rc < 0)
 799                 RETURN(rc);
 800
 801         if (rc > 0) {
 802                 rc = 0;
 803                 lustre_lma_swab(lma);
 804                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 805                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 806                               S_ISREG(inode->i_mode)))) {
 807                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 808                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 809                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 810                               PFID(rfid), inode->i_ino);
 811                         rc = -EOPNOTSUPP;
 812                 } else {
 813                         fid = &lma->lma_self_fid;
 814                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 815                             osd->od_is_ost)
 816                                 obj->oo_pfid_in_lma = 1;
 817                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 818                             !osd->od_is_ost)
 819                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 820                 }
 821         }
 822
 823         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 824                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 825                         struct ost_id   *oi   = &info->oti_ostid;
 826                         struct lu_fid   *fid1 = &info->oti_fid3;
 827                         __u32            idx  = fid_idif_ost_idx(rfid);
 828
 829                         /*
 830                          * For old IDIF, the OST index is not part of the IDIF,
 831                          * Means that different OSTs may have the same IDIFs.
 832                          * Under such case, we need to make some compatible
 833                          * check to make sure to trigger OI scrub properly.
 834                          */
 835                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 836                                 /* Given @rfid is new, LMA is old. */
 837                                 fid_to_ostid(fid, oi);
 838                                 ostid_to_fid(fid1, oi, idx);
 839                                 if (lu_fid_eq(fid1, rfid)) {
 840                                         if (osd->od_index_in_idif)
 841                                                 osd_lma_self_repair(info, osd,
 842                                                         inode, rfid,
 843                                                         LMAC_FID_ON_OST);
 844                                         RETURN(0);
 845                                 }
 846                         }
 847                 }
 848
 849                 rc = -EREMCHG;
 850         }
 851
 852         RETURN(rc);
 853 }
 854
 855 struct osd_check_lmv_buf {
 856         /* please keep it as first member */
 857         struct dir_context ctx;
 858         struct osd_thread_info *oclb_info;
 859         struct osd_device *oclb_dev;
 860         struct osd_idmap_cache *oclb_oic;
 861         int oclb_items;
 862         bool oclb_found;
 863 };
 864
 865 /**
 866  * It is called internally by ->iterate*() to filter out the
 867  * local slave object's FID of the striped directory.
 868  *
 869  * \retval      1 found the local slave's FID
 870  * \retval      0 continue to check next item
 871  * \retval      -ve for failure
 872  */
 873 #ifdef HAVE_FILLDIR_USE_CTX
 874 static int osd_stripe_dir_filldir(struct dir_context *buf,
 875 #else
 876 static int osd_stripe_dir_filldir(void *buf,
 877 #endif
 878                                   const char *name, int namelen,
 879                                   loff_t offset, __u64 ino, unsigned int d_type)
 880 {
 881         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 882         struct osd_thread_info *oti = oclb->oclb_info;
 883         struct lu_fid *fid = &oti->oti_fid3;
 884         struct osd_inode_id *id = &oti->oti_id3;
 885         struct osd_device *dev = oclb->oclb_dev;
 886         struct osd_idmap_cache *oic = oclb->oclb_oic;
 887         struct inode *inode;
 888
 889         oclb->oclb_items++;
 890
 891         if (name[0] == '.')
 892                 return 0;
 893
 894         fid_zero(fid);
 895         sscanf(name + 1, SFID, RFID(fid));
 896         if (!fid_is_sane(fid))
 897                 return 0;
 898
 899         if (osd_remote_fid(oti->oti_env, dev, fid))
 900                 return 0;
 901
 902         osd_id_gen(id, ino, OSD_OII_NOGEN);
 903         inode = osd_iget(oti, dev, id);
 904         if (IS_ERR(inode))
 905                 return PTR_ERR(inode);
 906
 907         iput(inode);
 908         osd_add_oi_cache(oti, dev, id, fid);
 909         oic->oic_fid = *fid;
 910         oic->oic_lid = *id;
 911         oic->oic_dev = dev;
 912         osd_oii_insert(dev, oic, true);
 913         oclb->oclb_found = true;
 914
 915         return 1;
 916 }
 917
 918 /*
 919  * When lookup item under striped directory, we need to locate the master
 920  * MDT-object of the striped directory firstly, then the client will send
 921  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 922  * and the item's name. If the system is restored from MDT file level backup,
 923  * then before the OI scrub completely built the OI files, the OI mappings of
 924  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 925  * not a problem for the master MDT-object. Because when locate the master
 926  * MDT-object, we will do name based lookup (for the striped directory itself)
 927  * firstly, during such process we can setup the correct OI mapping for the
 928  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 929  * the client will not trigger name based lookup on the MDT to locate the slave
 930  * MDT-object before locating item under the striped directory, then when
 931  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 932  * is invalid and does not know what the right OI mapping is, then the MDT has
 933  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 934  * the OI file, related OI mapping is unknown yet, please try again later. And
 935  * then client will re-try the RPC again and again until related OI mapping has
 936  * been updated. That is quite inefficient.
 937  *
 938  * To resolve above trouble, we will handle it as the following two cases:
 939  *
 940  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 941  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 942  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 943  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 944  *    directly. Please check osd_fid_lookup().
 945  *
 946  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 947  *    Under such case, during lookup the master MDT-object, we will lookup the
 948  *    slave MDT-object via readdir against the master MDT-object, because the
 949  *    slave MDT-objects information are stored as sub-directories with the name
 950  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 951  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 952  *    the correct OI mapping for the slave MDT-object.
 953  */
 954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 955                          struct inode *inode, struct osd_idmap_cache *oic)
 956 {
 957         struct lu_buf *buf = &oti->oti_big_buf;
 958         struct dentry *dentry = &oti->oti_obj_dentry;
 959         struct file *filp = &oti->oti_file;
 960         const struct file_operations *fops;
 961         struct lmv_mds_md_v1 *lmv1;
 962         struct osd_check_lmv_buf oclb = {
 963                 .ctx.actor = osd_stripe_dir_filldir,
 964                 .oclb_info = oti,
 965                 .oclb_dev = dev,
 966                 .oclb_oic = oic,
 967                 .oclb_found = false,
 968         };
 969         int rc = 0;
 970
 971         ENTRY;
 972
 973 again:
 974         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 975                              buf->lb_len);
 976         if (rc == -ERANGE) {
 977                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 978                 if (rc > 0) {
 979                         lu_buf_realloc(buf, rc);
 980                         if (buf->lb_buf == NULL)
 981                                 GOTO(out, rc = -ENOMEM);
 982
 983                         goto again;
 984                 }
 985         }
 986
 987         if (unlikely(rc == 0 || rc == -ENODATA))
 988                 GOTO(out, rc = 0);
 989
 990         if (rc < 0)
 991                 GOTO(out, rc);
 992
 993         if (unlikely(buf->lb_buf == NULL)) {
 994                 lu_buf_realloc(buf, rc);
 995                 if (buf->lb_buf == NULL)
 996                         GOTO(out, rc = -ENOMEM);
 997
 998                 goto again;
 999         }
1000
1001         lmv1 = buf->lb_buf;
1002         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1003                 GOTO(out, rc = 0);
1004
1005         fops = inode->i_fop;
1006         dentry->d_inode = inode;
1007         dentry->d_sb = inode->i_sb;
1008         filp->f_pos = 0;
1009         filp->f_path.dentry = dentry;
1010         filp->f_flags |= O_NOATIME;
1011         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012         filp->f_mapping = inode->i_mapping;
1013         filp->f_op = fops;
1014         filp->private_data = NULL;
1015         filp->f_cred = current_cred();
1016         filp->f_inode = inode;
1017         rc = osd_security_file_alloc(filp);
1018         if (rc)
1019                 goto out;
1020
1021         do {
1022                 oclb.oclb_items = 0;
1023                 rc = iterate_dir(filp, &oclb.ctx);
1024         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026         fops->release(inode, filp);
1027
1028 out:
1029         if (rc < 0)
1030                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031                        DFID": rc = %d\n", osd_ino2name(inode),
1032                        inode->i_ino, inode->i_generation,
1033                        PFID(&oic->oic_fid), rc);
1034         else
1035                 rc = 0;
1036
1037         RETURN(rc);
1038 }
1039
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041                           const struct lu_fid *fid,
1042                           const struct lu_object_conf *conf)
1043 {
1044         struct osd_thread_info *info;
1045         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046         struct osd_device *dev;
1047         struct osd_idmap_cache *oic;
1048         struct osd_inode_id *id;
1049         struct inode *inode = NULL;
1050         struct lustre_scrub *scrub;
1051         struct scrub_file *sf;
1052         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1053         __u32 saved_ino;
1054         __u32 saved_gen;
1055         int result = 0;
1056         int rc1 = 0;
1057         bool remote = false;
1058         bool trusted = true;
1059         bool updated = false;
1060         bool checked = false;
1061
1062         ENTRY;
1063
1064         LINVRNT(osd_invariant(obj));
1065         LASSERT(obj->oo_inode == NULL);
1066         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1067
1068         dev = osd_dev(ldev);
1069         scrub = &dev->od_scrub.os_scrub;
1070         sf = &scrub->os_file;
1071         info = osd_oti_get(env);
1072         LASSERT(info);
1073         oic = &info->oti_cache;
1074
1075         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1076                 RETURN(-ENOENT);
1077
1078         /*
1079          * For the object is created as locking anchor, or for the object to
1080          * be created on disk. No need to osd_oi_lookup() at here because FID
1081          * shouldn't never be re-used, if it's really a duplicate FID from
1082          * unexpected reason, we should be able to detect it later by calling
1083          * do_create->osd_oi_insert().
1084          */
1085         if (conf && conf->loc_flags & LOC_F_NEW)
1086                 GOTO(out, result = 0);
1087
1088         /* Search order: 1. per-thread cache. */
1089         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1090                 id = &oic->oic_lid;
1091                 goto iget;
1092         }
1093
1094         id = &info->oti_id;
1095         if (!list_empty(&scrub->os_inconsistent_items)) {
1096                 /* Search order: 2. OI scrub pending list. */
1097                 result = osd_oii_lookup(dev, fid, id);
1098                 if (!result)
1099                         goto iget;
1100         }
1101
1102         /*
1103          * The OI mapping in the OI file can be updated by the OI scrub
1104          * when we locate the inode via FID. So it may be not trustable.
1105          */
1106         trusted = false;
1107
1108         /* Search order: 3. OI files. */
1109         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110         if (result == -ENOENT) {
1111                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1114                                       sf->sf_oi_bitmap))
1115                         GOTO(out, result = 0);
1116
1117                 goto trigger;
1118         }
1119
1120         /* -ESTALE is returned if inode of OST object doesn't exist */
1121         if (result == -ESTALE &&
1122             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123                 GOTO(out, result = 0);
1124         }
1125
1126         if (result)
1127                 GOTO(out, result);
1128
1129 iget:
1130         obj->oo_inode = NULL;
1131         /* for later passes through checks, not true on first pass */
1132         if (!IS_ERR_OR_NULL(inode))
1133                 iput(inode);
1134
1135         inode = osd_iget_check(info, dev, fid, id, trusted);
1136         if (!IS_ERR(inode)) {
1137                 obj->oo_inode = inode;
1138                 result = 0;
1139                 if (remote)
1140                         goto trigger;
1141
1142                 goto check_lma;
1143         }
1144
1145         result = PTR_ERR(inode);
1146         if (result == -ENOENT || result == -ESTALE)
1147                 GOTO(out, result = 0);
1148
1149         if (result != -EREMCHG)
1150                 GOTO(out, result);
1151
1152 trigger:
1153         /*
1154          * We still have chance to get the valid inode: for the
1155          * object which is referenced by remote name entry, the
1156          * object on the local MDT will be linked under the dir
1157          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1158          *
1159          * We do not know whether the object for the given FID
1160          * is referenced by some remote name entry or not, and
1161          * especially for DNE II, a multiple-linked object may
1162          * have many name entries reside on many MDTs.
1163          *
1164          * To simplify the operation, OSD will not distinguish
1165          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166          * only happened for the RPC from other MDT during the
1167          * OI scrub, or for the client side RPC with FID only,
1168          * such as FID to path, or from old connected client.
1169          */
1170         if (!remote) {
1171                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1172                 if (!rc1) {
1173                         remote = true;
1174                         trusted = true;
1175                         flags |= SS_AUTO_PARTIAL;
1176                         flags &= ~SS_AUTO_FULL;
1177                         goto iget;
1178                 }
1179         }
1180
1181         if (thread_is_running(&scrub->os_thread)) {
1182                 if (scrub->os_partial_scan && !scrub->os_in_join)
1183                         goto join;
1184
1185                 if (IS_ERR_OR_NULL(inode) || result)
1186                         GOTO(out, result = -EINPROGRESS);
1187
1188                 LASSERT(remote);
1189                 LASSERT(obj->oo_inode == inode);
1190
1191                 osd_add_oi_cache(info, dev, id, fid);
1192                 osd_oii_insert(dev, oic, true);
1193                 goto found;
1194         }
1195
1196         if (dev->od_auto_scrub_interval == AS_NEVER) {
1197                 if (!remote)
1198                         GOTO(out, result = -EREMCHG);
1199
1200                 LASSERT(!result);
1201                 LASSERT(obj->oo_inode == inode);
1202
1203                 osd_add_oi_cache(info, dev, id, fid);
1204                 goto found;
1205         }
1206
1207 join:
1208         rc1 = osd_scrub_start(env, dev, flags);
1209         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211         if (rc1 && rc1 != -EALREADY)
1212                 GOTO(out, result = -EREMCHG);
1213
1214         if (IS_ERR_OR_NULL(inode) || result)
1215                 GOTO(out, result = -EINPROGRESS);
1216
1217         LASSERT(remote);
1218         LASSERT(obj->oo_inode == inode);
1219
1220         osd_add_oi_cache(info, dev, id, fid);
1221         osd_oii_insert(dev, oic, true);
1222         goto found;
1223
1224 check_lma:
1225         checked = true;
1226         if (unlikely(obj->oo_header))
1227                 goto found;
1228
1229         result = osd_check_lma(env, obj);
1230         if (!result)
1231                 goto found;
1232
1233         LASSERTF(id->oii_ino == inode->i_ino &&
1234                  id->oii_gen == inode->i_generation,
1235                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236                  PFID(fid), id->oii_ino, id->oii_gen,
1237                  inode->i_ino, inode->i_generation);
1238
1239         saved_ino = inode->i_ino;
1240         saved_gen = inode->i_generation;
1241
1242         if (unlikely(result == -ENODATA)) {
1243                 /*
1244                  * If the OI scrub updated the OI mapping by race, it
1245                  * must be valid. Trust the inode that has no LMA EA.
1246                  */
1247                 if (updated)
1248                         goto found;
1249
1250                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1251                 if (!result) {
1252                         /*
1253                          * The OI mapping is still there, the inode is still
1254                          * valid. It is just becaues the inode has no LMA EA.
1255                          */
1256                         if (saved_ino == id->oii_ino &&
1257                             saved_gen == id->oii_gen)
1258                                 goto found;
1259
1260                         /*
1261                          * It is the OI scrub updated the OI mapping by race.
1262                          * The new OI mapping must be valid.
1263                          */
1264                         trusted = true;
1265                         updated = true;
1266                         goto iget;
1267                 }
1268
1269                 /*
1270                  * "result == -ENOENT" means that the OI mappinghas been
1271                  * removed by race, so the inode belongs to other object.
1272                  *
1273                  * Others error can be returned  directly.
1274                  */
1275                 if (result == -ENOENT) {
1276                         LASSERT(trusted);
1277
1278                         obj->oo_inode = NULL;
1279                         result = 0;
1280                 }
1281         }
1282
1283         if (result != -EREMCHG)
1284                 GOTO(out, result);
1285
1286         LASSERT(!updated);
1287
1288         /*
1289          * if two OST objects map to the same inode, and inode mode is
1290          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291          * reserved by precreate, and not written yet, in this case, don't
1292          * set inode for the object whose FID mismatch, so that it can create
1293          * inode and not block precreate.
1294          */
1295         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297                 obj->oo_inode = NULL;
1298                 GOTO(out, result = 0);
1299         }
1300
1301         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1302         /*
1303          * "result == -ENOENT" means the cached OI mapping has been removed
1304          * from the OI file by race, above inode belongs to other object.
1305          */
1306         if (result == -ENOENT) {
1307                 LASSERT(trusted);
1308
1309                 obj->oo_inode = NULL;
1310                 GOTO(out, result = 0);
1311         }
1312
1313         if (result)
1314                 GOTO(out, result);
1315
1316         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1317                 result = -EREMCHG;
1318                 goto trigger;
1319         }
1320
1321         /*
1322          * It is the OI scrub updated the OI mapping by race.
1323          * The new OI mapping must be valid.
1324          */
1325         trusted = true;
1326         updated = true;
1327         goto iget;
1328
1329 found:
1330         if (!checked) {
1331                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1333
1334                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1335                 if (!result) {
1336                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1337                             dev->od_is_ost)
1338                                 obj->oo_pfid_in_lma = 1;
1339                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1340                             !dev->od_is_ost)
1341                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342                 } else if (result != -ENODATA) {
1343                         GOTO(out, result);
1344                 }
1345         }
1346
1347         obj->oo_compat_dot_created = 1;
1348         obj->oo_compat_dotdot_created = 1;
1349
1350         if (S_ISDIR(inode->i_mode) &&
1351             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352                 osd_check_lmv(info, dev, inode, oic);
1353
1354         result = osd_attach_jinode(inode);
1355         if (result)
1356                 GOTO(out, result);
1357
1358         if (!ldiskfs_pdo)
1359                 GOTO(out, result = 0);
1360
1361         LASSERT(!obj->oo_hl_head);
1362         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1363
1364         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1365
1366 out:
1367         if (result || !obj->oo_inode) {
1368                 if (!IS_ERR_OR_NULL(inode))
1369                         iput(inode);
1370
1371                 obj->oo_inode = NULL;
1372                 if (trusted)
1373                         fid_zero(&oic->oic_fid);
1374         }
1375
1376         LINVRNT(osd_invariant(obj));
1377         return result;
1378 }
1379
1380 /*
1381  * Concurrency: shouldn't matter.
1382  */
1383 static void osd_object_init0(struct osd_object *obj)
1384 {
1385         LASSERT(obj->oo_inode != NULL);
1386         obj->oo_dt.do_body_ops = &osd_body_ops;
1387         obj->oo_dt.do_lu.lo_header->loh_attr |=
1388                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1389 }
1390
1391 /*
1392  * Concurrency: no concurrent access is possible that early in object
1393  * life-cycle.
1394  */
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396                            const struct lu_object_conf *conf)
1397 {
1398         struct osd_object *obj = osd_obj(l);
1399         int result;
1400
1401         LINVRNT(osd_invariant(obj));
1402
1403         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1404                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1405                 l->lo_header->loh_attr |= LOHA_EXISTS;
1406                 return 0;
1407         }
1408
1409         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1410         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1411         if (result == 0 && obj->oo_inode != NULL) {
1412                 struct osd_thread_info *oti = osd_oti_get(env);
1413                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1414
1415                 osd_object_init0(obj);
1416                 if (unlikely(obj->oo_header))
1417                         return 0;
1418
1419                 result = osd_get_lma(oti, obj->oo_inode,
1420                                      &oti->oti_obj_dentry, loa);
1421                 if (!result) {
1422                         /*
1423                          * Convert LMAI flags to lustre LMA flags
1424                          * and cache it to oo_lma_flags
1425                          */
1426                         obj->oo_lma_flags =
1427                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1428                 } else if (result == -ENODATA) {
1429                         result = 0;
1430                 }
1431         }
1432
1433         LINVRNT(osd_invariant(obj));
1434         return result;
1435 }
1436
1437 /*
1438  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1439  * The left part is for value, binary mode.
1440  */
1441 struct osd_xattr_entry {
1442         struct list_head        oxe_list;
1443         size_t                  oxe_len;
1444         size_t                  oxe_namelen;
1445         bool                    oxe_exist;
1446         struct rcu_head         oxe_rcu;
1447         char                    oxe_buf[0];
1448 };
1449
1450 static int osd_oxc_get(struct osd_object *obj, const char *name,
1451                        struct lu_buf *buf)
1452 {
1453         struct osd_xattr_entry *tmp;
1454         struct osd_xattr_entry *oxe = NULL;
1455         size_t namelen = strlen(name);
1456         int rc;
1457
1458         rcu_read_lock();
1459         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1460                 if (namelen == tmp->oxe_namelen &&
1461                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1462                         oxe = tmp;
1463                         break;
1464                 }
1465         }
1466
1467         if (oxe == NULL)
1468                 GOTO(out, rc = -ENOENT);
1469
1470         if (!oxe->oxe_exist)
1471                 GOTO(out, rc = -ENODATA);
1472
1473         /* vallen */
1474         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1475         LASSERT(rc > 0);
1476
1477         if (buf->lb_buf == NULL)
1478                 GOTO(out, rc);
1479
1480         if (buf->lb_len < rc)
1481                 GOTO(out, rc = -ERANGE);
1482
1483         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1484 out:
1485         rcu_read_unlock();
1486
1487         return rc;
1488 }
1489
1490 static void osd_oxc_free(struct rcu_head *head)
1491 {
1492         struct osd_xattr_entry *oxe;
1493
1494         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1495         OBD_FREE(oxe, oxe->oxe_len);
1496 }
1497
1498 static void osd_oxc_add(struct osd_object *obj, const char *name,
1499                         const char *buf, int buflen)
1500 {
1501         struct osd_xattr_entry *oxe;
1502         struct osd_xattr_entry *old = NULL;
1503         struct osd_xattr_entry *tmp;
1504         size_t namelen = strlen(name);
1505         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1506
1507         OBD_ALLOC(oxe, len);
1508         if (oxe == NULL)
1509                 return;
1510
1511         INIT_LIST_HEAD(&oxe->oxe_list);
1512         oxe->oxe_len = len;
1513         oxe->oxe_namelen = namelen;
1514         memcpy(oxe->oxe_buf, name, namelen);
1515         if (buflen > 0) {
1516                 LASSERT(buf != NULL);
1517                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1518                 oxe->oxe_exist = true;
1519         } else {
1520                 oxe->oxe_exist = false;
1521         }
1522
1523         /* this should be rarely called, just remove old and add new */
1524         spin_lock(&obj->oo_guard);
1525         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1526                 if (namelen == tmp->oxe_namelen &&
1527                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1528                         old = tmp;
1529                         break;
1530                 }
1531         }
1532         if (old != NULL) {
1533                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1534                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1535         } else {
1536                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1537         }
1538         spin_unlock(&obj->oo_guard);
1539 }
1540
1541 static void osd_oxc_del(struct osd_object *obj, const char *name)
1542 {
1543         struct osd_xattr_entry *oxe;
1544         size_t namelen = strlen(name);
1545
1546         spin_lock(&obj->oo_guard);
1547         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1548                 if (namelen == oxe->oxe_namelen &&
1549                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1550                         list_del_rcu(&oxe->oxe_list);
1551                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1552                         break;
1553                 }
1554         }
1555         spin_unlock(&obj->oo_guard);
1556 }
1557
1558 static void osd_oxc_fini(struct osd_object *obj)
1559 {
1560         struct osd_xattr_entry *oxe, *next;
1561
1562         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1563                 list_del(&oxe->oxe_list);
1564                 OBD_FREE(oxe, oxe->oxe_len);
1565         }
1566 }
1567
1568 /*
1569  * Concurrency: no concurrent access is possible that late in object
1570  * life-cycle.
1571  */
1572 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1573 {
1574         struct osd_object *obj = osd_obj(l);
1575         struct lu_object_header *h = obj->oo_header;
1576
1577         LINVRNT(osd_invariant(obj));
1578
1579         osd_oxc_fini(obj);
1580         dt_object_fini(&obj->oo_dt);
1581         if (obj->oo_hl_head != NULL)
1582                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1583         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1584         OBD_FREE_PTR(obj);
1585         if (unlikely(h)) {
1586                 lu_object_header_fini(h);
1587                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1588                 kfree_rcu(h, loh_rcu);
1589         }
1590 }
1591
1592 /*
1593  * Concurrency: no concurrent access is possible that late in object
1594  * life-cycle.
1595  */
1596 static void osd_index_fini(struct osd_object *o)
1597 {
1598         struct iam_container *bag;
1599
1600         if (o->oo_dir != NULL) {
1601                 bag = &o->oo_dir->od_container;
1602                 if (o->oo_inode != NULL) {
1603                         if (bag->ic_object == o->oo_inode)
1604                                 iam_container_fini(bag);
1605                 }
1606                 OBD_FREE_PTR(o->oo_dir);
1607                 o->oo_dir = NULL;
1608         }
1609 }
1610
1611 /*
1612  * Concurrency: no concurrent access is possible that late in object
1613  * life-cycle (for all existing callers, that is. New callers have to provide
1614  * their own locking.)
1615  */
1616 static int osd_inode_unlinked(const struct inode *inode)
1617 {
1618         return inode->i_nlink == 0;
1619 }
1620
1621 enum {
1622         OSD_TXN_OI_DELETE_CREDITS    = 20,
1623         OSD_TXN_INODE_DELETE_CREDITS = 20
1624 };
1625
1626 /*
1627  * Journal
1628  */
1629
1630 #if OSD_THANDLE_STATS
1631 /**
1632  * Set time when the handle is allocated
1633  */
1634 static void osd_th_alloced(struct osd_thandle *oth)
1635 {
1636         oth->oth_alloced = ktime_get();
1637 }
1638
1639 /**
1640  * Set time when the handle started
1641  */
1642 static void osd_th_started(struct osd_thandle *oth)
1643 {
1644         oth->oth_started = ktime_get();
1645 }
1646
1647 /**
1648  * Check whether the we deal with this handle for too long.
1649  */
1650 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1651                                 ktime_t alloced, ktime_t started,
1652                                 ktime_t closed)
1653 {
1654         ktime_t now = ktime_get();
1655
1656         LASSERT(dev != NULL);
1657
1658         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1659                             ktime_us_delta(started, alloced));
1660         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1661                             ktime_us_delta(closed, started));
1662         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1663                             ktime_us_delta(now, closed));
1664
1665         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1666                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1667                                 oth, now, alloced, started, closed);
1668                 libcfs_debug_dumpstack(NULL);
1669         }
1670 }
1671
1672 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1673 {                                                                       \
1674         ktime_t __closed = ktime_get();                                 \
1675         ktime_t __alloced = oth->oth_alloced;                           \
1676         ktime_t __started = oth->oth_started;                           \
1677                                                                         \
1678         expr;                                                           \
1679         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1680 }
1681
1682 #else /* OSD_THANDLE_STATS */
1683
1684 #define osd_th_alloced(h)                  do {} while(0)
1685 #define osd_th_started(h)                  do {} while(0)
1686 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1687
1688 #endif /* OSD_THANDLE_STATS */
1689
1690 /*
1691  * Concurrency: doesn't access mutable data.
1692  */
1693 static int osd_param_is_not_sane(const struct osd_device *dev,
1694                                  const struct thandle *th)
1695 {
1696         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1697
1698         return oh->ot_credits > osd_transaction_size(dev);
1699 }
1700
1701 /*
1702  * Concurrency: shouldn't matter.
