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_PTR_ARRAY(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_PTR_ARRAY(oti->oti_ins_cache,
 219                                            oti->oti_ins_cache_used);
 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 (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1404             cfs_fail_val == 2) {
1405                 struct osd_thread_info *info = osd_oti_get(env);
1406                 struct osd_idmap_cache *oic = &info->oti_cache;
1407                 /* invalidate thread cache */
1408                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1409         }
1410         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1411                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1412                 l->lo_header->loh_attr |= LOHA_EXISTS;
1413                 return 0;
1414         }
1415
1416         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1417         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1418         if (result == 0 && obj->oo_inode != NULL) {
1419                 struct osd_thread_info *oti = osd_oti_get(env);
1420                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1421
1422                 osd_object_init0(obj);
1423                 if (unlikely(obj->oo_header))
1424                         return 0;
1425
1426                 result = osd_get_lma(oti, obj->oo_inode,
1427                                      &oti->oti_obj_dentry, loa);
1428                 if (!result) {
1429                         /*
1430                          * Convert LMAI flags to lustre LMA flags
1431                          * and cache it to oo_lma_flags
1432                          */
1433                         obj->oo_lma_flags =
1434                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1435                 } else if (result == -ENODATA) {
1436                         result = 0;
1437                 }
1438         }
1439         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1440
1441         LINVRNT(osd_invariant(obj));
1442         return result;
1443 }
1444
1445 /*
1446  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1447  * The left part is for value, binary mode.
1448  */
1449 struct osd_xattr_entry {
1450         struct list_head        oxe_list;
1451         size_t                  oxe_len;
1452         size_t                  oxe_namelen;
1453         bool                    oxe_exist;
1454         struct rcu_head         oxe_rcu;
1455         char                    oxe_buf[0];
1456 };
1457
1458 static int osd_oxc_get(struct osd_object *obj, const char *name,
1459                        struct lu_buf *buf)
1460 {
1461         struct osd_xattr_entry *tmp;
1462         struct osd_xattr_entry *oxe = NULL;
1463         size_t namelen = strlen(name);
1464         int rc;
1465
1466         rcu_read_lock();
1467         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1468                 if (namelen == tmp->oxe_namelen &&
1469                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1470                         oxe = tmp;
1471                         break;
1472                 }
1473         }
1474
1475         if (oxe == NULL)
1476                 GOTO(out, rc = -ENOENT);
1477
1478         if (!oxe->oxe_exist)
1479                 GOTO(out, rc = -ENODATA);
1480
1481         /* vallen */
1482         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1483         LASSERT(rc > 0);
1484
1485         if (buf->lb_buf == NULL)
1486                 GOTO(out, rc);
1487
1488         if (buf->lb_len < rc)
1489                 GOTO(out, rc = -ERANGE);
1490
1491         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1492 out:
1493         rcu_read_unlock();
1494
1495         return rc;
1496 }
1497
1498 static void osd_oxc_free(struct rcu_head *head)
1499 {
1500         struct osd_xattr_entry *oxe;
1501
1502         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1503         OBD_FREE(oxe, oxe->oxe_len);
1504 }
1505
1506 static void osd_oxc_add(struct osd_object *obj, const char *name,
1507                         const char *buf, int buflen)
1508 {
1509         struct osd_xattr_entry *oxe;
1510         struct osd_xattr_entry *old = NULL;
1511         struct osd_xattr_entry *tmp;
1512         size_t namelen = strlen(name);
1513         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1514
1515         OBD_ALLOC(oxe, len);
1516         if (oxe == NULL)
1517                 return;
1518
1519         INIT_LIST_HEAD(&oxe->oxe_list);
1520         oxe->oxe_len = len;
1521         oxe->oxe_namelen = namelen;
1522         memcpy(oxe->oxe_buf, name, namelen);
1523         if (buflen > 0) {
1524                 LASSERT(buf != NULL);
1525                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1526                 oxe->oxe_exist = true;
1527         } else {
1528                 oxe->oxe_exist = false;
1529         }
1530
1531         /* this should be rarely called, just remove old and add new */
1532         spin_lock(&obj->oo_guard);
1533         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1534                 if (namelen == tmp->oxe_namelen &&
1535                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1536                         old = tmp;
1537                         break;
1538                 }
1539         }
1540         if (old != NULL) {
1541                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1542                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1543         } else {
1544                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1545         }
1546         spin_unlock(&obj->oo_guard);
1547 }
1548
1549 static void osd_oxc_del(struct osd_object *obj, const char *name)
1550 {
1551         struct osd_xattr_entry *oxe;
1552         size_t namelen = strlen(name);
1553
1554         spin_lock(&obj->oo_guard);
1555         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1556                 if (namelen == oxe->oxe_namelen &&
1557                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1558                         list_del_rcu(&oxe->oxe_list);
1559                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1560                         break;
1561                 }
1562         }
1563         spin_unlock(&obj->oo_guard);
1564 }
1565
1566 static void osd_oxc_fini(struct osd_object *obj)
1567 {
1568         struct osd_xattr_entry *oxe, *next;
1569
1570         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1571                 list_del(&oxe->oxe_list);
1572                 OBD_FREE(oxe, oxe->oxe_len);
1573         }
1574 }
1575
1576 /*
1577  * Concurrency: no concurrent access is possible that late in object
1578  * life-cycle.
1579  */
1580 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1581 {
1582         struct osd_object *obj = osd_obj(l);
1583         struct lu_object_header *h = obj->oo_header;
1584
1585         LINVRNT(osd_invariant(obj));
1586
1587         osd_oxc_fini(obj);
1588         dt_object_fini(&obj->oo_dt);
1589         if (obj->oo_hl_head != NULL)
1590                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1591         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1592         OBD_FREE_PTR(obj);
1593         if (unlikely(h)) {
1594                 lu_object_header_fini(h);
1595                 OBD_FREE_PRE(h, sizeof(*h), "kfreed");
1596                 kfree_rcu(h, loh_rcu);
1597         }
1598 }
1599
1600 /*
1601  * Concurrency: no concurrent access is possible that late in object
1602  * life-cycle.
1603  */
1604 static void osd_index_fini(struct osd_object *o)
1605 {
1606         struct iam_container *bag;
1607
1608         if (o->oo_dir != NULL) {
1609                 bag = &o->oo_dir->od_container;
1610                 if (o->oo_inode != NULL) {
1611                         if (bag->ic_object == o->oo_inode)
1612                                 iam_container_fini(bag);
1613                 }
1614                 OBD_FREE_PTR(o->oo_dir);
1615                 o->oo_dir = NULL;
1616         }
1617 }
1618
1619 /*
1620  * Concurrency: no concurrent access is possible that late in object
1621  * life-cycle (for all existing callers, that is. New callers have to provide
1622  * their own locking.)
1623  */
1624 static int osd_inode_unlinked(const struct inode *inode)
1625 {
1626         return inode->i_nlink == 0;
1627 }
1628
1629 enum {
1630         OSD_TXN_OI_DELETE_CREDITS    = 20,
1631         OSD_TXN_INODE_DELETE_CREDITS = 20
1632 };
1633
1634 /*
1635  * Journal
1636  */
1637
1638 #if OSD_THANDLE_STATS
1639 /**
1640  * Set time when the handle is allocated
1641  */
1642 static void osd_th_alloced(struct osd_thandle *oth)
1643 {
1644         oth->oth_alloced = ktime_get();
1645 }
1646
1647 /**
1648  * Set time when the handle started
1649  */
1650 static void osd_th_started(struct osd_thandle *oth)
1651 {
1652         oth->oth_started = ktime_get();
1653 }
1654
1655 /**
1656  * Check whether the we deal with this handle for too long.
1657  */
1658 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1659                                 ktime_t alloced, ktime_t started,
1660                                 ktime_t closed)
1661 {
1662         ktime_t now = ktime_get();
1663
1664         LASSERT(dev != NULL);
1665
1666         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1667                             ktime_us_delta(started, alloced));
1668         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1669                             ktime_us_delta(closed, started));
1670         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1671                             ktime_us_delta(now, closed));
1672
1673         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1674                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1675                                 oth, now, alloced, started, closed);
1676                 libcfs_debug_dumpstack(NULL);
1677         }
1678 }
1679
1680 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1681 {                                                                       \
1682         ktime_t __closed = ktime_get();                                 \
1683         ktime_t __alloced = oth->oth_alloced;                           \
1684         ktime_t __started = oth->oth_started;                           \
1685                                                                         \
1686         expr;                                                           \
1687         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1688 }
1689
1690 #else /* OSD_THANDLE_STATS */
1691
1692 #define osd_th_alloced(h)                  do {} while(0)
1693 #define osd_th_started(h)                  do {} while(0)
1694 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1695
1696 #endif /* OSD_THANDLE_STATS */
1697
1698 /*
1699  * Concurrency: doesn't access mutable data.
1700  */
1701 static int osd_param_is_not_sane(const struct osd_device *dev,
1702                                  const struct thandle *th)
1703 {
1704         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1705
1706         return oh->ot_credits > osd_transaction_size(dev);
1707 }
1708
1709 /*
1710  * Concurrency: shouldn't matter.
1711  */
1712 static void osd_trans_commit_cb(struct super_block *sb,
1713                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1714 {
1715         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1716         struct thandle *th = &oh->ot_super;
1717         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1718         struct dt_txn_commit_cb *dcb, *tmp;
1719
1720         LASSERT(oh->ot_handle == NULL);
1721
1722         if (error)
1723                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1724
1725         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1726         /* call per-transaction callbacks if any */
1727         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1728                                  dcb_linkage) {
1729                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1730                          "commit callback entry: magic=%x name='%s'\n",
1731                          dcb->dcb_magic, dcb->dcb_name);
1732                 list_del_init(&dcb->dcb_linkage);
1733                 dcb->dcb_func(NULL, th, dcb, error);
1734         }
1735
1736         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1737         lu_device_put(lud);
1738         th->th_dev = NULL;
1739
1740         OBD_FREE_PTR(oh);
1741 }
1742
1743 static struct thandle *osd_trans_create(const struct lu_env *env,
1744                                         struct dt_device *d)
1745 {
1746         struct osd_thread_info *oti = osd_oti_get(env);
1747         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1748         struct osd_thandle *oh;
1749         struct thandle *th;
1750
1751         ENTRY;
1752
1753         if (d->dd_rdonly) {
1754                 CERROR("%s: someone try to start transaction under "
1755                        "readonly mode, should be disabled.\n",
1756                        osd_name(osd_dt_dev(d)));
1757                 dump_stack();
1758                 RETURN(ERR_PTR(-EROFS));
1759         }
1760
1761         /* on pending IO in this thread should left from prev. request */
1762         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1763
1764         sb_start_write(osd_sb(osd_dt_dev(d)));
1765
1766         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1767         if (!oh) {
1768                 sb_end_write(osd_sb(osd_dt_dev(d)));
1769                 RETURN(ERR_PTR(-ENOMEM));
1770         }
1771
1772         oh->ot_quota_trans = &oti->oti_quota_trans;
1773         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1774         th = &oh->ot_super;
1775         th->th_dev = d;
1776         th->th_result = 0;
1777         oh->ot_credits = 0;
1778         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1779         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1780         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1781         osd_th_alloced(oh);
1782
1783         memset(oti->oti_declare_ops, 0,
1784                sizeof(oti->oti_declare_ops));
1785         memset(oti->oti_declare_ops_cred, 0,
1786                sizeof(oti->oti_declare_ops_cred));
1787         memset(oti->oti_declare_ops_used, 0,
1788                sizeof(oti->oti_declare_ops_used));
1789
1790         oti->oti_ins_cache_depth++;
1791
1792         RETURN(th);
1793 }
1794
1795 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1796 {
1797         struct osd_thread_info *oti = osd_oti_get(env);
1798         struct osd_thandle *oh;
1799
1800         oh = container_of0(th, struct osd_thandle, ot_super);
1801         LASSERT(oh != NULL);
1802
1803         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1804               oti->oti_declare_ops[OSD_OT_CREATE],
1805               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1806               oti->oti_declare_ops_used[OSD_OT_CREATE],
1807               oti->oti_declare_ops[OSD_OT_DESTROY],
1808               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1809               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1810         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1811               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1812               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1813               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1814               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1815               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1816               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1817         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1818               oti->oti_declare_ops[OSD_OT_WRITE],
1819               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1820               oti->oti_declare_ops_used[OSD_OT_WRITE],
1821               oti->oti_declare_ops[OSD_OT_PUNCH],
1822               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1823               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1824               oti->oti_declare_ops[OSD_OT_QUOTA],
1825               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1826               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1827         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1828               oti->oti_declare_ops[OSD_OT_INSERT],
1829               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1830               oti->oti_declare_ops_used[OSD_OT_INSERT],
1831               oti->oti_declare_ops[OSD_OT_DELETE],
1832               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1833               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1834         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1835               oti->oti_declare_ops[OSD_OT_REF_ADD],
1836               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1837               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1838               oti->oti_declare_ops[OSD_OT_REF_DEL],
1839               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1840               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1841 }
1842
1843 /*
1844  * Concurrency: shouldn't matter.
1845  */
1846 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1847                            struct thandle *th)
1848 {
1849         struct osd_thread_info *oti = osd_oti_get(env);
1850         struct osd_device *dev = osd_dt_dev(d);
1851         handle_t *jh;
1852         struct osd_thandle *oh;
1853         int rc;
1854
1855         ENTRY;
1856
1857         LASSERT(current->journal_info == NULL);
1858
1859         oh = container_of0(th, struct osd_thandle, ot_super);
1860         LASSERT(oh != NULL);
1861         LASSERT(oh->ot_handle == NULL);
1862
1863         rc = dt_txn_hook_start(env, d, th);
1864         if (rc != 0)
1865                 GOTO(out, rc);
1866
1867         if (unlikely(osd_param_is_not_sane(dev, th))) {
1868                 static unsigned long last_printed;
1869                 static int last_credits;
1870
1871                 /*
1872                  * don't make noise on a tiny testing systems
1873                  * actual credits misuse will be caught anyway
1874                  */
1875                 if (last_credits != oh->ot_credits &&
1876                     time_after(jiffies, last_printed +
1877                                cfs_time_seconds(60)) &&
1878                     osd_transaction_size(dev) > 512) {
1879                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1880                               oh->ot_credits, osd_transaction_size(dev));
1881                         osd_trans_dump_creds(env, th);
1882                         libcfs_debug_dumpstack(NULL);
1883                         last_credits = oh->ot_credits;
1884                         last_printed = jiffies;
1885                 }
1886                 /*
1887                  * XXX Limit the credits to 'max_transaction_buffers', and
1888                  *     let the underlying filesystem to catch the error if
1889                  *     we really need so many credits.
1890                  *
1891                  *     This should be removed when we can calculate the
1892                  *     credits precisely.
1893                  */
1894                 oh->ot_credits = osd_transaction_size(dev);
1895         } else if (ldiskfs_track_declares_assert != 0) {
1896                 /*
1897                  * reserve few credits to prevent an assertion in JBD
1898                  * our debugging mechanism will be able to detected
1899                  * overuse. this can help to debug single-update
1900                  * transactions
1901                  */
1902                 oh->ot_credits += 10;
1903                 if (unlikely(osd_param_is_not_sane(dev, th)))
1904                         oh->ot_credits = osd_transaction_size(dev);
1905         }
1906
1907         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1908                 GOTO(out, rc = -EIO);
1909
1910         /*
1911          * XXX temporary stuff. Some abstraction layer should
1912          * be used.
1913          */
1914         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1915         osd_th_started(oh);
1916         if (!IS_ERR(jh)) {
1917                 oh->ot_handle = jh;
1918                 LASSERT(oti->oti_txns == 0);
1919
1920                 lu_device_get(&d->dd_lu_dev);
1921                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1922                               "osd-tx", th);
1923                 oti->oti_txns++;
1924                 rc = 0;
1925         } else {
1926                 rc = PTR_ERR(jh);
1927         }
1928 out:
1929         RETURN(rc);
1930 }
1931
1932 static int osd_seq_exists(const struct lu_env *env,
1933                           struct osd_device *osd, u64 seq)
1934 {
1935         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1936         struct seq_server_site *ss = osd_seq_site(osd);
1937         int rc;
1938
1939         ENTRY;
1940
1941         LASSERT(ss != NULL);
1942         LASSERT(ss->ss_server_fld != NULL);
1943
1944         rc = osd_fld_lookup(env, osd, seq, range);
1945         if (rc != 0) {
1946                 if (rc != -ENOENT)
1947                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1948                                osd_name(osd), seq, rc);
1949                 RETURN(0);
1950         }
1951
1952         RETURN(ss->ss_node_id == range->lsr_index);
1953 }
1954
1955 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1956 {
1957         struct dt_txn_commit_cb *dcb;
1958         struct dt_txn_commit_cb *tmp;
1959
1960         /* call per-transaction stop callbacks if any */
1961         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1962                                  dcb_linkage) {
1963                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1964                          "commit callback entry: magic=%x name='%s'\n",
1965                          dcb->dcb_magic, dcb->dcb_name);
1966                 list_del_init(&dcb->dcb_linkage);
1967                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1968         }
1969 }
1970
1971 /*
1972  * Concurrency: shouldn't matter.
1973  */
1974 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1975                           struct thandle *th)
1976 {
1977         struct osd_thread_info *oti = osd_oti_get(env);
1978         struct osd_thandle *oh;
1979         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1980         struct osd_device *osd = osd_dt_dev(th->th_dev);
1981         struct qsd_instance *qsd = osd_def_qsd(osd);
1982         struct lquota_trans *qtrans;
1983         LIST_HEAD(truncates);
1984         int rc = 0, remove_agents = 0;
1985
1986         ENTRY;
1987
1988         oh = container_of0(th, struct osd_thandle, ot_super);
1989
1990         remove_agents = oh->ot_remove_agents;
1991
1992         qtrans = oh->ot_quota_trans;
1993         oh->ot_quota_trans = NULL;
1994
1995         /* move locks to local list, stop tx, execute truncates */
1996         list_splice(&oh->ot_trunc_locks, &truncates);
1997
1998         if (oh->ot_handle != NULL) {
1999                 int rc2;
2000
2001                 handle_t *hdl = oh->ot_handle;
2002
2003                 /*
2004                  * add commit callback
2005                  * notice we don't do this in osd_trans_start()
2006                  * as underlying transaction can change during truncate
2007                  */
2008                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2009                                              &oh->ot_jcb);
2010
2011                 LASSERT(oti->oti_txns == 1);
2012                 oti->oti_txns--;
2013
2014                 rc = dt_txn_hook_stop(env, th);
2015                 if (rc != 0)
2016                         CERROR("%s: failed in transaction hook: rc = %d\n",
2017                                osd_name(osd), rc);
2018
2019                 osd_trans_stop_cb(oh, rc);
2020                 /* hook functions might modify th_sync */
2021                 hdl->h_sync = th->th_sync;
2022
2023                 oh->ot_handle = NULL;
2024                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2025                 if (rc2 != 0)
2026                         CERROR("%s: failed to stop transaction: rc = %d\n",
2027                                osd_name(osd), rc2);
2028                 if (!rc)
2029                         rc = rc2;
2030
2031                 osd_process_truncates(&truncates);
2032         } else {
2033                 osd_trans_stop_cb(oh, th->th_result);
2034                 OBD_FREE_PTR(oh);
2035         }
2036
2037         osd_trunc_unlock_all(env, &truncates);
2038
2039         /* inform the quota slave device that the transaction is stopping */
2040         qsd_op_end(env, qsd, qtrans);
2041
2042         /*
2043          * as we want IO to journal and data IO be concurrent, we don't block
2044          * awaiting data IO completion in osd_do_bio(), instead we wait here
2045          * once transaction is submitted to the journal. all reqular requests
2046          * don't do direct IO (except read/write), thus this wait_event becomes
2047          * no-op for them.
2048          *
2049          * IMPORTANT: we have to wait till any IO submited by the thread is
2050          * completed otherwise iobuf may be corrupted by different request
2051          */
2052         wait_event(iobuf->dr_wait,
2053                        atomic_read(&iobuf->dr_numreqs) == 0);
2054         osd_fini_iobuf(osd, iobuf);
2055         if (!rc)
2056                 rc = iobuf->dr_error;
2057
2058         if (unlikely(remove_agents != 0))
2059                 osd_process_scheduled_agent_removals(env, osd);
2060
2061         oti->oti_ins_cache_depth--;
2062         /* reset OI cache for safety */
2063         if (oti->oti_ins_cache_depth == 0)
2064                 oti->oti_ins_cache_used = 0;
2065
2066         sb_end_write(osd_sb(osd));
2067
2068         RETURN(rc);
2069 }
2070
2071 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2072 {
2073         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2074                                                ot_super);
2075
2076         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2077         LASSERT(&dcb->dcb_func != NULL);
2078         if (dcb->dcb_flags & DCB_TRANS_STOP)
2079                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2080         else
2081                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2082
2083         return 0;
2084 }
2085
2086 /*
2087  * Called just before object is freed. Releases all resources except for
2088  * object itself (that is released by osd_object_free()).
2089  *
2090  * Concurrency: no concurrent access is possible that late in object
2091  * life-cycle.
2092  */
2093 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2094 {
2095         struct osd_object *obj = osd_obj(l);
2096         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2097         struct inode *inode = obj->oo_inode;
2098         __u64 projid;
2099         qid_t uid;
2100         qid_t gid;
2101
2102         LINVRNT(osd_invariant(obj));
2103
2104         /*
2105          * If object is unlinked remove fid->ino mapping from object index.
2106          */
2107
2108         osd_index_fini(obj);
2109
2110         if (!inode)
2111                 return;
2112
2113         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2114                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2115
2116         uid = i_uid_read(inode);
2117         gid = i_gid_read(inode);
2118         projid = i_projid_read(inode);
2119
2120         obj->oo_inode = NULL;
2121         iput(inode);
2122
2123         /* do not rebalance quota if the caller needs to release memory
2124          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2125          * transaction and risk to deadlock - LU-12178 */
2126         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2127                 return;
2128
2129         if (!obj->oo_header && qsd) {
2130                 struct osd_thread_info *info = osd_oti_get(env);
2131                 struct lquota_id_info *qi = &info->oti_qi;
2132
2133                 /* Release granted quota to master if necessary */
2134                 qi->lqi_id.qid_uid = uid;
2135                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2136
2137                 qi->lqi_id.qid_uid = gid;
2138                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2139
2140                 qi->lqi_id.qid_uid = projid;
2141                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2142         }
2143 }
2144
2145 /*
2146  * Concurrency: ->loo_object_release() is called under site spin-lock.
2147  */
2148 static void osd_object_release(const struct lu_env *env,
2149                                struct lu_object *l)
2150 {
2151         struct osd_object *o = osd_obj(l);
2152
2153         /*
2154          * nobody should be releasing a non-destroyed object with nlink=0
2155          * the API allows this, but ldiskfs doesn't like and then report
2156          * this inode as deleted
2157          */
2158         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2159                   o->oo_inode->i_nlink == 0));
2160 }
2161
2162 /*
2163  * Concurrency: shouldn't matter.
2164  */
2165 static int osd_object_print(const struct lu_env *env, void *cookie,
2166                             lu_printer_t p, const struct lu_object *l)
2167 {
2168         struct osd_object *o = osd_obj(l);
2169         struct iam_descr *d;
2170
2171         if (o->oo_dir != NULL)
2172                 d = o->oo_dir->od_container.ic_descr;
2173         else
2174                 d = NULL;
2175         return (*p)(env, cookie,
2176                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2177                     o, o->oo_inode,
2178                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2179                     o->oo_inode ? o->oo_inode->i_generation : 0,
2180                     d ? d->id_ops->id_name : "plain");
2181 }
2182
2183 /*
2184  * Concurrency: shouldn't matter.
2185  */
2186 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2187                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2188 {
2189         struct osd_device *osd = osd_dt_dev(d);
2190         struct super_block *sb = osd_sb(osd);
2191         struct kstatfs *ksfs;
2192         __u64 reserved;
2193         int result = 0;
2194
2195         if (unlikely(osd->od_mnt == NULL))
2196                 return -EINPROGRESS;
2197
2198         /* osd_lproc.c call this without env, allocate ksfs for that case */
2199         if (unlikely(env == NULL)) {
2200                 OBD_ALLOC_PTR(ksfs);
2201                 if (ksfs == NULL)
2202                         return -ENOMEM;
2203         } else {
2204                 ksfs = &osd_oti_get(env)->oti_ksfs;
2205         }
2206
2207         result = sb->s_op->statfs(sb->s_root, ksfs);
2208         if (result)
2209                 goto out;
2210
2211         statfs_pack(sfs, ksfs);
2212         if (unlikely(sb->s_flags & SB_RDONLY))
2213                 sfs->os_state |= OS_STATE_READONLY;
2214
2215         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2216
2217         if (ldiskfs_has_feature_extents(sb))
2218                 sfs->os_maxbytes = sb->s_maxbytes;
2219         else
2220                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2221
2222         /*
2223          * Reserve some space so to avoid fragmenting the filesystem too much.
2224          * Fragmentation not only impacts performance, but can also increase
2225          * metadata overhead significantly, causing grant calculation to be
2226          * wrong.
2227          *
2228          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2229          */
2230         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2231         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2232         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2233                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2234
2235         sfs->os_blocks -= reserved;
2236         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2237         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2238
2239 out:
2240         if (unlikely(env == NULL))
2241                 OBD_FREE_PTR(ksfs);
2242         return result;
2243 }
2244
2245 /**
2246  * Estimate space needed for file creations. We assume the largest filename
2247  * which is 2^64 - 1, hence a filename of 20 chars.
2248  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2249  */
2250 #ifdef __LDISKFS_DIR_REC_LEN
2251 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2252 #else
2253 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2254 #endif
2255
2256 /*
2257  * Concurrency: doesn't access mutable data.
2258  */
2259 static void osd_conf_get(const struct lu_env *env,
2260                          const struct dt_device *dev,
2261                          struct dt_device_param *param)
2262 {
2263         struct osd_device *d = osd_dt_dev(dev);
2264         struct super_block *sb = osd_sb(d);
2265         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2266         const char *name;
2267         int ea_overhead;
2268
2269         /*
2270          * XXX should be taken from not-yet-existing fs abstraction layer.
2271          */
2272         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2273         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2274         param->ddp_symlink_max  = sb->s_blocksize;
2275         param->ddp_mount_type   = LDD_MT_LDISKFS;
2276         if (ldiskfs_has_feature_extents(sb))
2277                 param->ddp_maxbytes = sb->s_maxbytes;
2278         else
2279                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2280         /*
2281          * inode are statically allocated, so per-inode space consumption
2282          * is the space consumed by the directory entry
2283          */
2284         param->ddp_inodespace     = PER_OBJ_USAGE;
2285         /*
2286          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2287          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2288          * maximum length to not under-count the actual number of extents
2289          * needed for writing a file if there are sub-optimal block allocations.
2290          */
2291         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2292         /* worst-case extent insertion metadata overhead */
2293         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2294         param->ddp_mntopts = 0;
2295         if (test_opt(sb, XATTR_USER))
2296                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2297         if (test_opt(sb, POSIX_ACL))
2298                 param->ddp_mntopts |= MNTOPT_ACL;
2299
2300         /*
2301          * LOD might calculate the max stripe count based on max_ea_size,
2302          * so we need take account in the overhead as well,
2303          * xattr_header + magic + xattr_entry_head
2304          */
2305         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2306                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2307
2308 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2309         if (ldiskfs_has_feature_ea_inode(sb))
2310                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2311                                                                 ea_overhead;
2312         else
2313 #endif
2314                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2315
2316         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2317                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2318
2319         /*
2320          * Preferred RPC size for efficient disk IO.  4MB shows good
2321          * all-around performance for ldiskfs, but use bigalloc chunk size
2322          * by default if larger.
2323          */
2324 #if defined(LDISKFS_CLUSTER_SIZE)
2325         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2326                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2327         else
2328 #endif
2329                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2330
2331         param->ddp_t10_cksum_type = 0;
2332         if (bi) {
2333                 unsigned short interval = blk_integrity_interval(bi);
2334                 name = blk_integrity_name(bi);
2335                 /*
2336                  * Expected values:
2337                  * T10-DIF-TYPE1-CRC
2338                  * T10-DIF-TYPE3-CRC
2339                  * T10-DIF-TYPE1-IP
2340                  * T10-DIF-TYPE3-IP
2341                  */
2342                 if (strncmp(name, "T10-DIF-TYPE",
2343                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2344                         /* also skip "1/3-" at end */
2345                         const int type_off = sizeof("T10-DIF-TYPE.");
2346                         char type_number = name[type_off - 2];
2347
2348                         if (interval != 512 && interval != 4096) {
2349                                 CERROR("%s: unsupported T10PI sector size %u\n",
2350                                        d->od_svname, interval);
2351                         } else if (type_number != '1' && type_number != '3') {
2352                                 CERROR("%s: unsupported T10PI type %s\n",
2353                                        d->od_svname, name);
2354                         } else if (strcmp(name + type_off, "CRC") == 0) {
2355                                 d->od_t10_type = type_number == '1' ?
2356                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2357                                 param->ddp_t10_cksum_type = interval == 512 ?
2358                                         OBD_CKSUM_T10CRC512 :
2359                                         OBD_CKSUM_T10CRC4K;
2360                         } else if (strcmp(name + type_off, "IP") == 0) {
2361                                 d->od_t10_type = type_number == '1' ?
2362                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2363                                 param->ddp_t10_cksum_type = interval == 512 ?
2364                                         OBD_CKSUM_T10IP512 :
2365                                         OBD_CKSUM_T10IP4K;
2366                         } else {
2367                                 CERROR("%s: unsupported checksum type of "
2368                                        "T10PI type '%s'",
2369                                        d->od_svname, name);
2370                         }
2371
2372                 } else {
2373                         CERROR("%s: unsupported T10PI type '%s'",
2374                                d->od_svname, name);
2375                 }
2376         }
2377 }
2378
2379 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2380 {
2381         return osd_sb(osd_dt_dev(d));
2382 }
2383
2384 /*
2385  * Concurrency: shouldn't matter.
2386  */
2387 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2388 {
2389         int rc;
2390         struct super_block *s = osd_sb(osd_dt_dev(d));
2391         ENTRY;
2392
2393         down_read(&s->s_umount);
2394         rc = s->s_op->sync_fs(s, 1);
2395         up_read(&s->s_umount);
2396
2397         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2398                rc);
2399
2400         return rc;
2401 }
2402
2403 /**
2404  * Start commit for OSD device.
2405  *
2406  * An implementation of dt_commit_async method for OSD device.
2407  * Asychronously starts underlayng fs sync and thereby a transaction
2408  * commit.
2409  *
2410  * \param env environment
2411  * \param d dt device
2412  *
2413  * \see dt_device_operations
2414  */
2415 static int osd_commit_async(const struct lu_env *env,
2416                             struct dt_device *d)
2417 {
2418         struct super_block *s = osd_sb(osd_dt_dev(d));
2419         int rc;
2420
2421         ENTRY;
2422
2423         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2424         down_read(&s->s_umount);
2425         rc = s->s_op->sync_fs(s, 0);
2426         up_read(&s->s_umount);
2427
2428         RETURN(rc);
2429 }
2430
2431 static int (*priv_security_file_alloc)(struct file *file);
2432
2433 int osd_security_file_alloc(struct file *file)
2434 {
2435         if (priv_security_file_alloc)
2436                 return priv_security_file_alloc(file);
2437         return 0;
2438 }
2439
2440 /*
2441  * Concurrency: shouldn't matter.
2442  */
2443 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2444 {
2445         struct super_block *sb = osd_sb(osd_dt_dev(d));
2446         struct block_device *dev = sb->s_bdev;
2447         int rc = -EOPNOTSUPP;
2448
2449         ENTRY;
2450
2451         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2452                osd_dt_dev(d)->od_svname, (long)dev, rc);
2453
2454         RETURN(rc);
2455 }
2456
2457 /**
2458  * Note: we do not count into QUOTA here.
2459  * If we mount with --data_journal we may need more.
2460  */
2461 const int osd_dto_credits_noquota[DTO_NR] = {
2462         /**
2463          * Insert.
2464          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2465          * SINGLEDATA_TRANS_BLOCKS(8)
2466          * XXX Note: maybe iam need more, since iam have more level than
2467          *           EXT3 htree.
2468          */
2469         [DTO_INDEX_INSERT]  = 16,
2470         /**
2471          * Delete
2472          * just modify a single entry, probably merge few within a block
2473          */
2474         [DTO_INDEX_DELETE]  = 1,
2475         /**
2476          * Used for OI scrub
2477          */
2478         [DTO_INDEX_UPDATE]  = 16,
2479         /**
2480          * 4(inode, inode bits, groups, GDT)
2481          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2482          */
2483         [DTO_OBJECT_CREATE] = 4,
2484         /**
2485          * 4(inode, inode bits, groups, GDT)
2486          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2487          */
2488         [DTO_OBJECT_DELETE] = 4,
2489         /**
2490          * Attr set credits (inode)
2491          */
2492         [DTO_ATTR_SET_BASE] = 1,
2493         /**
2494          * Xattr set. The same as xattr of EXT3.
2495          * DATA_TRANS_BLOCKS(14)
2496          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2497          * are also counted in. Do not know why?
2498          */
2499         [DTO_XATTR_SET]     = 14,
2500         /**
2501          * credits for inode change during write.
2502          */
2503         [DTO_WRITE_BASE]    = 3,
2504         /**
2505          * credits for single block write.
2506          */
2507         [DTO_WRITE_BLOCK]   = 14,
2508         /**
2509          * Attr set credits for chown.
2510          * This is extra credits for setattr, and it is null without quota
2511          */
2512         [DTO_ATTR_SET_CHOWN] = 0
2513 };
2514
2515 static const struct dt_device_operations osd_dt_ops = {
2516         .dt_root_get       = osd_root_get,
2517         .dt_statfs         = osd_statfs,
2518         .dt_trans_create   = osd_trans_create,
2519         .dt_trans_start    = osd_trans_start,
2520         .dt_trans_stop     = osd_trans_stop,
2521         .dt_trans_cb_add   = osd_trans_cb_add,
2522         .dt_conf_get       = osd_conf_get,
2523         .dt_mnt_sb_get     = osd_mnt_sb_get,
2524         .dt_sync           = osd_sync,
2525         .dt_ro             = osd_ro,
2526         .dt_commit_async   = osd_commit_async,
2527 };
2528
2529 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2530                           unsigned int role)
2531 {
2532         struct osd_object *obj = osd_dt_obj(dt);
2533         struct osd_thread_info *oti = osd_oti_get(env);
2534
2535         LINVRNT(osd_invariant(obj));
2536
2537         LASSERT(obj->oo_owner != env);
2538         down_read_nested(&obj->oo_sem, role);
2539
2540         LASSERT(obj->oo_owner == NULL);
2541         oti->oti_r_locks++;
2542 }
2543
2544 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2545                            unsigned int role)
2546 {
2547         struct osd_object *obj = osd_dt_obj(dt);
2548         struct osd_thread_info *oti = osd_oti_get(env);
2549
2550         LINVRNT(osd_invariant(obj));
2551
2552         LASSERT(obj->oo_owner != env);
2553         down_write_nested(&obj->oo_sem, role);
2554
2555         LASSERT(obj->oo_owner == NULL);
2556         obj->oo_owner = env;
2557         oti->oti_w_locks++;
2558 }
2559
2560 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2561 {
2562         struct osd_object *obj = osd_dt_obj(dt);
2563         struct osd_thread_info *oti = osd_oti_get(env);
2564
2565         LINVRNT(osd_invariant(obj));
2566
2567         LASSERT(oti->oti_r_locks > 0);
2568         oti->oti_r_locks--;
2569         up_read(&obj->oo_sem);
2570 }
2571
2572 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2573 {
2574         struct osd_object *obj = osd_dt_obj(dt);
2575         struct osd_thread_info *oti = osd_oti_get(env);
2576
2577         LINVRNT(osd_invariant(obj));
2578
2579         LASSERT(obj->oo_owner == env);
2580         LASSERT(oti->oti_w_locks > 0);
2581         oti->oti_w_locks--;
2582         obj->oo_owner = NULL;
2583         up_write(&obj->oo_sem);
2584 }
2585
2586 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2587 {
2588         struct osd_object *obj = osd_dt_obj(dt);
2589
2590         LINVRNT(osd_invariant(obj));
2591
2592         return obj->oo_owner == env;
2593 }
2594
2595 static void osd_inode_getattr(const struct lu_env *env,
2596                               struct inode *inode, struct lu_attr *attr)
2597 {
2598         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2599                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2600                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2601                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2602
2603         attr->la_atime = inode->i_atime.tv_sec;
2604         attr->la_mtime = inode->i_mtime.tv_sec;
2605         attr->la_ctime = inode->i_ctime.tv_sec;
2606         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2607         attr->la_mode    = inode->i_mode;
2608         attr->la_size    = i_size_read(inode);
2609         attr->la_blocks  = inode->i_blocks;
2610         attr->la_uid     = i_uid_read(inode);
2611         attr->la_gid     = i_gid_read(inode);
2612         attr->la_projid  = i_projid_read(inode);
2613         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2614         attr->la_nlink   = inode->i_nlink;
2615         attr->la_rdev    = inode->i_rdev;
2616         attr->la_blksize = 1 << inode->i_blkbits;
2617         attr->la_blkbits = inode->i_blkbits;
2618         /*
2619          * Ext4 did not transfer inherit flags from raw inode
2620          * to inode flags, and ext4 internally test raw inode
2621          * @i_flags directly. Instead of patching ext4, we do it here.
2622          */
2623         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2624                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2625 }
2626
2627 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2628                             u64 *count)
2629 {
2630         struct osd_object *obj = osd_dt_obj(dt);
2631         const struct dt_it_ops *iops;
2632         struct dt_it *it;
2633         int rc;
2634
2635         ENTRY;
2636
2637         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2638         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2639
2640         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2641                 *count = obj->oo_dirent_count;
2642                 RETURN(0);
2643         }
2644
2645         /* directory not initialized yet */
2646         if (!dt->do_index_ops) {
2647                 *count = 0;
2648                 RETURN(0);
2649         }
2650
2651         iops = &dt->do_index_ops->dio_it;
2652         it = iops->init(env, dt, LUDA_64BITHASH);
2653         if (IS_ERR(it))
2654                 RETURN(PTR_ERR(it));
2655
2656         rc = iops->load(env, it, 0);
2657         if (rc < 0) {
2658                 if (rc == -ENODATA) {
2659                         rc = 0;
2660                         *count = 0;
2661                 }
2662                 GOTO(out, rc);
2663         }
2664         if (rc > 0)
2665                 rc = iops->next(env, it);
2666
2667         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2668                 if (rc == -ESTALE)
2669                         continue;
2670
2671                 if (iops->key_size(env, it) == 0)
2672                         continue;
2673
2674                 (*count)++;
2675         }
2676         if (rc == 1) {
2677                 obj->oo_dirent_count = *count;
2678                 rc = 0;
2679         }
2680 out:
2681         iops->put(env, it);
2682         iops->fini(env, it);
2683
2684         RETURN(rc);
2685 }
2686
2687 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2688                         struct lu_attr *attr)
2689 {
2690         struct osd_object *obj = osd_dt_obj(dt);
2691         int rc = 0;
2692
2693         if (unlikely(!dt_object_exists(dt)))
2694                 return -ENOENT;
2695         if (unlikely(obj->oo_destroyed))
2696                 return -ENOENT;
2697
2698         LASSERT(!dt_object_remote(dt));
2699         LINVRNT(osd_invariant(obj));
2700
2701         spin_lock(&obj->oo_guard);
2702         osd_inode_getattr(env, obj->oo_inode, attr);
2703         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2704                 attr->la_valid |= LA_FLAGS;
2705                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2706         }
2707         spin_unlock(&obj->oo_guard);
2708
2709         if (S_ISDIR(obj->oo_inode->i_mode) &&
2710             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2711                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2712
2713         return rc;
2714 }
2715
2716 static int osd_declare_attr_qid(const struct lu_env *env,
2717                                 struct osd_object *obj,
2718                                 struct osd_thandle *oh, long long bspace,
2719                                 qid_t old_id, qid_t new_id, bool enforce,
2720                                 unsigned int type, bool ignore_edquot)
2721 {
2722         int rc;
2723         struct osd_thread_info *info = osd_oti_get(env);
2724         struct lquota_id_info  *qi = &info->oti_qi;
2725
2726         qi->lqi_type = type;
2727         /* inode accounting */
2728         qi->lqi_is_blk = false;
2729
2730         /* one more inode for the new id ... */
2731         qi->lqi_id.qid_uid = new_id;
2732         qi->lqi_space      = 1;
2733         /* Reserve credits for the new id */
2734         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2735         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2736                 rc = 0;
2737         if (rc)
2738                 RETURN(rc);
2739
2740         /* and one less inode for the current id */
2741         qi->lqi_id.qid_uid = old_id;
2742         qi->lqi_space = -1;
2743         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2744         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2745                 rc = 0;
2746         if (rc)
2747                 RETURN(rc);
2748
2749         /* block accounting */
2750         qi->lqi_is_blk = true;
2751
2752         /* more blocks for the new id ... */
2753         qi->lqi_id.qid_uid = new_id;
2754         qi->lqi_space      = bspace;
2755         /*
2756          * Credits for the new uid has been reserved, re-use "obj"
2757          * to save credit reservation.
2758          */
2759         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2760         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2761                 rc = 0;
2762         if (rc)
2763                 RETURN(rc);
2764
2765         /* and finally less blocks for the current uid */
2766         qi->lqi_id.qid_uid = old_id;
2767         qi->lqi_space      = -bspace;
2768         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2769         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2770                 rc = 0;
2771
2772         RETURN(rc);
2773 }
2774
2775 static int osd_declare_attr_set(const struct lu_env *env,
2776                                 struct dt_object *dt,
2777                                 const struct lu_attr *attr,
2778                                 struct thandle *handle)
2779 {
2780         struct osd_thandle *oh;
2781         struct osd_object *obj;
2782         qid_t uid;
2783         qid_t gid;
2784         long long bspace;
2785         int rc = 0;
2786         bool enforce;
2787
2788         ENTRY;
2789
2790         LASSERT(dt != NULL);
2791         LASSERT(handle != NULL);
2792
2793         obj = osd_dt_obj(dt);
2794         LASSERT(osd_invariant(obj));
2795
2796         oh = container_of0(handle, struct osd_thandle, ot_super);
2797         LASSERT(oh->ot_handle == NULL);
2798
2799         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2800                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2801
2802         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2803                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2804
2805         if (attr == NULL || obj->oo_inode == NULL)
2806                 RETURN(rc);
2807
2808         bspace   = obj->oo_inode->i_blocks << 9;
2809         bspace   = toqb(bspace);
2810
2811         /*
2812          * Changing ownership is always preformed by super user, it should not
2813          * fail with EDQUOT unless required explicitly.
2814          *
2815          * We still need to call the osd_declare_qid() to calculate the journal
2816          * credits for updating quota accounting files and to trigger quota
2817          * space adjustment once the operation is completed.
2818          */
2819         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2820                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2821
2822                 if (!ignore_edquot)
2823                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2824                                "(the quota space is %lld)\n",
2825                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2826                                attr->la_gid, bspace);
2827
2828                 /* USERQUOTA */
2829                 uid = i_uid_read(obj->oo_inode);
2830                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2831                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2832                                           attr->la_uid, enforce, USRQUOTA,
2833                                           true);
2834                 if (rc)
2835                         RETURN(rc);
2836
2837                 gid = i_gid_read(obj->oo_inode);
2838                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2839                        attr->la_uid, gid, attr->la_gid);
2840                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2841                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2842                                           attr->la_gid, enforce, GRPQUOTA,
2843                                           ignore_edquot);
2844                 if (rc)
2845                         RETURN(rc);
2846
2847         }
2848 #ifdef HAVE_PROJECT_QUOTA
2849         if (attr->la_valid & LA_PROJID) {
2850                 __u32 projid = i_projid_read(obj->oo_inode);
2851
2852                 enforce = (attr->la_valid & LA_PROJID) &&
2853                                         (attr->la_projid != projid);
2854                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2855                                           (qid_t)projid, (qid_t)attr->la_projid,
2856                                           enforce, PRJQUOTA, true);
2857                 if (rc)
2858                         RETURN(rc);
2859         }
2860 #endif
2861         RETURN(rc);
2862 }
2863
2864 static int osd_inode_setattr(const struct lu_env *env,
2865                              struct inode *inode, const struct lu_attr *attr)
2866 {
2867         __u64 bits = attr->la_valid;
2868
2869         /* Only allow set size for regular file */
2870         if (!S_ISREG(inode->i_mode))
2871                 bits &= ~(LA_SIZE | LA_BLOCKS);
2872
2873         if (bits == 0)
2874                 return 0;
2875
2876         if (bits & LA_ATIME)
2877                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2878         if (bits & LA_CTIME)
2879                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2880         if (bits & LA_MTIME)
2881                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2882         if (bits & LA_SIZE) {
2883                 spin_lock(&inode->i_lock);
2884                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2885                 i_size_write(inode, attr->la_size);
2886                 spin_unlock(&inode->i_lock);
2887         }
2888
2889         /*
2890          * OSD should not change "i_blocks" which is used by quota.
2891          * "i_blocks" should be changed by ldiskfs only.
2892          */
2893         if (bits & LA_MODE)
2894                 inode->i_mode = (inode->i_mode & S_IFMT) |
2895                                 (attr->la_mode & ~S_IFMT);
2896         if (bits & LA_UID)
2897                 i_uid_write(inode, attr->la_uid);
2898         if (bits & LA_GID)
2899                 i_gid_write(inode, attr->la_gid);
2900         if (bits & LA_PROJID)
2901                 i_projid_write(inode, attr->la_projid);
2902         if (bits & LA_NLINK)
2903                 set_nlink(inode, attr->la_nlink);
2904         if (bits & LA_RDEV)
2905                 inode->i_rdev = attr->la_rdev;
2906
2907         if (bits & LA_FLAGS) {
2908                 /* always keep S_NOCMTIME */
2909                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2910                                  S_NOCMTIME;
2911                 /*
2912                  * Ext4 did not transfer inherit flags from
2913                  * @inode->i_flags to raw inode i_flags when writing
2914                  * flags, we do it explictly here.
2915                  */
2916                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2917                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2918                 else
2919                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2920         }
2921         return 0;
2922 }
2923
2924 #ifdef HAVE_PROJECT_QUOTA
2925 static int osd_transfer_project(struct inode *inode, __u32 projid)
2926 {
2927         struct super_block *sb = inode->i_sb;
2928         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2929         int err;
2930         kprojid_t kprojid;
2931         struct ldiskfs_iloc iloc;
2932         struct ldiskfs_inode *raw_inode;
2933         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2934
2935         if (!ldiskfs_has_feature_project(sb)) {
2936                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2937                         == LDISKFS_DEF_PROJID);
2938                 if (projid != LDISKFS_DEF_PROJID)
2939                         return -EOPNOTSUPP;
2940                 else
2941                         return 0;
2942         }
2943
2944         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2945                 return -EOPNOTSUPP;
2946
2947         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2948         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2949                 return 0;
2950
2951         err = ldiskfs_get_inode_loc(inode, &iloc);
2952         if (err)
2953                 return err;
2954
2955         raw_inode = ldiskfs_raw_inode(&iloc);
2956         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2957                 err = -EOVERFLOW;
2958                 brelse(iloc.bh);
2959                 return err;
2960         }
2961         brelse(iloc.bh);
2962
2963         dquot_initialize(inode);
2964         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2965         if (transfer_to[PRJQUOTA]) {
2966                 err = __dquot_transfer(inode, transfer_to);
2967                 dqput(transfer_to[PRJQUOTA]);
2968                 if (err)
2969                         return err;
2970         }
2971
2972         return err;
2973 }
2974 #endif
2975
2976 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2977 {
2978         int rc;
2979
2980         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2981             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2982                 struct iattr iattr;
2983
2984                 CDEBUG(D_QUOTA,
2985                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2986                        inode->i_ino, i_uid_read(inode), attr->la_uid,
2987                        i_gid_read(inode), attr->la_gid);
2988
2989                 dquot_initialize(inode);
2990                 iattr.ia_valid = 0;
2991                 if (attr->la_valid & LA_UID)
2992                         iattr.ia_valid |= ATTR_UID;
2993                 if (attr->la_valid & LA_GID)
2994                         iattr.ia_valid |= ATTR_GID;
2995                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2996                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2997
2998                 rc = dquot_transfer(inode, &iattr);
2999                 if (rc) {
3000                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
3001                                "enforcement enabled on the ldiskfs "
3002                                "filesystem?\n", inode->i_sb->s_id, rc);
3003                         return rc;
3004                 }
3005         }
3006
3007         /* Handle project id transfer here properly */
3008         if (attr->la_valid & LA_PROJID &&
3009             attr->la_projid != i_projid_read(inode)) {
3010 #ifdef HAVE_PROJECT_QUOTA
3011                 rc = osd_transfer_project(inode, attr->la_projid);
3012 #else
3013                 rc = -ENOTSUPP;
3014 #endif
3015                 if (rc) {
3016                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3017                                "enforcement enabled on the ldiskfs "
3018                                "filesystem?\n", inode->i_sb->s_id, rc);
3019                         return rc;
3020                 }
3021         }
3022         return 0;
3023 }
3024
3025 static int osd_attr_set(const struct lu_env *env,
3026                         struct dt_object *dt,
3027                         const struct lu_attr *attr,
3028                         struct thandle *handle)
3029 {
3030         struct osd_object *obj = osd_dt_obj(dt);
3031         struct inode *inode;
3032         int rc;
3033
3034         if (!dt_object_exists(dt))
3035                 return -ENOENT;
3036
3037         LASSERT(handle != NULL);
3038         LASSERT(!dt_object_remote(dt));
3039         LASSERT(osd_invariant(obj));
3040
3041         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3042
3043         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3044             !osd_obj2dev(obj)->od_is_ost) {
3045                 struct osd_thread_info *oti = osd_oti_get(env);
3046                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3047                 struct lu_fid *fid1 = &oti->oti_fid;
3048                 struct osd_inode_id *id = &oti->oti_id;
3049                 struct iam_path_descr *ipd;
3050                 struct iam_container *bag;
3051                 struct osd_thandle *oh;
3052                 int rc;
3053
3054                 fid_cpu_to_be(fid1, fid0);
3055                 memset(id, 1, sizeof(*id));
3056                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3057                                   fid0)->oi_dir.od_container;
3058                 ipd = osd_idx_ipd_get(env, bag);
3059                 if (unlikely(ipd == NULL))
3060                         RETURN(-ENOMEM);
3061
3062                 oh = container_of0(handle, struct osd_thandle, ot_super);
3063                 rc = iam_update(oh->ot_handle, bag,
3064                                 (const struct iam_key *)fid1,
3065                                 (const struct iam_rec *)id, ipd);
3066                 osd_ipd_put(env, bag, ipd);
3067                 return(rc > 0 ? 0 : rc);
3068         }
3069
3070         inode = obj->oo_inode;
3071
3072         rc = osd_quota_transfer(inode, attr);
3073         if (rc)
3074                 return rc;
3075
3076         spin_lock(&obj->oo_guard);
3077         rc = osd_inode_setattr(env, inode, attr);
3078         spin_unlock(&obj->oo_guard);
3079         if (rc != 0)
3080                 GOTO(out, rc);
3081
3082         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3083
3084         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3085
3086         if (!(attr->la_valid & LA_FLAGS))
3087                 GOTO(out, rc);
3088
3089         /* Let's check if there are extra flags need to be set into LMA */
3090         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3091                 struct osd_thread_info *info = osd_oti_get(env);
3092                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3093
3094                 LASSERT(!obj->oo_pfid_in_lma);
3095
3096                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3097                                  &info->oti_ost_attrs);
3098                 if (rc)
3099                         GOTO(out, rc);
3100
3101                 lma->lma_incompat |=
3102                         lustre_to_lma_flags(attr->la_flags);
3103                 lustre_lma_swab(lma);
3104
3105                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3106
3107                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3108                                      lma, sizeof(*lma), XATTR_REPLACE);
3109                 if (rc != 0) {
3110                         struct osd_device *osd = osd_obj2dev(obj);
3111
3112                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3113                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3114                               lma->lma_incompat, rc);
3115                 } else {
3116                         obj->oo_lma_flags =
3117                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3118                 }
3119                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3120         }
3121 out:
3122
3123         return rc;
3124 }
3125
3126 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3127                                            struct osd_object *obj,
3128                                            const char *name, const int namelen)
3129 {
3130         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3131 }
3132
3133 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3134                       umode_t mode, struct dt_allocation_hint *hint,
3135                       struct thandle *th, struct lu_attr *attr)
3136 {
3137         int result;
3138         struct osd_device *osd = osd_obj2dev(obj);
3139         struct osd_thandle *oth;
3140         struct dt_object *parent = NULL;
3141         struct inode *inode;
3142         uid_t owner[2] = {0, 0};
3143
3144         if (attr->la_valid & LA_UID)
3145                 owner[0] = attr->la_uid;
3146         if (attr->la_valid & LA_GID)
3147                 owner[1] = attr->la_gid;
3148
3149         LINVRNT(osd_invariant(obj));
3150         LASSERT(obj->oo_inode == NULL);
3151         LASSERT(obj->oo_hl_head == NULL);
3152
3153         if (S_ISDIR(mode) && ldiskfs_pdo) {
3154                 obj->oo_hl_head =
3155                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3156                 if (obj->oo_hl_head == NULL)
3157                         return -ENOMEM;
3158         }
3159
3160         oth = container_of(th, struct osd_thandle, ot_super);
3161         LASSERT(oth->ot_handle->h_transaction != NULL);
3162
3163         if (hint != NULL && hint->dah_parent != NULL &&
3164             !dt_object_remote(hint->dah_parent))
3165                 parent = hint->dah_parent;
3166
3167         inode = ldiskfs_create_inode(oth->ot_handle,
3168                                      parent ? osd_dt_obj(parent)->oo_inode :
3169                                               osd_sb(osd)->s_root->d_inode,
3170                                      mode, owner);
3171         if (!IS_ERR(inode)) {
3172                 /* Do not update file c/mtime in ldiskfs. */
3173                 inode->i_flags |= S_NOCMTIME;
3174
3175                 /*
3176                  * For new created object, it must be consistent,
3177                  * and it is unnecessary to scrub against it.