1703  */
1704 static void osd_trans_commit_cb(struct super_block *sb,
1705                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1706 {
1707         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1708         struct thandle *th = &oh->ot_super;
1709         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1710         struct dt_txn_commit_cb *dcb, *tmp;
1711
1712         LASSERT(oh->ot_handle == NULL);
1713
1714         if (error)
1715                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1716
1717         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1718         /* call per-transaction callbacks if any */
1719         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1720                                  dcb_linkage) {
1721                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1722                          "commit callback entry: magic=%x name='%s'\n",
1723                          dcb->dcb_magic, dcb->dcb_name);
1724                 list_del_init(&dcb->dcb_linkage);
1725                 dcb->dcb_func(NULL, th, dcb, error);
1726         }
1727
1728         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1729         lu_device_put(lud);
1730         th->th_dev = NULL;
1731
1732         OBD_FREE_PTR(oh);
1733 }
1734
1735 static struct thandle *osd_trans_create(const struct lu_env *env,
1736                                         struct dt_device *d)
1737 {
1738         struct osd_thread_info *oti = osd_oti_get(env);
1739         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1740         struct osd_thandle *oh;
1741         struct thandle *th;
1742
1743         ENTRY;
1744
1745         if (d->dd_rdonly) {
1746                 CERROR("%s: someone try to start transaction under "
1747                        "readonly mode, should be disabled.\n",
1748                        osd_name(osd_dt_dev(d)));
1749                 dump_stack();
1750                 RETURN(ERR_PTR(-EROFS));
1751         }
1752
1753         /* on pending IO in this thread should left from prev. request */
1754         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1755
1756         sb_start_write(osd_sb(osd_dt_dev(d)));
1757
1758         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1759         if (!oh) {
1760                 sb_end_write(osd_sb(osd_dt_dev(d)));
1761                 RETURN(ERR_PTR(-ENOMEM));
1762         }
1763
1764         oh->ot_quota_trans = &oti->oti_quota_trans;
1765         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1766         th = &oh->ot_super;
1767         th->th_dev = d;
1768         th->th_result = 0;
1769         oh->ot_credits = 0;
1770         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1771         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1772         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1773         osd_th_alloced(oh);
1774
1775         memset(oti->oti_declare_ops, 0,
1776                sizeof(oti->oti_declare_ops));
1777         memset(oti->oti_declare_ops_cred, 0,
1778                sizeof(oti->oti_declare_ops_cred));
1779         memset(oti->oti_declare_ops_used, 0,
1780                sizeof(oti->oti_declare_ops_used));
1781
1782         oti->oti_ins_cache_depth++;
1783
1784         RETURN(th);
1785 }
1786
1787 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1788 {
1789         struct osd_thread_info *oti = osd_oti_get(env);
1790         struct osd_thandle *oh;
1791
1792         oh = container_of0(th, struct osd_thandle, ot_super);
1793         LASSERT(oh != NULL);
1794
1795         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1796               oti->oti_declare_ops[OSD_OT_CREATE],
1797               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1798               oti->oti_declare_ops_used[OSD_OT_CREATE],
1799               oti->oti_declare_ops[OSD_OT_DESTROY],
1800               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1801               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1802         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1803               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1804               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1805               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1806               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1807               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1808               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1809         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1810               oti->oti_declare_ops[OSD_OT_WRITE],
1811               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1812               oti->oti_declare_ops_used[OSD_OT_WRITE],
1813               oti->oti_declare_ops[OSD_OT_PUNCH],
1814               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1815               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1816               oti->oti_declare_ops[OSD_OT_QUOTA],
1817               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1818               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1819         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1820               oti->oti_declare_ops[OSD_OT_INSERT],
1821               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1822               oti->oti_declare_ops_used[OSD_OT_INSERT],
1823               oti->oti_declare_ops[OSD_OT_DELETE],
1824               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1825               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1826         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1827               oti->oti_declare_ops[OSD_OT_REF_ADD],
1828               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1829               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1830               oti->oti_declare_ops[OSD_OT_REF_DEL],
1831               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1832               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1833 }
1834
1835 /*
1836  * Concurrency: shouldn't matter.
1837  */
1838 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1839                            struct thandle *th)
1840 {
1841         struct osd_thread_info *oti = osd_oti_get(env);
1842         struct osd_device *dev = osd_dt_dev(d);
1843         handle_t *jh;
1844         struct osd_thandle *oh;
1845         int rc;
1846
1847         ENTRY;
1848
1849         LASSERT(current->journal_info == NULL);
1850
1851         oh = container_of0(th, struct osd_thandle, ot_super);
1852         LASSERT(oh != NULL);
1853         LASSERT(oh->ot_handle == NULL);
1854
1855         rc = dt_txn_hook_start(env, d, th);
1856         if (rc != 0)
1857                 GOTO(out, rc);
1858
1859         if (unlikely(osd_param_is_not_sane(dev, th))) {
1860                 static unsigned long last_printed;
1861                 static int last_credits;
1862
1863                 /*
1864                  * don't make noise on a tiny testing systems
1865                  * actual credits misuse will be caught anyway
1866                  */
1867                 if (last_credits != oh->ot_credits &&
1868                     time_after(jiffies, last_printed +
1869                                cfs_time_seconds(60)) &&
1870                     osd_transaction_size(dev) > 512) {
1871                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1872                               oh->ot_credits, osd_transaction_size(dev));
1873                         osd_trans_dump_creds(env, th);
1874                         libcfs_debug_dumpstack(NULL);
1875                         last_credits = oh->ot_credits;
1876                         last_printed = jiffies;
1877                 }
1878                 /*
1879                  * XXX Limit the credits to 'max_transaction_buffers', and
1880                  *     let the underlying filesystem to catch the error if
1881                  *     we really need so many credits.
1882                  *
1883                  *     This should be removed when we can calculate the
1884                  *     credits precisely.
1885                  */
1886                 oh->ot_credits = osd_transaction_size(dev);
1887         } else if (ldiskfs_track_declares_assert != 0) {
1888                 /*
1889                  * reserve few credits to prevent an assertion in JBD
1890                  * our debugging mechanism will be able to detected
1891                  * overuse. this can help to debug single-update
1892                  * transactions
1893                  */
1894                 oh->ot_credits += 10;
1895                 if (unlikely(osd_param_is_not_sane(dev, th)))
1896                         oh->ot_credits = osd_transaction_size(dev);
1897         }
1898
1899         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1900                 GOTO(out, rc = -EIO);
1901
1902         /*
1903          * XXX temporary stuff. Some abstraction layer should
1904          * be used.
1905          */
1906         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1907         osd_th_started(oh);
1908         if (!IS_ERR(jh)) {
1909                 oh->ot_handle = jh;
1910                 LASSERT(oti->oti_txns == 0);
1911
1912                 lu_device_get(&d->dd_lu_dev);
1913                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1914                               "osd-tx", th);
1915                 oti->oti_txns++;
1916                 rc = 0;
1917         } else {
1918                 rc = PTR_ERR(jh);
1919         }
1920 out:
1921         RETURN(rc);
1922 }
1923
1924 static int osd_seq_exists(const struct lu_env *env,
1925                           struct osd_device *osd, u64 seq)
1926 {
1927         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1928         struct seq_server_site *ss = osd_seq_site(osd);
1929         int rc;
1930
1931         ENTRY;
1932
1933         LASSERT(ss != NULL);
1934         LASSERT(ss->ss_server_fld != NULL);
1935
1936         rc = osd_fld_lookup(env, osd, seq, range);
1937         if (rc != 0) {
1938                 if (rc != -ENOENT)
1939                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1940                                osd_name(osd), seq, rc);
1941                 RETURN(0);
1942         }
1943
1944         RETURN(ss->ss_node_id == range->lsr_index);
1945 }
1946
1947 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1948 {
1949         struct dt_txn_commit_cb *dcb;
1950         struct dt_txn_commit_cb *tmp;
1951
1952         /* call per-transaction stop callbacks if any */
1953         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1954                                  dcb_linkage) {
1955                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1956                          "commit callback entry: magic=%x name='%s'\n",
1957                          dcb->dcb_magic, dcb->dcb_name);
1958                 list_del_init(&dcb->dcb_linkage);
1959                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1960         }
1961 }
1962
1963 /*
1964  * Concurrency: shouldn't matter.
1965  */
1966 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1967                           struct thandle *th)
1968 {
1969         struct osd_thread_info *oti = osd_oti_get(env);
1970         struct osd_thandle *oh;
1971         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1972         struct osd_device *osd = osd_dt_dev(th->th_dev);
1973         struct qsd_instance *qsd = osd_def_qsd(osd);
1974         struct lquota_trans *qtrans;
1975         LIST_HEAD(truncates);
1976         int rc = 0, remove_agents = 0;
1977
1978         ENTRY;
1979
1980         oh = container_of0(th, struct osd_thandle, ot_super);
1981
1982         remove_agents = oh->ot_remove_agents;
1983
1984         qtrans = oh->ot_quota_trans;
1985         oh->ot_quota_trans = NULL;
1986
1987         /* move locks to local list, stop tx, execute truncates */
1988         list_splice(&oh->ot_trunc_locks, &truncates);
1989
1990         if (oh->ot_handle != NULL) {
1991                 int rc2;
1992
1993                 handle_t *hdl = oh->ot_handle;
1994
1995                 /*
1996                  * add commit callback
1997                  * notice we don't do this in osd_trans_start()
1998                  * as underlying transaction can change during truncate
1999                  */
2000                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2001                                              &oh->ot_jcb);
2002
2003                 LASSERT(oti->oti_txns == 1);
2004                 oti->oti_txns--;
2005
2006                 rc = dt_txn_hook_stop(env, th);
2007                 if (rc != 0)
2008                         CERROR("%s: failed in transaction hook: rc = %d\n",
2009                                osd_name(osd), rc);
2010
2011                 osd_trans_stop_cb(oh, rc);
2012                 /* hook functions might modify th_sync */
2013                 hdl->h_sync = th->th_sync;
2014
2015                 oh->ot_handle = NULL;
2016                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2017                 if (rc2 != 0)
2018                         CERROR("%s: failed to stop transaction: rc = %d\n",
2019                                osd_name(osd), rc2);
2020                 if (!rc)
2021                         rc = rc2;
2022
2023                 osd_process_truncates(&truncates);
2024         } else {
2025                 osd_trans_stop_cb(oh, th->th_result);
2026                 OBD_FREE_PTR(oh);
2027         }
2028
2029         osd_trunc_unlock_all(&truncates);
2030
2031         /* inform the quota slave device that the transaction is stopping */
2032         qsd_op_end(env, qsd, qtrans);
2033
2034         /*
2035          * as we want IO to journal and data IO be concurrent, we don't block
2036          * awaiting data IO completion in osd_do_bio(), instead we wait here
2037          * once transaction is submitted to the journal. all reqular requests
2038          * don't do direct IO (except read/write), thus this wait_event becomes
2039          * no-op for them.
2040          *
2041          * IMPORTANT: we have to wait till any IO submited by the thread is
2042          * completed otherwise iobuf may be corrupted by different request
2043          */
2044         wait_event(iobuf->dr_wait,
2045                        atomic_read(&iobuf->dr_numreqs) == 0);
2046         osd_fini_iobuf(osd, iobuf);
2047         if (!rc)
2048                 rc = iobuf->dr_error;
2049
2050         if (unlikely(remove_agents != 0))
2051                 osd_process_scheduled_agent_removals(env, osd);
2052
2053         oti->oti_ins_cache_depth--;
2054         /* reset OI cache for safety */
2055         if (oti->oti_ins_cache_depth == 0)
2056                 oti->oti_ins_cache_used = 0;
2057
2058         sb_end_write(osd_sb(osd));
2059
2060         RETURN(rc);
2061 }
2062
2063 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2064 {
2065         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2066                                                ot_super);
2067
2068         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2069         LASSERT(&dcb->dcb_func != NULL);
2070         if (dcb->dcb_flags & DCB_TRANS_STOP)
2071                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2072         else
2073                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2074
2075         return 0;
2076 }
2077
2078 /*
2079  * Called just before object is freed. Releases all resources except for
2080  * object itself (that is released by osd_object_free()).
2081  *
2082  * Concurrency: no concurrent access is possible that late in object
2083  * life-cycle.
2084  */
2085 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2086 {
2087         struct osd_object *obj = osd_obj(l);
2088         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2089         struct inode *inode = obj->oo_inode;
2090         __u64 projid;
2091         qid_t uid;
2092         qid_t gid;
2093
2094         LINVRNT(osd_invariant(obj));
2095
2096         /*
2097          * If object is unlinked remove fid->ino mapping from object index.
2098          */
2099
2100         osd_index_fini(obj);
2101
2102         if (!inode)
2103                 return;
2104
2105         uid = i_uid_read(inode);
2106         gid = i_gid_read(inode);
2107         projid = i_projid_read(inode);
2108
2109         obj->oo_inode = NULL;
2110         iput(inode);
2111
2112         /* do not rebalance quota if the caller needs to release memory
2113          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2114          * transaction and risk to deadlock - LU-12178 */
2115         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2116                 return;
2117
2118         if (!obj->oo_header && qsd) {
2119                 struct osd_thread_info *info = osd_oti_get(env);
2120                 struct lquota_id_info *qi = &info->oti_qi;
2121
2122                 /* Release granted quota to master if necessary */
2123                 qi->lqi_id.qid_uid = uid;
2124                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2125
2126                 qi->lqi_id.qid_uid = gid;
2127                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2128
2129                 qi->lqi_id.qid_uid = projid;
2130                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2131         }
2132 }
2133
2134 /*
2135  * Concurrency: ->loo_object_release() is called under site spin-lock.
2136  */
2137 static void osd_object_release(const struct lu_env *env,
2138                                struct lu_object *l)
2139 {
2140         struct osd_object *o = osd_obj(l);
2141
2142         /*
2143          * nobody should be releasing a non-destroyed object with nlink=0
2144          * the API allows this, but ldiskfs doesn't like and then report
2145          * this inode as deleted
2146          */
2147         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2148                   o->oo_inode->i_nlink == 0));
2149 }
2150
2151 /*
2152  * Concurrency: shouldn't matter.
2153  */
2154 static int osd_object_print(const struct lu_env *env, void *cookie,
2155                             lu_printer_t p, const struct lu_object *l)
2156 {
2157         struct osd_object *o = osd_obj(l);
2158         struct iam_descr *d;
2159
2160         if (o->oo_dir != NULL)
2161                 d = o->oo_dir->od_container.ic_descr;
2162         else
2163                 d = NULL;
2164         return (*p)(env, cookie,
2165                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2166                     o, o->oo_inode,
2167                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2168                     o->oo_inode ? o->oo_inode->i_generation : 0,
2169                     d ? d->id_ops->id_name : "plain");
2170 }
2171
2172 /*
2173  * Concurrency: shouldn't matter.
2174  */
2175 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2176                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2177 {
2178         struct osd_device *osd = osd_dt_dev(d);
2179         struct super_block *sb = osd_sb(osd);
2180         struct kstatfs *ksfs;
2181         __u64 reserved;
2182         int result = 0;
2183
2184         if (unlikely(osd->od_mnt == NULL))
2185                 return -EINPROGRESS;
2186
2187         /* osd_lproc.c call this without env, allocate ksfs for that case */
2188         if (unlikely(env == NULL)) {
2189                 OBD_ALLOC_PTR(ksfs);
2190                 if (ksfs == NULL)
2191                         return -ENOMEM;
2192         } else {
2193                 ksfs = &osd_oti_get(env)->oti_ksfs;
2194         }
2195
2196         result = sb->s_op->statfs(sb->s_root, ksfs);
2197         if (result)
2198                 goto out;
2199
2200         statfs_pack(sfs, ksfs);
2201         if (unlikely(sb->s_flags & SB_RDONLY))
2202                 sfs->os_state |= OS_STATE_READONLY;
2203
2204         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2205
2206         if (ldiskfs_has_feature_extents(sb))
2207                 sfs->os_maxbytes = sb->s_maxbytes;
2208         else
2209                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2210
2211         /*
2212          * Reserve some space so to avoid fragmenting the filesystem too much.
2213          * Fragmentation not only impacts performance, but can also increase
2214          * metadata overhead significantly, causing grant calculation to be
2215          * wrong.
2216          *
2217          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2218          */
2219         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2220         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2221         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2222                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2223
2224         sfs->os_blocks -= reserved;
2225         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2226         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2227
2228 out:
2229         if (unlikely(env == NULL))
2230                 OBD_FREE_PTR(ksfs);
2231         return result;
2232 }
2233
2234 /**
2235  * Estimate space needed for file creations. We assume the largest filename
2236  * which is 2^64 - 1, hence a filename of 20 chars.
2237  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2238  */
2239 #ifdef __LDISKFS_DIR_REC_LEN
2240 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2241 #else
2242 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2243 #endif
2244
2245 /*
2246  * Concurrency: doesn't access mutable data.
2247  */
2248 static void osd_conf_get(const struct lu_env *env,
2249                          const struct dt_device *dev,
2250                          struct dt_device_param *param)
2251 {
2252         struct osd_device *d = osd_dt_dev(dev);
2253         struct super_block *sb = osd_sb(d);
2254         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2255         const char *name;
2256         int ea_overhead;
2257
2258         /*
2259          * XXX should be taken from not-yet-existing fs abstraction layer.
2260          */
2261         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2262         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2263         param->ddp_symlink_max  = sb->s_blocksize;
2264         param->ddp_mount_type   = LDD_MT_LDISKFS;
2265         if (ldiskfs_has_feature_extents(sb))
2266                 param->ddp_maxbytes = sb->s_maxbytes;
2267         else
2268                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2269         /*
2270          * inode are statically allocated, so per-inode space consumption
2271          * is the space consumed by the directory entry
2272          */
2273         param->ddp_inodespace     = PER_OBJ_USAGE;
2274         /*
2275          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2276          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2277          * maximum length to not under-count the actual number of extents
2278          * needed for writing a file if there are sub-optimal block allocations.
2279          */
2280         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2281         /* worst-case extent insertion metadata overhead */
2282         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2283         param->ddp_mntopts = 0;
2284         if (test_opt(sb, XATTR_USER))
2285                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2286         if (test_opt(sb, POSIX_ACL))
2287                 param->ddp_mntopts |= MNTOPT_ACL;
2288
2289         /*
2290          * LOD might calculate the max stripe count based on max_ea_size,
2291          * so we need take account in the overhead as well,
2292          * xattr_header + magic + xattr_entry_head
2293          */
2294         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2295                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2296
2297 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2298         if (ldiskfs_has_feature_ea_inode(sb))
2299                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2300                                                                 ea_overhead;
2301         else
2302 #endif
2303                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2304
2305         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2306                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2307
2308         /*
2309          * Preferred RPC size for efficient disk IO.  4MB shows good
2310          * all-around performance for ldiskfs, but use bigalloc chunk size
2311          * by default if larger.
2312          */
2313 #if defined(LDISKFS_CLUSTER_SIZE)
2314         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2315                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2316         else
2317 #endif
2318                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2319
2320         param->ddp_t10_cksum_type = 0;
2321         if (bi) {
2322                 unsigned short interval = blk_integrity_interval(bi);
2323                 name = blk_integrity_name(bi);
2324                 /*
2325                  * Expected values:
2326                  * T10-DIF-TYPE1-CRC
2327                  * T10-DIF-TYPE3-CRC
2328                  * T10-DIF-TYPE1-IP
2329                  * T10-DIF-TYPE3-IP
2330                  */
2331                 if (strncmp(name, "T10-DIF-TYPE",
2332                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2333                         /* also skip "1/3-" at end */
2334                         const int type_off = sizeof("T10-DIF-TYPE.");
2335                         char type_number = name[type_off - 2];
2336
2337                         if (interval != 512 && interval != 4096) {
2338                                 CERROR("%s: unsupported T10PI sector size %u\n",
2339                                        d->od_svname, interval);
2340                         } else if (type_number != '1' && type_number != '3') {
2341                                 CERROR("%s: unsupported T10PI type %s\n",
2342                                        d->od_svname, name);
2343                         } else if (strcmp(name + type_off, "CRC") == 0) {
2344                                 d->od_t10_type = type_number == '1' ?
2345                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2346                                 param->ddp_t10_cksum_type = interval == 512 ?
2347                                         OBD_CKSUM_T10CRC512 :
2348                                         OBD_CKSUM_T10CRC4K;
2349                         } else if (strcmp(name + type_off, "IP") == 0) {
2350                                 d->od_t10_type = type_number == '1' ?
2351                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2352                                 param->ddp_t10_cksum_type = interval == 512 ?
2353                                         OBD_CKSUM_T10IP512 :
2354                                         OBD_CKSUM_T10IP4K;
2355                         } else {
2356                                 CERROR("%s: unsupported checksum type of "
2357                                        "T10PI type '%s'",
2358                                        d->od_svname, name);
2359                         }
2360
2361                 } else {
2362                         CERROR("%s: unsupported T10PI type '%s'",
2363                                d->od_svname, name);
2364                 }
2365         }
2366 }
2367
2368 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2369 {
2370         return osd_sb(osd_dt_dev(d));
2371 }
2372
2373 /*
2374  * Concurrency: shouldn't matter.
2375  */
2376 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2377 {
2378         int rc;
2379         struct super_block *s = osd_sb(osd_dt_dev(d));
2380         ENTRY;
2381
2382         down_read(&s->s_umount);
2383         rc = s->s_op->sync_fs(s, 1);
2384         up_read(&s->s_umount);
2385
2386         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2387                rc);
2388
2389         return rc;
2390 }
2391
2392 /**
2393  * Start commit for OSD device.
2394  *
2395  * An implementation of dt_commit_async method for OSD device.
2396  * Asychronously starts underlayng fs sync and thereby a transaction
2397  * commit.
2398  *
2399  * \param env environment
2400  * \param d dt device
2401  *
2402  * \see dt_device_operations
2403  */
2404 static int osd_commit_async(const struct lu_env *env,
2405                             struct dt_device *d)
2406 {
2407         struct super_block *s = osd_sb(osd_dt_dev(d));
2408         int rc;
2409
2410         ENTRY;
2411
2412         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2413         down_read(&s->s_umount);
2414         rc = s->s_op->sync_fs(s, 0);
2415         up_read(&s->s_umount);
2416
2417         RETURN(rc);
2418 }
2419
2420 /* Our own copy of the set readonly functions if present, or NU if not. */
2421 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2422 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2423 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2424 static int (*priv_security_file_alloc)(struct file *file);
2425
2426 int osd_security_file_alloc(struct file *file)
2427 {
2428         if (priv_security_file_alloc)
2429                 return priv_security_file_alloc(file);
2430         return 0;
2431 }
2432
2433 /*
2434  * Concurrency: shouldn't matter.
2435  */
2436 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2437 {
2438         struct super_block *sb = osd_sb(osd_dt_dev(d));
2439         struct block_device *dev = sb->s_bdev;
2440         int rc = -EOPNOTSUPP;
2441
2442         ENTRY;
2443
2444         if (priv_dev_set_rdonly) {
2445                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2446
2447                 rc = 0;
2448                 CERROR("*** setting %s read-only ***\n",
2449                        osd_dt_dev(d)->od_svname);
2450
2451                 if (sb->s_op->freeze_fs) {
2452                         rc = sb->s_op->freeze_fs(sb);
2453                         if (rc)
2454                                 goto out;
2455                 }
2456
2457                 if (jdev && (jdev != dev)) {
2458                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2459                                (long)jdev);
2460                         priv_dev_set_rdonly(jdev);
2461                 }
2462                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2463                 priv_dev_set_rdonly(dev);
2464
2465                 if (sb->s_op->unfreeze_fs)
2466                         sb->s_op->unfreeze_fs(sb);
2467         }
2468
2469 out:
2470         if (rc)
2471                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2472                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2473
2474         RETURN(rc);
2475 }
2476
2477 /**
2478  * Note: we do not count into QUOTA here.
2479  * If we mount with --data_journal we may need more.
2480  */
2481 const int osd_dto_credits_noquota[DTO_NR] = {
2482         /**
2483          * Insert.
2484          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2485          * SINGLEDATA_TRANS_BLOCKS(8)
2486          * XXX Note: maybe iam need more, since iam have more level than
2487          *           EXT3 htree.
2488          */
2489         [DTO_INDEX_INSERT]  = 16,
2490         /**
2491          * Delete
2492          * just modify a single entry, probably merge few within a block
2493          */
2494         [DTO_INDEX_DELETE]  = 1,
2495         /**
2496          * Used for OI scrub
2497          */
2498         [DTO_INDEX_UPDATE]  = 16,
2499         /**
2500          * 4(inode, inode bits, groups, GDT)
2501          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2502          */
2503         [DTO_OBJECT_CREATE] = 4,
2504         /**
2505          * 4(inode, inode bits, groups, GDT)
2506          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2507          */
2508         [DTO_OBJECT_DELETE] = 4,
2509         /**
2510          * Attr set credits (inode)
2511          */
2512         [DTO_ATTR_SET_BASE] = 1,
2513         /**
2514          * Xattr set. The same as xattr of EXT3.
2515          * DATA_TRANS_BLOCKS(14)
2516          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2517          * are also counted in. Do not know why?
2518          */
2519         [DTO_XATTR_SET]     = 14,
2520         /**
2521          * credits for inode change during write.
2522          */
2523         [DTO_WRITE_BASE]    = 3,
2524         /**
2525          * credits for single block write.
2526          */
2527         [DTO_WRITE_BLOCK]   = 14,
2528         /**
2529          * Attr set credits for chown.
2530          * This is extra credits for setattr, and it is null without quota
2531          */
2532         [DTO_ATTR_SET_CHOWN] = 0
2533 };
2534
2535 static const struct dt_device_operations osd_dt_ops = {
2536         .dt_root_get       = osd_root_get,
2537         .dt_statfs         = osd_statfs,
2538         .dt_trans_create   = osd_trans_create,
2539         .dt_trans_start    = osd_trans_start,
2540         .dt_trans_stop     = osd_trans_stop,
2541         .dt_trans_cb_add   = osd_trans_cb_add,
2542         .dt_conf_get       = osd_conf_get,
2543         .dt_mnt_sb_get     = osd_mnt_sb_get,
2544         .dt_sync           = osd_sync,
2545         .dt_ro             = osd_ro,
2546         .dt_commit_async   = osd_commit_async,
2547 };
2548
2549 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2550                           unsigned int role)
2551 {
2552         struct osd_object *obj = osd_dt_obj(dt);
2553         struct osd_thread_info *oti = osd_oti_get(env);
2554
2555         LINVRNT(osd_invariant(obj));
2556
2557         LASSERT(obj->oo_owner != env);
2558         down_read_nested(&obj->oo_sem, role);
2559
2560         LASSERT(obj->oo_owner == NULL);
2561         oti->oti_r_locks++;
2562 }
2563
2564 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2565                            unsigned int role)
2566 {
2567         struct osd_object *obj = osd_dt_obj(dt);
2568         struct osd_thread_info *oti = osd_oti_get(env);
2569
2570         LINVRNT(osd_invariant(obj));
2571
2572         LASSERT(obj->oo_owner != env);
2573         down_write_nested(&obj->oo_sem, role);
2574
2575         LASSERT(obj->oo_owner == NULL);
2576         obj->oo_owner = env;
2577         oti->oti_w_locks++;
2578 }
2579
2580 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2581 {
2582         struct osd_object *obj = osd_dt_obj(dt);
2583         struct osd_thread_info *oti = osd_oti_get(env);
2584
2585         LINVRNT(osd_invariant(obj));
2586
2587         LASSERT(oti->oti_r_locks > 0);
2588         oti->oti_r_locks--;
2589         up_read(&obj->oo_sem);
2590 }
2591
2592 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2593 {
2594         struct osd_object *obj = osd_dt_obj(dt);
2595         struct osd_thread_info *oti = osd_oti_get(env);
2596
2597         LINVRNT(osd_invariant(obj));
2598
2599         LASSERT(obj->oo_owner == env);
2600         LASSERT(oti->oti_w_locks > 0);
2601         oti->oti_w_locks--;
2602         obj->oo_owner = NULL;
2603         up_write(&obj->oo_sem);
2604 }
2605
2606 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2607 {
2608         struct osd_object *obj = osd_dt_obj(dt);
2609
2610         LINVRNT(osd_invariant(obj));
2611
2612         return obj->oo_owner == env;
2613 }
2614
2615 static void osd_inode_getattr(const struct lu_env *env,
2616                               struct inode *inode, struct lu_attr *attr)
2617 {
2618         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2619                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2620                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2621                            LA_BLKSIZE | LA_TYPE;
2622
2623         attr->la_atime = inode->i_atime.tv_sec;
2624         attr->la_mtime = inode->i_mtime.tv_sec;
2625         attr->la_ctime = inode->i_ctime.tv_sec;
2626         attr->la_mode    = inode->i_mode;
2627         attr->la_size    = i_size_read(inode);
2628         attr->la_blocks  = inode->i_blocks;
2629         attr->la_uid     = i_uid_read(inode);
2630         attr->la_gid     = i_gid_read(inode);
2631         attr->la_projid  = i_projid_read(inode);
2632         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2633         attr->la_nlink   = inode->i_nlink;
2634         attr->la_rdev    = inode->i_rdev;
2635         attr->la_blksize = 1 << inode->i_blkbits;
2636         attr->la_blkbits = inode->i_blkbits;
2637         /*
2638          * Ext4 did not transfer inherit flags from raw inode
2639          * to inode flags, and ext4 internally test raw inode
2640          * @i_flags directly. Instead of patching ext4, we do it here.