3178                  */
3179                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3180
3181                 obj->oo_inode = inode;
3182                 result = 0;
3183         } else {
3184                 if (obj->oo_hl_head != NULL) {
3185                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3186                         obj->oo_hl_head = NULL;
3187                 }
3188                 result = PTR_ERR(inode);
3189         }
3190         LINVRNT(osd_invariant(obj));
3191         return result;
3192 }
3193
3194 enum {
3195         OSD_NAME_LEN = 255
3196 };
3197
3198 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3199                      struct lu_attr *attr,
3200                      struct dt_allocation_hint *hint,
3201                      struct dt_object_format *dof,
3202                      struct thandle *th)
3203 {
3204         int result;
3205         struct osd_thandle *oth;
3206         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3207
3208         LASSERT(S_ISDIR(attr->la_mode));
3209
3210         oth = container_of(th, struct osd_thandle, ot_super);
3211         LASSERT(oth->ot_handle->h_transaction != NULL);
3212         if (fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)))
3213                 obj->oo_dirent_count = 0;
3214         result = osd_mkfile(info, obj, mode, hint, th, attr);
3215
3216         return result;
3217 }
3218
3219 static int osd_mk_index(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         int result;
3226         struct osd_thandle *oth;
3227         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3228
3229         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3230
3231         LASSERT(S_ISREG(attr->la_mode));
3232
3233         oth = container_of(th, struct osd_thandle, ot_super);
3234         LASSERT(oth->ot_handle->h_transaction != NULL);
3235
3236         result = osd_mkfile(info, obj, mode, hint, th, attr);
3237         if (result == 0) {
3238                 LASSERT(obj->oo_inode != NULL);
3239                 if (feat->dif_flags & DT_IND_VARKEY)
3240                         result = iam_lvar_create(obj->oo_inode,
3241                                                  feat->dif_keysize_max,
3242                                                  feat->dif_ptrsize,
3243                                                  feat->dif_recsize_max,
3244                                                  oth->ot_handle);
3245                 else
3246                         result = iam_lfix_create(obj->oo_inode,
3247                                                  feat->dif_keysize_max,
3248                                                  feat->dif_ptrsize,
3249                                                  feat->dif_recsize_max,
3250                                                  oth->ot_handle);
3251         }
3252         return result;
3253 }
3254
3255 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3256                      struct lu_attr *attr,
3257                      struct dt_allocation_hint *hint,
3258                      struct dt_object_format *dof,
3259                      struct thandle *th)
3260 {
3261         LASSERT(S_ISREG(attr->la_mode));
3262         return osd_mkfile(info, obj, (attr->la_mode &
3263                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3264                           attr);
3265 }
3266
3267 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3268                      struct lu_attr *attr,
3269                      struct dt_allocation_hint *hint,
3270                      struct dt_object_format *dof,
3271                      struct thandle *th)
3272 {
3273         LASSERT(S_ISLNK(attr->la_mode));
3274         return osd_mkfile(info, obj, (attr->la_mode &
3275                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3276                           attr);
3277 }
3278
3279 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3280                      struct lu_attr *attr,
3281                      struct dt_allocation_hint *hint,
3282                      struct dt_object_format *dof,
3283                      struct thandle *th)
3284 {
3285         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3286         int result;
3287
3288         LINVRNT(osd_invariant(obj));
3289         LASSERT(obj->oo_inode == NULL);
3290         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3291                 S_ISFIFO(mode) || S_ISSOCK(mode));
3292
3293         result = osd_mkfile(info, obj, mode, hint, th, attr);
3294         if (result == 0) {
3295                 LASSERT(obj->oo_inode != NULL);
3296                 /*
3297                  * This inode should be marked dirty for i_rdev.  Currently
3298                  * that is done in the osd_attr_init().
3299                  */
3300                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3301                                    attr->la_rdev);
3302         }
3303         LINVRNT(osd_invariant(obj));
3304         return result;
3305 }
3306
3307 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3308                               struct lu_attr *,
3309                               struct dt_allocation_hint *hint,
3310                               struct dt_object_format *dof,
3311                               struct thandle *);
3312
3313 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3314 {
3315         osd_obj_type_f result;
3316
3317         switch (type) {
3318         case DFT_DIR:
3319                 result = osd_mkdir;
3320                 break;
3321         case DFT_REGULAR:
3322                 result = osd_mkreg;
3323                 break;
3324         case DFT_SYM:
3325                 result = osd_mksym;
3326                 break;
3327         case DFT_NODE:
3328                 result = osd_mknod;
3329                 break;
3330         case DFT_INDEX:
3331                 result = osd_mk_index;
3332                 break;
3333
3334         default:
3335                 LBUG();
3336                 break;
3337         }
3338         return result;
3339 }
3340
3341
3342 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3343                         struct dt_object *parent, struct dt_object *child,
3344                         umode_t child_mode)
3345 {
3346         LASSERT(ah);
3347
3348         ah->dah_parent = parent;
3349         ah->dah_mode = child_mode;
3350
3351         if (parent != NULL && !dt_object_remote(parent)) {
3352                 /* will help to find FID->ino at dt_insert("..") */
3353                 struct osd_object *pobj = osd_dt_obj(parent);
3354
3355                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3356         }
3357 }
3358
3359 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3360                           struct lu_attr *attr, struct dt_object_format *dof)
3361 {
3362         struct inode *inode = obj->oo_inode;
3363         __u64 valid = attr->la_valid;
3364         int result;
3365
3366         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3367
3368         if (dof->dof_type != DFT_NODE)
3369                 attr->la_valid &= ~LA_RDEV;
3370         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3371                 attr->la_valid &= ~LA_ATIME;
3372         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3373                 attr->la_valid &= ~LA_CTIME;
3374         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3375                 attr->la_valid &= ~LA_MTIME;
3376
3377         result = osd_quota_transfer(inode, attr);
3378         if (result)
3379                 return;
3380
3381         if (attr->la_valid != 0) {
3382                 result = osd_inode_setattr(info->oti_env, inode, attr);
3383                 /*
3384                  * The osd_inode_setattr() should always succeed here.  The
3385                  * only error that could be returned is EDQUOT when we are
3386                  * trying to change the UID or GID of the inode. However, this
3387                  * should not happen since quota enforcement is no longer
3388                  * enabled on ldiskfs (lquota takes care of it).
3389                  */
3390                 LASSERTF(result == 0, "%d\n", result);
3391                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3392         }
3393
3394         attr->la_valid = valid;
3395 }
3396
3397 /**
3398  * Helper function for osd_create()
3399  *
3400  * \retval 0, on success
3401  */
3402 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3403                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3404                         struct dt_object_format *dof, struct thandle *th)
3405 {
3406         int result;
3407         __u32 umask;
3408
3409         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3410
3411         /* we drop umask so that permissions we pass are not affected */
3412         umask = current->fs->umask;
3413         current->fs->umask = 0;
3414
3415         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3416                                                   th);
3417         if (likely(obj->oo_inode != NULL)) {
3418                 LASSERT(obj->oo_inode->i_state & I_NEW);
3419
3420                 /*
3421                  * Unlock the inode before attr initialization to avoid
3422                  * unnecessary dqget operations. LU-6378
3423                  */
3424                 unlock_new_inode(obj->oo_inode);
3425         }
3426
3427         if (likely(result == 0)) {
3428                 osd_attr_init(info, obj, attr, dof);
3429                 osd_object_init0(obj);
3430         }
3431
3432         /* restore previous umask value */
3433         current->fs->umask = umask;
3434
3435         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3436
3437         return result;
3438 }
3439
3440 /**
3441  * Helper function for osd_create()
3442  *
3443  * \retval 0, on success
3444  */
3445 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3446                            const struct lu_fid *fid, struct thandle *th)
3447 {
3448         struct osd_thread_info *info = osd_oti_get(env);
3449         struct osd_inode_id    *id   = &info->oti_id;
3450         struct osd_device      *osd  = osd_obj2dev(obj);
3451         struct osd_thandle     *oh;
3452         int rc;
3453
3454         LASSERT(obj->oo_inode != NULL);
3455
3456         oh = container_of0(th, struct osd_thandle, ot_super);
3457         LASSERT(oh->ot_handle);
3458         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3459
3460         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3461         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3462                            OI_CHECK_FLD, NULL);
3463         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3464                 struct lu_fid next_fid = *fid;
3465
3466                 /* insert next object in advance, and map to the same inode */
3467                 next_fid.f_oid++;
3468                 if (next_fid.f_oid != 0) {
3469                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3470                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3471                                       OI_CHECK_FLD, NULL);
3472                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3473                 }
3474         }
3475
3476         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3477
3478         return rc;
3479 }
3480
3481 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3482                    u64 seq, struct lu_seq_range *range)
3483 {
3484         struct seq_server_site *ss = osd_seq_site(osd);
3485
3486         if (fid_seq_is_idif(seq)) {
3487                 fld_range_set_ost(range);
3488                 range->lsr_index = idif_ost_idx(seq);
3489                 return 0;
3490         }
3491
3492         if (!fid_seq_in_fldb(seq)) {
3493                 fld_range_set_mdt(range);
3494                 if (ss != NULL)
3495                         /*
3496                          * FIXME: If ss is NULL, it suppose not get lsr_index
3497                          * at all
3498                          */
3499                         range->lsr_index = ss->ss_node_id;
3500                 return 0;
3501         }
3502
3503         LASSERT(ss != NULL);
3504         fld_range_set_any(range);
3505         /* OSD will only do local fld lookup */
3506         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3507 }
3508
3509 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3510                               struct lu_attr *attr,
3511                               struct dt_allocation_hint *hint,
3512                               struct dt_object_format *dof,
3513                               struct thandle *handle)
3514 {
3515         struct osd_thandle *oh;
3516         int rc;
3517
3518         ENTRY;
3519
3520         LASSERT(handle != NULL);
3521
3522         oh = container_of0(handle, struct osd_thandle, ot_super);
3523         LASSERT(oh->ot_handle == NULL);
3524
3525         /*
3526          * EA object consumes more credits than regular object: osd_mk_index
3527          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3528          * leaf_node, could involves the block, block bitmap, groups, GDT
3529          * change for each block, so add 4 * 2 credits in that case.
3530          */
3531         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3532                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3533                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3534         /*
3535          * Reuse idle OI block may cause additional one OI block
3536          * to be changed.
3537          */
3538         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3539                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3540         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3541                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3542                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3543
3544         /* will help to find FID->ino mapping at dt_insert() */
3545         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3546                                    osd_dt_obj(dt));
3547         if (rc != 0)
3548                 RETURN(rc);
3549
3550         if (!attr)
3551                 RETURN(0);
3552
3553         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3554                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3555                                    NULL, OSD_QID_INODE);
3556         if (rc != 0)
3557                 RETURN(rc);
3558
3559         RETURN(rc);
3560 }
3561
3562 /**
3563  * Called to destroy on-disk representation of the object
3564  *
3565  * Concurrency: must be locked
3566  */
3567 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3568                                struct thandle *th)
3569 {
3570         struct osd_object *obj = osd_dt_obj(dt);
3571         struct inode *inode = obj->oo_inode;
3572         struct osd_thandle *oh;
3573         int rc;
3574
3575         ENTRY;
3576
3577         if (inode == NULL)
3578                 RETURN(-ENOENT);
3579
3580         oh = container_of0(th, struct osd_thandle, ot_super);
3581         LASSERT(oh->ot_handle == NULL);
3582
3583         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3584                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3585
3586         /* For removing agent entry */
3587         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3588                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3589
3590         /*
3591          * Recycle idle OI leaf may cause additional three OI blocks
3592          * to be changed.
3593          */
3594         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3595                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3596                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3597         /* one less inode */
3598         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3599                                    i_projid_read(inode), -1, oh, obj, NULL,
3600                                    OSD_QID_INODE);
3601         if (rc)
3602                 RETURN(rc);
3603         /* data to be truncated */
3604         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3605                                    i_projid_read(inode), 0, oh, obj, NULL,
3606                                    OSD_QID_BLK);
3607         if (rc)
3608                 RETURN(rc);
3609
3610         /*
3611          * will help to find FID->ino when this object is being
3612          * added to PENDING
3613          */
3614         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3615
3616         RETURN(rc);
3617 }
3618
3619 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3620                        struct thandle *th)
3621 {
3622         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3623         struct osd_object *obj = osd_dt_obj(dt);
3624         struct inode *inode = obj->oo_inode;
3625         struct osd_device *osd = osd_obj2dev(obj);
3626         struct osd_thandle *oh;
3627         int result;
3628
3629         ENTRY;
3630
3631         oh = container_of0(th, struct osd_thandle, ot_super);
3632         LASSERT(oh->ot_handle);
3633         LASSERT(inode);
3634         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3635
3636         if (unlikely(fid_is_acct(fid)))
3637                 RETURN(-EPERM);
3638
3639         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3640                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3641                 if (result != 0)
3642                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3643                                osd_name(osd), PFID(fid), result);
3644         }
3645
3646         if (S_ISDIR(inode->i_mode)) {
3647                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3648                         inode->i_nlink == 2);
3649
3650                 spin_lock(&obj->oo_guard);
3651                 clear_nlink(inode);
3652                 spin_unlock(&obj->oo_guard);
3653                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3654         }
3655
3656         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3657
3658         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3659
3660         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3661                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3662                                        oh->ot_handle, OI_CHECK_FLD);
3663
3664         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3665         /* XXX: add to ext3 orphan list */
3666         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3667
3668         /* not needed in the cache anymore */
3669         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3670         obj->oo_destroyed = 1;
3671
3672         RETURN(0);
3673 }
3674
3675 /**
3676  * Put the fid into lustre_mdt_attrs, and then place the structure
3677  * inode's ea. This fid should not be altered during the life time
3678  * of the inode.
3679  *
3680  * \retval +ve, on success
3681  * \retval -ve, on error
3682  *
3683  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3684  */
3685 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3686                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3687 {
3688         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3689         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3690         int rc;
3691
3692         ENTRY;
3693
3694         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3695                 RETURN(0);
3696
3697         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3698                 rc = -ENOMEM;
3699
3700         lustre_loa_init(loa, fid, compat, incompat);
3701         lustre_loa_swab(loa, false);
3702
3703         /*
3704          * For the OST device with 256 bytes inode size by default,
3705          * the PFID EA will be stored together with LMA EA to avoid
3706          * performance trouble. Otherwise the PFID EA can be stored
3707          * independently. LU-8998
3708          */
3709         if ((compat & LMAC_FID_ON_OST) &&
3710             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3711                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3712                                      sizeof(*loa), XATTR_CREATE);
3713         else
3714                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3715                                      sizeof(*lma), XATTR_CREATE);
3716         /*
3717          * LMA may already exist, but we need to check that all the
3718          * desired compat/incompat flags have been added.
3719          */
3720         if (unlikely(rc == -EEXIST)) {
3721                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3722                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3723                 if (rc < 0)
3724                         RETURN(rc);
3725
3726                 if (rc < sizeof(*lma))
3727                         RETURN(-EINVAL);
3728
3729                 lustre_loa_swab(loa, true);
3730                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3731                     ((compat == 0 && incompat == 0) ||
3732                      (!(~lma->lma_compat & compat) &&
3733                       !(~lma->lma_incompat & incompat))))
3734                         RETURN(0);
3735
3736                 lma->lma_self_fid = *fid;
3737                 lma->lma_compat |= compat;
3738                 lma->lma_incompat |= incompat;
3739                 if (rc == sizeof(*lma)) {
3740                         lustre_lma_swab(lma);
3741                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3742                                              sizeof(*lma), XATTR_REPLACE);
3743                 } else {
3744                         lustre_loa_swab(loa, false);
3745                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3746                                              sizeof(*loa), XATTR_REPLACE);
3747                 }
3748         }
3749
3750         RETURN(rc);
3751 }
3752
3753 /**
3754  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3755  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3756  * To have compatilibility with 1.8 ldiskfs driver we need to have
3757  * magic number at start of fid data.
3758  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3759  * its inmemory API.
3760  */
3761 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3762                                          const struct lu_fid *fid)
3763 {
3764         if (!fid_is_namespace_visible(fid) ||
3765             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3766                 param->edp_magic = 0;
3767                 return;
3768         }
3769
3770         param->edp_magic = LDISKFS_LUFID_MAGIC;
3771         param->edp_len =  sizeof(struct lu_fid) + 1;
3772         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3773 }
3774
3775 /**
3776  * Try to read the fid from inode ea into dt_rec.
3777  *
3778  * \param fid object fid.
3779  *
3780  * \retval 0 on success
3781  */
3782 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3783                           __u32 ino, struct lu_fid *fid,
3784                           struct osd_inode_id *id)
3785 {
3786         struct osd_thread_info *info  = osd_oti_get(env);
3787         struct inode *inode;
3788
3789         ENTRY;
3790
3791         osd_id_gen(id, ino, OSD_OII_NOGEN);
3792         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3793         if (IS_ERR(inode))
3794                 RETURN(PTR_ERR(inode));
3795
3796         iput(inode);
3797         RETURN(0);
3798 }
3799
3800 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3801                                         struct inode *dir,
3802                                         struct inode *parent_dir,
3803                                         const struct lu_fid *dot_fid,
3804                                         const struct lu_fid *dot_dot_fid,
3805                                         struct osd_thandle *oth)
3806 {
3807         struct ldiskfs_dentry_param *dot_ldp;
3808         struct ldiskfs_dentry_param *dot_dot_ldp;
3809         __u32 saved_nlink = dir->i_nlink;
3810         int rc;
3811
3812         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3813         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3814
3815         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3816         dot_ldp->edp_magic = 0;
3817
3818         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3819                                     dir, dot_ldp, dot_dot_ldp);
3820         /*
3821          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3822          * the subseqent ref_add() will increase the dir->i_nlink
3823          * as 3. That is incorrect for new created directory.
3824          *
3825          * It looks like hack, because we want to make the OSD API
3826          * to be order-independent for new created directory object
3827          * between dt_insert(..) and ref_add() operations.
3828          *
3829          * Here, we only restore the in-RAM dir-inode's nlink attr,
3830          * becuase if the nlink attr is not 2, then there will be
3831          * ref_add() called following the dt_insert(..), such call
3832          * will make both the in-RAM and on-disk dir-inode's nlink
3833          * attr to be set as 2. LU-7447
3834          */
3835         set_nlink(dir, saved_nlink);
3836         return rc;
3837 }
3838
3839 /**
3840  * Create an local agent inode for remote entry
3841  */
3842 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3843                                                   struct osd_device *osd,
3844                                                   struct osd_object *pobj,
3845                                                   const struct lu_fid *fid,
3846                                                   __u32 type,
3847                                                   struct thandle *th)
3848 {
3849         struct osd_thread_info *info = osd_oti_get(env);
3850         struct inode *local;
3851         struct osd_thandle *oh;
3852         int rc;
3853
3854         ENTRY;
3855
3856         LASSERT(th);
3857         oh = container_of(th, struct osd_thandle, ot_super);
3858         LASSERT(oh->ot_handle->h_transaction != NULL);
3859
3860         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3861                                      NULL);
3862         if (IS_ERR(local)) {
3863                 CERROR("%s: create local error %d\n", osd_name(osd),
3864                        (int)PTR_ERR(local));
3865                 RETURN(local);
3866         }
3867
3868         /*
3869          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3870          * correct gid on remote file, not agent here
3871          */
3872         local->i_gid = current_fsgid();
3873         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3874
3875         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3876          * That gives e2fsck something to look at and be happy, and allows
3877          * debugging if we need to determine where this symlink came from.
3878          */
3879         if (S_ISLNK(type)) {
3880                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3881                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3882                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3883
3884                 i_size_write(local, rc);
3885                 LDISKFS_I(local)->i_disksize = rc;
3886         }
3887         unlock_new_inode(local);
3888
3889         /* Agent inode should not have project ID */
3890 #ifdef  HAVE_PROJECT_QUOTA
3891         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3892             i_projid_read(pobj->oo_inode) != 0) {
3893                 rc = osd_transfer_project(local, 0);
3894                 if (rc) {
3895                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3896                                "quota enforcement enabled on the ldiskfs "
3897                                "filesystem?\n", local->i_sb->s_id, rc);
3898                         RETURN(ERR_PTR(rc));
3899                 }
3900         }
3901 #endif
3902         /* Set special LMA flag for local agent inode */
3903         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3904         if (rc != 0) {
3905                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3906                        osd_name(osd), PFID(fid), rc);
3907                 RETURN(ERR_PTR(rc));
3908         }
3909
3910         if (!S_ISDIR(type))
3911                 RETURN(local);
3912
3913         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3914                                          lu_object_fid(&pobj->oo_dt.do_lu),
3915                                          fid, oh);
3916         if (rc != 0) {
3917                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3918                         osd_name(osd), PFID(fid), rc);
3919                 RETURN(ERR_PTR(rc));
3920         }
3921
3922         RETURN(local);
3923 }
3924
3925 /**
3926  * when direntry is deleted, we have to take care of possible agent inode
3927  * referenced by that. unfortunately we can't do this at that point:
3928  * iget() within a running transaction leads to deadlock and we better do
3929  * not call that every delete declaration to save performance. so we put
3930  * a potention agent inode on a list and process that once the transaction
3931  * is over. Notice it's not any worse in terms of real orphans as regular
3932  * object destroy doesn't put inodes on the on-disk orphan list. this should
3933  * be addressed separately
3934  */
3935 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3936                                             struct osd_thandle *oh,
3937                                             __u32 ino)
3938 {
3939         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3940         struct osd_obj_orphan *oor;
3941
3942         OBD_ALLOC_PTR(oor);
3943         if (oor == NULL)
3944                 return -ENOMEM;
3945
3946         oor->oor_ino = ino;
3947         oor->oor_env = (struct lu_env *)env;
3948         spin_lock(&osd->od_osfs_lock);
3949         list_add(&oor->oor_list, &osd->od_orphan_list);
3950         spin_unlock(&osd->od_osfs_lock);
3951
3952         oh->ot_remove_agents = 1;
3953
3954         return 0;
3955
3956 }
3957
3958 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3959                                                 struct osd_device *osd)
3960 {
3961         struct osd_thread_info *info = osd_oti_get(env);
3962         struct osd_obj_orphan *oor, *tmp;
3963         struct osd_inode_id id;
3964         LIST_HEAD(list);
3965         struct inode *inode;
3966         struct lu_fid fid;
3967         handle_t *jh;
3968         __u32 ino;
3969
3970         spin_lock(&osd->od_osfs_lock);
3971         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3972                 if (oor->oor_env == env)
3973                         list_move(&oor->oor_list, &list);
3974         }
3975         spin_unlock(&osd->od_osfs_lock);
3976
3977         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3978
3979                 ino = oor->oor_ino;
3980
3981                 list_del(&oor->oor_list);
3982                 OBD_FREE_PTR(oor);
3983
3984                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3985                 inode = osd_iget_fid(info, osd, &id, &fid);
3986                 if (IS_ERR(inode))
3987                         continue;
3988
3989                 if (!osd_remote_fid(env, osd, &fid)) {
3990                         iput(inode);
3991                         continue;
3992                 }
3993
3994                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3995                 clear_nlink(inode);
3996                 mark_inode_dirty(inode);
3997                 ldiskfs_journal_stop(jh);
3998                 iput(inode);
3999         }
4000
4001         return 0;
4002 }
4003
4004 /**
4005  * OSD layer object create function for OST objects (b=11826).
4006  *
4007  * The FID is inserted into inode xattr here.
4008  *
4009  * \retval   0, on success
4010  * \retval -ve, on error
4011  */
4012 static int osd_create(const struct lu_env *env, struct dt_object *dt,
4013                       struct lu_attr *attr, struct dt_allocation_hint *hint,
4014                       struct dt_object_format *dof, struct thandle *th)
4015 {
4016         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4017         struct osd_object *obj = osd_dt_obj(dt);
4018         struct osd_thread_info *info = osd_oti_get(env);
4019         int result, on_ost = 0;
4020
4021         ENTRY;
4022
4023         if (dt_object_exists(dt))
4024                 RETURN(-EEXIST);
4025
4026         LINVRNT(osd_invariant(obj));
4027         LASSERT(!dt_object_remote(dt));
4028         LASSERT(osd_is_write_locked(env, obj));
4029         LASSERT(th != NULL);
4030
4031         if (unlikely(fid_is_acct(fid)))
4032                 /*
4033                  * Quota files can't be created from the kernel any more,
4034                  * 'tune2fs -O quota' will take care of creating them
4035                  */
4036                 RETURN(-EPERM);
4037
4038         result = __osd_create(info, obj, attr, hint, dof, th);
4039         if (result == 0) {
4040                 if (fid_is_idif(fid) &&
4041                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
4042                         struct lu_fid *tfid = &info->oti_fid;
4043                         struct ost_id *oi   = &info->oti_ostid;
4044
4045                         fid_to_ostid(fid, oi);
4046                         ostid_to_fid(tfid, oi, 0);
4047                         on_ost = 1;
4048                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4049                                                 LMAC_FID_ON_OST, 0);
4050                 } else {
4051                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4052                                                fid, OI_CHECK_FLD);
4053                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4054                                                 on_ost ? LMAC_FID_ON_OST : 0,
4055                                                 0);
4056                 }
4057                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4058                         obj->oo_dt.do_body_ops = &osd_body_ops;
4059         }
4060
4061         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4062                 result = __osd_oi_insert(env, obj, fid, th);
4063
4064         /*
4065          * a small optimization - dt_insert() isn't usually applied
4066          * to OST objects, so we don't need to cache OI mapping for
4067          * OST objects
4068          */
4069         if (result == 0 && on_ost == 0) {
4070                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4071
4072                 result = osd_idc_find_and_init(env, osd, obj);
4073                 LASSERT(result == 0);
4074         }
4075
4076         LASSERT(ergo(result == 0,
4077                      dt_object_exists(dt) && !dt_object_remote(dt)));
4078         LINVRNT(osd_invariant(obj));
4079         RETURN(result);
4080 }
4081
4082 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4083                                struct thandle *handle)
4084 {
4085         struct osd_thandle *oh;
4086
4087         /* it's possible that object doesn't exist yet */
4088         LASSERT(handle != NULL);
4089
4090         oh = container_of0(handle, struct osd_thandle, ot_super);
4091         LASSERT(oh->ot_handle == NULL);
4092
4093         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4094                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4095
4096         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4097
4098         return 0;
4099 }
4100
4101 /*
4102  * Concurrency: @dt is write locked.
4103  */
4104 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4105                        struct thandle *th)
4106 {
4107         struct osd_object *obj = osd_dt_obj(dt);
4108         struct inode *inode = obj->oo_inode;
4109         struct osd_thandle *oh;
4110         int rc = 0;
4111
4112         if (!dt_object_exists(dt) || obj->oo_destroyed)
4113                 return -ENOENT;
4114
4115         LINVRNT(osd_invariant(obj));
4116         LASSERT(!dt_object_remote(dt));
4117         LASSERT(osd_is_write_locked(env, obj));
4118         LASSERT(th != NULL);
4119
4120         oh = container_of0(th, struct osd_thandle, ot_super);
4121         LASSERT(oh->ot_handle != NULL);
4122
4123         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4124
4125         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4126                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4127         /*
4128          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4129          * (65000) subdirectories by storing "1" in i_nlink if the link count
4130          * would otherwise overflow. Directory tranversal tools understand
4131          * that (st_nlink == 1) indicates that the filesystem dose not track
4132          * hard links count on the directory, and will not abort subdirectory
4133          * scanning early once (st_nlink - 2) subdirs have been found.
4134          *
4135          * This also has to properly handle the case of inodes with nlink == 0
4136          * in case they are being linked into the PENDING directory
4137          */
4138         spin_lock(&obj->oo_guard);
4139         if (unlikely(inode->i_nlink == 0))
4140                 /* inc_nlink from 0 may cause WARN_ON */
4141                 set_nlink(inode, 1);
4142         else {
4143                 ldiskfs_inc_count(oh->ot_handle, inode);
4144                 if (!S_ISDIR(inode->i_mode))
4145                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4146         }
4147         spin_unlock(&obj->oo_guard);
4148
4149         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4150         LINVRNT(osd_invariant(obj));
4151
4152         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4153
4154         return rc;
4155 }
4156
4157 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4158                                struct thandle *handle)
4159 {
4160         struct osd_thandle *oh;
4161
4162         if (!dt_object_exists(dt))
4163                 return -ENOENT;
4164
4165         LASSERT(!dt_object_remote(dt));
4166         LASSERT(handle != NULL);
4167
4168         oh = container_of0(handle, struct osd_thandle, ot_super);
4169         LASSERT(oh->ot_handle == NULL);
4170
4171         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4172                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4173
4174         return 0;
4175 }
4176
4177 /*
4178  * Concurrency: @dt is write locked.
4179  */
4180 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4181                        struct thandle *th)
4182 {
4183         struct osd_object *obj = osd_dt_obj(dt);
4184         struct inode *inode = obj->oo_inode;
4185         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4186         struct osd_thandle *oh;
4187
4188         if (!dt_object_exists(dt))
4189                 return -ENOENT;
4190
4191         LINVRNT(osd_invariant(obj));
4192         LASSERT(!dt_object_remote(dt));
4193         LASSERT(osd_is_write_locked(env, obj));
4194         LASSERT(th != NULL);
4195
4196         oh = container_of0(th, struct osd_thandle, ot_super);
4197         LASSERT(oh->ot_handle != NULL);
4198
4199         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4200
4201         spin_lock(&obj->oo_guard);
4202         /*
4203          * That can be result of upgrade from old Lustre version and
4204          * applied only to local files.  Just skip this ref_del call.
4205          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4206          */
4207         if (inode->i_nlink == 0) {
4208                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4209                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4210                              ", maybe an upgraded file? (LU-3915)\n",
4211                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4212                 spin_unlock(&obj->oo_guard);
4213                 return 0;
4214         }
4215
4216         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4217                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4218
4219         ldiskfs_dec_count(oh->ot_handle, inode);
4220         spin_unlock(&obj->oo_guard);
4221
4222         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4223         LINVRNT(osd_invariant(obj));
4224
4225         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4226
4227         return 0;
4228 }
4229
4230 /*
4231  * Concurrency: @dt is read locked.
4232  */
4233 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4234                          struct lu_buf *buf, const char *name)
4235 {
4236         struct osd_object      *obj    = osd_dt_obj(dt);
4237         struct inode           *inode  = obj->oo_inode;
4238         struct osd_thread_info *info   = osd_oti_get(env);
4239         struct dentry          *dentry = &info->oti_obj_dentry;
4240         bool cache_xattr = false;
4241         int rc;
4242
4243         LASSERT(buf);
4244
4245         /* version get is not real XATTR but uses xattr API */
4246         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4247                 dt_obj_version_t *ver = buf->lb_buf;
4248
4249                 /*
4250                  * for version we are just using xattr API but change inode
4251                  * field instead
4252                  */
4253                 if (buf->lb_len == 0)
4254                         return sizeof(dt_obj_version_t);
4255
4256                 if (buf->lb_len < sizeof(dt_obj_version_t))
4257                         return -ERANGE;
4258
4259                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4260                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4261
4262                 *ver = LDISKFS_I(inode)->i_fs_version;
4263
4264                 return sizeof(dt_obj_version_t);
4265         }
4266
4267         if (!dt_object_exists(dt))
4268                 return -ENOENT;
4269
4270         LASSERT(!dt_object_remote(dt));
4271         LASSERT(inode->i_op != NULL);
4272 #ifdef HAVE_IOP_XATTR
4273         LASSERT(inode->i_op->getxattr != NULL);
4274 #endif
4275
4276         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4277             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4278                 cache_xattr = true;
4279
4280         if (cache_xattr) {
4281                 rc = osd_oxc_get(obj, name, buf);
4282                 if (rc != -ENOENT)
4283                         return rc;
4284         }
4285
4286         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4287                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4288                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4289                 struct filter_fid *ff;
4290                 struct ost_layout *ol;
4291
4292                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4293                 if (rc)
4294                         return rc;
4295
4296                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4297
4298                 rc = sizeof(*ff);
4299                 if (buf->lb_len == 0 || !buf->lb_buf)
4300                         return rc;
4301
4302                 if (buf->lb_len < rc)
4303                         return -ERANGE;
4304
4305                 ff = buf->lb_buf;
4306                 ol = &ff->ff_layout;
4307                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4308                                                   PFID_STRIPE_IDX_BITS);
4309                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4310                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4311                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4312                 if (lma->lma_compat & LMAC_COMP_INFO) {
4313                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4314                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4315                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4316                 } else {
4317                         ol->ol_comp_start = 0;
4318                         ol->ol_comp_end = 0;
4319                         ol->ol_comp_id = 0;
4320                 }
4321         } else {
4322                 rc = __osd_xattr_get(inode, dentry, name,
4323                                      buf->lb_buf, buf->lb_len);
4324         }
4325
4326         if (cache_xattr) {
4327                 if (rc == -ENOENT || rc == -ENODATA)
4328                         osd_oxc_add(obj, name, NULL, 0);
4329                 else if (rc > 0 && buf->lb_buf != NULL)
4330                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4331         }
4332
4333         return rc;
4334 }
4335
4336 static int osd_declare_xattr_set(const struct lu_env *env,
4337                                  struct dt_object *dt,
4338                                  const struct lu_buf *buf, const char *name,
4339                                  int fl, struct thandle *handle)
4340 {
4341         struct osd_thandle *oh;
4342         int credits = 0;
4343         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4344
4345         LASSERT(handle != NULL);
4346
4347         oh = container_of0(handle, struct osd_thandle, ot_super);
4348         LASSERT(oh->ot_handle == NULL);
4349
4350         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4351                 /*
4352                  * For non-upgrading case, the LMA is set first and
4353                  * usually fit inode. But for upgrade case, the LMA
4354                  * may be in another separated EA block.
4355                  */
4356                 if (dt_object_exists(dt)) {
4357                         if (fl == LU_XATTR_REPLACE)
4358                                 credits = 1;
4359                         else
4360                                 goto upgrade;
4361                 }
4362         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4363                 credits = 1;
4364         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4365                 /* We may need to delete the old PFID EA. */
4366                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4367                 if (fl == LU_XATTR_REPLACE)
4368                         credits += 1;
4369                 else
4370                         goto upgrade;
4371         } else {
4372                 /*
4373                  * If some name entry resides on remote MDT, then will create
4374                  * agent entry under remote parent. On the other hand, if the
4375                  * remote entry will be removed, then related agent entry may
4376                  * need to be removed from the remote parent. So there may be
4377                  * kinds of cases, let's declare enough credits. The credits
4378                  * for create agent entry is enough for remove case.
4379                  */
4380                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4381                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4382                         if (dt_object_exists(dt))
4383                                 credits += 1; /* For updating LMA */
4384                 }
4385
4386 upgrade:
4387                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4388
4389                 if (buf != NULL) {
4390                         ssize_t buflen;
4391
4392                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4393                                 /*
4394                                  * learn xattr size from osd_xattr_get if
4395                                  * attribute has not been read yet
4396                                  */
4397                                 buflen = __osd_xattr_get(
4398                                     osd_dt_obj(dt)->oo_inode,
4399                                     &osd_oti_get(env)->oti_obj_dentry,
4400                                     name, NULL, 0);
4401                                 if (buflen < 0)
4402                                         buflen = 0;
4403                         } else {
4404                                 buflen = buf->lb_len;
4405                         }
4406
4407                         if (buflen > sb->s_blocksize) {
4408                                 credits += osd_calc_bkmap_credits(
4409                                     sb, NULL, 0, -1,
4410                                     (buflen + sb->s_blocksize - 1) >>
4411                                     sb->s_blocksize_bits);
4412                         }
4413                 }
4414                 /*
4415                  * xattr set may involve inode quota change, reserve credits for
4416                  * dquot_initialize()
4417                  */
4418                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4419         }
4420
4421         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4422
4423         return 0;
4424 }
4425
4426 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4427                               const struct lu_buf *buf, int fl,
4428                               struct thandle *handle)
4429 {
4430         struct osd_thread_info *info = osd_oti_get(env);
4431         struct dentry *dentry = &info->oti_obj_dentry;
4432         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4433         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4434         struct inode *inode = obj->oo_inode;
4435         struct filter_fid *ff = buf->lb_buf;
4436         struct ost_layout *ol = &ff->ff_layout;
4437         int flags = XATTR_REPLACE;
4438         int rc;
4439
4440         ENTRY;
4441
4442         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4443                 RETURN(-EINVAL);
4444
4445         rc = osd_get_lma(info, inode, dentry, loa);
4446         if (rc == -ENODATA) {
4447                 /* Usually for upgarding from old device */
4448                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4449                                 LMAC_FID_ON_OST, 0);
4450                 flags = XATTR_CREATE;
4451         } else if (rc) {
4452                 RETURN(rc);
4453         }
4454
4455         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4456                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4457                         RETURN(-EEXIST);
4458
4459                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4460                         /* Separate PFID EA from LMA */
4461                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4462                         lustre_lma_swab(lma);
4463                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4464                                              sizeof(*lma), XATTR_REPLACE);
4465                         if (!rc) {
4466                                 obj->oo_pfid_in_lma = 0;
4467                                 rc = LU_XATTR_CREATE;
4468                         }
4469
4470                         RETURN(rc);
4471                 }
4472         } else {
4473                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4474                         RETURN(fl);
4475
4476                 /*
4477                  * Old client does not send stripe information,
4478                  * then store the PFID EA on disk separatedly.
4479                  */
4480                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4481                              ol->ol_stripe_size == 0))
4482                         RETURN(fl);
4483
4484                 /* Remove old PFID EA entry firstly. */
4485                 dquot_initialize(inode);
4486                 rc = ll_vfs_removexattr(dentry, inode, XATTR_NAME_FID);
4487                 if (rc == -ENODATA) {
4488                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4489                                 RETURN(rc);
4490                 } else if (rc) {
4491                         RETURN(rc);
4492                 }
4493         }
4494
4495         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4496         if (likely(ol->ol_stripe_size != 0)) {
4497                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4498                                              PFID_STRIPE_IDX_BITS;
4499                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4500                 lma->lma_compat |= LMAC_STRIPE_INFO;
4501                 if (ol->ol_comp_id != 0) {
4502                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4503                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4504                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4505                         lma->lma_compat |= LMAC_COMP_INFO;
4506                 }
4507         }
4508
4509         lustre_loa_swab(loa, false);
4510
4511         /* Store the PFID EA inside LMA. */
4512         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4513                              flags);
4514         if (!rc)
4515                 obj->oo_pfid_in_lma = 1;
4516
4517         RETURN(rc);
4518 }
4519
4520 /*
4521  * In DNE environment, the object (in spite of regular file or directory)
4522  * and its name entry may reside on different MDTs. Under such case, we will
4523  * create an agent entry on the MDT where the object resides. The agent entry
4524  * references the object locally, that makes the object to be visible to the
4525  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4526  * such as 'tar' can handle the object properly.
4527  *
4528  * We handle the agent entry during set linkEA that is the common interface
4529  * for both regular file and directroy, can handle kinds of cases, such as
4530  * create/link/unlink/rename, and so on.
4531  *
4532  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4533  *
4534  * XXX: There are two known issues:
4535  * 1. For one object, we will create at most one agent entry even if there
4536  *    may be more than one cross-MDTs hard links on the object. So the local
4537  *    e2fsck may claim that the object's nlink is larger than the name entries
4538  *    that reference such inode. And in further, the e2fsck will fix the nlink
4539  *    attribute to match the local references. Then it will cause the object's
4540  *    nlink attribute to be inconsistent with the global references. it is bad
4541  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4542  *    other hand, the global namespace LFSCK can repair the object's attribute
4543  *    according to the linkEA.
4544  * 2. There may be too many hard links on the object as to its linkEA overflow,
4545  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4546  *    case happened, then at this point, we do not know whether there are some
4547  *    cross-MDTs reference. But there are local references, it guarantees that
4548  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4549  */
4550 static int osd_xattr_handle_linkea(const struct lu_env *env,
4551                                    struct osd_device *osd,
4552                                    struct osd_object *obj,
4553                                    const struct lu_buf *buf,
4554                                    struct thandle *handle)
4555 {
4556         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4557         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4558         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4559         struct lu_name tmpname;
4560         struct osd_thandle *oh;
4561         int rc;
4562         bool remote = false;
4563
4564         ENTRY;
4565
4566         oh = container_of0(handle, struct osd_thandle, ot_super);
4567         LASSERT(oh->ot_handle != NULL);
4568
4569         rc = linkea_init_with_rec(&ldata);
4570         if (!rc) {
4571                 linkea_first_entry(&ldata);
4572                 while (ldata.ld_lee != NULL && !remote) {
4573                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4574                                             &tmpname, tfid);
4575                         if (osd_remote_fid(env, osd, tfid) > 0)
4576                                 remote = true;
4577                         else
4578                                 linkea_next_entry(&ldata);
4579                 }
4580         } else if (rc == -ENODATA) {
4581                 rc = 0;
4582         } else {
4583                 RETURN(rc);
4584         }
4585
4586         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4587                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4588                 if (rc)
4589                         CERROR("%s: failed to remove agent entry for "DFID
4590                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4591         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4592                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4593                 if (rc)
4594                         CERROR("%s: failed to create agent entry for "DFID
4595                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4596         }
4597
4598         RETURN(rc);
4599 }
4600
4601 /*
4602  * Concurrency: @dt is write locked.
4603  */
4604 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4605                          const struct lu_buf *buf, const char *name, int fl,
4606                          struct thandle *handle)
4607 {
4608         struct osd_object *obj = osd_dt_obj(dt);
4609         struct osd_device *osd = osd_obj2dev(obj);
4610         struct inode *inode = obj->oo_inode;
4611         struct osd_thread_info *info = osd_oti_get(env);
4612         int fs_flags = 0;
4613         int len;
4614         int rc;
4615
4616         ENTRY;
4617
4618         LASSERT(handle);
4619         LASSERT(buf);
4620
4621         /* version set is not real XATTR */
4622         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4623                 dt_obj_version_t *version = buf->lb_buf;
4624
4625                 /*
4626                  * for version we are just using xattr API but change inode
4627                  * field instead
4628                  */
4629                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4630
4631                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4632                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4633
4634                 LDISKFS_I(inode)->i_fs_version = *version;
4635                 /*
4636                  * Version is set after all inode operations are finished,
4637                  * so we should mark it dirty here
4638                  */
4639                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4640
4641                 RETURN(0);
4642         }
4643
4644         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4645                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4646
4647         len = buf->lb_len;
4648         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4649
4650         /*
4651          * For the OST device with 256 bytes inode size by default,
4652          * the PFID EA will be stored together with LMA EA to avoid
4653          * performance trouble. Otherwise the PFID EA can be stored
4654          * independently. LU-8998
4655          */
4656         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4657             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4658                 LASSERT(buf->lb_buf);
4659
4660                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4661                 if (fl <= 0)
4662                         RETURN(fl);
4663         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4664                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4665                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4666
4667                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4668                 if (rc)
4669                         RETURN(rc);
4670
4671                 lma->lma_incompat |= LMAI_STRIPED;
4672                 lustre_lma_swab(lma);
4673                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4674                                      sizeof(*lma), XATTR_REPLACE);
4675                 if (rc != 0)
4676                         RETURN(rc);
4677         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4678                 LASSERT(!osd->od_is_ost);
4679
4680                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4681                 if (rc)
4682                         RETURN(rc);
4683         }
4684
4685         if (fl & LU_XATTR_REPLACE)
4686                 fs_flags |= XATTR_REPLACE;
4687
4688         if (fl & LU_XATTR_CREATE)
4689                 fs_flags |= XATTR_CREATE;
4690
4691         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4692         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4693
4694         if (rc == 0 &&
4695             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4696              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4697                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4698
4699         return rc;
4700 }
4701
4702 /*
4703  * Concurrency: @dt is read locked.
4704  */
4705 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4706                           const struct lu_buf *buf)
4707 {
4708         struct osd_object *obj = osd_dt_obj(dt);
4709         struct inode *inode = obj->oo_inode;
4710         struct osd_thread_info *info = osd_oti_get(env);
4711         struct dentry *dentry = &info->oti_obj_dentry;
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(inode->i_op->listxattr != NULL);
4719
4720         dentry->d_inode = inode;
4721         dentry->d_sb = inode->i_sb;
4722         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4723 }
4724
4725 static int osd_declare_xattr_del(const struct lu_env *env,
4726                                  struct dt_object *dt, const char *name,
4727                                  struct thandle *handle)
4728 {
4729         struct osd_thandle *oh;
4730         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4731
4732         LASSERT(!dt_object_remote(dt));
4733         LASSERT(handle != NULL);
4734
4735         oh = container_of0(handle, struct osd_thandle, ot_super);
4736         LASSERT(oh->ot_handle == NULL);
4737
4738         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4739                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4740         /*
4741          * xattr del may involve inode quota change, reserve credits for
4742          * dquot_initialize()
4743          */
4744         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4745
4746         return 0;
4747 }
4748
4749 /*
4750  * Concurrency: @dt is write locked.
4751  */
4752 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4753                          const char *name, struct thandle *handle)
4754 {
4755         struct osd_object *obj = osd_dt_obj(dt);
4756         struct inode *inode = obj->oo_inode;
4757         struct osd_thread_info *info = osd_oti_get(env);
4758         struct dentry *dentry = &info->oti_obj_dentry;
4759         int rc;
4760
4761         if (!dt_object_exists(dt))
4762                 return -ENOENT;
4763
4764         LASSERT(!dt_object_remote(dt));
4765         LASSERT(inode->i_op != NULL);
4766         LASSERT(handle != NULL);
4767 #ifdef HAVE_IOP_XATTR
4768         LASSERT(inode->i_op->removexattr != NULL);
4769 #endif
4770
4771         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4772
4773         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4774                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4775
4776                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4777                                  &info->oti_ost_attrs);
4778                 if (!rc) {
4779                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4780
4781                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4782                         lustre_lma_swab(lma);
4783                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4784                                              sizeof(*lma), XATTR_REPLACE);
4785                         if (!rc)
4786                                 obj->oo_pfid_in_lma = 0;
4787                 }
4788         } else {
4789                 dquot_initialize(inode);
4790                 dentry->d_inode = inode;
4791                 dentry->d_sb = inode->i_sb;
4792                 rc = ll_vfs_removexattr(dentry, inode, name);
4793         }
4794
4795         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4796
4797         if (rc == 0 &&
4798             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4799              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4800                 osd_oxc_del(obj, name);
4801
4802         return rc;
4803 }
4804
4805 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4806                            __u64 start, __u64 end)
4807 {
4808         struct osd_object *obj = osd_dt_obj(dt);
4809         struct inode *inode = obj->oo_inode;
4810         struct osd_thread_info *info = osd_oti_get(env);
4811         struct dentry *dentry = &info->oti_obj_dentry;
4812         struct file *file = &info->oti_file;
4813         int rc;
4814
4815         ENTRY;
4816
4817         dentry->d_inode = inode;
4818         dentry->d_sb = inode->i_sb;
4819         file->f_path.dentry = dentry;
4820         file->f_mapping = inode->i_mapping;
4821         file->f_op = inode->i_fop;
4822         file->f_inode = inode;
4823
4824         rc = vfs_fsync_range(file, start, end, 0);
4825
4826         RETURN(rc);
4827 }
4828
4829 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4830 {
4831         return 0;
4832 }
4833
4834 /*
4835  * Index operations.
4836  */
4837
4838 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4839                                const struct dt_index_features *feat)
4840 {
4841         struct iam_descr *descr;
4842
4843         if (osd_object_is_root(o))
4844                 return feat == &dt_directory_features;
4845
4846         LASSERT(o->oo_dir != NULL);
4847
4848         descr = o->oo_dir->od_container.ic_descr;
4849         if (feat == &dt_directory_features) {
4850                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4851                         return 1;
4852                 else
4853                         return 0;
4854         } else {
4855                 return feat->dif_keysize_min <= descr->id_key_size &&
4856                        descr->id_key_size <= feat->dif_keysize_max &&
4857                        feat->dif_recsize_min <= descr->id_rec_size &&
4858                        descr->id_rec_size <= feat->dif_recsize_max &&
4859                        !(feat->dif_flags & (DT_IND_VARKEY |
4860                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4861                        ergo(feat->dif_flags & DT_IND_UPDATE,
4862                             1 /* XXX check that object (and fs) is writable */);
4863         }
4864 }
4865
4866 static int osd_iam_container_init(const struct lu_env *env,
4867                                   struct osd_object *obj,
4868                                   struct osd_directory *dir)
4869 {
4870         struct iam_container *bag = &dir->od_container;
4871         int result;
4872
4873         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4874         if (result != 0)
4875                 return result;
4876
4877         result = iam_container_setup(bag);
4878         if (result == 0)
4879                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4880         else
4881                 iam_container_fini(bag);
4882
4883         return result;
4884 }
4885
4886
4887 /*
4888  * Concurrency: no external locking is necessary.