2641          */
2642         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2643                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2644 }
2645
2646 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2647                         struct lu_attr *attr)
2648 {
2649         struct osd_object *obj = osd_dt_obj(dt);
2650
2651         if (unlikely(!dt_object_exists(dt)))
2652                 return -ENOENT;
2653         if (unlikely(obj->oo_destroyed))
2654                 return -ENOENT;
2655
2656         LASSERT(!dt_object_remote(dt));
2657         LINVRNT(osd_invariant(obj));
2658
2659         spin_lock(&obj->oo_guard);
2660         osd_inode_getattr(env, obj->oo_inode, attr);
2661         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2662                 attr->la_valid |= LA_FLAGS;
2663                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2664         }
2665         spin_unlock(&obj->oo_guard);
2666
2667         return 0;
2668 }
2669
2670 static int osd_declare_attr_qid(const struct lu_env *env,
2671                                 struct osd_object *obj,
2672                                 struct osd_thandle *oh, long long bspace,
2673                                 qid_t old_id, qid_t new_id, bool enforce,
2674                                 unsigned int type, bool ignore_edquot)
2675 {
2676         int rc;
2677         struct osd_thread_info *info = osd_oti_get(env);
2678         struct lquota_id_info  *qi = &info->oti_qi;
2679
2680         qi->lqi_type = type;
2681         /* inode accounting */
2682         qi->lqi_is_blk = false;
2683
2684         /* one more inode for the new id ... */
2685         qi->lqi_id.qid_uid = new_id;
2686         qi->lqi_space      = 1;
2687         /* Reserve credits for the new id */
2688         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2689         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2690                 rc = 0;
2691         if (rc)
2692                 RETURN(rc);
2693
2694         /* and one less inode for the current id */
2695         qi->lqi_id.qid_uid = old_id;
2696         qi->lqi_space = -1;
2697         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2698         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2699                 rc = 0;
2700         if (rc)
2701                 RETURN(rc);
2702
2703         /* block accounting */
2704         qi->lqi_is_blk = true;
2705
2706         /* more blocks for the new id ... */
2707         qi->lqi_id.qid_uid = new_id;
2708         qi->lqi_space      = bspace;
2709         /*
2710          * Credits for the new uid has been reserved, re-use "obj"
2711          * to save credit reservation.
2712          */
2713         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2714         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2715                 rc = 0;
2716         if (rc)
2717                 RETURN(rc);
2718
2719         /* and finally less blocks for the current uid */
2720         qi->lqi_id.qid_uid = old_id;
2721         qi->lqi_space      = -bspace;
2722         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2723         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2724                 rc = 0;
2725
2726         RETURN(rc);
2727 }
2728
2729 static int osd_declare_attr_set(const struct lu_env *env,
2730                                 struct dt_object *dt,
2731                                 const struct lu_attr *attr,
2732                                 struct thandle *handle)
2733 {
2734         struct osd_thandle *oh;
2735         struct osd_object *obj;
2736         qid_t uid;
2737         qid_t gid;
2738         long long bspace;
2739         int rc = 0;
2740         bool enforce;
2741
2742         ENTRY;
2743
2744         LASSERT(dt != NULL);
2745         LASSERT(handle != NULL);
2746
2747         obj = osd_dt_obj(dt);
2748         LASSERT(osd_invariant(obj));
2749
2750         oh = container_of0(handle, struct osd_thandle, ot_super);
2751         LASSERT(oh->ot_handle == NULL);
2752
2753         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2754                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2755
2756         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2757                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2758
2759         if (attr == NULL || obj->oo_inode == NULL)
2760                 RETURN(rc);
2761
2762         bspace   = obj->oo_inode->i_blocks << 9;
2763         bspace   = toqb(bspace);
2764
2765         /*
2766          * Changing ownership is always preformed by super user, it should not
2767          * fail with EDQUOT unless required explicitly.
2768          *
2769          * We still need to call the osd_declare_qid() to calculate the journal
2770          * credits for updating quota accounting files and to trigger quota
2771          * space adjustment once the operation is completed.
2772          */
2773         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2774                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2775
2776                 if (!ignore_edquot)
2777                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2778                                "(the quota space is %lld)\n",
2779                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2780                                attr->la_gid, bspace);
2781
2782                 /* USERQUOTA */
2783                 uid = i_uid_read(obj->oo_inode);
2784                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2785                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2786                                           attr->la_uid, enforce, USRQUOTA,
2787                                           true);
2788                 if (rc)
2789                         RETURN(rc);
2790
2791                 gid = i_gid_read(obj->oo_inode);
2792                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2793                        attr->la_uid, gid, attr->la_gid);
2794                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2795                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2796                                           attr->la_gid, enforce, GRPQUOTA,
2797                                           ignore_edquot);
2798                 if (rc)
2799                         RETURN(rc);
2800
2801         }
2802 #ifdef HAVE_PROJECT_QUOTA
2803         if (attr->la_valid & LA_PROJID) {
2804                 __u32 projid = i_projid_read(obj->oo_inode);
2805
2806                 enforce = (attr->la_valid & LA_PROJID) &&
2807                                         (attr->la_projid != projid);
2808                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2809                                           (qid_t)projid, (qid_t)attr->la_projid,
2810                                           enforce, PRJQUOTA, true);
2811                 if (rc)
2812                         RETURN(rc);
2813         }
2814 #endif
2815         RETURN(rc);
2816 }
2817
2818 static int osd_inode_setattr(const struct lu_env *env,
2819                              struct inode *inode, const struct lu_attr *attr)
2820 {
2821         __u64 bits = attr->la_valid;
2822
2823         /* Only allow set size for regular file */
2824         if (!S_ISREG(inode->i_mode))
2825                 bits &= ~(LA_SIZE | LA_BLOCKS);
2826
2827         if (bits == 0)
2828                 return 0;
2829
2830         if (bits & LA_ATIME)
2831                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2832         if (bits & LA_CTIME)
2833                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2834         if (bits & LA_MTIME)
2835                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2836         if (bits & LA_SIZE) {
2837                 spin_lock(&inode->i_lock);
2838                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2839                 i_size_write(inode, attr->la_size);
2840                 spin_unlock(&inode->i_lock);
2841         }
2842
2843         /*
2844          * OSD should not change "i_blocks" which is used by quota.
2845          * "i_blocks" should be changed by ldiskfs only.
2846          */
2847         if (bits & LA_MODE)
2848                 inode->i_mode = (inode->i_mode & S_IFMT) |
2849                                 (attr->la_mode & ~S_IFMT);
2850         if (bits & LA_UID)
2851                 i_uid_write(inode, attr->la_uid);
2852         if (bits & LA_GID)
2853                 i_gid_write(inode, attr->la_gid);
2854         if (bits & LA_PROJID)
2855                 i_projid_write(inode, attr->la_projid);
2856         if (bits & LA_NLINK)
2857                 set_nlink(inode, attr->la_nlink);
2858         if (bits & LA_RDEV)
2859                 inode->i_rdev = attr->la_rdev;
2860
2861         if (bits & LA_FLAGS) {
2862                 /* always keep S_NOCMTIME */
2863                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2864                                  S_NOCMTIME;
2865                 /*
2866                  * Ext4 did not transfer inherit flags from
2867                  * @inode->i_flags to raw inode i_flags when writing
2868                  * flags, we do it explictly here.
2869                  */
2870                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2871                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2872                 else
2873                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2874         }
2875         return 0;
2876 }
2877
2878 #ifdef HAVE_PROJECT_QUOTA
2879 static int osd_transfer_project(struct inode *inode, __u32 projid)
2880 {
2881         struct super_block *sb = inode->i_sb;
2882         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2883         int err;
2884         kprojid_t kprojid;
2885         struct ldiskfs_iloc iloc;
2886         struct ldiskfs_inode *raw_inode;
2887         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2888
2889         if (!ldiskfs_has_feature_project(sb)) {
2890                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2891                         == LDISKFS_DEF_PROJID);
2892                 if (projid != LDISKFS_DEF_PROJID)
2893                         return -EOPNOTSUPP;
2894                 else
2895                         return 0;
2896         }
2897
2898         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2899                 return -EOPNOTSUPP;
2900
2901         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2902         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2903                 return 0;
2904
2905         err = ldiskfs_get_inode_loc(inode, &iloc);
2906         if (err)
2907                 return err;
2908
2909         raw_inode = ldiskfs_raw_inode(&iloc);
2910         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2911                 err = -EOVERFLOW;
2912                 brelse(iloc.bh);
2913                 return err;
2914         }
2915         brelse(iloc.bh);
2916
2917         dquot_initialize(inode);
2918         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2919         if (transfer_to[PRJQUOTA]) {
2920                 err = __dquot_transfer(inode, transfer_to);
2921                 dqput(transfer_to[PRJQUOTA]);
2922                 if (err)
2923                         return err;
2924         }
2925
2926         return err;
2927 }
2928 #endif
2929
2930 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2931 {
2932         int rc;
2933
2934         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2935             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2936                 struct iattr iattr;
2937
2938                 CDEBUG(D_QUOTA,
2939                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2940                        inode->i_ino, i_uid_read(inode), attr->la_uid,
2941                        i_gid_read(inode), attr->la_gid);
2942
2943                 dquot_initialize(inode);
2944                 iattr.ia_valid = 0;
2945                 if (attr->la_valid & LA_UID)
2946                         iattr.ia_valid |= ATTR_UID;
2947                 if (attr->la_valid & LA_GID)
2948                         iattr.ia_valid |= ATTR_GID;
2949                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2950                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2951
2952                 rc = dquot_transfer(inode, &iattr);
2953                 if (rc) {
2954                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2955                                "enforcement enabled on the ldiskfs "
2956                                "filesystem?\n", inode->i_sb->s_id, rc);
2957                         return rc;
2958                 }
2959         }
2960
2961         /* Handle project id transfer here properly */
2962         if (attr->la_valid & LA_PROJID &&
2963             attr->la_projid != i_projid_read(inode)) {
2964 #ifdef HAVE_PROJECT_QUOTA
2965                 rc = osd_transfer_project(inode, attr->la_projid);
2966 #else
2967                 rc = -ENOTSUPP;
2968 #endif
2969                 if (rc) {
2970                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2971                                "enforcement enabled on the ldiskfs "
2972                                "filesystem?\n", inode->i_sb->s_id, rc);
2973                         return rc;
2974                 }
2975         }
2976         return 0;
2977 }
2978
2979 static int osd_attr_set(const struct lu_env *env,
2980                         struct dt_object *dt,
2981                         const struct lu_attr *attr,
2982                         struct thandle *handle)
2983 {
2984         struct osd_object *obj = osd_dt_obj(dt);
2985         struct inode *inode;
2986         int rc;
2987
2988         if (!dt_object_exists(dt))
2989                 return -ENOENT;
2990
2991         LASSERT(handle != NULL);
2992         LASSERT(!dt_object_remote(dt));
2993         LASSERT(osd_invariant(obj));
2994
2995         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2996
2997         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2998             !osd_obj2dev(obj)->od_is_ost) {
2999                 struct osd_thread_info *oti = osd_oti_get(env);
3000                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3001                 struct lu_fid *fid1 = &oti->oti_fid;
3002                 struct osd_inode_id *id = &oti->oti_id;
3003                 struct iam_path_descr *ipd;
3004                 struct iam_container *bag;
3005                 struct osd_thandle *oh;
3006                 int rc;
3007
3008                 fid_cpu_to_be(fid1, fid0);
3009                 memset(id, 1, sizeof(*id));
3010                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3011                                   fid0)->oi_dir.od_container;
3012                 ipd = osd_idx_ipd_get(env, bag);
3013                 if (unlikely(ipd == NULL))
3014                         RETURN(-ENOMEM);
3015
3016                 oh = container_of0(handle, struct osd_thandle, ot_super);
3017                 rc = iam_update(oh->ot_handle, bag,
3018                                 (const struct iam_key *)fid1,
3019                                 (const struct iam_rec *)id, ipd);
3020                 osd_ipd_put(env, bag, ipd);
3021                 return(rc > 0 ? 0 : rc);
3022         }
3023
3024         inode = obj->oo_inode;
3025
3026         rc = osd_quota_transfer(inode, attr);
3027         if (rc)
3028                 return rc;
3029
3030         spin_lock(&obj->oo_guard);
3031         rc = osd_inode_setattr(env, inode, attr);
3032         spin_unlock(&obj->oo_guard);
3033         if (rc != 0)
3034                 GOTO(out, rc);
3035
3036         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3037
3038         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3039
3040         if (!(attr->la_valid & LA_FLAGS))
3041                 GOTO(out, rc);
3042
3043         /* Let's check if there are extra flags need to be set into LMA */
3044         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3045                 struct osd_thread_info *info = osd_oti_get(env);
3046                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3047
3048                 LASSERT(!obj->oo_pfid_in_lma);
3049
3050                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3051                                  &info->oti_ost_attrs);
3052                 if (rc)
3053                         GOTO(out, rc);
3054
3055                 lma->lma_incompat |=
3056                         lustre_to_lma_flags(attr->la_flags);
3057                 lustre_lma_swab(lma);
3058
3059                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3060
3061                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3062                                      lma, sizeof(*lma), XATTR_REPLACE);
3063                 if (rc != 0) {
3064                         struct osd_device *osd = osd_obj2dev(obj);
3065
3066                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3067                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3068                               lma->lma_incompat, rc);
3069                 } else {
3070                         obj->oo_lma_flags =
3071                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3072                 }
3073                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3074         }
3075 out:
3076
3077         return rc;
3078 }
3079
3080 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3081                                            struct osd_object *obj,
3082                                            const char *name, const int namelen)
3083 {
3084         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3085 }
3086
3087 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3088                       umode_t mode, struct dt_allocation_hint *hint,
3089                       struct thandle *th, struct lu_attr *attr)
3090 {
3091         int result;
3092         struct osd_device *osd = osd_obj2dev(obj);
3093         struct osd_thandle *oth;
3094         struct dt_object *parent = NULL;
3095         struct inode *inode;
3096         uid_t owner[2] = {0, 0};
3097
3098         if (attr->la_valid & LA_UID)
3099                 owner[0] = attr->la_uid;
3100         if (attr->la_valid & LA_GID)
3101                 owner[1] = attr->la_gid;
3102
3103         LINVRNT(osd_invariant(obj));
3104         LASSERT(obj->oo_inode == NULL);
3105         LASSERT(obj->oo_hl_head == NULL);
3106
3107         if (S_ISDIR(mode) && ldiskfs_pdo) {
3108                 obj->oo_hl_head =
3109                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3110                 if (obj->oo_hl_head == NULL)
3111                         return -ENOMEM;
3112         }
3113
3114         oth = container_of(th, struct osd_thandle, ot_super);
3115         LASSERT(oth->ot_handle->h_transaction != NULL);
3116
3117         if (hint != NULL && hint->dah_parent != NULL &&
3118             !dt_object_remote(hint->dah_parent))
3119                 parent = hint->dah_parent;
3120
3121         inode = ldiskfs_create_inode(oth->ot_handle,
3122                                      parent ? osd_dt_obj(parent)->oo_inode :
3123                                               osd_sb(osd)->s_root->d_inode,
3124                                      mode, owner);
3125         if (!IS_ERR(inode)) {
3126                 /* Do not update file c/mtime in ldiskfs. */
3127                 inode->i_flags |= S_NOCMTIME;
3128
3129                 /*
3130                  * For new created object, it must be consistent,
3131                  * and it is unnecessary to scrub against it.
3132                  */
3133                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3134
3135                 obj->oo_inode = inode;
3136                 result = 0;
3137         } else {
3138                 if (obj->oo_hl_head != NULL) {
3139                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3140                         obj->oo_hl_head = NULL;
3141                 }
3142                 result = PTR_ERR(inode);
3143         }
3144         LINVRNT(osd_invariant(obj));
3145         return result;
3146 }
3147
3148 enum {
3149         OSD_NAME_LEN = 255
3150 };
3151
3152 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3153                      struct lu_attr *attr,
3154                      struct dt_allocation_hint *hint,
3155                      struct dt_object_format *dof,
3156                      struct thandle *th)
3157 {
3158         int result;
3159         struct osd_thandle *oth;
3160         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3161
3162         LASSERT(S_ISDIR(attr->la_mode));
3163
3164         oth = container_of(th, struct osd_thandle, ot_super);
3165         LASSERT(oth->ot_handle->h_transaction != NULL);
3166         result = osd_mkfile(info, obj, mode, hint, th, attr);
3167
3168         return result;
3169 }
3170
3171 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3172                         struct lu_attr *attr,
3173                         struct dt_allocation_hint *hint,
3174                         struct dt_object_format *dof,
3175                         struct thandle *th)
3176 {
3177         int result;
3178         struct osd_thandle *oth;
3179         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3180
3181         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3182
3183         LASSERT(S_ISREG(attr->la_mode));
3184
3185         oth = container_of(th, struct osd_thandle, ot_super);
3186         LASSERT(oth->ot_handle->h_transaction != NULL);
3187
3188         result = osd_mkfile(info, obj, mode, hint, th, attr);
3189         if (result == 0) {
3190                 LASSERT(obj->oo_inode != NULL);
3191                 if (feat->dif_flags & DT_IND_VARKEY)
3192                         result = iam_lvar_create(obj->oo_inode,
3193                                                  feat->dif_keysize_max,
3194                                                  feat->dif_ptrsize,
3195                                                  feat->dif_recsize_max,
3196                                                  oth->ot_handle);
3197                 else
3198                         result = iam_lfix_create(obj->oo_inode,
3199                                                  feat->dif_keysize_max,
3200                                                  feat->dif_ptrsize,
3201                                                  feat->dif_recsize_max,
3202                                                  oth->ot_handle);
3203         }
3204         return result;
3205 }
3206
3207 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3208                      struct lu_attr *attr,
3209                      struct dt_allocation_hint *hint,
3210                      struct dt_object_format *dof,
3211                      struct thandle *th)
3212 {
3213         LASSERT(S_ISREG(attr->la_mode));
3214         return osd_mkfile(info, obj, (attr->la_mode &
3215                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3216                           attr);
3217 }
3218
3219 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3220                      struct lu_attr *attr,
3221                      struct dt_allocation_hint *hint,
3222                      struct dt_object_format *dof,
3223                      struct thandle *th)
3224 {
3225         LASSERT(S_ISLNK(attr->la_mode));
3226         return osd_mkfile(info, obj, (attr->la_mode &
3227                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3228                           attr);
3229 }
3230
3231 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3232                      struct lu_attr *attr,
3233                      struct dt_allocation_hint *hint,
3234                      struct dt_object_format *dof,
3235                      struct thandle *th)
3236 {
3237         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3238         int result;
3239
3240         LINVRNT(osd_invariant(obj));
3241         LASSERT(obj->oo_inode == NULL);
3242         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3243                 S_ISFIFO(mode) || S_ISSOCK(mode));
3244
3245         result = osd_mkfile(info, obj, mode, hint, th, attr);
3246         if (result == 0) {
3247                 LASSERT(obj->oo_inode != NULL);
3248                 /*
3249                  * This inode should be marked dirty for i_rdev.  Currently
3250                  * that is done in the osd_attr_init().
3251                  */
3252                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3253                                    attr->la_rdev);
3254         }
3255         LINVRNT(osd_invariant(obj));
3256         return result;
3257 }
3258
3259 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3260                               struct lu_attr *,
3261                               struct dt_allocation_hint *hint,
3262                               struct dt_object_format *dof,
3263                               struct thandle *);
3264
3265 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3266 {
3267         osd_obj_type_f result;
3268
3269         switch (type) {
3270         case DFT_DIR:
3271                 result = osd_mkdir;
3272                 break;
3273         case DFT_REGULAR:
3274                 result = osd_mkreg;
3275                 break;
3276         case DFT_SYM:
3277                 result = osd_mksym;
3278                 break;
3279         case DFT_NODE:
3280                 result = osd_mknod;
3281                 break;
3282         case DFT_INDEX:
3283                 result = osd_mk_index;
3284                 break;
3285
3286         default:
3287                 LBUG();
3288                 break;
3289         }
3290         return result;
3291 }
3292
3293
3294 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3295                         struct dt_object *parent, struct dt_object *child,
3296                         umode_t child_mode)
3297 {
3298         LASSERT(ah);
3299
3300         ah->dah_parent = parent;
3301         ah->dah_mode = child_mode;
3302
3303         if (parent != NULL && !dt_object_remote(parent)) {
3304                 /* will help to find FID->ino at dt_insert("..") */
3305                 struct osd_object *pobj = osd_dt_obj(parent);
3306
3307                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3308         }
3309 }
3310
3311 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3312                           struct lu_attr *attr, struct dt_object_format *dof)
3313 {
3314         struct inode *inode = obj->oo_inode;
3315         __u64 valid = attr->la_valid;
3316         int result;
3317
3318         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3319
3320         if (dof->dof_type != DFT_NODE)
3321                 attr->la_valid &= ~LA_RDEV;
3322         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3323                 attr->la_valid &= ~LA_ATIME;
3324         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3325                 attr->la_valid &= ~LA_CTIME;
3326         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3327                 attr->la_valid &= ~LA_MTIME;
3328
3329         result = osd_quota_transfer(inode, attr);
3330         if (result)
3331                 return;
3332
3333         if (attr->la_valid != 0) {
3334                 result = osd_inode_setattr(info->oti_env, inode, attr);
3335                 /*
3336                  * The osd_inode_setattr() should always succeed here.  The
3337                  * only error that could be returned is EDQUOT when we are
3338                  * trying to change the UID or GID of the inode. However, this
3339                  * should not happen since quota enforcement is no longer
3340                  * enabled on ldiskfs (lquota takes care of it).
3341                  */
3342                 LASSERTF(result == 0, "%d\n", result);
3343                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3344         }
3345
3346         attr->la_valid = valid;
3347 }
3348
3349 /**
3350  * Helper function for osd_create()
3351  *
3352  * \retval 0, on success
3353  */
3354 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3355                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3356                         struct dt_object_format *dof, struct thandle *th)
3357 {
3358         int result;
3359         __u32 umask;
3360
3361         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3362
3363         /* we drop umask so that permissions we pass are not affected */
3364         umask = current->fs->umask;
3365         current->fs->umask = 0;
3366
3367         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3368                                                   th);
3369         if (likely(obj->oo_inode != NULL)) {
3370                 LASSERT(obj->oo_inode->i_state & I_NEW);
3371
3372                 /*
3373                  * Unlock the inode before attr initialization to avoid
3374                  * unnecessary dqget operations. LU-6378
3375                  */
3376                 unlock_new_inode(obj->oo_inode);
3377         }
3378
3379         if (likely(result == 0)) {
3380                 osd_attr_init(info, obj, attr, dof);
3381                 osd_object_init0(obj);
3382         }
3383
3384         /* restore previous umask value */
3385         current->fs->umask = umask;
3386
3387         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3388
3389         return result;
3390 }
3391
3392 /**
3393  * Helper function for osd_create()
3394  *
3395  * \retval 0, on success
3396  */
3397 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3398                            const struct lu_fid *fid, struct thandle *th)
3399 {
3400         struct osd_thread_info *info = osd_oti_get(env);
3401         struct osd_inode_id    *id   = &info->oti_id;
3402         struct osd_device      *osd  = osd_obj2dev(obj);
3403         struct osd_thandle     *oh;
3404         int rc;
3405
3406         LASSERT(obj->oo_inode != NULL);
3407
3408         oh = container_of0(th, struct osd_thandle, ot_super);
3409         LASSERT(oh->ot_handle);
3410         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3411
3412         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3413         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3414                            OI_CHECK_FLD, NULL);
3415         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3416                 struct lu_fid next_fid = *fid;
3417
3418                 /* insert next object in advance, and map to the same inode */
3419                 next_fid.f_oid++;
3420                 if (next_fid.f_oid != 0) {
3421                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3422                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3423                                       OI_CHECK_FLD, NULL);
3424                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3425                 }
3426         }
3427
3428         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3429
3430         return rc;
3431 }
3432
3433 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3434                    u64 seq, struct lu_seq_range *range)
3435 {
3436         struct seq_server_site *ss = osd_seq_site(osd);
3437
3438         if (fid_seq_is_idif(seq)) {
3439                 fld_range_set_ost(range);
3440                 range->lsr_index = idif_ost_idx(seq);
3441                 return 0;
3442         }
3443
3444         if (!fid_seq_in_fldb(seq)) {
3445                 fld_range_set_mdt(range);
3446                 if (ss != NULL)
3447                         /*
3448                          * FIXME: If ss is NULL, it suppose not get lsr_index
3449                          * at all
3450                          */
3451                         range->lsr_index = ss->ss_node_id;
3452                 return 0;
3453         }
3454
3455         LASSERT(ss != NULL);
3456         fld_range_set_any(range);
3457         /* OSD will only do local fld lookup */
3458         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3459 }
3460
3461 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3462                               struct lu_attr *attr,
3463                               struct dt_allocation_hint *hint,
3464                               struct dt_object_format *dof,
3465                               struct thandle *handle)
3466 {
3467         struct osd_thandle *oh;
3468         int rc;
3469
3470         ENTRY;
3471
3472         LASSERT(handle != NULL);
3473
3474         oh = container_of0(handle, struct osd_thandle, ot_super);
3475         LASSERT(oh->ot_handle == NULL);
3476
3477         /*
3478          * EA object consumes more credits than regular object: osd_mk_index
3479          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3480          * leaf_node, could involves the block, block bitmap, groups, GDT
3481          * change for each block, so add 4 * 2 credits in that case.
3482          */
3483         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3484                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3485                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3486         /*
3487          * Reuse idle OI block may cause additional one OI block
3488          * to be changed.
3489          */
3490         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3491                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3492         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3493                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3494                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3495
3496         /* will help to find FID->ino mapping at dt_insert() */
3497         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3498                                    osd_dt_obj(dt));
3499         if (rc != 0)
3500                 RETURN(rc);
3501
3502         if (!attr)
3503                 RETURN(0);
3504
3505         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3506                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3507                                    NULL, OSD_QID_INODE);
3508         if (rc != 0)
3509                 RETURN(rc);
3510
3511         RETURN(rc);
3512 }
3513
3514 /**
3515  * Called to destroy on-disk representation of the object
3516  *
3517  * Concurrency: must be locked
3518  */
3519 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3520                                struct thandle *th)
3521 {
3522         struct osd_object *obj = osd_dt_obj(dt);
3523         struct inode *inode = obj->oo_inode;
3524         struct osd_thandle *oh;
3525         int rc;
3526
3527         ENTRY;
3528
3529         if (inode == NULL)
3530                 RETURN(-ENOENT);
3531
3532         oh = container_of0(th, struct osd_thandle, ot_super);
3533         LASSERT(oh->ot_handle == NULL);
3534
3535         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3536                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3537
3538         /* For removing agent entry */
3539         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3540                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3541
3542         /*
3543          * Recycle idle OI leaf may cause additional three OI blocks
3544          * to be changed.
3545          */
3546         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3547                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3548                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3549         /* one less inode */
3550         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3551                                    i_projid_read(inode), -1, oh, obj, NULL,
3552                                    OSD_QID_INODE);
3553         if (rc)
3554                 RETURN(rc);
3555         /* data to be truncated */
3556         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3557                                    i_projid_read(inode), 0, oh, obj, NULL,
3558                                    OSD_QID_BLK);
3559         if (rc)
3560                 RETURN(rc);
3561
3562         /*
3563          * will help to find FID->ino when this object is being
3564          * added to PENDING
3565          */
3566         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3567
3568         RETURN(rc);
3569 }
3570
3571 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3572                        struct thandle *th)
3573 {
3574         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3575         struct osd_object *obj = osd_dt_obj(dt);
3576         struct inode *inode = obj->oo_inode;
3577         struct osd_device *osd = osd_obj2dev(obj);
3578         struct osd_thandle *oh;
3579         int result;
3580
3581         ENTRY;
3582
3583         oh = container_of0(th, struct osd_thandle, ot_super);
3584         LASSERT(oh->ot_handle);
3585         LASSERT(inode);
3586         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3587
3588         if (unlikely(fid_is_acct(fid)))
3589                 RETURN(-EPERM);
3590
3591         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3592                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3593                 if (result != 0)
3594                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3595                                osd_name(osd), PFID(fid), result);
3596         }
3597
3598         if (S_ISDIR(inode->i_mode)) {
3599                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3600                         inode->i_nlink == 2);
3601
3602                 spin_lock(&obj->oo_guard);
3603                 clear_nlink(inode);
3604                 spin_unlock(&obj->oo_guard);
3605                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3606         }
3607
3608         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3609
3610         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3611
3612         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3613                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3614                                        oh->ot_handle, OI_CHECK_FLD);
3615
3616         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3617         /* XXX: add to ext3 orphan list */
3618         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3619
3620         /* not needed in the cache anymore */
3621         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3622         obj->oo_destroyed = 1;
3623
3624         RETURN(0);
3625 }
3626
3627 /**
3628  * Put the fid into lustre_mdt_attrs, and then place the structure
3629  * inode's ea. This fid should not be altered during the life time
3630  * of the inode.