4889  */
4890 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4891                          const struct dt_index_features *feat)
4892 {
4893         int result;
4894         int skip_iam = 0;
4895         struct osd_object *obj = osd_dt_obj(dt);
4896
4897         LINVRNT(osd_invariant(obj));
4898
4899         if (osd_object_is_root(obj)) {
4900                 dt->do_index_ops = &osd_index_ea_ops;
4901                 result = 0;
4902         } else if (feat == &dt_directory_features) {
4903                 dt->do_index_ops = &osd_index_ea_ops;
4904                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4905                         result = 0;
4906                 else
4907                         result = -ENOTDIR;
4908                 skip_iam = 1;
4909         } else if (unlikely(feat == &dt_otable_features)) {
4910                 dt->do_index_ops = &osd_otable_ops;
4911                 return 0;
4912         } else if (unlikely(feat == &dt_acct_features)) {
4913                 dt->do_index_ops = &osd_acct_index_ops;
4914                 result = 0;
4915                 skip_iam = 1;
4916         } else if (!osd_has_index(obj)) {
4917                 struct osd_directory *dir;
4918                 struct osd_device *osd = osd_obj2dev(obj);
4919                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4920
4921                 OBD_ALLOC_PTR(dir);
4922                 if (dir) {
4923
4924                         spin_lock(&obj->oo_guard);
4925                         if (obj->oo_dir == NULL)
4926                                 obj->oo_dir = dir;
4927                         else
4928                                 /*
4929                                  * Concurrent thread allocated container data.
4930                                  */
4931                                 OBD_FREE_PTR(dir);
4932                         spin_unlock(&obj->oo_guard);
4933                         /*
4934                          * Now, that we have container data, serialize its
4935                          * initialization.
4936                          */
4937                         down_write(&obj->oo_ext_idx_sem);
4938                         /*
4939                          * recheck under lock.
4940                          */
4941
4942                         if (osd_has_index(obj)) {
4943                                 result = 0;
4944                                 goto unlock;
4945                         }
4946
4947                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4948                         if (result || feat == &dt_lfsck_namespace_features ||
4949                             feat == &dt_lfsck_layout_orphan_features ||
4950                             feat == &dt_lfsck_layout_dangling_features)
4951                                 goto unlock;
4952
4953                         result = osd_index_register(osd, fid,
4954                                                     feat->dif_keysize_max,
4955                                                     feat->dif_recsize_max);
4956                         if (result < 0)
4957                                 CWARN("%s: failed to register index "
4958                                       DFID": rc = %d\n",
4959                                       osd_name(osd), PFID(fid), result);
4960                         else if (result > 0)
4961                                 result = 0;
4962                         else
4963                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4964                                        " (%d/%d) registered\n",
4965                                        osd_name(osd), PFID(fid),
4966                                        (int)feat->dif_keysize_max,
4967                                        (int)feat->dif_recsize_max);
4968
4969 unlock:
4970                         up_write(&obj->oo_ext_idx_sem);
4971                 } else {
4972                         result = -ENOMEM;
4973                 }
4974         } else {
4975                 result = 0;
4976         }
4977
4978         if (result == 0 && skip_iam == 0) {
4979                 if (!osd_iam_index_probe(env, obj, feat))
4980                         result = -ENOTDIR;
4981         }
4982         LINVRNT(osd_invariant(obj));
4983
4984         return result;
4985 }
4986
4987 static int osd_otable_it_attr_get(const struct lu_env *env,
4988                                  struct dt_object *dt,
4989                                  struct lu_attr *attr)
4990 {
4991         attr->la_valid = 0;
4992         return 0;
4993 }
4994
4995 static const struct dt_object_operations osd_obj_ops = {
4996         .do_read_lock           = osd_read_lock,
4997         .do_write_lock          = osd_write_lock,
4998         .do_read_unlock         = osd_read_unlock,
4999         .do_write_unlock        = osd_write_unlock,
5000         .do_write_locked        = osd_write_locked,
5001         .do_attr_get            = osd_attr_get,
5002         .do_declare_attr_set    = osd_declare_attr_set,
5003         .do_attr_set            = osd_attr_set,
5004         .do_ah_init             = osd_ah_init,
5005         .do_declare_create      = osd_declare_create,
5006         .do_create              = osd_create,
5007         .do_declare_destroy     = osd_declare_destroy,
5008         .do_destroy             = osd_destroy,
5009         .do_index_try           = osd_index_try,
5010         .do_declare_ref_add     = osd_declare_ref_add,
5011         .do_ref_add             = osd_ref_add,
5012         .do_declare_ref_del     = osd_declare_ref_del,
5013         .do_ref_del             = osd_ref_del,
5014         .do_xattr_get           = osd_xattr_get,
5015         .do_declare_xattr_set   = osd_declare_xattr_set,
5016         .do_xattr_set           = osd_xattr_set,
5017         .do_declare_xattr_del   = osd_declare_xattr_del,
5018         .do_xattr_del           = osd_xattr_del,
5019         .do_xattr_list          = osd_xattr_list,
5020         .do_object_sync         = osd_object_sync,
5021         .do_invalidate          = osd_invalidate,
5022 };
5023
5024 static const struct dt_object_operations osd_obj_otable_it_ops = {
5025         .do_attr_get    = osd_otable_it_attr_get,
5026         .do_index_try   = osd_index_try,
5027 };
5028
5029 static int osd_index_declare_iam_delete(const struct lu_env *env,
5030                                         struct dt_object *dt,
5031                                         const struct dt_key *key,
5032                                         struct thandle *handle)
5033 {
5034         struct osd_thandle *oh;
5035
5036         oh = container_of0(handle, struct osd_thandle, ot_super);
5037         LASSERT(oh->ot_handle == NULL);
5038
5039         /* Recycle  may cause additional three blocks to be changed. */
5040         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
5041                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
5042
5043         return 0;
5044 }
5045
5046 /**
5047  *      delete a (key, value) pair from index \a dt specified by \a key
5048  *
5049  *      \param  dt      osd index object
5050  *      \param  key     key for index
5051  *      \param  rec     record reference
5052  *      \param  handle  transaction handler
5053  *
5054  *      \retval  0  success
5055  *      \retval -ve   failure
5056  */
5057 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5058                                 const struct dt_key *key,
5059                                 struct thandle *handle)
5060 {
5061         struct osd_thread_info *oti = osd_oti_get(env);
5062         struct osd_object *obj = osd_dt_obj(dt);
5063         struct osd_thandle *oh;
5064         struct iam_path_descr *ipd;
5065         struct iam_container *bag = &obj->oo_dir->od_container;
5066         int rc;
5067
5068         ENTRY;
5069
5070         if (!dt_object_exists(dt))
5071                 RETURN(-ENOENT);
5072
5073         LINVRNT(osd_invariant(obj));
5074         LASSERT(!dt_object_remote(dt));
5075         LASSERT(bag->ic_object == obj->oo_inode);
5076         LASSERT(handle != NULL);
5077
5078         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5079
5080         ipd = osd_idx_ipd_get(env, bag);
5081         if (unlikely(ipd == NULL))
5082                 RETURN(-ENOMEM);
5083
5084         oh = container_of0(handle, struct osd_thandle, ot_super);
5085         LASSERT(oh->ot_handle != NULL);
5086         LASSERT(oh->ot_handle->h_transaction != NULL);
5087
5088         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5089                 /* swab quota uid/gid provided by caller */
5090                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5091                 key = (const struct dt_key *)&oti->oti_quota_id;
5092         }
5093
5094         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5095         osd_ipd_put(env, bag, ipd);
5096         LINVRNT(osd_invariant(obj));
5097         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5098         RETURN(rc);
5099 }
5100
5101 static int osd_index_declare_ea_delete(const struct lu_env *env,
5102                                        struct dt_object *dt,
5103                                        const struct dt_key *key,
5104                                        struct thandle *handle)
5105 {
5106         struct osd_thandle *oh;
5107         struct inode *inode;
5108         int rc, credits;
5109
5110         ENTRY;
5111
5112         LASSERT(!dt_object_remote(dt));
5113         LASSERT(handle != NULL);
5114
5115         oh = container_of0(handle, struct osd_thandle, ot_super);
5116         LASSERT(oh->ot_handle == NULL);
5117
5118         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5119         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5120
5121         inode = osd_dt_obj(dt)->oo_inode;
5122         if (inode == NULL)
5123                 RETURN(-ENOENT);
5124
5125         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5126                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5127                                    NULL, OSD_QID_BLK);
5128         RETURN(rc);
5129 }
5130
5131 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5132                                           struct dt_rec *fid)
5133 {
5134         struct osd_fid_pack *rec;
5135         int rc = -ENODATA;
5136
5137         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5138                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5139                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5140                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5141                         rc = -EINVAL;
5142         }
5143         return rc;
5144 }
5145
5146 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5147                           const struct lu_fid *fid)
5148 {
5149         struct seq_server_site *ss = osd_seq_site(osd);
5150
5151         ENTRY;
5152
5153         /* FID seqs not in FLDB, must be local seq */
5154         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5155                 RETURN(0);
5156
5157         /*
5158          * If FLD is not being initialized yet, it only happens during the
5159          * initialization, likely during mgs initialization, and we assume
5160          * this is local FID.
5161          */
5162         if (ss == NULL || ss->ss_server_fld == NULL)
5163                 RETURN(0);
5164
5165         /* Only check the local FLDB here */
5166         if (osd_seq_exists(env, osd, fid_seq(fid)))
5167                 RETURN(0);
5168
5169         RETURN(1);
5170 }
5171
5172 static void osd_take_care_of_agent(const struct lu_env *env,
5173                                    struct osd_device *osd,
5174                                    struct osd_thandle *oh,
5175                                    struct ldiskfs_dir_entry_2 *de)
5176 {
5177         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5178         struct osd_idmap_cache *idc;
5179         int rc, schedule = 0;
5180
5181         LASSERT(de != NULL);
5182
5183         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5184         if (likely(rc == 0)) {
5185                 idc = osd_idc_find_or_init(env, osd, fid);
5186                 if (IS_ERR(idc) || idc->oic_remote)
5187                         schedule = 1;
5188         } else if (rc == -ENODATA) {
5189                 /*
5190                  * can't get FID, postpone to the end of the
5191                  * transaction when iget() is safe
5192                  */
5193                 schedule = 1;
5194         } else {
5195                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5196         }
5197         if (schedule)
5198                 osd_schedule_agent_inode_removal(env, oh,
5199                                                  le32_to_cpu(de->inode));
5200 }
5201
5202 /**
5203  * Index delete function for interoperability mode (b11826).
5204  * It will remove the directory entry added by osd_index_ea_insert().
5205  * This entry is needed to maintain name->fid mapping.
5206  *
5207  * \param key,  key i.e. file entry to be deleted
5208  *
5209  * \retval   0, on success
5210  * \retval -ve, on error
5211  */
5212 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5213                                const struct dt_key *key, struct thandle *handle)
5214 {
5215         struct osd_object *obj = osd_dt_obj(dt);
5216         struct inode *dir = obj->oo_inode;
5217         struct dentry *dentry;
5218         struct osd_thandle *oh;
5219         struct ldiskfs_dir_entry_2 *de = NULL;
5220         struct buffer_head *bh;
5221         struct htree_lock *hlock = NULL;
5222         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5223         int rc;
5224
5225         ENTRY;
5226
5227         if (!dt_object_exists(dt))
5228                 RETURN(-ENOENT);
5229
5230         LINVRNT(osd_invariant(obj));
5231         LASSERT(!dt_object_remote(dt));
5232         LASSERT(handle != NULL);
5233
5234         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5235
5236         oh = container_of(handle, struct osd_thandle, ot_super);
5237         LASSERT(oh->ot_handle != NULL);
5238         LASSERT(oh->ot_handle->h_transaction != NULL);
5239
5240         dquot_initialize(dir);
5241         dentry = osd_child_dentry_get(env, obj,
5242                                       (char *)key, strlen((char *)key));
5243
5244         if (obj->oo_hl_head != NULL) {
5245                 hlock = osd_oti_get(env)->oti_hlock;
5246                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5247                                    dir, LDISKFS_HLOCK_DEL);
5248         } else {
5249                 down_write(&obj->oo_ext_idx_sem);
5250         }
5251
5252         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5253         if (!IS_ERR(bh)) {
5254                 /*
5255                  * If this is not the ".." entry, it might be a remote DNE
5256                  * entry and  we need to check if the FID is for a remote
5257                  * MDT.  If the FID is  not in the directory entry (e.g.
5258                  * upgraded 1.8 filesystem without dirdata enabled) then
5259                  * we need to get the FID from the LMA. For a remote directory
5260                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5261                  * case.
5262                  *
5263                  * Delete the entry before the agent inode in order to
5264                  * simplify error handling.  At worst an error after deleting
5265                  * the entry first might leak the agent inode afterward. The
5266                  * reverse would need filesystem abort in case of error deleting
5267                  * the entry after the agent had been removed, or leave a
5268                  * dangling entry pointing at a random inode.
5269                  */
5270                 if (strcmp((char *)key, dotdot) != 0)
5271                         osd_take_care_of_agent(env, osd, oh, de);
5272                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5273                 brelse(bh);
5274         } else {
5275                 rc = PTR_ERR(bh);
5276         }
5277
5278         if (!rc && fid_is_namespace_visible(lu_object_fid(&dt->do_lu)) &&
5279             obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
5280                 /* NB, dirent count may not be accurate, because it's counted
5281                  * without lock.
5282                  */
5283                 if (obj->oo_dirent_count)
5284                         obj->oo_dirent_count--;
5285                 else
5286                         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
5287         }
5288         if (hlock != NULL)
5289                 ldiskfs_htree_unlock(hlock);
5290         else
5291                 up_write(&obj->oo_ext_idx_sem);
5292         GOTO(out, rc);
5293 out:
5294         LASSERT(osd_invariant(obj));
5295         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5296         RETURN(rc);
5297 }
5298
5299 /**
5300  *      Lookup index for \a key and copy record to \a rec.
5301  *
5302  *      \param  dt      osd index object
5303  *      \param  key     key for index
5304  *      \param  rec     record reference
5305  *
5306  *      \retval  +ve  success : exact mach
5307  *      \retval  0    return record with key not greater than \a key
5308  *      \retval -ve   failure
5309  */
5310 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5311                                 struct dt_rec *rec, const struct dt_key *key)
5312 {
5313         struct osd_object *obj = osd_dt_obj(dt);
5314         struct iam_path_descr *ipd;
5315         struct iam_container *bag = &obj->oo_dir->od_container;
5316         struct osd_thread_info *oti = osd_oti_get(env);
5317         struct iam_iterator *it = &oti->oti_idx_it;
5318         struct iam_rec *iam_rec;
5319         int rc;
5320
5321         ENTRY;
5322
5323         if (!dt_object_exists(dt))
5324                 RETURN(-ENOENT);
5325
5326         LASSERT(osd_invariant(obj));
5327         LASSERT(!dt_object_remote(dt));
5328         LASSERT(bag->ic_object == obj->oo_inode);
5329
5330         ipd = osd_idx_ipd_get(env, bag);
5331         if (IS_ERR(ipd))
5332                 RETURN(-ENOMEM);
5333
5334         /* got ipd now we can start iterator. */
5335         iam_it_init(it, bag, 0, ipd);
5336
5337         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5338                 /* swab quota uid/gid provided by caller */
5339                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5340                 key = (const struct dt_key *)&oti->oti_quota_id;
5341         }
5342
5343         rc = iam_it_get(it, (struct iam_key *)key);
5344         if (rc >= 0) {
5345                 if (S_ISDIR(obj->oo_inode->i_mode))
5346                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5347                 else
5348                         iam_rec = (struct iam_rec *)rec;
5349
5350                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5351
5352                 if (S_ISDIR(obj->oo_inode->i_mode))
5353                         osd_fid_unpack((struct lu_fid *)rec,
5354                                        (struct osd_fid_pack *)iam_rec);
5355                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5356                         osd_quota_unpack(obj, rec);
5357         }
5358
5359         iam_it_put(it);
5360         iam_it_fini(it);
5361         osd_ipd_put(env, bag, ipd);
5362
5363         LINVRNT(osd_invariant(obj));
5364
5365         RETURN(rc);
5366 }
5367
5368 static int osd_index_declare_iam_insert(const struct lu_env *env,
5369                                         struct dt_object *dt,
5370                                         const struct dt_rec *rec,
5371                                         const struct dt_key *key,
5372                                         struct thandle *handle)
5373 {
5374         struct osd_thandle *oh;
5375
5376         LASSERT(handle != NULL);
5377
5378         oh = container_of0(handle, struct osd_thandle, ot_super);
5379         LASSERT(oh->ot_handle == NULL);
5380
5381         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5382                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5383
5384         return 0;
5385 }
5386
5387 /**
5388  *      Inserts (key, value) pair in \a dt index object.
5389  *
5390  *      \param  dt      osd index object
5391  *      \param  key     key for index
5392  *      \param  rec     record reference
5393  *      \param  th      transaction handler
5394  *
5395  *      \retval  0  success
5396  *      \retval -ve failure
5397  */
5398 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5399                                 const struct dt_rec *rec,
5400                                 const struct dt_key *key, struct thandle *th)
5401 {
5402         struct osd_object *obj = osd_dt_obj(dt);
5403         struct iam_path_descr *ipd;
5404         struct osd_thandle *oh;
5405         struct iam_container *bag;
5406         struct osd_thread_info *oti = osd_oti_get(env);
5407         struct iam_rec *iam_rec;
5408         int rc;
5409
5410         ENTRY;
5411
5412         if (!dt_object_exists(dt))
5413                 RETURN(-ENOENT);
5414
5415         LINVRNT(osd_invariant(obj));
5416         LASSERT(!dt_object_remote(dt));
5417
5418         bag = &obj->oo_dir->od_container;
5419         LASSERT(bag->ic_object == obj->oo_inode);
5420         LASSERT(th != NULL);
5421
5422         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5423
5424         ipd = osd_idx_ipd_get(env, bag);
5425         if (unlikely(ipd == NULL))
5426                 RETURN(-ENOMEM);
5427
5428         oh = container_of0(th, struct osd_thandle, ot_super);
5429         LASSERT(oh->ot_handle != NULL);
5430         LASSERT(oh->ot_handle->h_transaction != NULL);
5431         if (S_ISDIR(obj->oo_inode->i_mode)) {
5432                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5433                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5434                              &oti->oti_fid);
5435         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5436                 /* pack quota uid/gid */
5437                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5438                 key = (const struct dt_key *)&oti->oti_quota_id;
5439                 /* pack quota record */
5440                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5441                 iam_rec = (struct iam_rec *)rec;
5442         } else {
5443                 iam_rec = (struct iam_rec *)rec;
5444         }
5445
5446         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5447                         iam_rec, ipd);
5448         osd_ipd_put(env, bag, ipd);
5449         LINVRNT(osd_invariant(obj));
5450         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5451         RETURN(rc);
5452 }
5453
5454 /**
5455  * Calls ldiskfs_add_entry() to add directory entry
5456  * into the directory. This is required for
5457  * interoperability mode (b11826)
5458  *
5459  * \retval   0, on success
5460  * \retval -ve, on error
5461  */
5462 static int __osd_ea_add_rec(struct osd_thread_info *info,
5463                             struct osd_object *pobj, struct inode  *cinode,
5464                             const char *name, const struct lu_fid *fid,
5465                             struct htree_lock *hlock, struct thandle *th)
5466 {
5467         struct ldiskfs_dentry_param *ldp;
5468         struct dentry *child;
5469         struct osd_thandle *oth;
5470         int rc;
5471
5472         oth = container_of(th, struct osd_thandle, ot_super);
5473         LASSERT(oth->ot_handle != NULL);
5474         LASSERT(oth->ot_handle->h_transaction != NULL);
5475         LASSERT(pobj->oo_inode);
5476
5477         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5478         if (unlikely(pobj->oo_inode ==
5479                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5480                 ldp->edp_magic = 0;
5481         else
5482                 osd_get_ldiskfs_dirent_param(ldp, fid);
5483         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5484         child->d_fsdata = (void *)ldp;
5485         dquot_initialize(pobj->oo_inode);
5486         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5487                                    child, cinode, hlock);
5488         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5489                 struct ldiskfs_dir_entry_2      *de;
5490                 struct buffer_head              *bh;
5491                 int                              rc1;
5492
5493                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5494                                             NULL, hlock);
5495                 if (!IS_ERR(bh)) {
5496                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5497                                                                bh);
5498                         if (rc1 == 0) {
5499                                 if (S_ISDIR(cinode->i_mode))
5500                                         de->file_type = LDISKFS_DIRENT_LUFID |
5501                                                         LDISKFS_FT_REG_FILE;
5502                                 else
5503                                         de->file_type = LDISKFS_DIRENT_LUFID |
5504                                                         LDISKFS_FT_DIR;
5505                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5506                                                               NULL, bh);
5507                         }
5508                         brelse(bh);
5509                 }
5510         }
5511
5512         RETURN(rc);
5513 }
5514
5515 /**
5516  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5517  * into the directory.Also sets flags into osd object to
5518  * indicate dot and dotdot are created. This is required for
5519  * interoperability mode (b11826)
5520  *
5521  * \param dir   directory for dot and dotdot fixup.
5522  * \param obj   child object for linking
5523  *
5524  * \retval   0, on success
5525  * \retval -ve, on error
5526  */
5527 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5528                               struct osd_object *dir,
5529                               struct inode *parent_dir, const char *name,
5530                               const struct lu_fid *dot_fid,
5531                               const struct lu_fid *dot_dot_fid,
5532                               struct thandle *th)
5533 {
5534         struct inode *inode = dir->oo_inode;
5535         struct osd_thandle *oth;
5536         int result = 0;
5537
5538         oth = container_of(th, struct osd_thandle, ot_super);
5539         LASSERT(oth->ot_handle->h_transaction != NULL);
5540         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5541
5542         if (strcmp(name, dot) == 0) {
5543                 if (dir->oo_compat_dot_created) {
5544                         result = -EEXIST;
5545                 } else {
5546                         LASSERT(inode->i_ino == parent_dir->i_ino);
5547                         dir->oo_compat_dot_created = 1;
5548                         result = 0;
5549                 }
5550         } else if (strcmp(name, dotdot) == 0) {
5551                 if (!dir->oo_compat_dot_created)
5552                         return -EINVAL;
5553                 /* in case of rename, dotdot is already created */
5554                 if (dir->oo_compat_dotdot_created) {
5555                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5556                                                 dot_dot_fid, NULL, th);
5557                 }
5558
5559                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5560                         struct lu_fid tfid = *dot_dot_fid;
5561
5562                         tfid.f_oid--;
5563                         result = osd_add_dot_dotdot_internal(info,
5564                                         dir->oo_inode, parent_dir, dot_fid,
5565                                         &tfid, oth);
5566                 } else {
5567                         result = osd_add_dot_dotdot_internal(info,
5568                                         dir->oo_inode, parent_dir, dot_fid,
5569                                         dot_dot_fid, oth);
5570                 }
5571
5572                 if (result == 0)
5573                         dir->oo_compat_dotdot_created = 1;
5574         }
5575
5576         return result;
5577 }
5578
5579
5580 /**
5581  * It will call the appropriate osd_add* function and return the
5582  * value, return by respective functions.
5583  */
5584 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5585                           struct inode *cinode, const char *name,
5586                           const struct lu_fid *fid, struct thandle *th)
5587 {
5588         struct osd_thread_info *info = osd_oti_get(env);
5589         struct htree_lock *hlock;
5590         int rc;
5591
5592         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5593
5594         if (name[0] == '.' && (name[1] == '\0' ||
5595                                (name[1] == '.' && name[2] == '\0'))) {
5596                 if (hlock != NULL) {
5597                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5598                                            pobj->oo_inode, 0);
5599                 } else {
5600                         down_write(&pobj->oo_ext_idx_sem);
5601                 }
5602
5603                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5604                                         lu_object_fid(&pobj->oo_dt.do_lu),
5605                                         fid, th);
5606         } else {
5607                 if (hlock != NULL) {
5608                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5609                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5610                 } else {
5611                         down_write(&pobj->oo_ext_idx_sem);
5612                 }
5613
5614                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5615                         struct lu_fid *tfid = &info->oti_fid;
5616
5617                         *tfid = *fid;
5618                         tfid->f_ver = ~0;
5619                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5620                                               tfid, hlock, th);
5621                 } else {
5622                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5623                                               hlock, th);
5624                 }
5625         }
5626         if (!rc && fid_is_namespace_visible(lu_object_fid(&pobj->oo_dt.do_lu))
5627             && pobj->oo_dirent_count != LU_DIRENT_COUNT_UNSET)
5628                 pobj->oo_dirent_count++;
5629
5630         if (hlock != NULL)
5631                 ldiskfs_htree_unlock(hlock);
5632         else
5633                 up_write(&pobj->oo_ext_idx_sem);
5634
5635         return rc;
5636 }
5637
5638 static int
5639 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5640                       struct osd_idmap_cache *oic)
5641 {
5642         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5643         struct lu_fid *fid = &oic->oic_fid;
5644         struct osd_inode_id *id = &oic->oic_lid;
5645         struct inode *inode = NULL;
5646         int once = 0;
5647         bool insert;
5648         int rc;
5649
5650         ENTRY;
5651
5652         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5653                 RETURN(0);
5654
5655         if (thread_is_running(&scrub->os_thread) &&
5656             scrub->os_pos_current > id->oii_ino)
5657                 RETURN(0);
5658
5659         if (dev->od_auto_scrub_interval == AS_NEVER ||
5660             ktime_get_real_seconds() <
5661             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5662                 RETURN(0);
5663
5664 again:
5665         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5666         if (rc == -ENOENT) {
5667                 __u32 gen = id->oii_gen;
5668
5669                 insert = true;
5670                 if (inode != NULL)
5671                         goto trigger;
5672
5673                 inode = osd_iget(oti, dev, id);
5674                 /* The inode has been removed (by race maybe). */
5675                 if (IS_ERR(inode)) {
5676                         rc = PTR_ERR(inode);
5677
5678                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5679                 }
5680
5681                 /* The OI mapping is lost. */
5682                 if (gen != OSD_OII_NOGEN)
5683                         goto trigger;
5684
5685                 /*
5686                  * The inode may has been reused by others, we do not know,
5687                  * leave it to be handled by subsequent osd_fid_lookup().
5688                  */
5689                 GOTO(out, rc = 0);
5690         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5691                 GOTO(out, rc);
5692         }
5693
5694         insert = false;
5695
5696 trigger:
5697         if (thread_is_running(&scrub->os_thread)) {
5698                 if (inode == NULL) {
5699                         inode = osd_iget(oti, dev, id);
5700                         /* The inode has been removed (by race maybe). */
5701                         if (IS_ERR(inode)) {
5702                                 rc = PTR_ERR(inode);
5703
5704                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5705                         }
5706                 }
5707
5708                 rc = osd_oii_insert(dev, oic, insert);
5709                 /*
5710                  * There is race condition between osd_oi_lookup and OI scrub.
5711                  * The OI scrub finished just after osd_oi_lookup() failure.
5712                  * Under such case, it is unnecessary to trigger OI scrub again,
5713                  * but try to call osd_oi_lookup() again.
5714                  */
5715                 if (unlikely(rc == -EAGAIN))
5716                         goto again;
5717
5718                 if (!S_ISDIR(inode->i_mode))
5719                         rc = 0;
5720                 else
5721                         rc = osd_check_lmv(oti, dev, inode, oic);
5722
5723                 GOTO(out, rc);
5724         }
5725
5726         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5727                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5728                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5729                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5730                        "%s: trigger partial OI scrub for RPC inconsistency "
5731                        "checking FID "DFID": rc = %d\n",
5732                        osd_dev2name(dev), PFID(fid), rc);
5733                 if (rc == 0 || rc == -EALREADY)
5734                         goto again;
5735         }
5736
5737         GOTO(out, rc);
5738
5739 out:
5740         if (inode)
5741                 iput(inode);
5742
5743         RETURN(rc);
5744 }
5745
5746 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5747                                struct osd_device *dev,
5748                                struct osd_idmap_cache *oic,
5749                                struct lu_fid *fid, __u32 ino)
5750 {
5751         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5752         struct inode *inode;
5753         int rc;
5754
5755         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5756         inode = osd_iget(oti, dev, &oic->oic_lid);
5757         if (IS_ERR(inode)) {
5758                 fid_zero(&oic->oic_fid);
5759                 return PTR_ERR(inode);
5760         }
5761
5762         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5763         iput(inode);
5764         if (rc != 0)
5765                 fid_zero(&oic->oic_fid);
5766         else
5767                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5768         return rc;
5769 }
5770
5771 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5772                       struct osd_inode_id *id, const struct lu_fid *fid)
5773 {
5774         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5775                id->oii_ino, id->oii_gen, info);
5776         info->oti_cache.oic_lid = *id;
5777         info->oti_cache.oic_fid = *fid;
5778         info->oti_cache.oic_dev = osd;
5779 }
5780
5781 /**
5782  * Get parent FID from the linkEA.
5783  *
5784  * For a directory which parent resides on remote MDT, to satisfy the
5785  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5786  * the other hand, to make the lookup(..) on the directory can return
5787  * the real parent FID, we append the real parent FID after its ".."
5788  * name entry in the /REMOTE_PARENT_DIR.
5789  *
5790  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5791  * backup. So after the restore, we cannot get the right parent FID from
5792  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5793  * we have stored the real parent FID in the directory object's linkEA,
5794  * we can parse the linkEA for the real parent FID.
5795  *
5796  * \param[in] env       pointer to the thread context
5797  * \param[in] obj       pointer to the object to be handled
5798  * \param[out]fid       pointer to the buffer to hold the parent FID
5799  *
5800  * \retval              0 for getting the real parent FID successfully
5801  * \retval              negative error number on failure
5802  */
5803 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5804                                     struct osd_object *obj,
5805                                     struct lu_fid *fid)
5806 {
5807         struct osd_thread_info *oti = osd_oti_get(env);
5808         struct lu_buf *buf = &oti->oti_big_buf;
5809         struct dentry *dentry = &oti->oti_obj_dentry;
5810         struct inode *inode = obj->oo_inode;
5811         struct linkea_data ldata = { NULL };
5812         int rc;
5813
5814         ENTRY;
5815
5816         fid_zero(fid);
5817         if (!S_ISDIR(inode->i_mode))
5818                 RETURN(-EIO);
5819
5820 again:
5821         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5822                              buf->lb_buf, buf->lb_len);
5823         if (rc == -ERANGE) {
5824                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5825                                      NULL, 0);
5826                 if (rc > 0) {
5827                         lu_buf_realloc(buf, rc);
5828                         if (buf->lb_buf == NULL)
5829                                 RETURN(-ENOMEM);
5830
5831                         goto again;
5832                 }
5833         }
5834
5835         if (unlikely(rc == 0))
5836                 RETURN(-ENODATA);
5837
5838         if (rc < 0)
5839                 RETURN(rc);
5840
5841         if (unlikely(buf->lb_buf == NULL)) {
5842                 lu_buf_realloc(buf, rc);
5843                 if (buf->lb_buf == NULL)
5844                         RETURN(-ENOMEM);
5845
5846                 goto again;
5847         }
5848
5849         ldata.ld_buf = buf;
5850         rc = linkea_init_with_rec(&ldata);
5851         if (!rc) {
5852                 linkea_first_entry(&ldata);
5853                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5854         }
5855
5856         RETURN(rc);
5857 }
5858
5859 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5860                                     struct inode *inode,
5861                                     const struct lu_fid *pfid,
5862                                     const char *name, const int namelen)
5863 {
5864         struct osd_thread_info *oti = osd_oti_get(env);
5865         struct lu_buf *buf = &oti->oti_big_buf;
5866         struct dentry *dentry = &oti->oti_obj_dentry;
5867         struct linkea_data ldata = { NULL };
5868         struct lu_name cname = { .ln_name = name,
5869                                  .ln_namelen = namelen };
5870         int rc;
5871
5872         ENTRY;
5873
5874 again:
5875         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5876                              buf->lb_buf, buf->lb_len);
5877         if (rc == -ERANGE)
5878                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5879
5880         if (rc < 0)
5881                 RETURN(rc);
5882
5883         if (unlikely(rc == 0))
5884                 RETURN(-ENODATA);
5885
5886         if (buf->lb_len < rc) {
5887                 lu_buf_realloc(buf, rc);
5888                 if (buf->lb_buf == NULL)
5889                         RETURN(-ENOMEM);
5890
5891                 goto again;
5892         }
5893
5894         ldata.ld_buf = buf;
5895         rc = linkea_init_with_rec(&ldata);
5896         if (!rc)
5897                 rc = linkea_links_find(&ldata, &cname, pfid);
5898
5899         RETURN(rc);
5900 }
5901
5902 /**
5903  * Calls ->lookup() to find dentry. From dentry get inode and
5904  * read inode's ea to get fid. This is required for  interoperability
5905  * mode (b11826)
5906  *
5907  * \retval   0, on success
5908  * \retval -ve, on error
5909  */
5910 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5911                              struct dt_rec *rec, const struct dt_key *key)
5912 {
5913         struct inode *dir = obj->oo_inode;
5914         struct dentry *dentry;
5915         struct ldiskfs_dir_entry_2 *de;
5916         struct buffer_head *bh;
5917         struct lu_fid *fid = (struct lu_fid *)rec;
5918         struct htree_lock *hlock = NULL;
5919         int ino;
5920         int rc;
5921
5922         ENTRY;
5923
5924         LASSERT(dir->i_op != NULL);
5925         LASSERT(dir->i_op->lookup != NULL);
5926
5927         dentry = osd_child_dentry_get(env, obj,
5928                                       (char *)key, strlen((char *)key));
5929
5930         if (obj->oo_hl_head != NULL) {
5931                 hlock = osd_oti_get(env)->oti_hlock;
5932                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5933                                    dir, LDISKFS_HLOCK_LOOKUP);
5934         } else {
5935                 down_read(&obj->oo_ext_idx_sem);
5936         }
5937
5938         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5939         if (!IS_ERR(bh)) {
5940                 struct osd_thread_info *oti = osd_oti_get(env);
5941                 struct osd_inode_id *id = &oti->oti_id;
5942                 struct osd_idmap_cache *oic = &oti->oti_cache;
5943                 struct osd_device *dev = osd_obj2dev(obj);
5944
5945                 ino = le32_to_cpu(de->inode);
5946                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5947                         brelse(bh);
5948                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5949                         GOTO(out, rc);
5950                 }
5951
5952                 rc = osd_get_fid_from_dentry(de, rec);
5953
5954                 /* done with de, release bh */
5955                 brelse(bh);
5956                 if (rc != 0) {
5957                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5958                                 const char *name = (const char *)key;
5959
5960                                 /*
5961                                  * If the parent is on remote MDT, and there
5962                                  * is no FID-in-dirent, then we have to get
5963                                  * the parent FID from the linkEA.
5964                                  */
5965                                 if (likely(strlen(name) == 2 &&
5966                                            name[0] == '.' && name[1] == '.'))
5967                                         rc = osd_get_pfid_from_linkea(env, obj,
5968                                                                       fid);
5969                         } else {
5970                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5971                         }
5972                 } else {
5973                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5974                 }
5975
5976                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5977                         fid_zero(&oic->oic_fid);
5978
5979                         GOTO(out, rc);
5980                 }
5981
5982                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5983                 rc = osd_consistency_check(oti, dev, oic);
5984                 if (rc == -ENOENT)
5985                         fid_zero(&oic->oic_fid);
5986                 else
5987                         /* Other error should not affect lookup result. */
5988                         rc = 0;
5989         } else {
5990                 rc = PTR_ERR(bh);
5991         }
5992
5993         GOTO(out, rc);
5994
5995 out:
5996         if (hlock != NULL)
5997                 ldiskfs_htree_unlock(hlock);
5998         else
5999                 up_read(&obj->oo_ext_idx_sem);
6000         return rc;
6001 }
6002
6003 static int osd_index_declare_ea_insert(const struct lu_env *env,
6004                                        struct dt_object *dt,
6005                                        const struct dt_rec *rec,
6006                                        const struct dt_key *key,
6007                                        struct thandle *handle)
6008 {
6009         struct osd_thandle *oh;
6010         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6011         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6012         const struct lu_fid *fid = rec1->rec_fid;
6013         int credits, rc = 0;
6014         struct osd_idmap_cache *idc;
6015
6016         ENTRY;
6017
6018         LASSERT(!dt_object_remote(dt));
6019         LASSERT(handle != NULL);
6020         LASSERT(fid != NULL);
6021         LASSERT(rec1->rec_type != 0);
6022
6023         oh = container_of0(handle, struct osd_thandle, ot_super);
6024         LASSERT(oh->ot_handle == NULL);
6025
6026         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
6027
6028         /*
6029          * we can't call iget() while a transactions is running
6030          * (this can lead to a deadlock), but we need to know
6031          * inum and object type. so we find this information at
6032          * declaration and cache in per-thread info
6033          */
6034         idc = osd_idc_find_or_init(env, osd, fid);
6035         if (IS_ERR(idc))
6036                 RETURN(PTR_ERR(idc));
6037         if (idc->oic_remote) {
6038                 /*
6039                  * a reference to remote inode is represented by an
6040                  * agent inode which we have to create
6041                  */
6042                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
6043                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
6044         }
6045
6046         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
6047
6048         if (osd_dt_obj(dt)->oo_inode != NULL) {
6049                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
6050
6051                 /*
6052                  * We ignore block quota on meta pool (MDTs), so needn't
6053                  * calculate how many blocks will be consumed by this index
6054                  * insert
6055                  */
6056                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
6057                                            i_gid_read(inode),
6058                                            i_projid_read(inode), 0,
6059                                            oh, osd_dt_obj(dt), NULL,
6060                                            OSD_QID_BLK);
6061                 if (rc)
6062                         RETURN(rc);
6063
6064 #ifdef HAVE_PROJECT_QUOTA
6065                 /*
6066                  * Reserve credits for local agent inode to transfer
6067                  * to 0, quota enforcement is ignored in this case.
6068                  */
6069                 if (idc->oic_remote &&
6070                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6071                     i_projid_read(inode) != 0)
6072                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6073                                                   0, i_projid_read(inode),
6074                                                   0, false, PRJQUOTA, true);
6075 #endif
6076         }
6077
6078         RETURN(rc);
6079 }
6080
6081 /**
6082  * Index add function for interoperability mode (b11826).
6083  * It will add the directory entry.This entry is needed to
6084  * maintain name->fid mapping.
6085  *
6086  * \param key it is key i.e. file entry to be inserted
6087  * \param rec it is value of given key i.e. fid
6088  *
6089  * \retval   0, on success
6090  * \retval -ve, on error
6091  */
6092 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6093                                const struct dt_rec *rec,
6094                                const struct dt_key *key, struct thandle *th)
6095 {
6096         struct osd_object *obj = osd_dt_obj(dt);
6097         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6098         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6099         const struct lu_fid *fid = rec1->rec_fid;
6100         const char *name = (const char *)key;
6101         struct osd_thread_info *oti = osd_oti_get(env);
6102         struct inode *child_inode = NULL;
6103         struct osd_idmap_cache *idc;
6104         int rc;
6105
6106         ENTRY;
6107
6108         if (!dt_object_exists(dt))
6109                 RETURN(-ENOENT);
6110
6111         LASSERT(osd_invariant(obj));
6112         LASSERT(!dt_object_remote(dt));
6113         LASSERT(th != NULL);
6114
6115         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6116
6117         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6118
6119         idc = osd_idc_find(env, osd, fid);
6120         if (unlikely(idc == NULL)) {
6121                 idc = osd_idc_find_or_init(env, osd, fid);
6122                 if (IS_ERR(idc)) {
6123                         /*
6124                          * this dt_insert() wasn't declared properly, so
6125                          * FID is missing in OI cache. we better do not
6126                          * lookup FID in FLDB/OI and don't risk to deadlock,
6127                          * but in some special cases (lfsck testing, etc)
6128                          * it's much simpler than fixing a caller.
6129                          *
6130                          * normally this error should be placed after the first
6131                          * find, but migrate may attach source stripes to
6132                          * target, which doesn't create stripes.
6133                          */
6134                         CERROR("%s: "DFID" wasn't declared for insert\n",
6135                                osd_name(osd), PFID(fid));
6136                         dump_stack();
6137                         RETURN(PTR_ERR(idc));
6138                 }
6139         }
6140
6141         if (idc->oic_remote) {
6142                 /* Insert remote entry */
6143                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6144                         child_inode =
6145                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6146                 } else {
6147                         child_inode = osd_create_local_agent_inode(env, osd,
6148                                         obj, fid, rec1->rec_type & S_IFMT, th);
6149                         if (IS_ERR(child_inode))
6150                                 RETURN(PTR_ERR(child_inode));
6151                 }
6152         } else {
6153                 /* Insert local entry */
6154                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6155                         /* for a reason OI cache wasn't filled properly */
6156                         CERROR("%s: OIC for "DFID" isn't filled\n",
6157                                osd_name(osd), PFID(fid));
6158                         RETURN(-EINVAL);
6159                 }
6160                 child_inode = oti->oti_inode;
6161                 if (unlikely(child_inode == NULL)) {
6162                         struct ldiskfs_inode_info *lii;
6163
6164                         OBD_ALLOC_PTR(lii);
6165                         if (lii == NULL)
6166                                 RETURN(-ENOMEM);
6167                         child_inode = oti->oti_inode = &lii->vfs_inode;
6168                 }
6169                 child_inode->i_sb = osd_sb(osd);
6170                 child_inode->i_ino = idc->oic_lid.oii_ino;
6171                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6172         }
6173
6174         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6175
6176         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6177                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6178
6179         if (child_inode && child_inode != oti->oti_inode)
6180                 iput(child_inode);
6181         LASSERT(osd_invariant(obj));
6182         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6183
6184         RETURN(rc);
6185 }
6186
6187 /**
6188  *  Initialize osd Iterator for given osd index object.
6189  *
6190  *  \param  dt      osd index object
6191  */
6192
6193 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6194                                      struct dt_object *dt,
6195                                      __u32 unused)
6196 {
6197         struct osd_it_iam *it;
6198         struct osd_object *obj = osd_dt_obj(dt);
6199         struct lu_object *lo = &dt->do_lu;
6200         struct iam_path_descr *ipd;
6201         struct iam_container *bag = &obj->oo_dir->od_container;
6202
6203         if (!dt_object_exists(dt))
6204                 return ERR_PTR(-ENOENT);
6205
6206         OBD_ALLOC_PTR(it);
6207         if (it == NULL)
6208                 return ERR_PTR(-ENOMEM);
6209
6210         ipd = osd_it_ipd_get(env, bag);
6211         if (likely(ipd != NULL)) {
6212                 it->oi_obj = obj;
6213                 it->oi_ipd = ipd;
6214                 lu_object_get(lo);
6215                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6216                 return (struct dt_it *)it;
6217         } else {
6218                 OBD_FREE_PTR(it);
6219                 return ERR_PTR(-ENOMEM);
6220         }
6221 }
6222
6223 /**
6224  * free given Iterator.
6225  */
6226 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6227 {
6228         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6229         struct osd_object *obj = it->oi_obj;
6230
6231         iam_it_fini(&it->oi_it);
6232         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6233         osd_object_put(env, obj);
6234         OBD_FREE_PTR(it);
6235 }
6236
6237 /**
6238  *  Move Iterator to record specified by \a key
6239  *
6240  *  \param  di      osd iterator
6241  *  \param  key     key for index
6242  *
6243  *  \retval +ve  di points to record with least key not larger than key
6244  *  \retval  0   di points to exact matched key
6245  *  \retval -ve  failure
6246  */
6247
6248 static int osd_it_iam_get(const struct lu_env *env,
6249                           struct dt_it *di, const struct dt_key *key)
6250 {
6251         struct osd_thread_info *oti = osd_oti_get(env);
6252         struct osd_it_iam *it = (struct osd_it_iam *)di;
6253
6254         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6255                 /* swab quota uid/gid */
6256                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6257                 key = (struct dt_key *)&oti->oti_quota_id;
6258         }
6259
6260         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6261 }
6262
6263 /**
6264  *  Release Iterator
6265  *
6266  *  \param  di      osd iterator
6267  */
6268 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6269 {
6270         struct osd_it_iam *it = (struct osd_it_iam *)di;
6271
6272         iam_it_put(&it->oi_it);
6273 }
6274
6275 /**
6276  *  Move iterator by one record
6277  *
6278  *  \param  di      osd iterator
6279  *
6280  *  \retval +1   end of container reached
6281  *  \retval  0   success
6282  *  \retval -ve  failure
6283  */
6284
6285 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6286 {
6287         struct osd_it_iam *it = (struct osd_it_iam *)di;
6288
6289         return iam_it_next(&it->oi_it);
6290 }
6291
6292 /**
6293  * Return pointer to the key under iterator.
6294  */
6295
6296 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6297                                      const struct dt_it *di)
6298 {
6299         struct osd_thread_info *oti = osd_oti_get(env);
6300         struct osd_it_iam *it = (struct osd_it_iam *)di;
6301         struct osd_object *obj = it->oi_obj;
6302         struct dt_key *key;
6303
6304         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6305
6306         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6307                 /* swab quota uid/gid */
6308                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6309                 key = (struct dt_key *)&oti->oti_quota_id;
6310         }
6311
6312         return key;
6313 }
6314
6315 /**
6316  * Return size of key under iterator (in bytes)
6317  */
6318
6319 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6320 {
6321         struct osd_it_iam *it = (struct osd_it_iam *)di;
6322
6323         return iam_it_key_size(&it->oi_it);
6324 }
6325
6326 static inline void
6327 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6328 {
6329         /* check if file type is required */
6330         if (ent->lde_attrs & LUDA_TYPE) {
6331                 struct luda_type *lt;
6332                 int align = sizeof(*lt) - 1;
6333
6334                 len = (len + align) & ~align;
6335                 lt = (struct luda_type *)(ent->lde_name + len);
6336                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6337         }
6338
6339         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6340 }
6341
6342 /**
6343  * build lu direct from backend fs dirent.
6344  */
6345
6346 static inline void
6347 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6348                    char *name, __u16 namelen, __u16 type, __u32 attr)
6349 {
6350         ent->lde_attrs = attr | LUDA_FID;
6351         fid_cpu_to_le(&ent->lde_fid, fid);
6352
6353         ent->lde_hash = cpu_to_le64(offset);
6354         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6355
6356         strncpy(ent->lde_name, name, namelen);
6357         ent->lde_name[namelen] = '\0';
6358         ent->lde_namelen = cpu_to_le16(namelen);
6359
6360         /* append lustre attributes */
6361         osd_it_append_attrs(ent, namelen, type);
6362 }
6363
6364 /**
6365  * Return pointer to the record under iterator.
6366  */
6367 static int osd_it_iam_rec(const struct lu_env *env,
6368                           const struct dt_it *di,
6369                           struct dt_rec *dtrec, __u32 attr)
6370 {
6371         struct osd_it_iam *it = (struct osd_it_iam *)di;
6372         struct osd_thread_info *info = osd_oti_get(env);
6373
6374         ENTRY;
6375
6376         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6377                 const struct osd_fid_pack *rec;
6378                 struct lu_fid *fid = &info->oti_fid;
6379                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6380                 char *name;
6381                 int namelen;
6382                 __u64 hash;
6383                 int rc;
6384
6385                 name = (char *)iam_it_key_get(&it->oi_it);
6386                 if (IS_ERR(name))
6387                         RETURN(PTR_ERR(name));
6388
6389                 namelen = iam_it_key_size(&it->oi_it);
6390
6391                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6392                 if (IS_ERR(rec))
6393                         RETURN(PTR_ERR(rec));
6394
6395                 rc = osd_fid_unpack(fid, rec);
6396                 if (rc)
6397                         RETURN(rc);
6398
6399                 hash = iam_it_store(&it->oi_it);
6400
6401                 /* IAM does not store object type in IAM index (dir) */
6402                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6403                                    0, LUDA_FID);
6404         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6405                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6406                            (struct iam_rec *)dtrec);
6407                 osd_quota_unpack(it->oi_obj, dtrec);
6408         } else {
6409                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6410                            (struct iam_rec *)dtrec);
6411         }
6412
6413         RETURN(0);
6414 }
6415
6416 /**
6417  * Returns cookie for current Iterator position.
6418  */
6419 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6420 {
6421         struct osd_it_iam *it = (struct osd_it_iam *)di;
6422
6423         return iam_it_store(&it->oi_it);
6424 }
6425
6426 /**
6427  * Restore iterator from cookie.
6428  *
6429  * \param  di      osd iterator
6430  * \param  hash    Iterator location cookie
6431  *
6432  * \retval +ve  di points to record with least key not larger than key.
6433  * \retval  0   di points to exact matched key
6434  * \retval -ve  failure
6435  */
6436
6437 static int osd_it_iam_load(const struct lu_env *env,
6438                            const struct dt_it *di, __u64 hash)
6439 {
6440         struct osd_it_iam *it = (struct osd_it_iam *)di;
6441
6442         return iam_it_load(&it->oi_it, hash);
6443 }
6444
6445 static const struct dt_index_operations osd_index_iam_ops = {
6446         .dio_lookup         = osd_index_iam_lookup,
6447         .dio_declare_insert = osd_index_declare_iam_insert,
6448         .dio_insert         = osd_index_iam_insert,
6449         .dio_declare_delete = osd_index_declare_iam_delete,
6450         .dio_delete         = osd_index_iam_delete,
6451         .dio_it     = {
6452                 .init     = osd_it_iam_init,
6453                 .fini     = osd_it_iam_fini,
6454                 .get      = osd_it_iam_get,
6455                 .put      = osd_it_iam_put,
6456                 .next     = osd_it_iam_next,
6457                 .key      = osd_it_iam_key,
6458                 .key_size = osd_it_iam_key_size,
6459                 .rec      = osd_it_iam_rec,
6460                 .store    = osd_it_iam_store,
6461                 .load     = osd_it_iam_load
6462         }
6463 };
6464
6465
6466 /**
6467  * Creates or initializes iterator context.
6468  *
6469  * \retval struct osd_it_ea, iterator structure on success
6470  *
6471  */
6472 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6473                                     struct dt_object *dt,
6474                                     __u32 attr)
6475 {
6476         struct osd_object *obj = osd_dt_obj(dt);
6477         struct osd_thread_info *info = osd_oti_get(env);
6478         struct osd_it_ea *oie;
6479         struct file *file;
6480         struct lu_object *lo = &dt->do_lu;
6481         struct dentry *obj_dentry;
6482
6483         ENTRY;
6484
6485         if (!dt_object_exists(dt) || obj->oo_destroyed)
6486                 RETURN(ERR_PTR(-ENOENT));
6487
6488         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6489         if (oie == NULL)
6490                 RETURN(ERR_PTR(-ENOMEM));
6491         obj_dentry = &oie->oie_dentry;
6492
6493         obj_dentry->d_inode = obj->oo_inode;
6494         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6495         obj_dentry->d_name.hash = 0;
6496
6497         oie->oie_rd_dirent       = 0;
6498         oie->oie_it_dirent       = 0;
6499         oie->oie_dirent          = NULL;
6500         if (unlikely(!info->oti_it_ea_buf_used)) {
6501                 oie->oie_buf = info->oti_it_ea_buf;
6502                 info->oti_it_ea_buf_used = 1;
6503         } else {
6504                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6505                 if (oie->oie_buf == NULL)
6506                         RETURN(ERR_PTR(-ENOMEM));
6507         }
6508         oie->oie_obj = obj;
6509
6510         file = &oie->oie_file;
6511
6512         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6513         if (attr & LUDA_64BITHASH)
6514                 file->f_mode    = FMODE_64BITHASH;
6515         else
6516                 file->f_mode    = FMODE_32BITHASH;
6517         file->f_path.dentry     = obj_dentry;
6518         file->f_mapping         = obj->oo_inode->i_mapping;
6519         file->f_op              = obj->oo_inode->i_fop;
6520         file->f_inode = obj->oo_inode;
6521
6522         lu_object_get(lo);
6523         RETURN((struct dt_it *)oie);
6524 }
6525
6526 /**
6527  * Destroy or finishes iterator context.