3631  *
3632  * \retval +ve, on success
3633  * \retval -ve, on error
3634  *
3635  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3636  */
3637 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3638                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3639 {
3640         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3641         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3642         int rc;
3643
3644         ENTRY;
3645
3646         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3647                 RETURN(0);
3648
3649         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3650                 rc = -ENOMEM;
3651
3652         lustre_loa_init(loa, fid, compat, incompat);
3653         lustre_loa_swab(loa, false);
3654
3655         /*
3656          * For the OST device with 256 bytes inode size by default,
3657          * the PFID EA will be stored together with LMA EA to avoid
3658          * performance trouble. Otherwise the PFID EA can be stored
3659          * independently. LU-8998
3660          */
3661         if ((compat & LMAC_FID_ON_OST) &&
3662             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3663                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3664                                      sizeof(*loa), XATTR_CREATE);
3665         else
3666                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3667                                      sizeof(*lma), XATTR_CREATE);
3668         /*
3669          * LMA may already exist, but we need to check that all the
3670          * desired compat/incompat flags have been added.
3671          */
3672         if (unlikely(rc == -EEXIST)) {
3673                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3674                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3675                 if (rc < 0)
3676                         RETURN(rc);
3677
3678                 if (rc < sizeof(*lma))
3679                         RETURN(-EINVAL);
3680
3681                 lustre_loa_swab(loa, true);
3682                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3683                     ((compat == 0 && incompat == 0) ||
3684                      (!(~lma->lma_compat & compat) &&
3685                       !(~lma->lma_incompat & incompat))))
3686                         RETURN(0);
3687
3688                 lma->lma_self_fid = *fid;
3689                 lma->lma_compat |= compat;
3690                 lma->lma_incompat |= incompat;
3691                 if (rc == sizeof(*lma)) {
3692                         lustre_lma_swab(lma);
3693                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3694                                              sizeof(*lma), XATTR_REPLACE);
3695                 } else {
3696                         lustre_loa_swab(loa, false);
3697                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3698                                              sizeof(*loa), XATTR_REPLACE);
3699                 }
3700         }
3701
3702         RETURN(rc);
3703 }
3704
3705 /**
3706  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3707  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3708  * To have compatilibility with 1.8 ldiskfs driver we need to have
3709  * magic number at start of fid data.
3710  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3711  * its inmemory API.
3712  */
3713 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3714                                          const struct lu_fid *fid)
3715 {
3716         if (!fid_is_namespace_visible(fid) ||
3717             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3718                 param->edp_magic = 0;
3719                 return;
3720         }
3721
3722         param->edp_magic = LDISKFS_LUFID_MAGIC;
3723         param->edp_len =  sizeof(struct lu_fid) + 1;
3724         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3725 }
3726
3727 /**
3728  * Try to read the fid from inode ea into dt_rec.
3729  *
3730  * \param fid object fid.
3731  *
3732  * \retval 0 on success
3733  */
3734 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3735                           __u32 ino, struct lu_fid *fid,
3736                           struct osd_inode_id *id)
3737 {
3738         struct osd_thread_info *info  = osd_oti_get(env);
3739         struct inode *inode;
3740
3741         ENTRY;
3742
3743         osd_id_gen(id, ino, OSD_OII_NOGEN);
3744         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3745         if (IS_ERR(inode))
3746                 RETURN(PTR_ERR(inode));
3747
3748         iput(inode);
3749         RETURN(0);
3750 }
3751
3752 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3753                                         struct inode *dir,
3754                                         struct inode *parent_dir,
3755                                         const struct lu_fid *dot_fid,
3756                                         const struct lu_fid *dot_dot_fid,
3757                                         struct osd_thandle *oth)
3758 {
3759         struct ldiskfs_dentry_param *dot_ldp;
3760         struct ldiskfs_dentry_param *dot_dot_ldp;
3761         __u32 saved_nlink = dir->i_nlink;
3762         int rc;
3763
3764         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3765         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3766
3767         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3768         dot_ldp->edp_magic = 0;
3769
3770         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3771                                     dir, dot_ldp, dot_dot_ldp);
3772         /*
3773          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3774          * the subseqent ref_add() will increase the dir->i_nlink
3775          * as 3. That is incorrect for new created directory.
3776          *
3777          * It looks like hack, because we want to make the OSD API
3778          * to be order-independent for new created directory object
3779          * between dt_insert(..) and ref_add() operations.
3780          *
3781          * Here, we only restore the in-RAM dir-inode's nlink attr,
3782          * becuase if the nlink attr is not 2, then there will be
3783          * ref_add() called following the dt_insert(..), such call
3784          * will make both the in-RAM and on-disk dir-inode's nlink
3785          * attr to be set as 2. LU-7447
3786          */
3787         set_nlink(dir, saved_nlink);
3788         return rc;
3789 }
3790
3791 /**
3792  * Create an local agent inode for remote entry
3793  */
3794 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3795                                                   struct osd_device *osd,
3796                                                   struct osd_object *pobj,
3797                                                   const struct lu_fid *fid,
3798                                                   __u32 type,
3799                                                   struct thandle *th)
3800 {
3801         struct osd_thread_info *info = osd_oti_get(env);
3802         struct inode *local;
3803         struct osd_thandle *oh;
3804         int rc;
3805
3806         ENTRY;
3807
3808         LASSERT(th);
3809         oh = container_of(th, struct osd_thandle, ot_super);
3810         LASSERT(oh->ot_handle->h_transaction != NULL);
3811
3812         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3813                                      NULL);
3814         if (IS_ERR(local)) {
3815                 CERROR("%s: create local error %d\n", osd_name(osd),
3816                        (int)PTR_ERR(local));
3817                 RETURN(local);
3818         }
3819
3820         /*
3821          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3822          * correct gid on remote file, not agent here
3823          */
3824         local->i_gid = current_fsgid();
3825         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3826
3827         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3828          * That gives e2fsck something to look at and be happy, and allows
3829          * debugging if we need to determine where this symlink came from.
3830          */
3831         if (S_ISLNK(type)) {
3832                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3833                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3834                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3835
3836                 i_size_write(local, rc);
3837                 LDISKFS_I(local)->i_disksize = rc;
3838         }
3839         unlock_new_inode(local);
3840
3841         /* Agent inode should not have project ID */
3842 #ifdef  HAVE_PROJECT_QUOTA
3843         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3844             i_projid_read(pobj->oo_inode) != 0) {
3845                 rc = osd_transfer_project(local, 0);
3846                 if (rc) {
3847                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3848                                "quota enforcement enabled on the ldiskfs "
3849                                "filesystem?\n", local->i_sb->s_id, rc);
3850                         RETURN(ERR_PTR(rc));
3851                 }
3852         }
3853 #endif
3854         /* Set special LMA flag for local agent inode */
3855         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3856         if (rc != 0) {
3857                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3858                        osd_name(osd), PFID(fid), rc);
3859                 RETURN(ERR_PTR(rc));
3860         }
3861
3862         if (!S_ISDIR(type))
3863                 RETURN(local);
3864
3865         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3866                                          lu_object_fid(&pobj->oo_dt.do_lu),
3867                                          fid, oh);
3868         if (rc != 0) {
3869                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3870                         osd_name(osd), PFID(fid), rc);
3871                 RETURN(ERR_PTR(rc));
3872         }
3873
3874         RETURN(local);
3875 }
3876
3877 /**
3878  * when direntry is deleted, we have to take care of possible agent inode
3879  * referenced by that. unfortunately we can't do this at that point:
3880  * iget() within a running transaction leads to deadlock and we better do
3881  * not call that every delete declaration to save performance. so we put
3882  * a potention agent inode on a list and process that once the transaction
3883  * is over. Notice it's not any worse in terms of real orphans as regular
3884  * object destroy doesn't put inodes on the on-disk orphan list. this should
3885  * be addressed separately
3886  */
3887 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3888                                             struct osd_thandle *oh,
3889                                             __u32 ino)
3890 {
3891         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3892         struct osd_obj_orphan *oor;
3893
3894         OBD_ALLOC_PTR(oor);
3895         if (oor == NULL)
3896                 return -ENOMEM;
3897
3898         oor->oor_ino = ino;
3899         oor->oor_env = (struct lu_env *)env;
3900         spin_lock(&osd->od_osfs_lock);
3901         list_add(&oor->oor_list, &osd->od_orphan_list);
3902         spin_unlock(&osd->od_osfs_lock);
3903
3904         oh->ot_remove_agents = 1;
3905
3906         return 0;
3907
3908 }
3909
3910 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3911                                                 struct osd_device *osd)
3912 {
3913         struct osd_thread_info *info = osd_oti_get(env);
3914         struct osd_obj_orphan *oor, *tmp;
3915         struct osd_inode_id id;
3916         LIST_HEAD(list);
3917         struct inode *inode;
3918         struct lu_fid fid;
3919         handle_t *jh;
3920         __u32 ino;
3921
3922         spin_lock(&osd->od_osfs_lock);
3923         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3924                 if (oor->oor_env == env)
3925                         list_move(&oor->oor_list, &list);
3926         }
3927         spin_unlock(&osd->od_osfs_lock);
3928
3929         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3930
3931                 ino = oor->oor_ino;
3932
3933                 list_del(&oor->oor_list);
3934                 OBD_FREE_PTR(oor);
3935
3936                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3937                 inode = osd_iget_fid(info, osd, &id, &fid);
3938                 if (IS_ERR(inode))
3939                         continue;
3940
3941                 if (!osd_remote_fid(env, osd, &fid)) {
3942                         iput(inode);
3943                         continue;
3944                 }
3945
3946                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3947                 clear_nlink(inode);
3948                 mark_inode_dirty(inode);
3949                 ldiskfs_journal_stop(jh);
3950                 iput(inode);
3951         }
3952
3953         return 0;
3954 }
3955
3956 /**
3957  * OSD layer object create function for OST objects (b=11826).
3958  *
3959  * The FID is inserted into inode xattr here.
3960  *
3961  * \retval   0, on success
3962  * \retval -ve, on error
3963  */
3964 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3965                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3966                       struct dt_object_format *dof, struct thandle *th)
3967 {
3968         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3969         struct osd_object *obj = osd_dt_obj(dt);
3970         struct osd_thread_info *info = osd_oti_get(env);
3971         int result, on_ost = 0;
3972
3973         ENTRY;
3974
3975         if (dt_object_exists(dt))
3976                 RETURN(-EEXIST);
3977
3978         LINVRNT(osd_invariant(obj));
3979         LASSERT(!dt_object_remote(dt));
3980         LASSERT(osd_is_write_locked(env, obj));
3981         LASSERT(th != NULL);
3982
3983         if (unlikely(fid_is_acct(fid)))
3984                 /*
3985                  * Quota files can't be created from the kernel any more,
3986                  * 'tune2fs -O quota' will take care of creating them
3987                  */
3988                 RETURN(-EPERM);
3989
3990         result = __osd_create(info, obj, attr, hint, dof, th);
3991         if (result == 0) {
3992                 if (fid_is_idif(fid) &&
3993                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3994                         struct lu_fid *tfid = &info->oti_fid;
3995                         struct ost_id *oi   = &info->oti_ostid;
3996
3997                         fid_to_ostid(fid, oi);
3998                         ostid_to_fid(tfid, oi, 0);
3999                         on_ost = 1;
4000                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4001                                                 LMAC_FID_ON_OST, 0);
4002                 } else {
4003                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4004                                                fid, OI_CHECK_FLD);
4005                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4006                                                 on_ost ? LMAC_FID_ON_OST : 0,
4007                                                 0);
4008                 }
4009                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4010                         obj->oo_dt.do_body_ops = &osd_body_ops;
4011         }
4012
4013         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4014                 result = __osd_oi_insert(env, obj, fid, th);
4015
4016         /*
4017          * a small optimization - dt_insert() isn't usually applied
4018          * to OST objects, so we don't need to cache OI mapping for
4019          * OST objects
4020          */
4021         if (result == 0 && on_ost == 0) {
4022                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4023
4024                 result = osd_idc_find_and_init(env, osd, obj);
4025                 LASSERT(result == 0);
4026         }
4027
4028         LASSERT(ergo(result == 0,
4029                      dt_object_exists(dt) && !dt_object_remote(dt)));
4030         LINVRNT(osd_invariant(obj));
4031         RETURN(result);
4032 }
4033
4034 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4035                                struct thandle *handle)
4036 {
4037         struct osd_thandle *oh;
4038
4039         /* it's possible that object doesn't exist yet */
4040         LASSERT(handle != NULL);
4041
4042         oh = container_of0(handle, struct osd_thandle, ot_super);
4043         LASSERT(oh->ot_handle == NULL);
4044
4045         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4046                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4047
4048         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4049
4050         return 0;
4051 }
4052
4053 /*
4054  * Concurrency: @dt is write locked.
4055  */
4056 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4057                        struct thandle *th)
4058 {
4059         struct osd_object *obj = osd_dt_obj(dt);
4060         struct inode *inode = obj->oo_inode;
4061         struct osd_thandle *oh;
4062         int rc = 0;
4063
4064         if (!dt_object_exists(dt) || obj->oo_destroyed)
4065                 return -ENOENT;
4066
4067         LINVRNT(osd_invariant(obj));
4068         LASSERT(!dt_object_remote(dt));
4069         LASSERT(osd_is_write_locked(env, obj));
4070         LASSERT(th != NULL);
4071
4072         oh = container_of0(th, struct osd_thandle, ot_super);
4073         LASSERT(oh->ot_handle != NULL);
4074
4075         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4076
4077         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4078                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4079         /*
4080          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4081          * (65000) subdirectories by storing "1" in i_nlink if the link count
4082          * would otherwise overflow. Directory tranversal tools understand
4083          * that (st_nlink == 1) indicates that the filesystem dose not track
4084          * hard links count on the directory, and will not abort subdirectory
4085          * scanning early once (st_nlink - 2) subdirs have been found.
4086          *
4087          * This also has to properly handle the case of inodes with nlink == 0
4088          * in case they are being linked into the PENDING directory
4089          */
4090         spin_lock(&obj->oo_guard);
4091         if (unlikely(inode->i_nlink == 0))
4092                 /* inc_nlink from 0 may cause WARN_ON */
4093                 set_nlink(inode, 1);
4094         else {
4095                 ldiskfs_inc_count(oh->ot_handle, inode);
4096                 if (!S_ISDIR(inode->i_mode))
4097                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4098         }
4099         spin_unlock(&obj->oo_guard);
4100
4101         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4102         LINVRNT(osd_invariant(obj));
4103
4104         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4105
4106         return rc;
4107 }
4108
4109 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4110                                struct thandle *handle)
4111 {
4112         struct osd_thandle *oh;
4113
4114         if (!dt_object_exists(dt))
4115                 return -ENOENT;
4116
4117         LASSERT(!dt_object_remote(dt));
4118         LASSERT(handle != NULL);
4119
4120         oh = container_of0(handle, struct osd_thandle, ot_super);
4121         LASSERT(oh->ot_handle == NULL);
4122
4123         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4124                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4125
4126         return 0;
4127 }
4128
4129 /*
4130  * Concurrency: @dt is write locked.
4131  */
4132 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4133                        struct thandle *th)
4134 {
4135         struct osd_object *obj = osd_dt_obj(dt);
4136         struct inode *inode = obj->oo_inode;
4137         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4138         struct osd_thandle *oh;
4139
4140         if (!dt_object_exists(dt))
4141                 return -ENOENT;
4142
4143         LINVRNT(osd_invariant(obj));
4144         LASSERT(!dt_object_remote(dt));
4145         LASSERT(osd_is_write_locked(env, obj));
4146         LASSERT(th != NULL);
4147
4148         oh = container_of0(th, struct osd_thandle, ot_super);
4149         LASSERT(oh->ot_handle != NULL);
4150
4151         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4152
4153         spin_lock(&obj->oo_guard);
4154         /*
4155          * That can be result of upgrade from old Lustre version and
4156          * applied only to local files.  Just skip this ref_del call.
4157          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4158          */
4159         if (inode->i_nlink == 0) {
4160                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4161                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4162                              ", maybe an upgraded file? (LU-3915)\n",
4163                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4164                 spin_unlock(&obj->oo_guard);
4165                 return 0;
4166         }
4167
4168         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4169                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4170
4171         ldiskfs_dec_count(oh->ot_handle, inode);
4172         spin_unlock(&obj->oo_guard);
4173
4174         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4175         LINVRNT(osd_invariant(obj));
4176
4177         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4178
4179         return 0;
4180 }
4181
4182 /*
4183  * Concurrency: @dt is read locked.
4184  */
4185 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4186                          struct lu_buf *buf, const char *name)
4187 {
4188         struct osd_object      *obj    = osd_dt_obj(dt);
4189         struct inode           *inode  = obj->oo_inode;
4190         struct osd_thread_info *info   = osd_oti_get(env);
4191         struct dentry          *dentry = &info->oti_obj_dentry;
4192         bool cache_xattr = false;
4193         int rc;
4194
4195         LASSERT(buf);
4196
4197         /* version get is not real XATTR but uses xattr API */
4198         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4199                 dt_obj_version_t *ver = buf->lb_buf;
4200
4201                 /*
4202                  * for version we are just using xattr API but change inode
4203                  * field instead
4204                  */
4205                 if (buf->lb_len == 0)
4206                         return sizeof(dt_obj_version_t);
4207
4208                 if (buf->lb_len < sizeof(dt_obj_version_t))
4209                         return -ERANGE;
4210
4211                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4212                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4213
4214                 *ver = LDISKFS_I(inode)->i_fs_version;
4215
4216                 return sizeof(dt_obj_version_t);
4217         }
4218
4219         if (!dt_object_exists(dt))
4220                 return -ENOENT;
4221
4222         LASSERT(!dt_object_remote(dt));
4223         LASSERT(inode->i_op != NULL);
4224 #ifdef HAVE_IOP_XATTR
4225         LASSERT(inode->i_op->getxattr != NULL);
4226 #endif
4227
4228         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4229             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4230                 cache_xattr = true;
4231
4232         if (cache_xattr) {
4233                 rc = osd_oxc_get(obj, name, buf);
4234                 if (rc != -ENOENT)
4235                         return rc;
4236         }
4237
4238         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4239                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4240                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4241                 struct filter_fid *ff;
4242                 struct ost_layout *ol;
4243
4244                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4245                 if (rc)
4246                         return rc;
4247
4248                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4249
4250                 rc = sizeof(*ff);
4251                 if (buf->lb_len == 0 || !buf->lb_buf)
4252                         return rc;
4253
4254                 if (buf->lb_len < rc)
4255                         return -ERANGE;
4256
4257                 ff = buf->lb_buf;
4258                 ol = &ff->ff_layout;
4259                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4260                                                   PFID_STRIPE_IDX_BITS);
4261                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4262                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4263                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4264                 if (lma->lma_compat & LMAC_COMP_INFO) {
4265                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4266                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4267                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4268                 } else {
4269                         ol->ol_comp_start = 0;
4270                         ol->ol_comp_end = 0;
4271                         ol->ol_comp_id = 0;
4272                 }
4273         } else {
4274                 rc = __osd_xattr_get(inode, dentry, name,
4275                                      buf->lb_buf, buf->lb_len);
4276         }
4277
4278         if (cache_xattr) {
4279                 if (rc == -ENOENT || rc == -ENODATA)
4280                         osd_oxc_add(obj, name, NULL, 0);
4281                 else if (rc > 0 && buf->lb_buf != NULL)
4282                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4283         }
4284
4285         return rc;
4286 }
4287
4288 static int osd_declare_xattr_set(const struct lu_env *env,
4289                                  struct dt_object *dt,
4290                                  const struct lu_buf *buf, const char *name,
4291                                  int fl, struct thandle *handle)
4292 {
4293         struct osd_thandle *oh;
4294         int credits = 0;
4295         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4296
4297         LASSERT(handle != NULL);
4298
4299         oh = container_of0(handle, struct osd_thandle, ot_super);
4300         LASSERT(oh->ot_handle == NULL);
4301
4302         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4303                 /*
4304                  * For non-upgrading case, the LMA is set first and
4305                  * usually fit inode. But for upgrade case, the LMA
4306                  * may be in another separated EA block.
4307                  */
4308                 if (dt_object_exists(dt)) {
4309                         if (fl == LU_XATTR_REPLACE)
4310                                 credits = 1;
4311                         else
4312                                 goto upgrade;
4313                 }
4314         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4315                 credits = 1;
4316         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4317                 /* We may need to delete the old PFID EA. */
4318                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4319                 if (fl == LU_XATTR_REPLACE)
4320                         credits += 1;
4321                 else
4322                         goto upgrade;
4323         } else {
4324                 /*
4325                  * If some name entry resides on remote MDT, then will create
4326                  * agent entry under remote parent. On the other hand, if the
4327                  * remote entry will be removed, then related agent entry may
4328                  * need to be removed from the remote parent. So there may be
4329                  * kinds of cases, let's declare enough credits. The credits
4330                  * for create agent entry is enough for remove case.
4331                  */
4332                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4333                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4334                         if (dt_object_exists(dt))
4335                                 credits += 1; /* For updating LMA */
4336                 }
4337
4338 upgrade:
4339                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4340
4341                 if (buf != NULL) {
4342                         ssize_t buflen;
4343
4344                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4345                                 /*
4346                                  * learn xattr size from osd_xattr_get if
4347                                  * attribute has not been read yet
4348                                  */
4349                                 buflen = __osd_xattr_get(
4350                                     osd_dt_obj(dt)->oo_inode,
4351                                     &osd_oti_get(env)->oti_obj_dentry,
4352                                     name, NULL, 0);
4353                                 if (buflen < 0)
4354                                         buflen = 0;
4355                         } else {
4356                                 buflen = buf->lb_len;
4357                         }
4358
4359                         if (buflen > sb->s_blocksize) {
4360                                 credits += osd_calc_bkmap_credits(
4361                                     sb, NULL, 0, -1,
4362                                     (buflen + sb->s_blocksize - 1) >>
4363                                     sb->s_blocksize_bits);
4364                         }
4365                 }
4366                 /*
4367                  * xattr set may involve inode quota change, reserve credits for
4368                  * dquot_initialize()
4369                  */
4370                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4371         }
4372
4373         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4374
4375         return 0;
4376 }
4377
4378 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4379                               const struct lu_buf *buf, int fl,
4380                               struct thandle *handle)
4381 {
4382         struct osd_thread_info *info = osd_oti_get(env);
4383         struct dentry *dentry = &info->oti_obj_dentry;
4384         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4385         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4386         struct inode *inode = obj->oo_inode;
4387         struct filter_fid *ff = buf->lb_buf;
4388         struct ost_layout *ol = &ff->ff_layout;
4389         int flags = XATTR_REPLACE;
4390         int rc;
4391
4392         ENTRY;
4393
4394         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4395                 RETURN(-EINVAL);
4396
4397         rc = osd_get_lma(info, inode, dentry, loa);
4398         if (rc == -ENODATA) {
4399                 /* Usually for upgarding from old device */
4400                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4401                                 LMAC_FID_ON_OST, 0);
4402                 flags = XATTR_CREATE;
4403         } else if (rc) {
4404                 RETURN(rc);
4405         }
4406
4407         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4408                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4409                         RETURN(-EEXIST);
4410
4411                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4412                         /* Separate PFID EA from LMA */
4413                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4414                         lustre_lma_swab(lma);
4415                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4416                                              sizeof(*lma), XATTR_REPLACE);
4417                         if (!rc) {
4418                                 obj->oo_pfid_in_lma = 0;
4419                                 rc = LU_XATTR_CREATE;
4420                         }
4421
4422                         RETURN(rc);
4423                 }
4424         } else {
4425                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4426                         RETURN(fl);
4427
4428                 /*
4429                  * Old client does not send stripe information,
4430                  * then store the PFID EA on disk separatedly.
4431                  */
4432                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4433                              ol->ol_stripe_size == 0))
4434                         RETURN(fl);
4435
4436                 /* Remove old PFID EA entry firstly. */
4437                 dquot_initialize(inode);
4438                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4439                 if (rc == -ENODATA) {
4440                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4441                                 RETURN(rc);
4442                 } else if (rc) {
4443                         RETURN(rc);
4444                 }
4445         }
4446
4447         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4448         if (likely(ol->ol_stripe_size != 0)) {
4449                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4450                                              PFID_STRIPE_IDX_BITS;
4451                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4452                 lma->lma_compat |= LMAC_STRIPE_INFO;
4453                 if (ol->ol_comp_id != 0) {
4454                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4455                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4456                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4457                         lma->lma_compat |= LMAC_COMP_INFO;
4458                 }
4459         }
4460
4461         lustre_loa_swab(loa, false);
4462
4463         /* Store the PFID EA inside LMA. */
4464         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4465                              flags);
4466         if (!rc)
4467                 obj->oo_pfid_in_lma = 1;
4468
4469         RETURN(rc);
4470 }
4471
4472 /*
4473  * In DNE environment, the object (in spite of regular file or directory)
4474  * and its name entry may reside on different MDTs. Under such case, we will
4475  * create an agent entry on the MDT where the object resides. The agent entry
4476  * references the object locally, that makes the object to be visible to the
4477  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4478  * such as 'tar' can handle the object properly.
4479  *
4480  * We handle the agent entry during set linkEA that is the common interface
4481  * for both regular file and directroy, can handle kinds of cases, such as
4482  * create/link/unlink/rename, and so on.
4483  *
4484  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4485  *
4486  * XXX: There are two known issues:
4487  * 1. For one object, we will create at most one agent entry even if there
4488  *    may be more than one cross-MDTs hard links on the object. So the local
4489  *    e2fsck may claim that the object's nlink is larger than the name entries
4490  *    that reference such inode. And in further, the e2fsck will fix the nlink
4491  *    attribute to match the local references. Then it will cause the object's
4492  *    nlink attribute to be inconsistent with the global references. it is bad
4493  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4494  *    other hand, the global namespace LFSCK can repair the object's attribute
4495  *    according to the linkEA.
4496  * 2. There may be too many hard links on the object as to its linkEA overflow,
4497  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4498  *    case happened, then at this point, we do not know whether there are some
4499  *    cross-MDTs reference. But there are local references, it guarantees that
4500  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4501  */
4502 static int osd_xattr_handle_linkea(const struct lu_env *env,
4503                                    struct osd_device *osd,
4504                                    struct osd_object *obj,
4505                                    const struct lu_buf *buf,
4506                                    struct thandle *handle)
4507 {
4508         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4509         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4510         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4511         struct lu_name tmpname;
4512         struct osd_thandle *oh;
4513         int rc;
4514         bool remote = false;
4515
4516         ENTRY;
4517
4518         oh = container_of0(handle, struct osd_thandle, ot_super);
4519         LASSERT(oh->ot_handle != NULL);
4520
4521         rc = linkea_init_with_rec(&ldata);
4522         if (!rc) {
4523                 linkea_first_entry(&ldata);
4524                 while (ldata.ld_lee != NULL && !remote) {
4525                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4526                                             &tmpname, tfid);
4527                         if (osd_remote_fid(env, osd, tfid) > 0)
4528                                 remote = true;
4529                         else
4530                                 linkea_next_entry(&ldata);
4531                 }
4532         } else if (rc == -ENODATA) {
4533                 rc = 0;
4534         } else {
4535                 RETURN(rc);
4536         }
4537
4538         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4539                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4540                 if (rc)
4541                         CERROR("%s: failed to remove agent entry for "DFID
4542                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4543         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4544                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4545                 if (rc)
4546                         CERROR("%s: failed to create agent entry for "DFID
4547                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4548         }
4549
4550         RETURN(rc);
4551 }
4552
4553 /*
4554  * Concurrency: @dt is write locked.