6528  *
6529  * \param di iterator structure to be destroyed
6530  */
6531 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6532 {
6533         struct osd_thread_info *info = osd_oti_get(env);
6534         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6535         struct osd_object *obj = oie->oie_obj;
6536         struct inode *inode = obj->oo_inode;
6537
6538         ENTRY;
6539         oie->oie_file.f_op->release(inode, &oie->oie_file);
6540         osd_object_put(env, obj);
6541         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6542                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6543         else
6544                 info->oti_it_ea_buf_used = 0;
6545         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6546         EXIT;
6547 }
6548
6549 /**
6550  * It position the iterator at given key, so that next lookup continues from
6551  * that key Or it is similar to dio_it->load() but based on a key,
6552  * rather than file position.
6553  *
6554  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6555  * to the beginning.
6556  *
6557  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6558  */
6559 static int osd_it_ea_get(const struct lu_env *env,
6560                          struct dt_it *di, const struct dt_key *key)
6561 {
6562         struct osd_it_ea *it = (struct osd_it_ea *)di;
6563
6564         ENTRY;
6565         LASSERT(((const char *)key)[0] == '\0');
6566         it->oie_file.f_pos = 0;
6567         it->oie_rd_dirent = 0;
6568         it->oie_it_dirent = 0;
6569         it->oie_dirent = NULL;
6570
6571         RETURN(+1);
6572 }
6573
6574 /**
6575  * Does nothing
6576  */
6577 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6578 {
6579 }
6580
6581 struct osd_filldir_cbs {
6582         struct dir_context ctx;
6583         struct osd_it_ea  *it;
6584 };
6585 /**
6586  * It is called internally by ->iterate*(). It fills the
6587  * iterator's in-memory data structure with required
6588  * information i.e. name, namelen, rec_size etc.
6589  *
6590  * \param buf in which information to be filled in.
6591  * \param name name of the file in given dir
6592  *
6593  * \retval 0 on success
6594  * \retval 1 on buffer full
6595  */
6596 #ifdef HAVE_FILLDIR_USE_CTX
6597 static int osd_ldiskfs_filldir(struct dir_context *buf,
6598 #else
6599 static int osd_ldiskfs_filldir(void *buf,
6600 #endif
6601                                const char *name, int namelen,
6602                                loff_t offset, __u64 ino, unsigned int d_type)
6603 {
6604         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6605         struct osd_object *obj = it->oie_obj;
6606         struct osd_it_ea_dirent *ent = it->oie_dirent;
6607         struct lu_fid *fid = &ent->oied_fid;
6608         struct osd_fid_pack *rec;
6609
6610         ENTRY;
6611
6612 /* this should never happen */
6613         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6614                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6615                 RETURN(-EIO);
6616         }
6617
6618         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6619             OSD_IT_EA_BUFSIZE)
6620                 RETURN(1);
6621
6622         /* "." is just the object itself. */
6623         if (namelen == 1 && name[0] == '.') {
6624                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6625         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6626                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6627                 if (osd_fid_unpack(fid, rec) != 0)
6628                         fid_zero(fid);
6629         } else {
6630                 fid_zero(fid);
6631         }
6632         d_type &= ~LDISKFS_DIRENT_LUFID;
6633
6634         /* NOT export local root. */
6635         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6636                 ino = obj->oo_inode->i_ino;
6637                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6638         }
6639
6640         ent->oied_ino     = ino;
6641         ent->oied_off     = offset;
6642         ent->oied_namelen = namelen;
6643         ent->oied_type    = d_type;
6644
6645         memcpy(ent->oied_name, name, namelen);
6646
6647         it->oie_rd_dirent++;
6648         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6649         RETURN(0);
6650 }
6651
6652 /**
6653  * Calls ->iterate*() to load a directory entry at a time
6654  * and stored it in iterator's in-memory data structure.
6655  *
6656  * \param di iterator's in memory structure
6657  *
6658  * \retval   0 on success
6659  * \retval -ve on error
6660  * \retval +1 reach the end of entry
6661  */
6662 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6663                                const struct dt_it *di)
6664 {
6665         struct osd_it_ea *it = (struct osd_it_ea *)di;
6666         struct osd_object *obj = it->oie_obj;
6667         struct inode *inode = obj->oo_inode;
6668         struct htree_lock *hlock = NULL;
6669         struct file *filp = &it->oie_file;
6670         int rc = 0;
6671         struct osd_filldir_cbs buf = {
6672                 .ctx.actor = osd_ldiskfs_filldir,
6673                 .it = it
6674         };
6675
6676         ENTRY;
6677         it->oie_dirent = it->oie_buf;
6678         it->oie_rd_dirent = 0;
6679
6680         if (obj->oo_hl_head != NULL) {
6681                 hlock = osd_oti_get(env)->oti_hlock;
6682                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6683                                    inode, LDISKFS_HLOCK_READDIR);
6684         } else {
6685                 down_read(&obj->oo_ext_idx_sem);
6686         }
6687
6688         filp->f_cred = current_cred();
6689         rc = osd_security_file_alloc(filp);
6690         if (rc)
6691                 RETURN(rc);
6692
6693         filp->f_flags |= O_NOATIME;
6694         filp->f_mode |= FMODE_NONOTIFY;
6695         rc = iterate_dir(filp, &buf.ctx);
6696         if (rc)
6697                 RETURN(rc);
6698
6699         if (hlock != NULL)
6700                 ldiskfs_htree_unlock(hlock);
6701         else
6702                 up_read(&obj->oo_ext_idx_sem);
6703
6704         if (it->oie_rd_dirent == 0) {
6705                 /*
6706                  * If it does not get any dirent, it means it has been reached
6707                  * to the end of the dir
6708                  */
6709                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6710                 if (rc == 0)
6711                         rc = 1;
6712         } else {
6713                 it->oie_dirent = it->oie_buf;
6714                 it->oie_it_dirent = 1;
6715         }
6716
6717         RETURN(rc);
6718 }
6719
6720 /**
6721  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6722  * to load a directory entry at a time and stored it in
6723  * iterator's in-memory data structure.
6724  *
6725  * \param di iterator's in memory structure
6726  *
6727  * \retval +ve iterator reached to end
6728  * \retval   0 iterator not reached to end
6729  * \retval -ve on error
6730  */
6731 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6732 {
6733         struct osd_it_ea *it = (struct osd_it_ea *)di;
6734         int rc;
6735
6736         ENTRY;
6737
6738         if (it->oie_it_dirent < it->oie_rd_dirent) {
6739                 it->oie_dirent =
6740                         (void *)it->oie_dirent +
6741                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6742                                        it->oie_dirent->oied_namelen);
6743                 it->oie_it_dirent++;
6744                 rc = 0;
6745         } else {
6746                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6747                         rc = 1;
6748                 else
6749                         rc = osd_ldiskfs_it_fill(env, di);
6750         }
6751
6752         RETURN(rc);
6753 }
6754
6755 /**
6756  * Returns the key at current position from iterator's in memory structure.
6757  *
6758  * \param di iterator's in memory structure
6759  *
6760  * \retval key i.e. struct dt_key on success
6761  */
6762 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6763                                     const struct dt_it *di)
6764 {
6765         struct osd_it_ea *it = (struct osd_it_ea *)di;
6766
6767         return (struct dt_key *)it->oie_dirent->oied_name;
6768 }
6769
6770 /**
6771  * Returns key's size at current position from iterator's in memory structure.
6772  *
6773  * \param di iterator's in memory structure
6774  *
6775  * \retval key_size i.e. struct dt_key on success
6776  */
6777 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6778 {
6779         struct osd_it_ea *it = (struct osd_it_ea *)di;
6780
6781         return it->oie_dirent->oied_namelen;
6782 }
6783
6784 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6785 #undef LDISKFS_DIR_REC_LEN
6786 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6787 #endif
6788
6789 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6790 {
6791         if (LDISKFS_DIR_REC_LEN(de) >=
6792             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6793                 return true;
6794
6795         return false;
6796 }
6797
6798 static inline bool
6799 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6800                      unsigned int blocksize, bool dotdot)
6801 {
6802         if (dotdot)
6803                 return osd_dotdot_has_space(de);
6804
6805         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6806             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6807                 return true;
6808
6809         return false;
6810 }
6811
6812 static int
6813 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6814                     handle_t *jh, struct dentry *dentry,
6815                     const struct lu_fid *fid, struct buffer_head *bh,
6816                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6817                     bool dotdot)
6818 {
6819         struct inode *dir = dentry->d_parent->d_inode;
6820         struct inode *inode = dentry->d_inode;
6821         struct osd_fid_pack *rec;
6822         struct ldiskfs_dentry_param *ldp;
6823         int namelen = dentry->d_name.len;
6824         int rc;
6825         struct osd_thread_info *info = osd_oti_get(env);
6826
6827         ENTRY;
6828
6829         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6830                 RETURN(0);
6831
6832         /* There is enough space to hold the FID-in-dirent. */
6833         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6834                 rc = ldiskfs_journal_get_write_access(jh, bh);
6835                 if (rc != 0)
6836                         RETURN(rc);
6837
6838                 de->name[namelen] = 0;
6839                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6840                 rec->fp_len = sizeof(struct lu_fid) + 1;
6841                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6842                 de->file_type |= LDISKFS_DIRENT_LUFID;
6843                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6844
6845                 RETURN(rc);
6846         }
6847
6848         LASSERT(!dotdot);
6849
6850         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6851         if (rc != 0)
6852                 RETURN(rc);
6853
6854         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6855         osd_get_ldiskfs_dirent_param(ldp, fid);
6856         dentry->d_fsdata = (void *)ldp;
6857         dquot_initialize(dir);
6858         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6859         /*
6860          * It is too bad, we cannot reinsert the name entry back.
6861          * That means we lose it!
6862          */
6863         if (rc != 0)
6864                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6865                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6866                        osd_ino2name(inode),
6867                        dir->i_ino, dir->i_generation, namelen,
6868                        dentry->d_name.name, PFID(fid), rc);
6869
6870         RETURN(rc);
6871 }
6872
6873 static int
6874 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6875                         struct osd_it_ea *it, struct lu_fid *fid,
6876                         struct osd_inode_id *id, __u32 *attr)
6877 {
6878         struct osd_thread_info *info = osd_oti_get(env);
6879         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6880         struct osd_device *dev = osd_obj2dev(obj);
6881         struct super_block *sb = osd_sb(dev);
6882         const char *devname = osd_name(dev);
6883         struct osd_it_ea_dirent *ent = it->oie_dirent;
6884         struct inode *dir = obj->oo_inode;
6885         struct htree_lock *hlock = NULL;
6886         struct buffer_head *bh = NULL;
6887         handle_t *jh = NULL;
6888         struct ldiskfs_dir_entry_2 *de;
6889         struct dentry *dentry;
6890         struct inode *inode;
6891         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6892         int credits;
6893         int rc;
6894         bool dotdot = false;
6895         bool dirty = false;
6896
6897         ENTRY;
6898
6899         if (ent->oied_name[0] == '.') {
6900                 if (ent->oied_namelen == 1)
6901                         RETURN(0);
6902
6903                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6904                         dotdot = true;
6905         }
6906
6907         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6908         inode = osd_iget(info, dev, id);
6909         if (IS_ERR(inode)) {
6910                 rc = PTR_ERR(inode);
6911                 if (rc == -ENOENT || rc == -ESTALE) {
6912                         /*
6913                          * Maybe dangling name entry, or
6914                          * corrupted directory entry.
6915                          */
6916                         *attr |= LUDA_UNKNOWN;
6917                         rc = 0;
6918                 } else {
6919                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6920                                "check_repair, dir = %lu/%u, name = %.*s, "
6921                                "ino = %llu, rc = %d\n",
6922                                devname, dir->i_ino, dir->i_generation,
6923                                ent->oied_namelen, ent->oied_name,
6924                                ent->oied_ino, rc);
6925                 }
6926
6927                 RETURN(rc);
6928         }
6929
6930         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6931                                            ent->oied_namelen);
6932         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6933         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6934                 lma = NULL;
6935         else if (rc != 0)
6936                 GOTO(out, rc);
6937
6938         /*
6939          * We need to ensure that the name entry is still valid.
6940          * Because it may be removed or renamed by other already.
6941          *
6942          * The unlink or rename operation will start journal before PDO lock,
6943          * so to avoid deadlock, here we need to start journal handle before
6944          * related PDO lock also. But because we do not know whether there
6945          * will be something to be repaired before PDO lock, we just start
6946          * journal without conditions.
6947          *
6948          * We may need to remove the name entry firstly, then insert back.
6949          * One credit is for user quota file update.
6950          * One credit is for group quota file update.
6951          * Two credits are for dirty inode.
6952          */
6953         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6954                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6955
6956         if (dev->od_dirent_journal != 0) {
6957
6958 again:
6959                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6960                 if (IS_ERR(jh)) {
6961                         rc = PTR_ERR(jh);
6962                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6963                                "check_repair, dir = %lu/%u, credits = %d, "
6964                                "name = %.*s, ino = %llu: rc = %d\n",
6965                                devname, dir->i_ino, dir->i_generation, credits,
6966                                ent->oied_namelen, ent->oied_name,
6967                                ent->oied_ino, rc);
6968
6969                         GOTO(out_inode, rc);
6970                 }
6971
6972                 if (obj->oo_hl_head != NULL) {
6973                         hlock = osd_oti_get(env)->oti_hlock;
6974                         /*
6975                          * "0" means exclusive lock for the whole directory.
6976                          * We need to prevent others access such name entry
6977                          * during the delete + insert. Neither HLOCK_ADD nor
6978                          * HLOCK_DEL cannot guarantee the atomicity.
6979                          */
6980                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6981                 } else {
6982                         down_write(&obj->oo_ext_idx_sem);
6983                 }
6984         } else {
6985                 if (obj->oo_hl_head != NULL) {
6986                         hlock = osd_oti_get(env)->oti_hlock;
6987                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6988                                            LDISKFS_HLOCK_LOOKUP);
6989                 } else {
6990                         down_read(&obj->oo_ext_idx_sem);
6991                 }
6992         }
6993
6994         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6995         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6996                 *attr |= LUDA_IGNORE;
6997
6998                 GOTO(out, rc = 0);
6999         }
7000
7001         /*
7002          * For dotdot entry, if there is not enough space to hold the
7003          * FID-in-dirent, just keep them there. It only happens when the
7004          * device upgraded from 1.8 or restored from MDT file-level backup.
7005          * For the whole directory, only dotdot entry have no FID-in-dirent
7006          * and needs to get FID from LMA when readdir, it will not affect the
7007          * performance much.
7008          */
7009         if (dotdot && !osd_dotdot_has_space(de)) {
7010                 *attr |= LUDA_UNKNOWN;
7011
7012                 GOTO(out, rc = 0);
7013         }
7014
7015         if (lma != NULL) {
7016                 if (lu_fid_eq(fid, &lma->lma_self_fid))
7017                         GOTO(out, rc = 0);
7018
7019                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
7020                         struct lu_fid *tfid = &lma->lma_self_fid;
7021
7022                         if (likely(dotdot &&
7023                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
7024                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
7025                                 /*
7026                                  * It must be REMOTE_PARENT_DIR and as the
7027                                  * 'dotdot' entry of remote directory
7028                                  */
7029                                 *attr |= LUDA_IGNORE;
7030                         } else {
7031                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
7032                                        "parent directory, but got %.*s under "
7033                                        "dir = %lu/%u with the FID "DFID"\n",
7034                                        devname, ent->oied_namelen,
7035                                        ent->oied_name, dir->i_ino,
7036                                        dir->i_generation, PFID(tfid));
7037
7038                                 *attr |= LUDA_UNKNOWN;
7039                         }
7040
7041                         GOTO(out, rc = 0);
7042                 }
7043         }
7044
7045         if (!fid_is_zero(fid)) {
7046                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
7047                                               ent->oied_namelen);
7048                 if (rc == -ENOENT ||
7049                     (rc == -ENODATA &&
7050                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
7051                         /*
7052                          * linkEA does not recognize the dirent entry,
7053                          * it may because the dirent entry corruption
7054                          * and points to other's inode.
7055                          */
7056                         CDEBUG(D_LFSCK, "%s: the target inode does not "
7057                                "recognize the dirent, dir = %lu/%u, "
7058                                " name = %.*s, ino = %llu, "
7059                                DFID": rc = %d\n", devname, dir->i_ino,
7060                                dir->i_generation, ent->oied_namelen,
7061                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7062                         *attr |= LUDA_UNKNOWN;
7063
7064                         GOTO(out, rc = 0);
7065                 }
7066
7067                 if (rc && rc != -ENODATA) {
7068                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7069                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7070                                DFID": rc = %d\n", devname, dir->i_ino,
7071                                dir->i_generation, ent->oied_namelen,
7072                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7073                         *attr |= LUDA_UNKNOWN;
7074
7075                         GOTO(out, rc = 0);
7076                 }
7077         }
7078
7079         if (lma != NULL) {
7080                 /*
7081                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7082                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7083                  */
7084                 if (*attr & LUDA_VERIFY_DRYRUN) {
7085                         *fid = lma->lma_self_fid;
7086                         *attr |= LUDA_REPAIR;
7087
7088                         GOTO(out, rc = 0);
7089                 }
7090
7091                 if (jh == NULL) {
7092                         brelse(bh);
7093                         dev->od_dirent_journal = 1;
7094                         if (hlock != NULL) {
7095                                 ldiskfs_htree_unlock(hlock);
7096                                 hlock = NULL;
7097                         } else {
7098                                 up_read(&obj->oo_ext_idx_sem);
7099                         }
7100
7101                         goto again;
7102                 }
7103
7104                 *fid = lma->lma_self_fid;
7105                 dirty = true;
7106                 /* Update or append the FID-in-dirent. */
7107                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7108                                          bh, de, hlock, dotdot);
7109                 if (rc == 0)
7110                         *attr |= LUDA_REPAIR;
7111                 else
7112                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7113                                "the dirent, dir = %lu/%u, name = %.*s, "
7114                                "ino = %llu, "DFID": rc = %d\n",
7115                                devname, dir->i_ino, dir->i_generation,
7116                                ent->oied_namelen, ent->oied_name,
7117                                ent->oied_ino, PFID(fid), rc);
7118         } else {
7119                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7120                 if (*attr & LUDA_VERIFY_DRYRUN) {
7121                         if (fid_is_sane(fid)) {
7122                                 *attr |= LUDA_REPAIR;
7123                         } else if (dev->od_index == 0) {
7124                                 lu_igif_build(fid, inode->i_ino,
7125                                               inode->i_generation);
7126                                 *attr |= LUDA_UPGRADE;
7127                         }
7128
7129                         GOTO(out, rc = 0);
7130                 }
7131
7132                 if (jh == NULL) {
7133                         brelse(bh);
7134                         dev->od_dirent_journal = 1;
7135                         if (hlock != NULL) {
7136                                 ldiskfs_htree_unlock(hlock);
7137                                 hlock = NULL;
7138                         } else {
7139                                 up_read(&obj->oo_ext_idx_sem);
7140                         }
7141
7142                         goto again;
7143                 }
7144
7145                 dirty = true;
7146                 if (unlikely(fid_is_sane(fid))) {
7147                         /*
7148                          * FID-in-dirent exists, but FID-in-LMA is lost.
7149                          * Trust the FID-in-dirent, and add FID-in-LMA.
7150                          */
7151                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7152                         if (rc == 0)
7153                                 *attr |= LUDA_REPAIR;
7154                         else
7155                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7156                                        "update dirent, dir = %lu/%u, "
7157                                        "name = %.*s, ino = %llu, "
7158                                        DFID": rc = %d\n",
7159                                        devname, dir->i_ino, dir->i_generation,
7160                                        ent->oied_namelen, ent->oied_name,
7161                                        ent->oied_ino, PFID(fid), rc);
7162                 } else if (dev->od_index == 0) {
7163                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7164                         /*
7165                          * It is probably IGIF object. Only aappend the
7166                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7167                          */
7168                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7169                                                  bh, de, hlock, dotdot);
7170                         if (rc == 0)
7171                                 *attr |= LUDA_UPGRADE;
7172                         else
7173                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7174                                        "after the dirent, dir = %lu/%u, "
7175                                        "name = %.*s, ino = %llu, "
7176                                        DFID": rc = %d\n",
7177                                        devname, dir->i_ino, dir->i_generation,
7178                                        ent->oied_namelen, ent->oied_name,
7179                                        ent->oied_ino, PFID(fid), rc);
7180                 }
7181         }
7182
7183         GOTO(out, rc);
7184
7185 out:
7186         if (!IS_ERR(bh))
7187                 brelse(bh);
7188         if (hlock != NULL) {
7189                 ldiskfs_htree_unlock(hlock);
7190         } else {
7191                 if (dev->od_dirent_journal != 0)
7192                         up_write(&obj->oo_ext_idx_sem);
7193                 else
7194                         up_read(&obj->oo_ext_idx_sem);
7195         }
7196
7197         if (jh != NULL)
7198                 ldiskfs_journal_stop(jh);
7199
7200 out_inode:
7201         iput(inode);
7202         if (rc >= 0 && !dirty)
7203                 dev->od_dirent_journal = 0;
7204
7205         return rc;
7206 }
7207
7208 /**
7209  * Returns the value at current position from iterator's in memory structure.
7210  *
7211  * \param di struct osd_it_ea, iterator's in memory structure
7212  * \param attr attr requested for dirent.
7213  * \param lde lustre dirent
7214  *
7215  * \retval   0 no error and \param lde has correct lustre dirent.
7216  * \retval -ve on error
7217  */
7218 static inline int osd_it_ea_rec(const struct lu_env *env,
7219                                 const struct dt_it *di,
7220                                 struct dt_rec *dtrec, __u32 attr)
7221 {
7222         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7223         struct osd_object      *obj   = it->oie_obj;
7224         struct osd_device      *dev   = osd_obj2dev(obj);
7225         struct osd_thread_info *oti   = osd_oti_get(env);
7226         struct osd_inode_id    *id    = &oti->oti_id;
7227         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7228         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7229         __u32 ino = it->oie_dirent->oied_ino;
7230         int rc = 0;
7231
7232         ENTRY;
7233
7234         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7235
7236         if (attr & LUDA_VERIFY) {
7237                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7238                         attr |= LUDA_IGNORE;
7239                         /*
7240                          * If the parent is on remote MDT, and there
7241                          * is no FID-in-dirent, then we have to get
7242                          * the parent FID from the linkEA.
7243                          */
7244                         if (!fid_is_sane(fid) &&
7245                             it->oie_dirent->oied_namelen == 2 &&
7246                             it->oie_dirent->oied_name[0] == '.' &&
7247                             it->oie_dirent->oied_name[1] == '.')
7248                                 osd_get_pfid_from_linkea(env, obj, fid);
7249                 } else {
7250                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7251                                                      &attr);
7252                 }
7253
7254                 if (!fid_is_sane(fid))
7255                         attr |= LUDA_UNKNOWN;
7256         } else {
7257                 attr &= ~LU_DIRENT_ATTRS_MASK;
7258                 if (!fid_is_sane(fid)) {
7259                         bool is_dotdot = false;
7260
7261                         if (it->oie_dirent->oied_namelen == 2 &&
7262                             it->oie_dirent->oied_name[0] == '.' &&
7263                             it->oie_dirent->oied_name[1] == '.')
7264                                 is_dotdot = true;
7265                         /*
7266                          * If the parent is on remote MDT, and there
7267                          * is no FID-in-dirent, then we have to get
7268                          * the parent FID from the linkEA.
7269                          */
7270                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7271                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7272                         } else {
7273                                 if (is_dotdot == false &&
7274                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7275                                         RETURN(-ENOENT);
7276
7277                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7278                         }
7279                 } else {
7280                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7281                 }
7282         }
7283
7284         /* Pack the entry anyway, at least the offset is right. */
7285         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7286                            it->oie_dirent->oied_name,
7287                            it->oie_dirent->oied_namelen,
7288                            it->oie_dirent->oied_type, attr);
7289
7290         if (rc < 0)
7291                 RETURN(rc);
7292
7293         if (osd_remote_fid(env, dev, fid))
7294                 RETURN(0);
7295
7296         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7297                 osd_add_oi_cache(oti, dev, id, fid);
7298
7299         RETURN(rc > 0 ? 0 : rc);
7300 }
7301
7302 /**
7303  * Returns the record size size at current position.
7304  *
7305  * This function will return record(lu_dirent) size in bytes.
7306  *
7307  * \param[in] env       execution environment
7308  * \param[in] di        iterator's in memory structure
7309  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7310  *                      calculate the lu_dirent size.
7311  *
7312  * \retval      record size(in bytes & in memory) of the current lu_dirent
7313  *              entry.
7314  */
7315 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7316                               __u32 attr)
7317 {
7318         struct osd_it_ea *it = (struct osd_it_ea *)di;
7319
7320         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7321 }
7322
7323 /**
7324  * Returns a cookie for current position of the iterator head, so that
7325  * user can use this cookie to load/start the iterator next time.
7326  *
7327  * \param di iterator's in memory structure
7328  *
7329  * \retval cookie for current position, on success
7330  */
7331 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7332 {
7333         struct osd_it_ea *it = (struct osd_it_ea *)di;
7334
7335         return it->oie_dirent->oied_off;
7336 }
7337
7338 /**
7339  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7340  * to load a directory entry at a time and stored it i inn,
7341  * in iterator's in-memory data structure.
7342  *
7343  * \param di struct osd_it_ea, iterator's in memory structure
7344  *
7345  * \retval +ve on success
7346  * \retval -ve on error
7347  */
7348 static int osd_it_ea_load(const struct lu_env *env,
7349                           const struct dt_it *di, __u64 hash)
7350 {
7351         struct osd_it_ea *it = (struct osd_it_ea *)di;
7352         int rc;
7353
7354         ENTRY;
7355         it->oie_file.f_pos = hash;
7356
7357         rc =  osd_ldiskfs_it_fill(env, di);
7358         if (rc > 0)
7359                 rc = -ENODATA;
7360
7361         if (rc == 0)
7362                 rc = 1;
7363
7364         RETURN(rc);
7365 }
7366
7367 /**
7368  * Index lookup function for interoperability mode (b11826).