4555  */
4556 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4557                          const struct lu_buf *buf, const char *name, int fl,
4558                          struct thandle *handle)
4559 {
4560         struct osd_object *obj = osd_dt_obj(dt);
4561         struct osd_device *osd = osd_obj2dev(obj);
4562         struct inode *inode = obj->oo_inode;
4563         struct osd_thread_info *info = osd_oti_get(env);
4564         int fs_flags = 0;
4565         int len;
4566         int rc;
4567
4568         ENTRY;
4569
4570         LASSERT(handle);
4571         LASSERT(buf);
4572
4573         /* version set is not real XATTR */
4574         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4575                 dt_obj_version_t *version = buf->lb_buf;
4576
4577                 /*
4578                  * for version we are just using xattr API but change inode
4579                  * field instead
4580                  */
4581                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4582
4583                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4584                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4585
4586                 LDISKFS_I(inode)->i_fs_version = *version;
4587                 /*
4588                  * Version is set after all inode operations are finished,
4589                  * so we should mark it dirty here
4590                  */
4591                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4592
4593                 RETURN(0);
4594         }
4595
4596         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4597                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4598
4599         len = buf->lb_len;
4600         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4601
4602         /*
4603          * For the OST device with 256 bytes inode size by default,
4604          * the PFID EA will be stored together with LMA EA to avoid
4605          * performance trouble. Otherwise the PFID EA can be stored
4606          * independently. LU-8998
4607          */
4608         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4609             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4610                 LASSERT(buf->lb_buf);
4611
4612                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4613                 if (fl <= 0)
4614                         RETURN(fl);
4615         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4616                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4617                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4618
4619                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4620                 if (rc)
4621                         RETURN(rc);
4622
4623                 lma->lma_incompat |= LMAI_STRIPED;
4624                 lustre_lma_swab(lma);
4625                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4626                                      sizeof(*lma), XATTR_REPLACE);
4627                 if (rc != 0)
4628                         RETURN(rc);
4629         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4630                 LASSERT(!osd->od_is_ost);
4631
4632                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4633                 if (rc)
4634                         RETURN(rc);
4635         }
4636
4637         if (fl & LU_XATTR_REPLACE)
4638                 fs_flags |= XATTR_REPLACE;
4639
4640         if (fl & LU_XATTR_CREATE)
4641                 fs_flags |= XATTR_CREATE;
4642
4643         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4644         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4645
4646         if (rc == 0 &&
4647             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4648              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4649                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4650
4651         return rc;
4652 }
4653
4654 /*
4655  * Concurrency: @dt is read locked.
4656  */
4657 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4658                           const struct lu_buf *buf)
4659 {
4660         struct osd_object *obj = osd_dt_obj(dt);
4661         struct inode *inode = obj->oo_inode;
4662         struct osd_thread_info *info = osd_oti_get(env);
4663         struct dentry *dentry = &info->oti_obj_dentry;
4664
4665         if (!dt_object_exists(dt))
4666                 return -ENOENT;
4667
4668         LASSERT(!dt_object_remote(dt));
4669         LASSERT(inode->i_op != NULL);
4670         LASSERT(inode->i_op->listxattr != NULL);
4671
4672         dentry->d_inode = inode;
4673         dentry->d_sb = inode->i_sb;
4674         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4675 }
4676
4677 static int osd_declare_xattr_del(const struct lu_env *env,
4678                                  struct dt_object *dt, const char *name,
4679                                  struct thandle *handle)
4680 {
4681         struct osd_thandle *oh;
4682         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4683
4684         LASSERT(!dt_object_remote(dt));
4685         LASSERT(handle != NULL);
4686
4687         oh = container_of0(handle, struct osd_thandle, ot_super);
4688         LASSERT(oh->ot_handle == NULL);
4689
4690         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4691                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4692         /*
4693          * xattr del may involve inode quota change, reserve credits for
4694          * dquot_initialize()
4695          */
4696         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4697
4698         return 0;
4699 }
4700
4701 /*
4702  * Concurrency: @dt is write locked.
4703  */
4704 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4705                          const char *name, struct thandle *handle)
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         int rc;
4712
4713         if (!dt_object_exists(dt))
4714                 return -ENOENT;
4715
4716         LASSERT(!dt_object_remote(dt));
4717         LASSERT(inode->i_op != NULL);
4718         LASSERT(handle != NULL);
4719 #ifdef HAVE_IOP_XATTR
4720         LASSERT(inode->i_op->removexattr != NULL);
4721 #endif
4722
4723         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4724
4725         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4726                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4727
4728                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4729                                  &info->oti_ost_attrs);
4730                 if (!rc) {
4731                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4732
4733                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4734                         lustre_lma_swab(lma);
4735                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4736                                              sizeof(*lma), XATTR_REPLACE);
4737                         if (!rc)
4738                                 obj->oo_pfid_in_lma = 0;
4739                 }
4740         } else {
4741                 dquot_initialize(inode);
4742                 dentry->d_inode = inode;
4743                 dentry->d_sb = inode->i_sb;
4744                 rc = osd_removexattr(dentry, inode, name);
4745         }
4746
4747         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4748
4749         if (rc == 0 &&
4750             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4751              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4752                 osd_oxc_del(obj, name);
4753
4754         return rc;
4755 }
4756
4757 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4758                            __u64 start, __u64 end)
4759 {
4760         struct osd_object *obj = osd_dt_obj(dt);
4761         struct inode *inode = obj->oo_inode;
4762         struct osd_thread_info *info = osd_oti_get(env);
4763         struct dentry *dentry = &info->oti_obj_dentry;
4764         struct file *file = &info->oti_file;
4765         int rc;
4766
4767         ENTRY;
4768
4769         dentry->d_inode = inode;
4770         dentry->d_sb = inode->i_sb;
4771         file->f_path.dentry = dentry;
4772         file->f_mapping = inode->i_mapping;
4773         file->f_op = inode->i_fop;
4774         file->f_inode = inode;
4775
4776         rc = vfs_fsync_range(file, start, end, 0);
4777
4778         RETURN(rc);
4779 }
4780
4781 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4782 {
4783         return 0;
4784 }
4785
4786 /*
4787  * Index operations.
4788  */
4789
4790 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4791                                const struct dt_index_features *feat)
4792 {
4793         struct iam_descr *descr;
4794
4795         if (osd_object_is_root(o))
4796                 return feat == &dt_directory_features;
4797
4798         LASSERT(o->oo_dir != NULL);
4799
4800         descr = o->oo_dir->od_container.ic_descr;
4801         if (feat == &dt_directory_features) {
4802                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4803                         return 1;
4804                 else
4805                         return 0;
4806         } else {
4807                 return feat->dif_keysize_min <= descr->id_key_size &&
4808                        descr->id_key_size <= feat->dif_keysize_max &&
4809                        feat->dif_recsize_min <= descr->id_rec_size &&
4810                        descr->id_rec_size <= feat->dif_recsize_max &&
4811                        !(feat->dif_flags & (DT_IND_VARKEY |
4812                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4813                        ergo(feat->dif_flags & DT_IND_UPDATE,
4814                             1 /* XXX check that object (and fs) is writable */);
4815         }
4816 }
4817
4818 static int osd_iam_container_init(const struct lu_env *env,
4819                                   struct osd_object *obj,
4820                                   struct osd_directory *dir)
4821 {
4822         struct iam_container *bag = &dir->od_container;
4823         int result;
4824
4825         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4826         if (result != 0)
4827                 return result;
4828
4829         result = iam_container_setup(bag);
4830         if (result == 0)
4831                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4832         else
4833                 iam_container_fini(bag);
4834
4835         return result;
4836 }
4837
4838
4839 /*
4840  * Concurrency: no external locking is necessary.
4841  */
4842 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4843                          const struct dt_index_features *feat)
4844 {
4845         int result;
4846         int skip_iam = 0;
4847         struct osd_object *obj = osd_dt_obj(dt);
4848
4849         LINVRNT(osd_invariant(obj));
4850
4851         if (osd_object_is_root(obj)) {
4852                 dt->do_index_ops = &osd_index_ea_ops;
4853                 result = 0;
4854         } else if (feat == &dt_directory_features) {
4855                 dt->do_index_ops = &osd_index_ea_ops;
4856                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4857                         result = 0;
4858                 else
4859                         result = -ENOTDIR;
4860                 skip_iam = 1;
4861         } else if (unlikely(feat == &dt_otable_features)) {
4862                 dt->do_index_ops = &osd_otable_ops;
4863                 return 0;
4864         } else if (unlikely(feat == &dt_acct_features)) {
4865                 dt->do_index_ops = &osd_acct_index_ops;
4866                 result = 0;
4867                 skip_iam = 1;
4868         } else if (!osd_has_index(obj)) {
4869                 struct osd_directory *dir;
4870                 struct osd_device *osd = osd_obj2dev(obj);
4871                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4872
4873                 OBD_ALLOC_PTR(dir);
4874                 if (dir) {
4875
4876                         spin_lock(&obj->oo_guard);
4877                         if (obj->oo_dir == NULL)
4878                                 obj->oo_dir = dir;
4879                         else
4880                                 /*
4881                                  * Concurrent thread allocated container data.
4882                                  */
4883                                 OBD_FREE_PTR(dir);
4884                         spin_unlock(&obj->oo_guard);
4885                         /*
4886                          * Now, that we have container data, serialize its
4887                          * initialization.
4888                          */
4889                         down_write(&obj->oo_ext_idx_sem);
4890                         /*
4891                          * recheck under lock.
4892                          */
4893
4894                         if (osd_has_index(obj)) {
4895                                 result = 0;
4896                                 goto unlock;
4897                         }
4898
4899                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4900                         if (result || feat == &dt_lfsck_namespace_features ||
4901                             feat == &dt_lfsck_layout_orphan_features ||
4902                             feat == &dt_lfsck_layout_dangling_features)
4903                                 goto unlock;
4904
4905                         result = osd_index_register(osd, fid,
4906                                                     feat->dif_keysize_max,
4907                                                     feat->dif_recsize_max);
4908                         if (result < 0)
4909                                 CWARN("%s: failed to register index "
4910                                       DFID": rc = %d\n",
4911                                       osd_name(osd), PFID(fid), result);
4912                         else if (result > 0)
4913                                 result = 0;
4914                         else
4915                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4916                                        " (%d/%d) registered\n",
4917                                        osd_name(osd), PFID(fid),
4918                                        (int)feat->dif_keysize_max,
4919                                        (int)feat->dif_recsize_max);
4920
4921 unlock:
4922                         up_write(&obj->oo_ext_idx_sem);
4923                 } else {
4924                         result = -ENOMEM;
4925                 }
4926         } else {
4927                 result = 0;
4928         }
4929
4930         if (result == 0 && skip_iam == 0) {
4931                 if (!osd_iam_index_probe(env, obj, feat))
4932                         result = -ENOTDIR;
4933         }
4934         LINVRNT(osd_invariant(obj));
4935
4936         return result;
4937 }
4938
4939 static int osd_otable_it_attr_get(const struct lu_env *env,
4940                                  struct dt_object *dt,
4941                                  struct lu_attr *attr)
4942 {
4943         attr->la_valid = 0;
4944         return 0;
4945 }
4946
4947 static const struct dt_object_operations osd_obj_ops = {
4948         .do_read_lock           = osd_read_lock,
4949         .do_write_lock          = osd_write_lock,
4950         .do_read_unlock         = osd_read_unlock,
4951         .do_write_unlock        = osd_write_unlock,
4952         .do_write_locked        = osd_write_locked,
4953         .do_attr_get            = osd_attr_get,
4954         .do_declare_attr_set    = osd_declare_attr_set,
4955         .do_attr_set            = osd_attr_set,
4956         .do_ah_init             = osd_ah_init,
4957         .do_declare_create      = osd_declare_create,
4958         .do_create              = osd_create,
4959         .do_declare_destroy     = osd_declare_destroy,
4960         .do_destroy             = osd_destroy,
4961         .do_index_try           = osd_index_try,
4962         .do_declare_ref_add     = osd_declare_ref_add,
4963         .do_ref_add             = osd_ref_add,
4964         .do_declare_ref_del     = osd_declare_ref_del,
4965         .do_ref_del             = osd_ref_del,
4966         .do_xattr_get           = osd_xattr_get,
4967         .do_declare_xattr_set   = osd_declare_xattr_set,
4968         .do_xattr_set           = osd_xattr_set,
4969         .do_declare_xattr_del   = osd_declare_xattr_del,
4970         .do_xattr_del           = osd_xattr_del,
4971         .do_xattr_list          = osd_xattr_list,
4972         .do_object_sync         = osd_object_sync,
4973         .do_invalidate          = osd_invalidate,
4974 };
4975
4976 static const struct dt_object_operations osd_obj_otable_it_ops = {
4977         .do_attr_get    = osd_otable_it_attr_get,
4978         .do_index_try   = osd_index_try,
4979 };
4980
4981 static int osd_index_declare_iam_delete(const struct lu_env *env,
4982                                         struct dt_object *dt,
4983                                         const struct dt_key *key,
4984                                         struct thandle *handle)
4985 {
4986         struct osd_thandle *oh;
4987
4988         oh = container_of0(handle, struct osd_thandle, ot_super);
4989         LASSERT(oh->ot_handle == NULL);
4990
4991         /* Recycle  may cause additional three blocks to be changed. */
4992         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4993                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4994
4995         return 0;
4996 }
4997
4998 /**
4999  *      delete a (key, value) pair from index \a dt specified by \a key
5000  *
5001  *      \param  dt      osd index object
5002  *      \param  key     key for index
5003  *      \param  rec     record reference
5004  *      \param  handle  transaction handler
5005  *
5006  *      \retval  0  success
5007  *      \retval -ve   failure
5008  */
5009 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5010                                 const struct dt_key *key,
5011                                 struct thandle *handle)
5012 {
5013         struct osd_thread_info *oti = osd_oti_get(env);
5014         struct osd_object *obj = osd_dt_obj(dt);
5015         struct osd_thandle *oh;
5016         struct iam_path_descr *ipd;
5017         struct iam_container *bag = &obj->oo_dir->od_container;
5018         int rc;
5019
5020         ENTRY;
5021
5022         if (!dt_object_exists(dt))
5023                 RETURN(-ENOENT);
5024
5025         LINVRNT(osd_invariant(obj));
5026         LASSERT(!dt_object_remote(dt));
5027         LASSERT(bag->ic_object == obj->oo_inode);
5028         LASSERT(handle != NULL);
5029
5030         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5031
5032         ipd = osd_idx_ipd_get(env, bag);
5033         if (unlikely(ipd == NULL))
5034                 RETURN(-ENOMEM);
5035
5036         oh = container_of0(handle, struct osd_thandle, ot_super);
5037         LASSERT(oh->ot_handle != NULL);
5038         LASSERT(oh->ot_handle->h_transaction != NULL);
5039
5040         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5041                 /* swab quota uid/gid provided by caller */
5042                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5043                 key = (const struct dt_key *)&oti->oti_quota_id;
5044         }
5045
5046         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5047         osd_ipd_put(env, bag, ipd);
5048         LINVRNT(osd_invariant(obj));
5049         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5050         RETURN(rc);
5051 }
5052
5053 static int osd_index_declare_ea_delete(const struct lu_env *env,
5054                                        struct dt_object *dt,
5055                                        const struct dt_key *key,
5056                                        struct thandle *handle)
5057 {
5058         struct osd_thandle *oh;
5059         struct inode *inode;
5060         int rc, credits;
5061
5062         ENTRY;
5063
5064         LASSERT(!dt_object_remote(dt));
5065         LASSERT(handle != NULL);
5066
5067         oh = container_of0(handle, struct osd_thandle, ot_super);
5068         LASSERT(oh->ot_handle == NULL);
5069
5070         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5071         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5072
5073         inode = osd_dt_obj(dt)->oo_inode;
5074         if (inode == NULL)
5075                 RETURN(-ENOENT);
5076
5077         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5078                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5079                                    NULL, OSD_QID_BLK);
5080         RETURN(rc);
5081 }
5082
5083 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5084                                           struct dt_rec *fid)
5085 {
5086         struct osd_fid_pack *rec;
5087         int rc = -ENODATA;
5088
5089         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5090                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5091                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5092                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5093                         rc = -EINVAL;
5094         }
5095         return rc;
5096 }
5097
5098 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5099                           const struct lu_fid *fid)
5100 {
5101         struct seq_server_site *ss = osd_seq_site(osd);
5102
5103         ENTRY;
5104
5105         /* FID seqs not in FLDB, must be local seq */
5106         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5107                 RETURN(0);
5108
5109         /*
5110          * If FLD is not being initialized yet, it only happens during the
5111          * initialization, likely during mgs initialization, and we assume
5112          * this is local FID.
5113          */
5114         if (ss == NULL || ss->ss_server_fld == NULL)
5115                 RETURN(0);
5116
5117         /* Only check the local FLDB here */
5118         if (osd_seq_exists(env, osd, fid_seq(fid)))
5119                 RETURN(0);
5120
5121         RETURN(1);
5122 }
5123
5124 static void osd_take_care_of_agent(const struct lu_env *env,
5125                                    struct osd_device *osd,
5126                                    struct osd_thandle *oh,
5127                                    struct ldiskfs_dir_entry_2 *de)
5128 {
5129         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5130         struct osd_idmap_cache *idc;
5131         int rc, schedule = 0;
5132
5133         LASSERT(de != NULL);
5134
5135         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5136         if (likely(rc == 0)) {
5137                 idc = osd_idc_find_or_init(env, osd, fid);
5138                 if (IS_ERR(idc) || idc->oic_remote)
5139                         schedule = 1;
5140         } else if (rc == -ENODATA) {
5141                 /*
5142                  * can't get FID, postpone to the end of the
5143                  * transaction when iget() is safe
5144                  */
5145                 schedule = 1;
5146         } else {
5147                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5148         }
5149         if (schedule)
5150                 osd_schedule_agent_inode_removal(env, oh,
5151                                                  le32_to_cpu(de->inode));
5152 }
5153
5154 /**
5155  * Index delete function for interoperability mode (b11826).
5156  * It will remove the directory entry added by osd_index_ea_insert().
5157  * This entry is needed to maintain name->fid mapping.
5158  *
5159  * \param key,  key i.e. file entry to be deleted
5160  *
5161  * \retval   0, on success
5162  * \retval -ve, on error
5163  */
5164 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5165                                const struct dt_key *key, struct thandle *handle)
5166 {
5167         struct osd_object *obj = osd_dt_obj(dt);
5168         struct inode *dir = obj->oo_inode;
5169         struct dentry *dentry;
5170         struct osd_thandle *oh;
5171         struct ldiskfs_dir_entry_2 *de = NULL;
5172         struct buffer_head *bh;
5173         struct htree_lock *hlock = NULL;
5174         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5175         int rc;
5176
5177         ENTRY;
5178
5179         if (!dt_object_exists(dt))
5180                 RETURN(-ENOENT);
5181
5182         LINVRNT(osd_invariant(obj));
5183         LASSERT(!dt_object_remote(dt));
5184         LASSERT(handle != NULL);
5185
5186         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5187
5188         oh = container_of(handle, struct osd_thandle, ot_super);
5189         LASSERT(oh->ot_handle != NULL);
5190         LASSERT(oh->ot_handle->h_transaction != NULL);
5191
5192         dquot_initialize(dir);
5193         dentry = osd_child_dentry_get(env, obj,
5194                                       (char *)key, strlen((char *)key));
5195
5196         if (obj->oo_hl_head != NULL) {
5197                 hlock = osd_oti_get(env)->oti_hlock;
5198                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5199                                    dir, LDISKFS_HLOCK_DEL);
5200         } else {
5201                 down_write(&obj->oo_ext_idx_sem);
5202         }
5203
5204         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5205         if (!IS_ERR(bh)) {
5206                 /*
5207                  * If this is not the ".." entry, it might be a remote DNE
5208                  * entry and  we need to check if the FID is for a remote
5209                  * MDT.  If the FID is  not in the directory entry (e.g.
5210                  * upgraded 1.8 filesystem without dirdata enabled) then
5211                  * we need to get the FID from the LMA. For a remote directory
5212                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5213                  * case.
5214                  *
5215                  * Delete the entry before the agent inode in order to
5216                  * simplify error handling.  At worst an error after deleting
5217                  * the entry first might leak the agent inode afterward. The
5218                  * reverse would need filesystem abort in case of error deleting
5219                  * the entry after the agent had been removed, or leave a
5220                  * dangling entry pointing at a random inode.
5221                  */
5222                 if (strcmp((char *)key, dotdot) != 0)
5223                         osd_take_care_of_agent(env, osd, oh, de);
5224                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5225                 brelse(bh);
5226         } else {
5227                 rc = PTR_ERR(bh);
5228         }
5229         if (hlock != NULL)
5230                 ldiskfs_htree_unlock(hlock);
5231         else
5232                 up_write(&obj->oo_ext_idx_sem);
5233
5234         GOTO(out, rc);
5235 out:
5236         LASSERT(osd_invariant(obj));
5237         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5238         RETURN(rc);
5239 }
5240
5241 /**
5242  *      Lookup index for \a key and copy record to \a rec.
5243  *
5244  *      \param  dt      osd index object
5245  *      \param  key     key for index
5246  *      \param  rec     record reference
5247  *
5248  *      \retval  +ve  success : exact mach
5249  *      \retval  0    return record with key not greater than \a key
5250  *      \retval -ve   failure
5251  */
5252 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5253                                 struct dt_rec *rec, const struct dt_key *key)
5254 {
5255         struct osd_object *obj = osd_dt_obj(dt);
5256         struct iam_path_descr *ipd;
5257         struct iam_container *bag = &obj->oo_dir->od_container;
5258         struct osd_thread_info *oti = osd_oti_get(env);
5259         struct iam_iterator *it = &oti->oti_idx_it;
5260         struct iam_rec *iam_rec;
5261         int rc;
5262
5263         ENTRY;
5264
5265         if (!dt_object_exists(dt))
5266                 RETURN(-ENOENT);
5267
5268         LASSERT(osd_invariant(obj));
5269         LASSERT(!dt_object_remote(dt));
5270         LASSERT(bag->ic_object == obj->oo_inode);
5271
5272         ipd = osd_idx_ipd_get(env, bag);
5273         if (IS_ERR(ipd))
5274                 RETURN(-ENOMEM);
5275
5276         /* got ipd now we can start iterator. */
5277         iam_it_init(it, bag, 0, ipd);
5278
5279         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5280                 /* swab quota uid/gid provided by caller */
5281                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5282                 key = (const struct dt_key *)&oti->oti_quota_id;
5283         }
5284
5285         rc = iam_it_get(it, (struct iam_key *)key);
5286         if (rc >= 0) {
5287                 if (S_ISDIR(obj->oo_inode->i_mode))
5288                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5289                 else
5290                         iam_rec = (struct iam_rec *)rec;
5291
5292                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5293
5294                 if (S_ISDIR(obj->oo_inode->i_mode))
5295                         osd_fid_unpack((struct lu_fid *)rec,
5296                                        (struct osd_fid_pack *)iam_rec);
5297                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5298                         osd_quota_unpack(obj, rec);
5299         }
5300
5301         iam_it_put(it);
5302         iam_it_fini(it);
5303         osd_ipd_put(env, bag, ipd);
5304
5305         LINVRNT(osd_invariant(obj));
5306
5307         RETURN(rc);
5308 }
5309
5310 static int osd_index_declare_iam_insert(const struct lu_env *env,
5311                                         struct dt_object *dt,
5312                                         const struct dt_rec *rec,
5313                                         const struct dt_key *key,
5314                                         struct thandle *handle)
5315 {
5316         struct osd_thandle *oh;
5317
5318         LASSERT(handle != NULL);
5319
5320         oh = container_of0(handle, struct osd_thandle, ot_super);
5321         LASSERT(oh->ot_handle == NULL);
5322
5323         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5324                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5325
5326         return 0;
5327 }
5328
5329 /**
5330  *      Inserts (key, value) pair in \a dt index object.
5331  *
5332  *      \param  dt      osd index object
5333  *      \param  key     key for index
5334  *      \param  rec     record reference
5335  *      \param  th      transaction handler
5336  *
5337  *      \retval  0  success
5338  *      \retval -ve failure
5339  */
5340 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5341                                 const struct dt_rec *rec,
5342                                 const struct dt_key *key, struct thandle *th)
5343 {
5344         struct osd_object *obj = osd_dt_obj(dt);
5345         struct iam_path_descr *ipd;
5346         struct osd_thandle *oh;
5347         struct iam_container *bag;
5348         struct osd_thread_info *oti = osd_oti_get(env);
5349         struct iam_rec *iam_rec;
5350         int rc;
5351
5352         ENTRY;
5353
5354         if (!dt_object_exists(dt))
5355                 RETURN(-ENOENT);
5356
5357         LINVRNT(osd_invariant(obj));
5358         LASSERT(!dt_object_remote(dt));
5359
5360         bag = &obj->oo_dir->od_container;
5361         LASSERT(bag->ic_object == obj->oo_inode);
5362         LASSERT(th != NULL);
5363
5364         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5365
5366         ipd = osd_idx_ipd_get(env, bag);
5367         if (unlikely(ipd == NULL))
5368                 RETURN(-ENOMEM);
5369
5370         oh = container_of0(th, struct osd_thandle, ot_super);
5371         LASSERT(oh->ot_handle != NULL);
5372         LASSERT(oh->ot_handle->h_transaction != NULL);
5373         if (S_ISDIR(obj->oo_inode->i_mode)) {
5374                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5375                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5376                              &oti->oti_fid);
5377         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5378                 /* pack quota uid/gid */
5379                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5380                 key = (const struct dt_key *)&oti->oti_quota_id;
5381                 /* pack quota record */
5382                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5383                 iam_rec = (struct iam_rec *)rec;
5384         } else {
5385                 iam_rec = (struct iam_rec *)rec;
5386         }
5387
5388         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5389                         iam_rec, ipd);
5390         osd_ipd_put(env, bag, ipd);
5391         LINVRNT(osd_invariant(obj));
5392         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5393         RETURN(rc);
5394 }
5395
5396 /**
5397  * Calls ldiskfs_add_entry() to add directory entry
5398  * into the directory. This is required for
5399  * interoperability mode (b11826)
5400  *
5401  * \retval   0, on success
5402  * \retval -ve, on error
5403  */
5404 static int __osd_ea_add_rec(struct osd_thread_info *info,
5405                             struct osd_object *pobj, struct inode  *cinode,
5406                             const char *name, const struct lu_fid *fid,
5407                             struct htree_lock *hlock, struct thandle *th)
5408 {
5409         struct ldiskfs_dentry_param *ldp;
5410         struct dentry *child;
5411         struct osd_thandle *oth;
5412         int rc;
5413
5414         oth = container_of(th, struct osd_thandle, ot_super);
5415         LASSERT(oth->ot_handle != NULL);
5416         LASSERT(oth->ot_handle->h_transaction != NULL);
5417         LASSERT(pobj->oo_inode);
5418
5419         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5420         if (unlikely(pobj->oo_inode ==
5421                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5422                 ldp->edp_magic = 0;
5423         else
5424                 osd_get_ldiskfs_dirent_param(ldp, fid);
5425         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5426         child->d_fsdata = (void *)ldp;
5427         dquot_initialize(pobj->oo_inode);
5428         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5429                                    child, cinode, hlock);
5430         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5431                 struct ldiskfs_dir_entry_2      *de;
5432                 struct buffer_head              *bh;
5433                 int                              rc1;
5434
5435                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5436                                             NULL, hlock);
5437                 if (!IS_ERR(bh)) {
5438                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5439                                                                bh);
5440                         if (rc1 == 0) {
5441                                 if (S_ISDIR(cinode->i_mode))
5442                                         de->file_type = LDISKFS_DIRENT_LUFID |
5443                                                         LDISKFS_FT_REG_FILE;
5444                                 else
5445                                         de->file_type = LDISKFS_DIRENT_LUFID |
5446                                                         LDISKFS_FT_DIR;
5447                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5448                                                               NULL, bh);
5449                         }
5450                         brelse(bh);
5451                 }
5452         }
5453
5454         RETURN(rc);
5455 }
5456
5457 /**
5458  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5459  * into the directory.Also sets flags into osd object to
5460  * indicate dot and dotdot are created. This is required for
5461  * interoperability mode (b11826)
5462  *
5463  * \param dir   directory for dot and dotdot fixup.
5464  * \param obj   child object for linking
5465  *
5466  * \retval   0, on success
5467  * \retval -ve, on error
5468  */
5469 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5470                               struct osd_object *dir,
5471                               struct inode *parent_dir, const char *name,
5472                               const struct lu_fid *dot_fid,
5473                               const struct lu_fid *dot_dot_fid,
5474                               struct thandle *th)
5475 {
5476         struct inode *inode = dir->oo_inode;
5477         struct osd_thandle *oth;
5478         int result = 0;
5479
5480         oth = container_of(th, struct osd_thandle, ot_super);
5481         LASSERT(oth->ot_handle->h_transaction != NULL);
5482         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5483
5484         if (strcmp(name, dot) == 0) {
5485                 if (dir->oo_compat_dot_created) {
5486                         result = -EEXIST;
5487                 } else {
5488                         LASSERT(inode->i_ino == parent_dir->i_ino);
5489                         dir->oo_compat_dot_created = 1;
5490                         result = 0;
5491                 }
5492         } else if (strcmp(name, dotdot) == 0) {
5493                 if (!dir->oo_compat_dot_created)
5494                         return -EINVAL;
5495                 /* in case of rename, dotdot is already created */
5496                 if (dir->oo_compat_dotdot_created) {
5497                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5498                                                 dot_dot_fid, NULL, th);
5499                 }
5500
5501                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5502                         struct lu_fid tfid = *dot_dot_fid;
5503
5504                         tfid.f_oid--;
5505                         result = osd_add_dot_dotdot_internal(info,
5506                                         dir->oo_inode, parent_dir, dot_fid,
5507                                         &tfid, oth);
5508                 } else {
5509                         result = osd_add_dot_dotdot_internal(info,
5510                                         dir->oo_inode, parent_dir, dot_fid,
5511                                         dot_dot_fid, oth);
5512                 }
5513
5514                 if (result == 0)
5515                         dir->oo_compat_dotdot_created = 1;
5516         }
5517
5518         return result;
5519 }
5520
5521
5522 /**
5523  * It will call the appropriate osd_add* function and return the
5524  * value, return by respective functions.
5525  */
5526 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5527                           struct inode *cinode, const char *name,
5528                           const struct lu_fid *fid, struct thandle *th)
5529 {
5530         struct osd_thread_info *info = osd_oti_get(env);
5531         struct htree_lock *hlock;
5532         int rc;
5533
5534         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5535
5536         if (name[0] == '.' && (name[1] == '\0' ||
5537                                (name[1] == '.' && name[2] == '\0'))) {
5538                 if (hlock != NULL) {
5539                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5540                                            pobj->oo_inode, 0);
5541                 } else {
5542                         down_write(&pobj->oo_ext_idx_sem);
5543                 }
5544
5545                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5546                                         lu_object_fid(&pobj->oo_dt.do_lu),
5547                                         fid, th);
5548         } else {
5549                 if (hlock != NULL) {
5550                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5551                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5552                 } else {
5553                         down_write(&pobj->oo_ext_idx_sem);
5554                 }
5555
5556                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5557                         struct lu_fid *tfid = &info->oti_fid;
5558
5559                         *tfid = *fid;
5560                         tfid->f_ver = ~0;
5561                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5562                                               tfid, hlock, th);
5563                 } else {
5564                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5565                                               hlock, th);
5566                 }
5567         }
5568         if (hlock != NULL)
5569                 ldiskfs_htree_unlock(hlock);
5570         else
5571                 up_write(&pobj->oo_ext_idx_sem);
5572
5573         return rc;
5574 }
5575
5576 static int
5577 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5578                       struct osd_idmap_cache *oic)
5579 {
5580         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5581         struct lu_fid *fid = &oic->oic_fid;
5582         struct osd_inode_id *id = &oic->oic_lid;
5583         struct inode *inode = NULL;
5584         int once = 0;
5585         bool insert;
5586         int rc;
5587
5588         ENTRY;
5589
5590         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5591                 RETURN(0);
5592
5593         if (thread_is_running(&scrub->os_thread) &&
5594             scrub->os_pos_current > id->oii_ino)
5595                 RETURN(0);
5596
5597         if (dev->od_auto_scrub_interval == AS_NEVER ||
5598             ktime_get_real_seconds() <
5599             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5600                 RETURN(0);
5601
5602 again:
5603         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5604         if (rc == -ENOENT) {
5605                 __u32 gen = id->oii_gen;
5606
5607                 insert = true;
5608                 if (inode != NULL)
5609                         goto trigger;
5610
5611                 inode = osd_iget(oti, dev, id);
5612                 /* The inode has been removed (by race maybe). */
5613                 if (IS_ERR(inode)) {
5614                         rc = PTR_ERR(inode);
5615
5616                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5617                 }
5618
5619                 /* The OI mapping is lost. */
5620                 if (gen != OSD_OII_NOGEN)
5621                         goto trigger;
5622
5623                 /*
5624                  * The inode may has been reused by others, we do not know,
5625                  * leave it to be handled by subsequent osd_fid_lookup().
5626                  */
5627                 GOTO(out, rc = 0);
5628         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5629                 GOTO(out, rc);
5630         }
5631
5632         insert = false;
5633
5634 trigger:
5635         if (thread_is_running(&scrub->os_thread)) {
5636                 if (inode == NULL) {
5637                         inode = osd_iget(oti, dev, id);
5638                         /* The inode has been removed (by race maybe). */
5639                         if (IS_ERR(inode)) {
5640                                 rc = PTR_ERR(inode);
5641
5642                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5643                         }
5644                 }
5645
5646                 rc = osd_oii_insert(dev, oic, insert);
5647                 /*
5648                  * There is race condition between osd_oi_lookup and OI scrub.
5649                  * The OI scrub finished just after osd_oi_lookup() failure.
5650                  * Under such case, it is unnecessary to trigger OI scrub again,
5651                  * but try to call osd_oi_lookup() again.
5652                  */
5653                 if (unlikely(rc == -EAGAIN))
5654                         goto again;
5655
5656                 if (!S_ISDIR(inode->i_mode))
5657                         rc = 0;
5658                 else
5659                         rc = osd_check_lmv(oti, dev, inode, oic);
5660
5661                 GOTO(out, rc);
5662         }
5663
5664         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5665                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5666                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5667                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5668                        "%s: trigger partial OI scrub for RPC inconsistency "
5669                        "checking FID "DFID": rc = %d\n",
5670                        osd_dev2name(dev), PFID(fid), rc);
5671                 if (rc == 0 || rc == -EALREADY)
5672                         goto again;
5673         }
5674
5675         GOTO(out, rc);
5676
5677 out:
5678         if (inode)
5679                 iput(inode);
5680
5681         RETURN(rc);
5682 }
5683
5684 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5685                                struct osd_device *dev,
5686                                struct osd_idmap_cache *oic,
5687                                struct lu_fid *fid, __u32 ino)
5688 {
5689         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5690         struct inode *inode;
5691         int rc;
5692
5693         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5694         inode = osd_iget(oti, dev, &oic->oic_lid);
5695         if (IS_ERR(inode)) {
5696                 fid_zero(&oic->oic_fid);
5697                 return PTR_ERR(inode);
5698         }
5699
5700         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5701         iput(inode);
5702         if (rc != 0)
5703                 fid_zero(&oic->oic_fid);
5704         else
5705                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5706         return rc;
5707 }
5708
5709 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5710                       struct osd_inode_id *id, const struct lu_fid *fid)
5711 {
5712         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5713                id->oii_ino, id->oii_gen, info);
5714         info->oti_cache.oic_lid = *id;
5715         info->oti_cache.oic_fid = *fid;
5716         info->oti_cache.oic_dev = osd;
5717 }
5718
5719 /**
5720  * Get parent FID from the linkEA.
5721  *
5722  * For a directory which parent resides on remote MDT, to satisfy the
5723  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5724  * the other hand, to make the lookup(..) on the directory can return
5725  * the real parent FID, we append the real parent FID after its ".."
5726  * name entry in the /REMOTE_PARENT_DIR.
5727  *
5728  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5729  * backup. So after the restore, we cannot get the right parent FID from
5730  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5731  * we have stored the real parent FID in the directory object's linkEA,
5732  * we can parse the linkEA for the real parent FID.
5733  *
5734  * \param[in] env       pointer to the thread context
5735  * \param[in] obj       pointer to the object to be handled
5736  * \param[out]fid       pointer to the buffer to hold the parent FID
5737  *
5738  * \retval              0 for getting the real parent FID successfully
5739  * \retval              negative error number on failure
5740  */
5741 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5742                                     struct osd_object *obj,
5743                                     struct lu_fid *fid)
5744 {
5745         struct osd_thread_info *oti = osd_oti_get(env);
5746         struct lu_buf *buf = &oti->oti_big_buf;
5747         struct dentry *dentry = &oti->oti_obj_dentry;
5748         struct inode *inode = obj->oo_inode;
5749         struct linkea_data ldata = { NULL };
5750         int rc;
5751
5752         ENTRY;
5753
5754         fid_zero(fid);
5755         if (!S_ISDIR(inode->i_mode))
5756                 RETURN(-EIO);
5757
5758 again:
5759         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5760                              buf->lb_buf, buf->lb_len);
5761         if (rc == -ERANGE) {
5762                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5763                                      NULL, 0);
5764                 if (rc > 0) {
5765                         lu_buf_realloc(buf, rc);
5766                         if (buf->lb_buf == NULL)
5767                                 RETURN(-ENOMEM);
5768
5769                         goto again;
5770                 }
5771         }
5772
5773         if (unlikely(rc == 0))
5774                 RETURN(-ENODATA);
5775
5776         if (rc < 0)
5777                 RETURN(rc);
5778
5779         if (unlikely(buf->lb_buf == NULL)) {
5780                 lu_buf_realloc(buf, rc);
5781                 if (buf->lb_buf == NULL)
5782                         RETURN(-ENOMEM);
5783
5784                 goto again;
5785         }
5786
5787         ldata.ld_buf = buf;
5788         rc = linkea_init_with_rec(&ldata);
5789         if (!rc) {
5790                 linkea_first_entry(&ldata);
5791                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5792         }
5793
5794         RETURN(rc);
5795 }
5796
5797 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5798                                     struct inode *inode,
5799                                     const struct lu_fid *pfid,
5800                                     const char *name, const int namelen)
5801 {
5802         struct osd_thread_info *oti = osd_oti_get(env);
5803         struct lu_buf *buf = &oti->oti_big_buf;
5804         struct dentry *dentry = &oti->oti_obj_dentry;
5805         struct linkea_data ldata = { NULL };
5806         struct lu_name cname = { .ln_name = name,
5807                                  .ln_namelen = namelen };
5808         int rc;
5809
5810         ENTRY;
5811
5812 again:
5813         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5814                              buf->lb_buf, buf->lb_len);
5815         if (rc == -ERANGE)
5816                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5817
5818         if (rc < 0)
5819                 RETURN(rc);
5820
5821         if (unlikely(rc == 0))
5822                 RETURN(-ENODATA);
5823
5824         if (buf->lb_len < rc) {
5825                 lu_buf_realloc(buf, rc);
5826                 if (buf->lb_buf == NULL)
5827                         RETURN(-ENOMEM);
5828
5829                 goto again;
5830         }
5831
5832         ldata.ld_buf = buf;
5833         rc = linkea_init_with_rec(&ldata);
5834         if (!rc)
5835                 rc = linkea_links_find(&ldata, &cname, pfid);
5836
5837         RETURN(rc);
5838 }
5839
5840 /**
5841  * Calls ->lookup() to find dentry. From dentry get inode and
5842  * read inode's ea to get fid. This is required for  interoperability
5843  * mode (b11826)
5844  *
5845  * \retval   0, on success
5846  * \retval -ve, on error
5847  */
5848 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5849                              struct dt_rec *rec, const struct dt_key *key)
5850 {
5851         struct inode *dir = obj->oo_inode;
5852         struct dentry *dentry;
5853         struct ldiskfs_dir_entry_2 *de;
5854         struct buffer_head *bh;
5855         struct lu_fid *fid = (struct lu_fid *)rec;
5856         struct htree_lock *hlock = NULL;
5857         int ino;
5858         int rc;
5859
5860         ENTRY;
5861
5862         LASSERT(dir->i_op != NULL);
5863         LASSERT(dir->i_op->lookup != NULL);
5864
5865         dentry = osd_child_dentry_get(env, obj,
5866                                       (char *)key, strlen((char *)key));
5867
5868         if (obj->oo_hl_head != NULL) {
5869                 hlock = osd_oti_get(env)->oti_hlock;
5870                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5871                                    dir, LDISKFS_HLOCK_LOOKUP);
5872         } else {
5873                 down_read(&obj->oo_ext_idx_sem);
5874         }
5875
5876         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5877         if (!IS_ERR(bh)) {
5878                 struct osd_thread_info *oti = osd_oti_get(env);
5879                 struct osd_inode_id *id = &oti->oti_id;
5880                 struct osd_idmap_cache *oic = &oti->oti_cache;
5881                 struct osd_device *dev = osd_obj2dev(obj);
5882
5883                 ino = le32_to_cpu(de->inode);
5884                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5885                         brelse(bh);
5886                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5887                         GOTO(out, rc);
5888                 }
5889
5890                 rc = osd_get_fid_from_dentry(de, rec);
5891
5892                 /* done with de, release bh */
5893                 brelse(bh);
5894                 if (rc != 0) {
5895                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5896                                 const char *name = (const char *)key;
5897
5898                                 /*
5899                                  * If the parent is on remote MDT, and there
5900                                  * is no FID-in-dirent, then we have to get
5901                                  * the parent FID from the linkEA.
5902                                  */
5903                                 if (likely(strlen(name) == 2 &&
5904                                            name[0] == '.' && name[1] == '.'))
5905                                         rc = osd_get_pfid_from_linkea(env, obj,
5906                                                                       fid);
5907                         } else {
5908                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5909                         }
5910                 } else {
5911                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5912                 }
5913
5914                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5915                         fid_zero(&oic->oic_fid);
5916
5917                         GOTO(out, rc);
5918                 }
5919
5920                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5921                 rc = osd_consistency_check(oti, dev, oic);
5922                 if (rc == -ENOENT)
5923                         fid_zero(&oic->oic_fid);
5924                 else
5925                         /* Other error should not affect lookup result. */
5926                         rc = 0;
5927         } else {
5928                 rc = PTR_ERR(bh);
5929         }
5930
5931         GOTO(out, rc);
5932
5933 out:
5934         if (hlock != NULL)
5935                 ldiskfs_htree_unlock(hlock);
5936         else
5937                 up_read(&obj->oo_ext_idx_sem);
5938         return rc;
5939 }
5940
5941 static int osd_index_declare_ea_insert(const struct lu_env *env,
5942                                        struct dt_object *dt,
5943                                        const struct dt_rec *rec,
5944                                        const struct dt_key *key,
5945                                        struct thandle *handle)
5946 {
5947         struct osd_thandle *oh;
5948         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5949         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5950         const struct lu_fid *fid = rec1->rec_fid;
5951         int credits, rc = 0;
5952         struct osd_idmap_cache *idc;
5953
5954         ENTRY;
5955
5956         LASSERT(!dt_object_remote(dt));
5957         LASSERT(handle != NULL);
5958         LASSERT(fid != NULL);
5959         LASSERT(rec1->rec_type != 0);
5960
5961         oh = container_of0(handle, struct osd_thandle, ot_super);
5962         LASSERT(oh->ot_handle == NULL);
5963
5964         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5965
5966         /*
5967          * we can't call iget() while a transactions is running
5968          * (this can lead to a deadlock), but we need to know
5969          * inum and object type. so we find this information at
5970          * declaration and cache in per-thread info
5971          */
5972         idc = osd_idc_find_or_init(env, osd, fid);
5973         if (IS_ERR(idc))
5974                 RETURN(PTR_ERR(idc));
5975         if (idc->oic_remote) {
5976                 /*
5977                  * a reference to remote inode is represented by an
5978                  * agent inode which we have to create
5979                  */
5980                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5981                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5982         }
5983
5984         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5985
5986         if (osd_dt_obj(dt)->oo_inode != NULL) {
5987                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5988
5989                 /*
5990                  * We ignore block quota on meta pool (MDTs), so needn't
5991                  * calculate how many blocks will be consumed by this index
5992                  * insert
5993                  */
5994                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5995                                            i_gid_read(inode),
5996                                            i_projid_read(inode), 0,
5997                                            oh, osd_dt_obj(dt), NULL,
5998                                            OSD_QID_BLK);
5999                 if (rc)
6000                         RETURN(rc);
6001
6002 #ifdef HAVE_PROJECT_QUOTA
6003                 /*
6004                  * Reserve credits for local agent inode to transfer
6005                  * to 0, quota enforcement is ignored in this case.
6006                  */
6007                 if (idc->oic_remote &&
6008                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6009                     i_projid_read(inode) != 0)
6010                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6011                                                   0, i_projid_read(inode),
6012                                                   0, false, PRJQUOTA, true);
6013 #endif
6014         }
6015
6016         RETURN(rc);
6017 }
6018
6019 /**
6020  * Index add function for interoperability mode (b11826).
6021  * It will add the directory entry.This entry is needed to
6022  * maintain name->fid mapping.
6023  *
6024  * \param key it is key i.e. file entry to be inserted
6025  * \param rec it is value of given key i.e. fid
6026  *
6027  * \retval   0, on success
6028  * \retval -ve, on error
6029  */
6030 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6031                                const struct dt_rec *rec,
6032                                const struct dt_key *key, struct thandle *th)
6033 {
6034         struct osd_object *obj = osd_dt_obj(dt);
6035         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6036         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6037         const struct lu_fid *fid = rec1->rec_fid;
6038         const char *name = (const char *)key;
6039         struct osd_thread_info *oti = osd_oti_get(env);
6040         struct inode *child_inode = NULL;
6041         struct osd_idmap_cache *idc;
6042         int rc;
6043
6044         ENTRY;
6045
6046         if (!dt_object_exists(dt))
6047                 RETURN(-ENOENT);
6048
6049         LASSERT(osd_invariant(obj));
6050         LASSERT(!dt_object_remote(dt));
6051         LASSERT(th != NULL);
6052
6053         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6054
6055         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6056
6057         idc = osd_idc_find(env, osd, fid);
6058         if (unlikely(idc == NULL)) {
6059                 idc = osd_idc_find_or_init(env, osd, fid);
6060                 if (IS_ERR(idc)) {
6061                         /*
6062                          * this dt_insert() wasn't declared properly, so
6063                          * FID is missing in OI cache. we better do not
6064                          * lookup FID in FLDB/OI and don't risk to deadlock,
6065                          * but in some special cases (lfsck testing, etc)
6066                          * it's much simpler than fixing a caller.
6067                          *
6068                          * normally this error should be placed after the first
6069                          * find, but migrate may attach source stripes to
6070                          * target, which doesn't create stripes.
6071                          */
6072                         CERROR("%s: "DFID" wasn't declared for insert\n",
6073                                osd_name(osd), PFID(fid));
6074                         dump_stack();
6075                         RETURN(PTR_ERR(idc));
6076                 }
6077         }
6078
6079         if (idc->oic_remote) {
6080                 /* Insert remote entry */
6081                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6082                         child_inode =
6083                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6084                 } else {
6085                         child_inode = osd_create_local_agent_inode(env, osd,
6086                                         obj, fid, rec1->rec_type & S_IFMT, th);
6087                         if (IS_ERR(child_inode))
6088                                 RETURN(PTR_ERR(child_inode));
6089                 }
6090         } else {
6091                 /* Insert local entry */
6092                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6093                         /* for a reason OI cache wasn't filled properly */
6094                         CERROR("%s: OIC for "DFID" isn't filled\n",
6095                                osd_name(osd), PFID(fid));
6096                         RETURN(-EINVAL);
6097                 }
6098                 child_inode = oti->oti_inode;
6099                 if (unlikely(child_inode == NULL)) {
6100                         struct ldiskfs_inode_info *lii;
6101
6102                         OBD_ALLOC_PTR(lii);
6103                         if (lii == NULL)
6104                                 RETURN(-ENOMEM);
6105                         child_inode = oti->oti_inode = &lii->vfs_inode;
6106                 }
6107                 child_inode->i_sb = osd_sb(osd);
6108                 child_inode->i_ino = idc->oic_lid.oii_ino;
6109                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6110         }
6111
6112         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6113
6114         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6115                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6116
6117         if (child_inode && child_inode != oti->oti_inode)
6118                 iput(child_inode);
6119         LASSERT(osd_invariant(obj));
6120         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6121         RETURN(rc);
6122 }
6123
6124 /**
6125  *  Initialize osd Iterator for given osd index object.
6126  *
6127  *  \param  dt      osd index object
6128  */
6129
6130 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6131                                      struct dt_object *dt,
6132                                      __u32 unused)
6133 {
6134         struct osd_it_iam *it;
6135         struct osd_object *obj = osd_dt_obj(dt);
6136         struct lu_object *lo = &dt->do_lu;
6137         struct iam_path_descr *ipd;
6138         struct iam_container *bag = &obj->oo_dir->od_container;
6139
6140         if (!dt_object_exists(dt))
6141                 return ERR_PTR(-ENOENT);
6142
6143         OBD_ALLOC_PTR(it);
6144         if (it == NULL)
6145                 return ERR_PTR(-ENOMEM);
6146
6147         ipd = osd_it_ipd_get(env, bag);
6148         if (likely(ipd != NULL)) {
6149                 it->oi_obj = obj;
6150                 it->oi_ipd = ipd;
6151                 lu_object_get(lo);
6152                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6153                 return (struct dt_it *)it;
6154         } else {
6155                 OBD_FREE_PTR(it);
6156                 return ERR_PTR(-ENOMEM);
6157         }
6158 }
6159
6160 /**
6161  * free given Iterator.
6162  */
6163 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6164 {
6165         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6166         struct osd_object *obj = it->oi_obj;
6167
6168         iam_it_fini(&it->oi_it);
6169         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6170         osd_object_put(env, obj);
6171         OBD_FREE_PTR(it);
6172 }
6173
6174 /**
6175  *  Move Iterator to record specified by \a key
6176  *
6177  *  \param  di      osd iterator
6178  *  \param  key     key for index
6179  *
6180  *  \retval +ve  di points to record with least key not larger than key
6181  *  \retval  0   di points to exact matched key
6182  *  \retval -ve  failure
6183  */
6184
6185 static int osd_it_iam_get(const struct lu_env *env,
6186                           struct dt_it *di, const struct dt_key *key)
6187 {
6188         struct osd_thread_info *oti = osd_oti_get(env);
6189         struct osd_it_iam *it = (struct osd_it_iam *)di;
6190
6191         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6192                 /* swab quota uid/gid */
6193                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6194                 key = (struct dt_key *)&oti->oti_quota_id;
6195         }
6196
6197         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6198 }
6199
6200 /**
6201  *  Release Iterator
6202  *
6203  *  \param  di      osd iterator
6204  */
6205 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6206 {
6207         struct osd_it_iam *it = (struct osd_it_iam *)di;
6208
6209         iam_it_put(&it->oi_it);
6210 }
6211
6212 /**
6213  *  Move iterator by one record
6214  *
6215  *  \param  di      osd iterator
6216  *
6217  *  \retval +1   end of container reached
6218  *  \retval  0   success
6219  *  \retval -ve  failure
6220  */
6221
6222 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6223 {
6224         struct osd_it_iam *it = (struct osd_it_iam *)di;
6225
6226         return iam_it_next(&it->oi_it);
6227 }
6228
6229 /**
6230  * Return pointer to the key under iterator.
6231  */
6232
6233 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6234                                      const struct dt_it *di)
6235 {
6236         struct osd_thread_info *oti = osd_oti_get(env);
6237         struct osd_it_iam *it = (struct osd_it_iam *)di;
6238         struct osd_object *obj = it->oi_obj;
6239         struct dt_key *key;
6240
6241         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6242
6243         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6244                 /* swab quota uid/gid */
6245                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6246                 key = (struct dt_key *)&oti->oti_quota_id;
6247         }
6248
6249         return key;
6250 }
6251
6252 /**
6253  * Return size of key under iterator (in bytes)
6254  */
6255
6256 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6257 {
6258         struct osd_it_iam *it = (struct osd_it_iam *)di;
6259
6260         return iam_it_key_size(&it->oi_it);
6261 }
6262
6263 static inline void
6264 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6265 {
6266         /* check if file type is required */
6267         if (ent->lde_attrs & LUDA_TYPE) {
6268                 struct luda_type *lt;
6269                 int align = sizeof(*lt) - 1;
6270
6271                 len = (len + align) & ~align;
6272                 lt = (struct luda_type *)(ent->lde_name + len);
6273                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6274         }
6275
6276         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6277 }
6278
6279 /**
6280  * build lu direct from backend fs dirent.
6281  */
6282
6283 static inline void
6284 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6285                    char *name, __u16 namelen, __u16 type, __u32 attr)
6286 {
6287         ent->lde_attrs = attr | LUDA_FID;
6288         fid_cpu_to_le(&ent->lde_fid, fid);
6289
6290         ent->lde_hash = cpu_to_le64(offset);
6291         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6292
6293         strncpy(ent->lde_name, name, namelen);
6294         ent->lde_name[namelen] = '\0';
6295         ent->lde_namelen = cpu_to_le16(namelen);
6296
6297         /* append lustre attributes */
6298         osd_it_append_attrs(ent, namelen, type);
6299 }
6300
6301 /**
6302  * Return pointer to the record under iterator.
6303  */
6304 static int osd_it_iam_rec(const struct lu_env *env,
6305                           const struct dt_it *di,
6306                           struct dt_rec *dtrec, __u32 attr)
6307 {
6308         struct osd_it_iam *it = (struct osd_it_iam *)di;
6309         struct osd_thread_info *info = osd_oti_get(env);
6310
6311         ENTRY;
6312
6313         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6314                 const struct osd_fid_pack *rec;
6315                 struct lu_fid *fid = &info->oti_fid;
6316                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6317                 char *name;
6318                 int namelen;
6319                 __u64 hash;
6320                 int rc;
6321
6322                 name = (char *)iam_it_key_get(&it->oi_it);
6323                 if (IS_ERR(name))
6324                         RETURN(PTR_ERR(name));
6325
6326                 namelen = iam_it_key_size(&it->oi_it);
6327
6328                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6329                 if (IS_ERR(rec))
6330                         RETURN(PTR_ERR(rec));
6331
6332                 rc = osd_fid_unpack(fid, rec);
6333                 if (rc)
6334                         RETURN(rc);
6335
6336                 hash = iam_it_store(&it->oi_it);
6337
6338                 /* IAM does not store object type in IAM index (dir) */
6339                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6340                                    0, LUDA_FID);
6341         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6342                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6343                            (struct iam_rec *)dtrec);
6344                 osd_quota_unpack(it->oi_obj, dtrec);
6345         } else {
6346                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6347                            (struct iam_rec *)dtrec);
6348         }
6349
6350         RETURN(0);
6351 }
6352
6353 /**
6354  * Returns cookie for current Iterator position.
6355  */
6356 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6357 {
6358         struct osd_it_iam *it = (struct osd_it_iam *)di;
6359
6360         return iam_it_store(&it->oi_it);
6361 }
6362
6363 /**
6364  * Restore iterator from cookie.
6365  *
6366  * \param  di      osd iterator
6367  * \param  hash    Iterator location cookie
6368  *
6369  * \retval +ve  di points to record with least key not larger than key.
6370  * \retval  0   di points to exact matched key
6371  * \retval -ve  failure
6372  */
6373
6374 static int osd_it_iam_load(const struct lu_env *env,
6375                            const struct dt_it *di, __u64 hash)
6376 {
6377         struct osd_it_iam *it = (struct osd_it_iam *)di;
6378
6379         return iam_it_load(&it->oi_it, hash);
6380 }
6381
6382 static const struct dt_index_operations osd_index_iam_ops = {
6383         .dio_lookup         = osd_index_iam_lookup,
6384         .dio_declare_insert = osd_index_declare_iam_insert,
6385         .dio_insert         = osd_index_iam_insert,
6386         .dio_declare_delete = osd_index_declare_iam_delete,
6387         .dio_delete         = osd_index_iam_delete,
6388         .dio_it     = {
6389                 .init     = osd_it_iam_init,
6390                 .fini     = osd_it_iam_fini,
6391                 .get      = osd_it_iam_get,
6392                 .put      = osd_it_iam_put,
6393                 .next     = osd_it_iam_next,
6394                 .key      = osd_it_iam_key,
6395                 .key_size = osd_it_iam_key_size,
6396                 .rec      = osd_it_iam_rec,
6397                 .store    = osd_it_iam_store,
6398                 .load     = osd_it_iam_load
6399         }
6400 };
6401
6402
6403 /**
6404  * Creates or initializes iterator context.
6405  *
6406  * \retval struct osd_it_ea, iterator structure on success
6407  *
6408  */
6409 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6410                                     struct dt_object *dt,
6411                                     __u32 attr)
6412 {
6413         struct osd_object *obj = osd_dt_obj(dt);
6414         struct osd_thread_info *info = osd_oti_get(env);
6415         struct osd_it_ea *oie;
6416         struct file *file;
6417         struct lu_object *lo = &dt->do_lu;
6418         struct dentry *obj_dentry;
6419
6420         ENTRY;
6421
6422         if (!dt_object_exists(dt) || obj->oo_destroyed)
6423                 RETURN(ERR_PTR(-ENOENT));
6424
6425         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6426         if (oie == NULL)
6427                 RETURN(ERR_PTR(-ENOMEM));
6428         obj_dentry = &oie->oie_dentry;
6429
6430         obj_dentry->d_inode = obj->oo_inode;
6431         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6432         obj_dentry->d_name.hash = 0;
6433
6434         oie->oie_rd_dirent       = 0;
6435         oie->oie_it_dirent       = 0;
6436         oie->oie_dirent          = NULL;
6437         if (unlikely(!info->oti_it_ea_buf_used)) {
6438                 oie->oie_buf = info->oti_it_ea_buf;
6439                 info->oti_it_ea_buf_used = 1;
6440         } else {
6441                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6442                 if (oie->oie_buf == NULL)
6443                         RETURN(ERR_PTR(-ENOMEM));
6444         }
6445         oie->oie_obj = obj;
6446
6447         file = &oie->oie_file;
6448
6449         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6450         if (attr & LUDA_64BITHASH)
6451                 file->f_mode    = FMODE_64BITHASH;
6452         else
6453                 file->f_mode    = FMODE_32BITHASH;
6454         file->f_path.dentry     = obj_dentry;
6455         file->f_mapping         = obj->oo_inode->i_mapping;
6456         file->f_op              = obj->oo_inode->i_fop;
6457         file->f_inode = obj->oo_inode;
6458
6459         lu_object_get(lo);
6460         RETURN((struct dt_it *)oie);
6461 }
6462
6463 /**
6464  * Destroy or finishes iterator context.
6465  *
6466  * \param di iterator structure to be destroyed
6467  */
6468 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6469 {
6470         struct osd_thread_info *info = osd_oti_get(env);
6471         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6472         struct osd_object *obj = oie->oie_obj;
6473         struct inode *inode = obj->oo_inode;
6474
6475         ENTRY;
6476         oie->oie_file.f_op->release(inode, &oie->oie_file);
6477         osd_object_put(env, obj);
6478         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6479                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6480         else
6481                 info->oti_it_ea_buf_used = 0;
6482         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6483         EXIT;
6484 }
6485
6486 /**
6487  * It position the iterator at given key, so that next lookup continues from
6488  * that key Or it is similar to dio_it->load() but based on a key,
6489  * rather than file position.
6490  *
6491  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6492  * to the beginning.
6493  *
6494  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6495  */
6496 static int osd_it_ea_get(const struct lu_env *env,
6497                          struct dt_it *di, const struct dt_key *key)
6498 {
6499         struct osd_it_ea *it = (struct osd_it_ea *)di;
6500
6501         ENTRY;
6502         LASSERT(((const char *)key)[0] == '\0');
6503         it->oie_file.f_pos = 0;
6504         it->oie_rd_dirent = 0;
6505         it->oie_it_dirent = 0;
6506         it->oie_dirent = NULL;
6507
6508         RETURN(+1);
6509 }
6510
6511 /**
6512  * Does nothing
6513  */
6514 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6515 {
6516 }
6517
6518 struct osd_filldir_cbs {
6519         struct dir_context ctx;
6520         struct osd_it_ea  *it;
6521 };
6522 /**
6523  * It is called internally by ->iterate*(). It fills the
6524  * iterator's in-memory data structure with required
6525  * information i.e. name, namelen, rec_size etc.
6526  *
6527  * \param buf in which information to be filled in.
6528  * \param name name of the file in given dir
6529  *
6530  * \retval 0 on success
6531  * \retval 1 on buffer full
6532  */
6533 #ifdef HAVE_FILLDIR_USE_CTX
6534 static int osd_ldiskfs_filldir(struct dir_context *buf,
6535 #else
6536 static int osd_ldiskfs_filldir(void *buf,
6537 #endif
6538                                const char *name, int namelen,
6539                                loff_t offset, __u64 ino, unsigned int d_type)
6540 {
6541         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6542         struct osd_object *obj = it->oie_obj;
6543         struct osd_it_ea_dirent *ent = it->oie_dirent;
6544         struct lu_fid *fid = &ent->oied_fid;
6545         struct osd_fid_pack *rec;
6546
6547         ENTRY;
6548
6549 /* this should never happen */
6550         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6551                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6552                 RETURN(-EIO);
6553         }
6554
6555         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6556             OSD_IT_EA_BUFSIZE)
6557                 RETURN(1);
6558
6559         /* "." is just the object itself. */
6560         if (namelen == 1 && name[0] == '.') {
6561                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6562         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6563                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6564                 if (osd_fid_unpack(fid, rec) != 0)
6565                         fid_zero(fid);
6566         } else {
6567                 fid_zero(fid);
6568         }
6569         d_type &= ~LDISKFS_DIRENT_LUFID;
6570
6571         /* NOT export local root. */
6572         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6573                 ino = obj->oo_inode->i_ino;
6574                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6575         }
6576
6577         ent->oied_ino     = ino;
6578         ent->oied_off     = offset;
6579         ent->oied_namelen = namelen;
6580         ent->oied_type    = d_type;
6581
6582         memcpy(ent->oied_name, name, namelen);
6583
6584         it->oie_rd_dirent++;
6585         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6586         RETURN(0);
6587 }
6588
6589 /**
6590  * Calls ->iterate*() to load a directory entry at a time
6591  * and stored it in iterator's in-memory data structure.
6592  *
6593  * \param di iterator's in memory structure
6594  *
6595  * \retval   0 on success
6596  * \retval -ve on error
6597  * \retval +1 reach the end of entry
6598  */
6599 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6600                                const struct dt_it *di)
6601 {
6602         struct osd_it_ea *it = (struct osd_it_ea *)di;
6603         struct osd_object *obj = it->oie_obj;
6604         struct inode *inode = obj->oo_inode;
6605         struct htree_lock *hlock = NULL;
6606         struct file *filp = &it->oie_file;
6607         int rc = 0;
6608         struct osd_filldir_cbs buf = {
6609                 .ctx.actor = osd_ldiskfs_filldir,
6610                 .it = it
6611         };
6612
6613         ENTRY;
6614         it->oie_dirent = it->oie_buf;
6615         it->oie_rd_dirent = 0;
6616
6617         if (obj->oo_hl_head != NULL) {
6618                 hlock = osd_oti_get(env)->oti_hlock;
6619                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6620                                    inode, LDISKFS_HLOCK_READDIR);
6621         } else {
6622                 down_read(&obj->oo_ext_idx_sem);
6623         }
6624
6625         filp->f_cred = current_cred();
6626         rc = osd_security_file_alloc(filp);
6627         if (rc)
6628                 RETURN(rc);
6629
6630         filp->f_flags |= O_NOATIME;
6631         filp->f_mode |= FMODE_NONOTIFY;
6632         rc = iterate_dir(filp, &buf.ctx);
6633         if (rc)
6634                 RETURN(rc);
6635
6636         if (hlock != NULL)
6637                 ldiskfs_htree_unlock(hlock);
6638         else
6639                 up_read(&obj->oo_ext_idx_sem);
6640
6641         if (it->oie_rd_dirent == 0) {
6642                 /*
6643                  * If it does not get any dirent, it means it has been reached
6644                  * to the end of the dir
6645                  */
6646                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6647                 if (rc == 0)
6648                         rc = 1;
6649         } else {
6650                 it->oie_dirent = it->oie_buf;
6651                 it->oie_it_dirent = 1;
6652         }
6653
6654         RETURN(rc);
6655 }
6656
6657 /**
6658  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6659  * to load a directory entry at a time and stored it in
6660  * iterator's in-memory data structure.
6661  *
6662  * \param di iterator's in memory structure
6663  *
6664  * \retval +ve iterator reached to end
6665  * \retval   0 iterator not reached to end
6666  * \retval -ve on error
6667  */
6668 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6669 {
6670         struct osd_it_ea *it = (struct osd_it_ea *)di;
6671         int rc;
6672
6673         ENTRY;
6674
6675         if (it->oie_it_dirent < it->oie_rd_dirent) {
6676                 it->oie_dirent =
6677                         (void *)it->oie_dirent +
6678                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6679                                        it->oie_dirent->oied_namelen);
6680                 it->oie_it_dirent++;
6681                 rc = 0;
6682         } else {
6683                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6684                         rc = 1;
6685                 else
6686                         rc = osd_ldiskfs_it_fill(env, di);
6687         }
6688
6689         RETURN(rc);
6690 }
6691
6692 /**
6693  * Returns the key at current position from iterator's in memory structure.
6694  *
6695  * \param di iterator's in memory structure
6696  *
6697  * \retval key i.e. struct dt_key on success
6698  */
6699 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6700                                     const struct dt_it *di)
6701 {
6702         struct osd_it_ea *it = (struct osd_it_ea *)di;
6703
6704         return (struct dt_key *)it->oie_dirent->oied_name;
6705 }
6706
6707 /**
6708  * Returns key's size at current position from iterator's in memory structure.
6709  *
6710  * \param di iterator's in memory structure
6711  *
6712  * \retval key_size i.e. struct dt_key on success
6713  */
6714 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6715 {
6716         struct osd_it_ea *it = (struct osd_it_ea *)di;
6717
6718         return it->oie_dirent->oied_namelen;
6719 }
6720
6721 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6722 #undef LDISKFS_DIR_REC_LEN
6723 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6724 #endif
6725
6726 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6727 {
6728         if (LDISKFS_DIR_REC_LEN(de) >=
6729             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6730                 return true;
6731
6732         return false;
6733 }
6734
6735 static inline bool
6736 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6737                      unsigned int blocksize, bool dotdot)
6738 {
6739         if (dotdot)
6740                 return osd_dotdot_has_space(de);
6741
6742         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6743             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6744                 return true;
6745
6746         return false;
6747 }
6748
6749 static int
6750 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6751                     handle_t *jh, struct dentry *dentry,
6752                     const struct lu_fid *fid, struct buffer_head *bh,
6753                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6754                     bool dotdot)
6755 {
6756         struct inode *dir = dentry->d_parent->d_inode;
6757         struct inode *inode = dentry->d_inode;
6758         struct osd_fid_pack *rec;
6759         struct ldiskfs_dentry_param *ldp;
6760         int namelen = dentry->d_name.len;
6761         int rc;
6762         struct osd_thread_info *info = osd_oti_get(env);
6763
6764         ENTRY;
6765
6766         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6767                 RETURN(0);
6768
6769         /* There is enough space to hold the FID-in-dirent. */
6770         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6771                 rc = ldiskfs_journal_get_write_access(jh, bh);
6772                 if (rc != 0)
6773                         RETURN(rc);
6774
6775                 de->name[namelen] = 0;
6776                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6777                 rec->fp_len = sizeof(struct lu_fid) + 1;
6778                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6779                 de->file_type |= LDISKFS_DIRENT_LUFID;
6780                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6781
6782                 RETURN(rc);
6783         }
6784
6785         LASSERT(!dotdot);
6786
6787         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6788         if (rc != 0)
6789                 RETURN(rc);
6790
6791         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6792         osd_get_ldiskfs_dirent_param(ldp, fid);
6793         dentry->d_fsdata = (void *)ldp;
6794         dquot_initialize(dir);
6795         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6796         /*
6797          * It is too bad, we cannot reinsert the name entry back.
6798          * That means we lose it!
6799          */
6800         if (rc != 0)
6801                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6802                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6803                        osd_ino2name(inode),
6804                        dir->i_ino, dir->i_generation, namelen,
6805                        dentry->d_name.name, PFID(fid), rc);
6806
6807         RETURN(rc);
6808 }
6809
6810 static int
6811 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6812                         struct osd_it_ea *it, struct lu_fid *fid,
6813                         struct osd_inode_id *id, __u32 *attr)
6814 {
6815         struct osd_thread_info *info = osd_oti_get(env);
6816         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6817         struct osd_device *dev = osd_obj2dev(obj);
6818         struct super_block *sb = osd_sb(dev);
6819         const char *devname = osd_name(dev);
6820         struct osd_it_ea_dirent *ent = it->oie_dirent;
6821         struct inode *dir = obj->oo_inode;
6822         struct htree_lock *hlock = NULL;
6823         struct buffer_head *bh = NULL;
6824         handle_t *jh = NULL;
6825         struct ldiskfs_dir_entry_2 *de;
6826         struct dentry *dentry;
6827         struct inode *inode;
6828         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6829         int credits;
6830         int rc;
6831         bool dotdot = false;
6832         bool dirty = false;
6833
6834         ENTRY;
6835
6836         if (ent->oied_name[0] == '.') {
6837                 if (ent->oied_namelen == 1)
6838                         RETURN(0);
6839
6840                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6841                         dotdot = true;
6842         }
6843
6844         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6845         inode = osd_iget(info, dev, id);
6846         if (IS_ERR(inode)) {
6847                 rc = PTR_ERR(inode);
6848                 if (rc == -ENOENT || rc == -ESTALE) {
6849                         /*
6850                          * Maybe dangling name entry, or
6851                          * corrupted directory entry.
6852                          */
6853                         *attr |= LUDA_UNKNOWN;
6854                         rc = 0;
6855                 } else {
6856                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6857                                "check_repair, dir = %lu/%u, name = %.*s, "
6858                                "ino = %llu, rc = %d\n",
6859                                devname, dir->i_ino, dir->i_generation,
6860                                ent->oied_namelen, ent->oied_name,
6861                                ent->oied_ino, rc);
6862                 }
6863
6864                 RETURN(rc);
6865         }
6866
6867         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6868                                            ent->oied_namelen);
6869         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6870         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6871                 lma = NULL;
6872         else if (rc != 0)
6873                 GOTO(out, rc);
6874
6875         /*
6876          * We need to ensure that the name entry is still valid.
6877          * Because it may be removed or renamed by other already.
6878          *
6879          * The unlink or rename operation will start journal before PDO lock,
6880          * so to avoid deadlock, here we need to start journal handle before
6881          * related PDO lock also. But because we do not know whether there
6882          * will be something to be repaired before PDO lock, we just start
6883          * journal without conditions.
6884          *
6885          * We may need to remove the name entry firstly, then insert back.
6886          * One credit is for user quota file update.
6887          * One credit is for group quota file update.
6888          * Two credits are for dirty inode.
6889          */
6890         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6891                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6892
6893         if (dev->od_dirent_journal != 0) {
6894
6895 again:
6896                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6897                 if (IS_ERR(jh)) {
6898                         rc = PTR_ERR(jh);
6899                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6900                                "check_repair, dir = %lu/%u, credits = %d, "
6901                                "name = %.*s, ino = %llu: rc = %d\n",
6902                                devname, dir->i_ino, dir->i_generation, credits,
6903                                ent->oied_namelen, ent->oied_name,
6904                                ent->oied_ino, rc);
6905
6906                         GOTO(out_inode, rc);
6907                 }
6908
6909                 if (obj->oo_hl_head != NULL) {
6910                         hlock = osd_oti_get(env)->oti_hlock;
6911                         /*
6912                          * "0" means exclusive lock for the whole directory.
6913                          * We need to prevent others access such name entry
6914                          * during the delete + insert. Neither HLOCK_ADD nor
6915                          * HLOCK_DEL cannot guarantee the atomicity.
6916                          */
6917                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6918                 } else {
6919                         down_write(&obj->oo_ext_idx_sem);
6920                 }
6921         } else {
6922                 if (obj->oo_hl_head != NULL) {
6923                         hlock = osd_oti_get(env)->oti_hlock;
6924                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6925                                            LDISKFS_HLOCK_LOOKUP);
6926                 } else {
6927                         down_read(&obj->oo_ext_idx_sem);
6928                 }
6929         }
6930
6931         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6932         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6933                 *attr |= LUDA_IGNORE;
6934
6935                 GOTO(out, rc = 0);
6936         }
6937
6938         /*
6939          * For dotdot entry, if there is not enough space to hold the
6940          * FID-in-dirent, just keep them there. It only happens when the
6941          * device upgraded from 1.8 or restored from MDT file-level backup.
6942          * For the whole directory, only dotdot entry have no FID-in-dirent
6943          * and needs to get FID from LMA when readdir, it will not affect the
6944          * performance much.
6945          */
6946         if (dotdot && !osd_dotdot_has_space(de)) {
6947                 *attr |= LUDA_UNKNOWN;
6948
6949                 GOTO(out, rc = 0);
6950         }
6951
6952         if (lma != NULL) {
6953                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6954                         GOTO(out, rc = 0);
6955
6956                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6957                         struct lu_fid *tfid = &lma->lma_self_fid;
6958
6959                         if (likely(dotdot &&
6960                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6961                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6962                                 /*
6963                                  * It must be REMOTE_PARENT_DIR and as the
6964                                  * 'dotdot' entry of remote directory
6965                                  */
6966                                 *attr |= LUDA_IGNORE;
6967                         } else {
6968                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6969                                        "parent directory, but got %.*s under "
6970                                        "dir = %lu/%u with the FID "DFID"\n",
6971                                        devname, ent->oied_namelen,
6972                                        ent->oied_name, dir->i_ino,
6973                                        dir->i_generation, PFID(tfid));
6974
6975                                 *attr |= LUDA_UNKNOWN;
6976                         }
6977
6978                         GOTO(out, rc = 0);
6979                 }
6980         }
6981
6982         if (!fid_is_zero(fid)) {
6983                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6984                                               ent->oied_namelen);
6985                 if (rc == -ENOENT ||
6986                     (rc == -ENODATA &&
6987                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6988                         /*
6989                          * linkEA does not recognize the dirent entry,
6990                          * it may because the dirent entry corruption
6991                          * and points to other's inode.
6992                          */
6993                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6994                                "recognize the dirent, dir = %lu/%u, "
6995                                " name = %.*s, ino = %llu, "
6996                                DFID": rc = %d\n", devname, dir->i_ino,
6997                                dir->i_generation, ent->oied_namelen,
6998                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
6999                         *attr |= LUDA_UNKNOWN;
7000
7001                         GOTO(out, rc = 0);
7002                 }
7003
7004                 if (rc && rc != -ENODATA) {
7005                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7006                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7007                                DFID": rc = %d\n", devname, dir->i_ino,
7008                                dir->i_generation, ent->oied_namelen,
7009                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7010                         *attr |= LUDA_UNKNOWN;
7011
7012                         GOTO(out, rc = 0);
7013                 }
7014         }
7015
7016         if (lma != NULL) {
7017                 /*
7018                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7019                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7020                  */
7021                 if (*attr & LUDA_VERIFY_DRYRUN) {
7022                         *fid = lma->lma_self_fid;
7023                         *attr |= LUDA_REPAIR;
7024
7025                         GOTO(out, rc = 0);
7026                 }
7027
7028                 if (jh == NULL) {
7029                         brelse(bh);
7030                         dev->od_dirent_journal = 1;
7031                         if (hlock != NULL) {
7032                                 ldiskfs_htree_unlock(hlock);
7033                                 hlock = NULL;
7034                         } else {
7035                                 up_read(&obj->oo_ext_idx_sem);
7036                         }
7037
7038                         goto again;
7039                 }
7040
7041                 *fid = lma->lma_self_fid;
7042                 dirty = true;
7043                 /* Update or append the FID-in-dirent. */
7044                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7045                                          bh, de, hlock, dotdot);
7046                 if (rc == 0)
7047                         *attr |= LUDA_REPAIR;
7048                 else
7049                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7050                                "the dirent, dir = %lu/%u, name = %.*s, "
7051                                "ino = %llu, "DFID": rc = %d\n",
7052                                devname, dir->i_ino, dir->i_generation,
7053                                ent->oied_namelen, ent->oied_name,
7054                                ent->oied_ino, PFID(fid), rc);
7055         } else {
7056                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7057                 if (*attr & LUDA_VERIFY_DRYRUN) {
7058                         if (fid_is_sane(fid)) {
7059                                 *attr |= LUDA_REPAIR;
7060                         } else if (dev->od_index == 0) {
7061                                 lu_igif_build(fid, inode->i_ino,
7062                                               inode->i_generation);
7063                                 *attr |= LUDA_UPGRADE;
7064                         }
7065
7066                         GOTO(out, rc = 0);
7067                 }
7068
7069                 if (jh == NULL) {
7070                         brelse(bh);
7071                         dev->od_dirent_journal = 1;
7072                         if (hlock != NULL) {
7073                                 ldiskfs_htree_unlock(hlock);
7074                                 hlock = NULL;
7075                         } else {
7076                                 up_read(&obj->oo_ext_idx_sem);
7077                         }
7078
7079                         goto again;
7080                 }
7081
7082                 dirty = true;
7083                 if (unlikely(fid_is_sane(fid))) {
7084                         /*
7085                          * FID-in-dirent exists, but FID-in-LMA is lost.
7086                          * Trust the FID-in-dirent, and add FID-in-LMA.
7087                          */
7088                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7089                         if (rc == 0)
7090                                 *attr |= LUDA_REPAIR;
7091                         else
7092                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7093                                        "update dirent, dir = %lu/%u, "
7094                                        "name = %.*s, ino = %llu, "
7095                                        DFID": rc = %d\n",
7096                                        devname, dir->i_ino, dir->i_generation,
7097                                        ent->oied_namelen, ent->oied_name,
7098                                        ent->oied_ino, PFID(fid), rc);
7099                 } else if (dev->od_index == 0) {
7100                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7101                         /*
7102                          * It is probably IGIF object. Only aappend the
7103                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7104                          */
7105                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7106                                                  bh, de, hlock, dotdot);
7107                         if (rc == 0)
7108                                 *attr |= LUDA_UPGRADE;
7109                         else
7110                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7111                                        "after the dirent, dir = %lu/%u, "
7112                                        "name = %.*s, ino = %llu, "
7113                                        DFID": rc = %d\n",
7114                                        devname, dir->i_ino, dir->i_generation,
7115                                        ent->oied_namelen, ent->oied_name,
7116                                        ent->oied_ino, PFID(fid), rc);
7117                 }
7118         }
7119
7120         GOTO(out, rc);
7121
7122 out:
7123         if (!IS_ERR(bh))
7124                 brelse(bh);
7125         if (hlock != NULL) {
7126                 ldiskfs_htree_unlock(hlock);
7127         } else {
7128                 if (dev->od_dirent_journal != 0)
7129                         up_write(&obj->oo_ext_idx_sem);
7130                 else
7131                         up_read(&obj->oo_ext_idx_sem);
7132         }
7133
7134         if (jh != NULL)
7135                 ldiskfs_journal_stop(jh);
7136
7137 out_inode:
7138         iput(inode);
7139         if (rc >= 0 && !dirty)
7140                 dev->od_dirent_journal = 0;
7141
7142         return rc;
7143 }
7144
7145 /**
7146  * Returns the value at current position from iterator's in memory structure.
7147  *
7148  * \param di struct osd_it_ea, iterator's in memory structure
7149  * \param attr attr requested for dirent.
7150  * \param lde lustre dirent
7151  *
7152  * \retval   0 no error and \param lde has correct lustre dirent.
7153  * \retval -ve on error
7154  */
7155 static inline int osd_it_ea_rec(const struct lu_env *env,
7156                                 const struct dt_it *di,
7157                                 struct dt_rec *dtrec, __u32 attr)
7158 {
7159         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7160         struct osd_object      *obj   = it->oie_obj;
7161         struct osd_device      *dev   = osd_obj2dev(obj);
7162         struct osd_thread_info *oti   = osd_oti_get(env);
7163         struct osd_inode_id    *id    = &oti->oti_id;
7164         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7165         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7166         __u32 ino = it->oie_dirent->oied_ino;
7167         int rc = 0;
7168
7169         ENTRY;
7170
7171         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7172
7173         if (attr & LUDA_VERIFY) {
7174                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7175                         attr |= LUDA_IGNORE;
7176                         /*
7177                          * If the parent is on remote MDT, and there
7178                          * is no FID-in-dirent, then we have to get
7179                          * the parent FID from the linkEA.
7180                          */
7181                         if (!fid_is_sane(fid) &&
7182                             it->oie_dirent->oied_namelen == 2 &&
7183                             it->oie_dirent->oied_name[0] == '.' &&
7184                             it->oie_dirent->oied_name[1] == '.')
7185                                 osd_get_pfid_from_linkea(env, obj, fid);
7186                 } else {
7187                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7188                                                      &attr);
7189                 }
7190
7191                 if (!fid_is_sane(fid))
7192                         attr |= LUDA_UNKNOWN;
7193         } else {
7194                 attr &= ~LU_DIRENT_ATTRS_MASK;
7195                 if (!fid_is_sane(fid)) {
7196                         bool is_dotdot = false;
7197
7198                         if (it->oie_dirent->oied_namelen == 2 &&
7199                             it->oie_dirent->oied_name[0] == '.' &&
7200                             it->oie_dirent->oied_name[1] == '.')
7201                                 is_dotdot = true;
7202                         /*
7203                          * If the parent is on remote MDT, and there
7204                          * is no FID-in-dirent, then we have to get
7205                          * the parent FID from the linkEA.
7206                          */
7207                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7208                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7209                         } else {
7210                                 if (is_dotdot == false &&
7211                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7212                                         RETURN(-ENOENT);
7213
7214                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7215                         }
7216                 } else {
7217                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7218                 }
7219         }
7220
7221         /* Pack the entry anyway, at least the offset is right. */
7222         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7223                            it->oie_dirent->oied_name,
7224                            it->oie_dirent->oied_namelen,
7225                            it->oie_dirent->oied_type, attr);
7226
7227         if (rc < 0)
7228                 RETURN(rc);
7229
7230         if (osd_remote_fid(env, dev, fid))
7231                 RETURN(0);
7232
7233         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7234                 osd_add_oi_cache(oti, dev, id, fid);
7235
7236         RETURN(rc > 0 ? 0 : rc);
7237 }
7238
7239 /**
7240  * Returns the record size size at current position.
7241  *
7242  * This function will return record(lu_dirent) size in bytes.
7243  *
7244  * \param[in] env       execution environment
7245  * \param[in] di        iterator's in memory structure
7246  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7247  *                      calculate the lu_dirent size.
7248  *
7249  * \retval      record size(in bytes & in memory) of the current lu_dirent
7250  *              entry.
7251  */
7252 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7253                               __u32 attr)
7254 {
7255         struct osd_it_ea *it = (struct osd_it_ea *)di;
7256
7257         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7258 }
7259
7260 /**
7261  * Returns a cookie for current position of the iterator head, so that
7262  * user can use this cookie to load/start the iterator next time.
7263  *
7264  * \param di iterator's in memory structure
7265  *
7266  * \retval cookie for current position, on success
7267  */
7268 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7269 {
7270         struct osd_it_ea *it = (struct osd_it_ea *)di;
7271
7272         return it->oie_dirent->oied_off;
7273 }
7274
7275 /**
7276  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7277  * to load a directory entry at a time and stored it i inn,
7278  * in iterator's in-memory data structure.
7279  *
7280  * \param di struct osd_it_ea, iterator's in memory structure
7281  *
7282  * \retval +ve on success
7283  * \retval -ve on error
7284  */
7285 static int osd_it_ea_load(const struct lu_env *env,
7286                           const struct dt_it *di, __u64 hash)
7287 {
7288         struct osd_it_ea *it = (struct osd_it_ea *)di;
7289         int rc;
7290
7291         ENTRY;
7292         it->oie_file.f_pos = hash;
7293
7294         rc =  osd_ldiskfs_it_fill(env, di);
7295         if (rc > 0)
7296                 rc = -ENODATA;
7297
7298         if (rc == 0)
7299                 rc = 1;
7300
7301         RETURN(rc);
7302 }
7303
7304 /**
7305  * Index lookup function for interoperability mode (b11826).
7306  *
7307  * \param key,  key i.e. file name to be searched
7308  *
7309  * \retval +ve, on success
7310  * \retval -ve, on error
7311  */
7312 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7313                                struct dt_rec *rec, const struct dt_key *key)
7314 {
7315         struct osd_object *obj = osd_dt_obj(dt);
7316         int rc = 0;
7317
7318         ENTRY;
7319
7320         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7321         LINVRNT(osd_invariant(obj));
7322
7323         rc = osd_ea_lookup_rec(env, obj, rec, key);
7324         if (rc == 0)
7325                 rc = 1;
7326         RETURN(rc);
7327 }
7328
7329 /**
7330  * Index and Iterator operations for interoperability
7331  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7332  */
7333 static const struct dt_index_operations osd_index_ea_ops = {
7334         .dio_lookup         = osd_index_ea_lookup,
7335         .dio_declare_insert = osd_index_declare_ea_insert,
7336         .dio_insert         = osd_index_ea_insert,
7337         .dio_declare_delete = osd_index_declare_ea_delete,
7338         .dio_delete         = osd_index_ea_delete,
7339         .dio_it     = {
7340                 .init     = osd_it_ea_init,
7341                 .fini     = osd_it_ea_fini,
7342                 .get      = osd_it_ea_get,
7343                 .put      = osd_it_ea_put,
7344                 .next     = osd_it_ea_next,
7345                 .key      = osd_it_ea_key,
7346                 .key_size = osd_it_ea_key_size,
7347                 .rec      = osd_it_ea_rec,
7348                 .rec_size = osd_it_ea_rec_size,
7349                 .store    = osd_it_ea_store,
7350                 .load     = osd_it_ea_load
7351         }
7352 };
7353
7354 static void *osd_key_init(const struct lu_context *ctx,
7355                           struct lu_context_key *key)
7356 {
7357         struct osd_thread_info *info;
7358
7359         OBD_ALLOC_PTR(info);
7360         if (info == NULL)
7361                 return ERR_PTR(-ENOMEM);
7362
7363         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7364         if (info->oti_it_ea_buf == NULL)
7365                 goto out_free_info;
7366
7367         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7368
7369         info->oti_hlock = ldiskfs_htree_lock_alloc();
7370         if (info->oti_hlock == NULL)
7371                 goto out_free_ea;
7372
7373         return info;
7374
7375 out_free_ea:
7376         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7377 out_free_info:
7378         OBD_FREE_PTR(info);
7379         return ERR_PTR(-ENOMEM);
7380 }
7381
7382 static void osd_key_fini(const struct lu_context *ctx,
7383                          struct lu_context_key *key, void *data)
7384 {
7385         struct osd_thread_info *info = data;
7386         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7387         struct osd_idmap_cache *idc = info->oti_ins_cache;
7388
7389         if (info->oti_dio_pages) {
7390                 int i;
7391                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7392                         if (info->oti_dio_pages[i])
7393                                 __free_page(info->oti_dio_pages[i]);
7394                 }
7395                 OBD_FREE(info->oti_dio_pages,
7396                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7397         }
7398
7399         if (info->oti_inode != NULL)
7400                 OBD_FREE_PTR(lli);
7401         if (info->oti_hlock != NULL)
7402                 ldiskfs_htree_lock_free(info->oti_hlock);
7403         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7404         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7405         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7406         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7407         lu_buf_free(&info->oti_big_buf);
7408         if (idc != NULL) {
7409                 LASSERT(info->oti_ins_cache_size > 0);
7410                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7411                 info->oti_ins_cache = NULL;
7412                 info->oti_ins_cache_size = 0;
7413         }
7414         OBD_FREE_PTR(info);
7415 }
7416
7417 static void osd_key_exit(const struct lu_context *ctx,
7418                          struct lu_context_key *key, void *data)
7419 {
7420         struct osd_thread_info *info = data;
7421
7422         LASSERT(info->oti_r_locks == 0);
7423         LASSERT(info->oti_w_locks == 0);
7424         LASSERT(info->oti_txns    == 0);
7425 }
7426
7427 /* type constructor/destructor: osd_type_init, osd_type_fini */
7428 LU_TYPE_INIT_FINI(osd, &osd_key);
7429
7430 struct lu_context_key osd_key = {
7431         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7432         .lct_init = osd_key_init,
7433         .lct_fini = osd_key_fini,
7434         .lct_exit = osd_key_exit
7435 };
7436
7437
7438 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7439                            const char *name, struct lu_device *next)
7440 {
7441         struct osd_device *osd = osd_dev(d);
7442
7443         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7444             sizeof(osd->od_svname))
7445                 return -E2BIG;
7446         return osd_procfs_init(osd, name);
7447 }
7448
7449 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7450 {
7451         struct seq_server_site *ss = osd_seq_site(osd);
7452         int rc;
7453
7454         ENTRY;
7455
7456         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7457                 RETURN(0);
7458
7459         if (unlikely(ss == NULL))
7460                 RETURN(-ENODEV);
7461
7462         OBD_ALLOC_PTR(osd->od_cl_seq);
7463         if (osd->od_cl_seq == NULL)
7464                 RETURN(-ENOMEM);
7465
7466         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7467                              osd->od_svname, ss->ss_server_seq);
7468         if (rc != 0) {
7469                 OBD_FREE_PTR(osd->od_cl_seq);
7470                 osd->od_cl_seq = NULL;
7471                 RETURN(rc);
7472         }
7473
7474         if (ss->ss_node_id == 0) {
7475                 /*
7476                  * If the OSD on the sequence controller(MDT0), then allocate
7477                  * sequence here, otherwise allocate sequence after connected
7478                  * to MDT0 (see mdt_register_lwp_callback()).
7479                  */
7480                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7481                                    &osd->od_cl_seq->lcs_space, env);
7482         }
7483
7484         RETURN(rc);
7485 }
7486
7487 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7488 {
7489         if (osd->od_cl_seq == NULL)
7490                 return;
7491
7492         seq_client_fini(osd->od_cl_seq);
7493         OBD_FREE_PTR(osd->od_cl_seq);
7494         osd->od_cl_seq = NULL;
7495 }
7496
7497 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7498 {
7499         ENTRY;
7500
7501         /* shutdown quota slave instance associated with the device */
7502         if (o->od_quota_slave_md != NULL) {
7503                 struct qsd_instance *qsd = o->od_quota_slave_md;
7504
7505                 o->od_quota_slave_md = NULL;
7506                 qsd_fini(env, qsd);
7507         }
7508
7509         if (o->od_quota_slave_dt != NULL) {
7510                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7511
7512                 o->od_quota_slave_dt = NULL;
7513                 qsd_fini(env, qsd);
7514         }
7515
7516         osd_fid_fini(env, o);
7517         osd_scrub_cleanup(env, o);
7518
7519         RETURN(0);
7520 }
7521
7522 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7523 {
7524         ENTRY;
7525
7526         if (o->od_mnt != NULL) {
7527                 shrink_dcache_sb(osd_sb(o));
7528                 osd_sync(env, &o->od_dt_dev);
7529
7530                 mntput(o->od_mnt);
7531                 o->od_mnt = NULL;
7532         }
7533
7534         EXIT;
7535 }
7536
7537 static int osd_mount(const struct lu_env *env,
7538                      struct osd_device *o, struct lustre_cfg *cfg)
7539 {
7540         const char *name = lustre_cfg_string(cfg, 0);
7541         const char *dev = lustre_cfg_string(cfg, 1);
7542         const char *opts;
7543         unsigned long page, s_flags, lmd_flags = 0;
7544         struct page *__page;
7545         struct file_system_type *type;
7546         char *options = NULL;
7547         char *str;
7548         struct osd_thread_info *info = osd_oti_get(env);
7549         struct lu_fid *fid = &info->oti_fid;
7550         struct inode *inode;
7551         int rc = 0, force_over_1024tb = 0;
7552
7553         ENTRY;
7554
7555         if (o->od_mnt != NULL)
7556                 RETURN(0);
7557
7558         if (strlen(dev) >= sizeof(o->od_mntdev))
7559                 RETURN(-E2BIG);
7560         strcpy(o->od_mntdev, dev);
7561
7562         str = lustre_cfg_string(cfg, 2);
7563         s_flags = simple_strtoul(str, NULL, 0);
7564         str = strstr(str, ":");
7565         if (str)
7566                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7567         opts = lustre_cfg_string(cfg, 3);
7568 #ifdef __BIG_ENDIAN
7569         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7570                 CERROR("%s: device %s extents feature is not guaranteed to "
7571                        "work on big-endian systems. Use \"bigendian_extents\" "
7572                        "mount option to override.\n", name, dev);
7573                 RETURN(-EINVAL);
7574         }
7575 #endif
7576 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7577         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7578                 CWARN("force_over_128tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force option. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7579         }
7580 #endif
7581 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7582         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7583                 CWARN("force_over_256tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7584         }
7585 #endif
7586 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7587         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7588                 CWARN("force_over_512tb option is deprecated.  Filesystems smaller than 1024TB can be created without any force options. Use force_over_1024tb option for filesystems larger than 1024TB.\n");
7589         }
7590 #endif
7591
7592         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7593                 force_over_1024tb = 1;
7594
7595         __page = alloc_page(GFP_KERNEL);
7596         if (__page == NULL)
7597                 GOTO(out, rc = -ENOMEM);
7598         page = (unsigned long)page_address(__page);
7599         options = (char *)page;
7600         *options = '\0';
7601         if (opts != NULL) {
7602                 /* strip out the options for back compatiblity */
7603                 static char *sout[] = {
7604                         "mballoc",
7605                         "iopen",
7606                         "noiopen",
7607                         "iopen_nopriv",
7608                         "extents",
7609                         "noextents",
7610                         /* strip out option we processed in osd */
7611                         "bigendian_extents",
7612                         "force_over_128tb",
7613                         "force_over_256tb",
7614                         "force_over_512tb",
7615                         "force_over_1024tb",
7616                         NULL
7617                 };
7618                 strncat(options, opts, PAGE_SIZE);
7619                 for (rc = 0, str = options; sout[rc]; ) {
7620                         char *op = strstr(str, sout[rc]);
7621
7622                         if (op == NULL) {
7623                                 rc++;
7624                                 str = options;
7625                                 continue;
7626                         }
7627                         if (op == options || *(op - 1) == ',') {
7628                                 str = op + strlen(sout[rc]);
7629                                 if (*str == ',' || *str == '\0') {
7630                                         *str == ',' ? str++ : str;
7631                                         memmove(op, str, strlen(str) + 1);
7632                                 }
7633                         }
7634                         for (str = op; *str != ',' && *str != '\0'; str++)
7635                                 ;
7636                 }
7637         } else {
7638                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7639         }
7640
7641         /* Glom up mount options */
7642         if (*options != '\0')
7643                 strncat(options, ",", PAGE_SIZE);
7644         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7645
7646         type = get_fs_type("ldiskfs");
7647         if (!type) {
7648                 CERROR("%s: cannot find ldiskfs module\n", name);
7649                 GOTO(out, rc = -ENODEV);
7650         }
7651
7652         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7653         module_put(type->owner);
7654
7655         if (IS_ERR(o->od_mnt)) {
7656                 rc = PTR_ERR(o->od_mnt);
7657                 o->od_mnt = NULL;
7658                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7659                 GOTO(out, rc);
7660         }
7661
7662         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7663                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7664                                  force_over_1024tb == 0) {
7665                 CERROR("%s: device %s LDISKFS has not been tested on filesystems larger than 1024TB and may cause data corruption. Use 'force_over_1024tb' mount option to override.\n",
7666                        name, dev);
7667                 GOTO(out_mnt, rc = -EINVAL);
7668         }
7669
7670         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7671                 if (priv_dev_set_rdonly) {
7672                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7673                         o->od_dt_dev.dd_rdonly = 1;
7674                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7675                                       name);
7676                 } else {
7677                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7678                                       name);
7679
7680                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7681                 }
7682         } else if (priv_dev_check_rdonly &&
7683                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7684                 CERROR("%s: underlying device %s is marked as "
7685                        "read-only. Setup failed\n", name, dev);
7686
7687                 GOTO(out_mnt, rc = -EROFS);
7688         }
7689
7690         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7691                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7692                 GOTO(out_mnt, rc = -EINVAL);
7693         }
7694
7695 #ifdef LDISKFS_MOUNT_DIRDATA
7696         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7697                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7698         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7699                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7700                       "enabling the dirdata feature. If you do not want to "
7701                       "downgrade to Lustre-1.x again, you can enable it via "
7702                       "'tune2fs -O dirdata device'\n", name, dev);
7703 #endif
7704         inode = osd_sb(o)->s_root->d_inode;
7705         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7706         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7707         if (rc != 0) {
7708                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7709                 GOTO(out_mnt, rc);
7710         }
7711
7712         if (lmd_flags & LMD_FLG_NOSCRUB)
7713                 o->od_auto_scrub_interval = AS_NEVER;
7714
7715         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7716                 /* do not use pagecache with flash-backed storage */
7717                 o->od_writethrough_cache = 0;
7718                 o->od_read_cache = 0;
7719         }
7720
7721         GOTO(out, rc = 0);
7722
7723 out_mnt:
7724         mntput(o->od_mnt);
7725         o->od_mnt = NULL;
7726
7727 out:
7728         if (__page)
7729                 __free_page(__page);
7730
7731         return rc;
7732 }
7733
7734 static struct lu_device *osd_device_fini(const struct lu_env *env,
7735                                          struct lu_device *d)
7736 {
7737         struct osd_device *o = osd_dev(d);
7738
7739         ENTRY;
7740
7741         osd_index_backup(env, o, false);
7742         osd_shutdown(env, o);
7743         osd_procfs_fini(o);
7744         osd_obj_map_fini(o);
7745         osd_umount(env, o);
7746
7747         RETURN(NULL);
7748 }
7749
7750 static int osd_device_init0(const struct lu_env *env,
7751                             struct osd_device *o,
7752                             struct lustre_cfg *cfg)
7753 {
7754         struct lu_device *l = osd2lu_dev(o);
7755         struct osd_thread_info *info;
7756         int rc;
7757         int cplen = 0;
7758
7759         /* if the module was re-loaded, env can loose its keys */
7760         rc = lu_env_refill((struct lu_env *)env);
7761         if (rc)
7762                 GOTO(out, rc);
7763         info = osd_oti_get(env);
7764         LASSERT(info);
7765
7766         l->ld_ops = &osd_lu_ops;
7767         o->od_dt_dev.dd_ops = &osd_dt_ops;
7768
7769         spin_lock_init(&o->od_osfs_lock);
7770         mutex_init(&o->od_otable_mutex);
7771         INIT_LIST_HEAD(&o->od_orphan_list);
7772         INIT_LIST_HEAD(&o->od_index_backup_list);
7773         INIT_LIST_HEAD(&o->od_index_restore_list);
7774         spin_lock_init(&o->od_lock);
7775         o->od_index_backup_policy = LIBP_NONE;
7776         o->od_t10_type = 0;
7777
7778         o->od_read_cache = 1;
7779         o->od_writethrough_cache = 1;
7780         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7781         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7782         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7783         o->od_auto_scrub_interval = AS_DEFAULT;
7784
7785         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7786                         sizeof(o->od_svname));
7787         if (cplen >= sizeof(o->od_svname)) {
7788                 rc = -E2BIG;
7789                 GOTO(out, rc);
7790         }
7791
7792         o->od_index_backup_stop = 0;
7793         o->od_index = -1; /* -1 means index is invalid */
7794         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7795         if (rc == LDD_F_SV_TYPE_OST)
7796                 o->od_is_ost = 1;
7797
7798         o->od_full_scrub_ratio = OFSR_DEFAULT;
7799         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7800         rc = osd_mount(env, o, cfg);
7801         if (rc != 0)
7802                 GOTO(out, rc);
7803
7804         /* Can only check block device after mount */
7805         o->od_nonrotational =
7806                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7807
7808         rc = osd_obj_map_init(env, o);
7809         if (rc != 0)
7810                 GOTO(out_mnt, rc);
7811
7812         rc = lu_site_init(&o->od_site, l);
7813         if (rc != 0)
7814                 GOTO(out_compat, rc);
7815         o->od_site.ls_bottom_dev = l;
7816
7817         rc = lu_site_init_finish(&o->od_site);
7818         if (rc != 0)
7819                 GOTO(out_site, rc);
7820
7821         INIT_LIST_HEAD(&o->od_ios_list);
7822         /* setup scrub, including OI files initialization */
7823         o->od_in_init = 1;
7824         rc = osd_scrub_setup(env, o);
7825         o->od_in_init = 0;
7826         if (rc < 0)
7827                 GOTO(out_site, rc);
7828
7829         rc = osd_procfs_init(o, o->od_svname);
7830         if (rc != 0) {
7831                 CERROR("%s: can't initialize procfs: rc = %d\n",
7832                        o->od_svname, rc);
7833                 GOTO(out_scrub, rc);
7834         }
7835
7836         LASSERT(l->ld_site->ls_linkage.next != NULL);
7837         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7838
7839         /* initialize quota slave instance */
7840         /* currently it's no need to prepare qsd_instance_md for OST */
7841         if (!o->od_is_ost) {
7842                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7843                                                 &o->od_dt_dev,
7844                                                 o->od_proc_entry, true);
7845                 if (IS_ERR(o->od_quota_slave_md)) {
7846                         rc = PTR_ERR(o->od_quota_slave_md);
7847                         o->od_quota_slave_md = NULL;
7848                         GOTO(out_procfs, rc);
7849                 }
7850         }
7851
7852         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7853                                         o->od_proc_entry, false);
7854
7855         if (IS_ERR(o->od_quota_slave_dt)) {
7856                 if (o->od_quota_slave_md != NULL) {
7857                         qsd_fini(env, o->od_quota_slave_md);
7858                         o->od_quota_slave_md = NULL;
7859                 }
7860
7861                 rc = PTR_ERR(o->od_quota_slave_dt);
7862                 o->od_quota_slave_dt = NULL;
7863                 GOTO(out_procfs, rc);
7864         }
7865
7866         RETURN(0);
7867
7868 out_procfs:
7869         osd_procfs_fini(o);
7870 out_scrub:
7871         osd_scrub_cleanup(env, o);
7872 out_site:
7873         lu_site_fini(&o->od_site);
7874 out_compat:
7875         osd_obj_map_fini(o);
7876 out_mnt:
7877         osd_umount(env, o);
7878 out:
7879         return rc;
7880 }
7881
7882 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7883                                           struct lu_device_type *t,
7884                                           struct lustre_cfg *cfg)
7885 {
7886         struct osd_device *o;
7887         int rc;
7888
7889         OBD_ALLOC_PTR(o);
7890         if (o == NULL)
7891                 return ERR_PTR(-ENOMEM);
7892
7893         rc = dt_device_init(&o->od_dt_dev, t);
7894         if (rc == 0) {
7895                 /*
7896                  * Because the ctx might be revived in dt_device_init,
7897                  * refill the env here
7898                  */
7899                 lu_env_refill((struct lu_env *)env);
7900                 rc = osd_device_init0(env, o, cfg);
7901                 if (rc)
7902                         dt_device_fini(&o->od_dt_dev);
7903         }
7904
7905         if (unlikely(rc != 0))
7906                 OBD_FREE_PTR(o);
7907
7908         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7909 }
7910
7911 static struct lu_device *osd_device_free(const struct lu_env *env,
7912                                          struct lu_device *d)
7913 {
7914         struct osd_device *o = osd_dev(d);
7915
7916         ENTRY;
7917
7918         /* XXX: make osd top device in order to release reference */
7919         d->ld_site->ls_top_dev = d;
7920         lu_site_purge(env, d->ld_site, -1);
7921         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7922                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7923                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7924         }
7925         lu_site_fini(&o->od_site);
7926         dt_device_fini(&o->od_dt_dev);
7927         OBD_FREE_PTR(o);
7928         RETURN(NULL);
7929 }
7930
7931 static int osd_process_config(const struct lu_env *env,
7932                               struct lu_device *d, struct lustre_cfg *cfg)
7933 {
7934         struct osd_device *o = osd_dev(d);
7935         ssize_t count;
7936         int rc;
7937
7938         ENTRY;
7939
7940         switch (cfg->lcfg_command) {
7941         case LCFG_SETUP:
7942                 rc = osd_mount(env, o, cfg);
7943                 break;
7944         case LCFG_CLEANUP:
7945                 /*
7946                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7947                  * that may happend if hit failure during mount process.
7948                  */
7949                 osd_index_backup(env, o, false);
7950                 lu_dev_del_linkage(d->ld_site, d);
7951                 rc = osd_shutdown(env, o);
7952                 break;
7953         case LCFG_PARAM:
7954                 LASSERT(&o->od_dt_dev);
7955                 count  = class_modify_config(cfg, PARAM_OSD,
7956                                              &o->od_dt_dev.dd_kobj);
7957                 if (count < 0)
7958                         count = class_modify_config(cfg, PARAM_OST,
7959                                                     &o->od_dt_dev.dd_kobj);
7960                 rc = count > 0 ? 0 : count;
7961                 break;
7962         case LCFG_PRE_CLEANUP:
7963                 osd_scrub_stop(o);
7964                 osd_index_backup(env, o,
7965                                  o->od_index_backup_policy != LIBP_NONE);
7966                 rc = 0;
7967                 break;
7968         default:
7969                 rc = -ENOSYS;
7970         }
7971
7972         RETURN(rc);
7973 }
7974
7975 static int osd_recovery_complete(const struct lu_env *env,
7976                                  struct lu_device *d)
7977 {
7978         struct osd_device *osd = osd_dev(d);
7979         int rc = 0;
7980
7981         ENTRY;
7982
7983         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7984                 RETURN(0);
7985
7986         /*
7987          * start qsd instance on recovery completion, this notifies the quota
7988          * slave code that we are about to process new requests now
7989          */
7990         rc = qsd_start(env, osd->od_quota_slave_dt);
7991         if (rc == 0 && osd->od_quota_slave_md != NULL)
7992                 rc = qsd_start(env, osd->od_quota_slave_md);
7993
7994         RETURN(rc);
7995 }
7996
7997 /*
7998  * we use exports to track all osd users
7999  */
8000 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8001                            struct obd_device *obd, struct obd_uuid *cluuid,
8002                            struct obd_connect_data *data, void *localdata)
8003 {
8004         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8005         struct lustre_handle conn;
8006         int rc;
8007
8008         ENTRY;
8009
8010         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8011
8012         rc = class_connect(&conn, obd, cluuid);
8013         if (rc)
8014                 RETURN(rc);
8015
8016         *exp = class_conn2export(&conn);
8017
8018         spin_lock(&osd->od_osfs_lock);
8019         osd->od_connects++;
8020         spin_unlock(&osd->od_osfs_lock);
8021
8022         RETURN(0);
8023 }
8024
8025 /*
8026  * once last export (we don't count self-export) disappeared
8027  * osd can be released
8028  */
8029 static int osd_obd_disconnect(struct obd_export *exp)
8030 {
8031         struct obd_device *obd = exp->exp_obd;
8032         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8033         int rc, release = 0;
8034
8035         ENTRY;
8036
8037         /* Only disconnect the underlying layers on the final disconnect. */
8038         spin_lock(&osd->od_osfs_lock);
8039         osd->od_connects--;
8040         if (osd->od_connects == 0)
8041                 release = 1;
8042         spin_unlock(&osd->od_osfs_lock);
8043
8044         rc = class_disconnect(exp); /* bz 9811 */
8045
8046         if (rc == 0 && release)
8047                 class_manual_cleanup(obd);
8048         RETURN(rc);
8049 }
8050
8051 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8052                        struct lu_device *dev)
8053 {
8054         struct osd_device *osd = osd_dev(dev);
8055         struct lr_server_data *lsd =
8056                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8057         int result = 0;
8058
8059         ENTRY;
8060
8061         if (osd->od_quota_slave_md != NULL) {
8062                 /* set up quota slave objects for inode */
8063                 result = qsd_prepare(env, osd->od_quota_slave_md);
8064                 if (result != 0)
8065                         RETURN(result);
8066         }
8067
8068         if (osd->od_quota_slave_dt != NULL) {
8069                 /* set up quota slave objects for block */
8070                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8071                 if (result != 0)
8072                         RETURN(result);
8073         }
8074
8075
8076         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8077 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8078                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8079                         osd->od_index_in_idif = 1;
8080                 } else {
8081                         osd->od_index_in_idif = 0;
8082                         result = osd_register_proc_index_in_idif(osd);
8083                         if (result != 0)
8084                                 RETURN(result);
8085                 }
8086 #else
8087                 osd->od_index_in_idif = 1;
8088 #endif
8089         }
8090
8091         result = osd_fid_init(env, osd);
8092
8093         RETURN(result);
8094 }
8095
8096 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8097                          struct lu_fid *fid, struct md_op_data *op_data)
8098 {
8099         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8100
8101         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8102 }
8103
8104 static const struct lu_object_operations osd_lu_obj_ops = {
8105         .loo_object_init      = osd_object_init,
8106         .loo_object_delete    = osd_object_delete,
8107         .loo_object_release   = osd_object_release,
8108         .loo_object_free      = osd_object_free,
8109         .loo_object_print     = osd_object_print,
8110         .loo_object_invariant = osd_object_invariant
8111 };
8112
8113 const struct lu_device_operations osd_lu_ops = {
8114         .ldo_object_alloc      = osd_object_alloc,
8115         .ldo_process_config    = osd_process_config,
8116         .ldo_recovery_complete = osd_recovery_complete,
8117         .ldo_prepare           = osd_prepare,
8118 };
8119
8120 static const struct lu_device_type_operations osd_device_type_ops = {
8121         .ldto_init = osd_type_init,
8122         .ldto_fini = osd_type_fini,
8123
8124         .ldto_start = osd_type_start,
8125         .ldto_stop  = osd_type_stop,
8126
8127         .ldto_device_alloc = osd_device_alloc,
8128         .ldto_device_free  = osd_device_free,
8129
8130         .ldto_device_init = osd_device_init,
8131         .ldto_device_fini = osd_device_fini
8132 };
8133
8134 static struct lu_device_type osd_device_type = {
8135         .ldt_tags     = LU_DEVICE_DT,
8136         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8137         .ldt_ops      = &osd_device_type_ops,
8138         .ldt_ctx_tags = LCT_LOCAL,
8139 };
8140
8141 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8142 {
8143         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8144         struct super_block *sb = osd_sb(osd);
8145
8146         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8147 }
8148
8149 /*
8150  * lprocfs legacy support.
8151  */
8152 static const struct obd_ops osd_obd_device_ops = {
8153         .o_owner = THIS_MODULE,
8154         .o_connect      = osd_obd_connect,
8155         .o_disconnect   = osd_obd_disconnect,
8156         .o_fid_alloc    = osd_fid_alloc,
8157         .o_health_check = osd_health_check,
8158 };
8159
8160 static ssize_t track_declares_assert_show(struct kobject *kobj,
8161                                    struct attribute *attr,
8162                                    char *buf)
8163 {
8164         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8165 }
8166
8167 static ssize_t track_declares_assert_store(struct kobject *kobj,
8168                                            struct attribute *attr,
8169                                            const char *buffer, size_t count)
8170 {
8171         bool track_declares_assert;
8172         int rc;
8173
8174         rc = kstrtobool(buffer, &track_declares_assert);
8175         if (rc)
8176                 return rc;
8177
8178         ldiskfs_track_declares_assert = track_declares_assert;
8179
8180         return count;
8181 }
8182 LUSTRE_RW_ATTR(track_declares_assert);
8183
8184 static int __init osd_init(void)
8185 {
8186         struct kobject *kobj;
8187         int rc;
8188
8189         BUILD_BUG_ON(BH_DXLock >=
8190                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8191 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8192         /* please, try to keep osd_thread_info smaller than a page */
8193         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8194 #endif
8195
8196         osd_oi_mod_init();
8197
8198         rc = lu_kmem_init(ldiskfs_caches);
8199         if (rc)
8200                 return rc;
8201
8202 #ifdef CONFIG_KALLSYMS
8203         priv_security_file_alloc =
8204                 (void *)kallsyms_lookup_name("security_file_alloc");
8205         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8206         priv_dev_check_rdonly =
8207                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8208 #endif
8209
8210         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8211                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8212         if (rc) {
8213                 lu_kmem_fini(ldiskfs_caches);
8214                 return rc;
8215         }
8216
8217         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8218         if (kobj) {
8219                 rc = sysfs_create_file(kobj,
8220                                        &lustre_attr_track_declares_assert.attr);
8221                 kobject_put(kobj);
8222                 if (rc) {
8223                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8224                         rc = 0;
8225                 }
8226         }
8227         return rc;
8228 }
8229
8230 static void __exit osd_exit(void)
8231 {
8232         struct kobject *kobj;
8233
8234         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8235         if (kobj) {
8236                 sysfs_remove_file(kobj,
8237                                   &lustre_attr_track_declares_assert.attr);
8238                 kobject_put(kobj);
8239         }
8240         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8241         lu_kmem_fini(ldiskfs_caches);
8242 }
8243
8244 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8245 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8246 MODULE_VERSION(LUSTRE_VERSION_STRING);
8247 MODULE_LICENSE("GPL");
8248
8249 module_init(osd_init);
8250 module_exit(osd_exit);