7369  *
7370  * \param key,  key i.e. file name to be searched
7371  *
7372  * \retval +ve, on success
7373  * \retval -ve, on error
7374  */
7375 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7376                                struct dt_rec *rec, const struct dt_key *key)
7377 {
7378         struct osd_object *obj = osd_dt_obj(dt);
7379         int rc = 0;
7380
7381         ENTRY;
7382
7383         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7384         LINVRNT(osd_invariant(obj));
7385
7386         rc = osd_ea_lookup_rec(env, obj, rec, key);
7387         if (rc == 0)
7388                 rc = 1;
7389         RETURN(rc);
7390 }
7391
7392 /**
7393  * Index and Iterator operations for interoperability
7394  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7395  */
7396 static const struct dt_index_operations osd_index_ea_ops = {
7397         .dio_lookup         = osd_index_ea_lookup,
7398         .dio_declare_insert = osd_index_declare_ea_insert,
7399         .dio_insert         = osd_index_ea_insert,
7400         .dio_declare_delete = osd_index_declare_ea_delete,
7401         .dio_delete         = osd_index_ea_delete,
7402         .dio_it     = {
7403                 .init     = osd_it_ea_init,
7404                 .fini     = osd_it_ea_fini,
7405                 .get      = osd_it_ea_get,
7406                 .put      = osd_it_ea_put,
7407                 .next     = osd_it_ea_next,
7408                 .key      = osd_it_ea_key,
7409                 .key_size = osd_it_ea_key_size,
7410                 .rec      = osd_it_ea_rec,
7411                 .rec_size = osd_it_ea_rec_size,
7412                 .store    = osd_it_ea_store,
7413                 .load     = osd_it_ea_load
7414         }
7415 };
7416
7417 static void *osd_key_init(const struct lu_context *ctx,
7418                           struct lu_context_key *key)
7419 {
7420         struct osd_thread_info *info;
7421
7422         OBD_ALLOC_PTR(info);
7423         if (info == NULL)
7424                 return ERR_PTR(-ENOMEM);
7425
7426         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7427         if (info->oti_it_ea_buf == NULL)
7428                 goto out_free_info;
7429
7430         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7431
7432         info->oti_hlock = ldiskfs_htree_lock_alloc();
7433         if (info->oti_hlock == NULL)
7434                 goto out_free_ea;
7435
7436         return info;
7437
7438 out_free_ea:
7439         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7440 out_free_info:
7441         OBD_FREE_PTR(info);
7442         return ERR_PTR(-ENOMEM);
7443 }
7444
7445 static void osd_key_fini(const struct lu_context *ctx,
7446                          struct lu_context_key *key, void *data)
7447 {
7448         struct osd_thread_info *info = data;
7449         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7450         struct osd_idmap_cache *idc = info->oti_ins_cache;
7451
7452         if (info->oti_dio_pages) {
7453                 int i;
7454                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7455                         struct page *page = info->oti_dio_pages[i];
7456                         if (page) {
7457                                 LASSERT(PagePrivate2(page));
7458                                 LASSERT(PageLocked(page));
7459                                 ClearPagePrivate2(page);
7460                                 unlock_page(page);
7461                                 __free_page(page);
7462                         }
7463                 }
7464                 OBD_FREE_PTR_ARRAY(info->oti_dio_pages, PTLRPC_MAX_BRW_PAGES);
7465         }
7466
7467         if (info->oti_inode != NULL)
7468                 OBD_FREE_PTR(lli);
7469         if (info->oti_hlock != NULL)
7470                 ldiskfs_htree_lock_free(info->oti_hlock);
7471         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7472         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7473         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7474         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7475         lu_buf_free(&info->oti_big_buf);
7476         if (idc != NULL) {
7477                 LASSERT(info->oti_ins_cache_size > 0);
7478                 OBD_FREE_PTR_ARRAY(idc, info->oti_ins_cache_size);
7479                 info->oti_ins_cache = NULL;
7480                 info->oti_ins_cache_size = 0;
7481         }
7482         OBD_FREE_PTR(info);
7483 }
7484
7485 static void osd_key_exit(const struct lu_context *ctx,
7486                          struct lu_context_key *key, void *data)
7487 {
7488         struct osd_thread_info *info = data;
7489
7490         LASSERT(info->oti_r_locks == 0);
7491         LASSERT(info->oti_w_locks == 0);
7492         LASSERT(info->oti_txns    == 0);
7493 }
7494
7495 /* type constructor/destructor: osd_type_init, osd_type_fini */
7496 LU_TYPE_INIT_FINI(osd, &osd_key);
7497
7498 struct lu_context_key osd_key = {
7499         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7500         .lct_init = osd_key_init,
7501         .lct_fini = osd_key_fini,
7502         .lct_exit = osd_key_exit
7503 };
7504
7505
7506 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7507                            const char *name, struct lu_device *next)
7508 {
7509         struct osd_device *osd = osd_dev(d);
7510
7511         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7512             sizeof(osd->od_svname))
7513                 return -E2BIG;
7514         return osd_procfs_init(osd, name);
7515 }
7516
7517 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7518 {
7519         struct seq_server_site *ss = osd_seq_site(osd);
7520         int rc = 0;
7521
7522         ENTRY;
7523
7524         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7525                 RETURN(0);
7526
7527         if (unlikely(ss == NULL))
7528                 RETURN(-ENODEV);
7529
7530         OBD_ALLOC_PTR(osd->od_cl_seq);
7531         if (osd->od_cl_seq == NULL)
7532                 RETURN(-ENOMEM);
7533
7534         seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7535                         osd->od_svname, ss->ss_server_seq);
7536
7537         if (ss->ss_node_id == 0) {
7538                 /*
7539                  * If the OSD on the sequence controller(MDT0), then allocate
7540                  * sequence here, otherwise allocate sequence after connected
7541                  * to MDT0 (see mdt_register_lwp_callback()).
7542                  */
7543                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7544                                    &osd->od_cl_seq->lcs_space, env);
7545         }
7546
7547         RETURN(rc);
7548 }
7549
7550 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7551 {
7552         if (osd->od_cl_seq == NULL)
7553                 return;
7554
7555         seq_client_fini(osd->od_cl_seq);
7556         OBD_FREE_PTR(osd->od_cl_seq);
7557         osd->od_cl_seq = NULL;
7558 }
7559
7560 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7561 {
7562         ENTRY;
7563
7564         /* shutdown quota slave instance associated with the device */
7565         if (o->od_quota_slave_md != NULL) {
7566                 struct qsd_instance *qsd = o->od_quota_slave_md;
7567
7568                 o->od_quota_slave_md = NULL;
7569                 qsd_fini(env, qsd);
7570         }
7571
7572         if (o->od_quota_slave_dt != NULL) {
7573                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7574
7575                 o->od_quota_slave_dt = NULL;
7576                 qsd_fini(env, qsd);
7577         }
7578
7579         osd_fid_fini(env, o);
7580         osd_scrub_cleanup(env, o);
7581
7582         RETURN(0);
7583 }
7584
7585 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7586 {
7587         ENTRY;
7588
7589         if (o->od_mnt != NULL) {
7590                 shrink_dcache_sb(osd_sb(o));
7591                 osd_sync(env, &o->od_dt_dev);
7592
7593                 mntput(o->od_mnt);
7594                 o->od_mnt = NULL;
7595         }
7596
7597         EXIT;
7598 }
7599
7600 static int osd_mount(const struct lu_env *env,
7601                      struct osd_device *o, struct lustre_cfg *cfg)
7602 {
7603         const char *name = lustre_cfg_string(cfg, 0);
7604         const char *dev = lustre_cfg_string(cfg, 1);
7605         const char *opts;
7606         unsigned long page, s_flags, lmd_flags = 0;
7607         struct page *__page;
7608         struct file_system_type *type;
7609         char *options = NULL;
7610         char *str;
7611         struct osd_thread_info *info = osd_oti_get(env);
7612         struct lu_fid *fid = &info->oti_fid;
7613         struct inode *inode;
7614         int rc = 0, force_over_1024tb = 0;
7615
7616         ENTRY;
7617
7618         if (o->od_mnt != NULL)
7619                 RETURN(0);
7620
7621         if (strlen(dev) >= sizeof(o->od_mntdev))
7622                 RETURN(-E2BIG);
7623         strcpy(o->od_mntdev, dev);
7624
7625         str = lustre_cfg_string(cfg, 2);
7626         s_flags = simple_strtoul(str, NULL, 0);
7627         str = strstr(str, ":");
7628         if (str)
7629                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7630         opts = lustre_cfg_string(cfg, 3);
7631 #ifdef __BIG_ENDIAN
7632         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7633                 CERROR("%s: device %s extents feature is not guaranteed to "
7634                        "work on big-endian systems. Use \"bigendian_extents\" "
7635                        "mount option to override.\n", name, dev);
7636                 RETURN(-EINVAL);
7637         }
7638 #endif
7639 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7640         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7641                 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");
7642         }
7643 #endif
7644 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7645         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7646                 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");
7647         }
7648 #endif
7649 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 53, 0)
7650         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL) {
7651                 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");
7652         }
7653 #endif
7654
7655         if (opts != NULL && strstr(opts, "force_over_1024tb") != NULL)
7656                 force_over_1024tb = 1;
7657
7658         __page = alloc_page(GFP_KERNEL);
7659         if (__page == NULL)
7660                 GOTO(out, rc = -ENOMEM);
7661         page = (unsigned long)page_address(__page);
7662         options = (char *)page;
7663         *options = '\0';
7664         if (opts != NULL) {
7665                 /* strip out the options for back compatiblity */
7666                 static const char * const sout[] = {
7667                         "mballoc",
7668                         "iopen",
7669                         "noiopen",
7670                         "iopen_nopriv",
7671                         "extents",
7672                         "noextents",
7673                         /* strip out option we processed in osd */
7674                         "bigendian_extents",
7675                         "force_over_128tb",
7676                         "force_over_256tb",
7677                         "force_over_512tb",
7678                         "force_over_1024tb",
7679                         NULL
7680                 };
7681                 strncat(options, opts, PAGE_SIZE);
7682                 for (rc = 0, str = options; sout[rc]; ) {
7683                         char *op = strstr(str, sout[rc]);
7684
7685                         if (op == NULL) {
7686                                 rc++;
7687                                 str = options;
7688                                 continue;
7689                         }
7690                         if (op == options || *(op - 1) == ',') {
7691                                 str = op + strlen(sout[rc]);
7692                                 if (*str == ',' || *str == '\0') {
7693                                         *str == ',' ? str++ : str;
7694                                         memmove(op, str, strlen(str) + 1);
7695                                 }
7696                         }
7697                         for (str = op; *str != ',' && *str != '\0'; str++)
7698                                 ;
7699                 }
7700         } else {
7701                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7702         }
7703
7704         /* Glom up mount options */
7705         if (*options != '\0')
7706                 strncat(options, ",", PAGE_SIZE);
7707         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7708
7709         type = get_fs_type("ldiskfs");
7710         if (!type) {
7711                 CERROR("%s: cannot find ldiskfs module\n", name);
7712                 GOTO(out, rc = -ENODEV);
7713         }
7714
7715         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7716         module_put(type->owner);
7717
7718         if (IS_ERR(o->od_mnt)) {
7719                 rc = PTR_ERR(o->od_mnt);
7720                 o->od_mnt = NULL;
7721                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7722                 GOTO(out, rc);
7723         }
7724
7725         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7726                                  osd_sb(o)->s_blocksize_bits > 1024ULL << 40 &&
7727                                  force_over_1024tb == 0) {
7728                 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",
7729                        name, dev);
7730                 GOTO(out_mnt, rc = -EINVAL);
7731         }
7732
7733         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7734                 LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7735                               name);
7736
7737                 GOTO(out_mnt, rc = -EOPNOTSUPP);
7738         }
7739
7740         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7741                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7742                 GOTO(out_mnt, rc = -EINVAL);
7743         }
7744
7745 #ifdef LDISKFS_MOUNT_DIRDATA
7746         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7747                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7748         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7749                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7750                       "enabling the dirdata feature. If you do not want to "
7751                       "downgrade to Lustre-1.x again, you can enable it via "
7752                       "'tune2fs -O dirdata device'\n", name, dev);
7753 #endif
7754         inode = osd_sb(o)->s_root->d_inode;
7755         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7756         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7757         if (rc != 0) {
7758                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7759                 GOTO(out_mnt, rc);
7760         }
7761
7762         if (lmd_flags & LMD_FLG_NOSCRUB)
7763                 o->od_auto_scrub_interval = AS_NEVER;
7764
7765         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7766                 /* do not use pagecache with flash-backed storage */
7767                 o->od_writethrough_cache = 0;
7768                 o->od_read_cache = 0;
7769         }
7770
7771         GOTO(out, rc = 0);
7772
7773 out_mnt:
7774         mntput(o->od_mnt);
7775         o->od_mnt = NULL;
7776
7777 out:
7778         if (__page)
7779                 __free_page(__page);
7780
7781         return rc;
7782 }
7783
7784 static struct lu_device *osd_device_fini(const struct lu_env *env,
7785                                          struct lu_device *d)
7786 {
7787         struct osd_device *o = osd_dev(d);
7788
7789         ENTRY;
7790
7791         osd_index_backup(env, o, false);
7792         osd_shutdown(env, o);
7793         osd_procfs_fini(o);
7794         if (o->od_oi_table != NULL)
7795                 osd_oi_fini(osd_oti_get(env), o);
7796         osd_obj_map_fini(o);
7797         osd_umount(env, o);
7798
7799         RETURN(NULL);
7800 }
7801
7802 static int osd_device_init0(const struct lu_env *env,
7803                             struct osd_device *o,
7804                             struct lustre_cfg *cfg)
7805 {
7806         struct lu_device *l = osd2lu_dev(o);
7807         struct osd_thread_info *info;
7808         int rc;
7809         int cplen = 0;
7810
7811         /* if the module was re-loaded, env can loose its keys */
7812         rc = lu_env_refill((struct lu_env *)env);
7813         if (rc)
7814                 GOTO(out, rc);
7815         info = osd_oti_get(env);
7816         LASSERT(info);
7817
7818         l->ld_ops = &osd_lu_ops;
7819         o->od_dt_dev.dd_ops = &osd_dt_ops;
7820
7821         spin_lock_init(&o->od_osfs_lock);
7822         mutex_init(&o->od_otable_mutex);
7823         INIT_LIST_HEAD(&o->od_orphan_list);
7824         INIT_LIST_HEAD(&o->od_index_backup_list);
7825         INIT_LIST_HEAD(&o->od_index_restore_list);
7826         spin_lock_init(&o->od_lock);
7827         o->od_index_backup_policy = LIBP_NONE;
7828         o->od_t10_type = 0;
7829
7830         o->od_read_cache = 1;
7831         o->od_writethrough_cache = 1;
7832         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7833         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7834         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7835         o->od_auto_scrub_interval = AS_DEFAULT;
7836
7837         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7838                         sizeof(o->od_svname));
7839         if (cplen >= sizeof(o->od_svname)) {
7840                 rc = -E2BIG;
7841                 GOTO(out, rc);
7842         }
7843
7844         o->od_index_backup_stop = 0;
7845         o->od_index = -1; /* -1 means index is invalid */
7846         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7847         if (rc == LDD_F_SV_TYPE_OST)
7848                 o->od_is_ost = 1;
7849
7850         o->od_full_scrub_ratio = OFSR_DEFAULT;
7851         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7852         rc = osd_mount(env, o, cfg);
7853         if (rc != 0)
7854                 GOTO(out, rc);
7855
7856         /* Can only check block device after mount */
7857         o->od_nonrotational =
7858                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7859
7860         rc = osd_obj_map_init(env, o);
7861         if (rc != 0)
7862                 GOTO(out_mnt, rc);
7863
7864         rc = lu_site_init(&o->od_site, l);
7865         if (rc != 0)
7866                 GOTO(out_compat, rc);
7867         o->od_site.ls_bottom_dev = l;
7868
7869         rc = lu_site_init_finish(&o->od_site);
7870         if (rc != 0)
7871                 GOTO(out_site, rc);
7872
7873         INIT_LIST_HEAD(&o->od_ios_list);
7874         /* setup scrub, including OI files initialization */
7875         o->od_in_init = 1;
7876         rc = osd_scrub_setup(env, o);
7877         o->od_in_init = 0;
7878         if (rc < 0)
7879                 GOTO(out_site, rc);
7880
7881         rc = osd_procfs_init(o, o->od_svname);
7882         if (rc != 0) {
7883                 CERROR("%s: can't initialize procfs: rc = %d\n",
7884                        o->od_svname, rc);
7885                 GOTO(out_scrub, rc);
7886         }
7887
7888         LASSERT(l->ld_site->ls_linkage.next != NULL);
7889         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7890
7891         /* initialize quota slave instance */
7892         /* currently it's no need to prepare qsd_instance_md for OST */
7893         if (!o->od_is_ost) {
7894                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7895                                                 &o->od_dt_dev,
7896                                                 o->od_proc_entry, true);
7897                 if (IS_ERR(o->od_quota_slave_md)) {
7898                         rc = PTR_ERR(o->od_quota_slave_md);
7899                         o->od_quota_slave_md = NULL;
7900                         GOTO(out_procfs, rc);
7901                 }
7902         }
7903
7904         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7905                                         o->od_proc_entry, false);
7906
7907         if (IS_ERR(o->od_quota_slave_dt)) {
7908                 if (o->od_quota_slave_md != NULL) {
7909                         qsd_fini(env, o->od_quota_slave_md);
7910                         o->od_quota_slave_md = NULL;
7911                 }
7912
7913                 rc = PTR_ERR(o->od_quota_slave_dt);
7914                 o->od_quota_slave_dt = NULL;
7915                 GOTO(out_procfs, rc);
7916         }
7917
7918         RETURN(0);
7919
7920 out_procfs:
7921         osd_procfs_fini(o);
7922 out_scrub:
7923         osd_scrub_cleanup(env, o);
7924 out_site:
7925         lu_site_fini(&o->od_site);
7926 out_compat:
7927         osd_obj_map_fini(o);
7928 out_mnt:
7929         osd_umount(env, o);
7930 out:
7931         return rc;
7932 }
7933
7934 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7935                                           struct lu_device_type *t,
7936                                           struct lustre_cfg *cfg)
7937 {
7938         struct osd_device *o;
7939         int rc;
7940
7941         OBD_ALLOC_PTR(o);
7942         if (o == NULL)
7943                 return ERR_PTR(-ENOMEM);
7944
7945         rc = dt_device_init(&o->od_dt_dev, t);
7946         if (rc == 0) {
7947                 /*
7948                  * Because the ctx might be revived in dt_device_init,
7949                  * refill the env here
7950                  */
7951                 lu_env_refill((struct lu_env *)env);
7952                 rc = osd_device_init0(env, o, cfg);
7953                 if (rc)
7954                         dt_device_fini(&o->od_dt_dev);
7955         }
7956
7957         if (unlikely(rc != 0))
7958                 OBD_FREE_PTR(o);
7959
7960         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7961 }
7962
7963 static struct lu_device *osd_device_free(const struct lu_env *env,
7964                                          struct lu_device *d)
7965 {
7966         struct osd_device *o = osd_dev(d);
7967
7968         ENTRY;
7969
7970         /* XXX: make osd top device in order to release reference */
7971         d->ld_site->ls_top_dev = d;
7972         lu_site_purge(env, d->ld_site, -1);
7973         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7974                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7975                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7976         }
7977         lu_site_fini(&o->od_site);
7978         dt_device_fini(&o->od_dt_dev);
7979         OBD_FREE_PTR(o);
7980         RETURN(NULL);
7981 }
7982
7983 static int osd_process_config(const struct lu_env *env,
7984                               struct lu_device *d, struct lustre_cfg *cfg)
7985 {
7986         struct osd_device *o = osd_dev(d);
7987         ssize_t count;
7988         int rc;
7989
7990         ENTRY;
7991
7992         switch (cfg->lcfg_command) {
7993         case LCFG_SETUP:
7994                 rc = osd_mount(env, o, cfg);
7995                 break;
7996         case LCFG_CLEANUP:
7997                 /*
7998                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7999                  * that may happend if hit failure during mount process.
8000                  */
8001                 osd_index_backup(env, o, false);
8002                 lu_dev_del_linkage(d->ld_site, d);
8003                 rc = osd_shutdown(env, o);
8004                 break;
8005         case LCFG_PARAM:
8006                 LASSERT(&o->od_dt_dev);
8007                 count  = class_modify_config(cfg, PARAM_OSD,
8008                                              &o->od_dt_dev.dd_kobj);
8009                 if (count < 0)
8010                         count = class_modify_config(cfg, PARAM_OST,
8011                                                     &o->od_dt_dev.dd_kobj);
8012                 rc = count > 0 ? 0 : count;
8013                 break;
8014         case LCFG_PRE_CLEANUP:
8015                 osd_scrub_stop(o);
8016                 osd_index_backup(env, o,
8017                                  o->od_index_backup_policy != LIBP_NONE);
8018                 rc = 0;
8019                 break;
8020         default:
8021                 rc = -ENOSYS;
8022         }
8023
8024         RETURN(rc);
8025 }
8026
8027 static int osd_recovery_complete(const struct lu_env *env,
8028                                  struct lu_device *d)
8029 {
8030         struct osd_device *osd = osd_dev(d);
8031         int rc = 0;
8032
8033         ENTRY;
8034
8035         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
8036                 RETURN(0);
8037
8038         /*
8039          * start qsd instance on recovery completion, this notifies the quota
8040          * slave code that we are about to process new requests now
8041          */
8042         rc = qsd_start(env, osd->od_quota_slave_dt);
8043         if (rc == 0 && osd->od_quota_slave_md != NULL)
8044                 rc = qsd_start(env, osd->od_quota_slave_md);
8045
8046         RETURN(rc);
8047 }
8048
8049 /*
8050  * we use exports to track all osd users
8051  */
8052 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8053                            struct obd_device *obd, struct obd_uuid *cluuid,
8054                            struct obd_connect_data *data, void *localdata)
8055 {
8056         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8057         struct lustre_handle conn;
8058         int rc;
8059
8060         ENTRY;
8061
8062         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8063
8064         rc = class_connect(&conn, obd, cluuid);
8065         if (rc)
8066                 RETURN(rc);
8067
8068         *exp = class_conn2export(&conn);
8069
8070         spin_lock(&osd->od_osfs_lock);
8071         osd->od_connects++;
8072         spin_unlock(&osd->od_osfs_lock);
8073
8074         RETURN(0);
8075 }
8076
8077 /*
8078  * once last export (we don't count self-export) disappeared
8079  * osd can be released
8080  */
8081 static int osd_obd_disconnect(struct obd_export *exp)
8082 {
8083         struct obd_device *obd = exp->exp_obd;
8084         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8085         int rc, release = 0;
8086
8087         ENTRY;
8088
8089         /* Only disconnect the underlying layers on the final disconnect. */
8090         spin_lock(&osd->od_osfs_lock);
8091         osd->od_connects--;
8092         if (osd->od_connects == 0)
8093                 release = 1;
8094         spin_unlock(&osd->od_osfs_lock);
8095
8096         rc = class_disconnect(exp); /* bz 9811 */
8097
8098         if (rc == 0 && release)
8099                 class_manual_cleanup(obd);
8100         RETURN(rc);
8101 }
8102
8103 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8104                        struct lu_device *dev)
8105 {
8106         struct osd_device *osd = osd_dev(dev);
8107         struct lr_server_data *lsd =
8108                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8109         int result = 0;
8110
8111         ENTRY;
8112
8113         if (osd->od_quota_slave_md != NULL) {
8114                 /* set up quota slave objects for inode */
8115                 result = qsd_prepare(env, osd->od_quota_slave_md);
8116                 if (result != 0)
8117                         RETURN(result);
8118         }
8119
8120         if (osd->od_quota_slave_dt != NULL) {
8121                 /* set up quota slave objects for block */
8122                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8123                 if (result != 0)
8124                         RETURN(result);
8125         }
8126
8127
8128         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8129 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8130                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8131                         osd->od_index_in_idif = 1;
8132                 } else {
8133                         osd->od_index_in_idif = 0;
8134                         result = osd_register_proc_index_in_idif(osd);
8135                         if (result != 0)
8136                                 RETURN(result);
8137                 }
8138 #else
8139                 osd->od_index_in_idif = 1;
8140 #endif
8141         }
8142
8143         result = osd_fid_init(env, osd);
8144
8145         RETURN(result);
8146 }
8147
8148 /**
8149  * Implementation of lu_device_operations::ldo_fid_alloc() for OSD
8150  *
8151  * Allocate FID.
8152  *
8153  * see include/lu_object.h for the details.
8154  */
8155 static int osd_fid_alloc(const struct lu_env *env, struct lu_device *d,
8156                          struct lu_fid *fid, struct lu_object *parent,
8157                          const struct lu_name *name)
8158 {
8159         struct osd_device *osd = osd_dev(d);
8160
8161         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8162 }
8163
8164 static const struct lu_object_operations osd_lu_obj_ops = {
8165         .loo_object_init      = osd_object_init,
8166         .loo_object_delete    = osd_object_delete,
8167         .loo_object_release   = osd_object_release,
8168         .loo_object_free      = osd_object_free,
8169         .loo_object_print     = osd_object_print,
8170         .loo_object_invariant = osd_object_invariant
8171 };
8172
8173 const struct lu_device_operations osd_lu_ops = {
8174         .ldo_object_alloc      = osd_object_alloc,
8175         .ldo_process_config    = osd_process_config,
8176         .ldo_recovery_complete = osd_recovery_complete,
8177         .ldo_prepare           = osd_prepare,
8178         .ldo_fid_alloc         = osd_fid_alloc,
8179 };
8180
8181 static const struct lu_device_type_operations osd_device_type_ops = {
8182         .ldto_init = osd_type_init,
8183         .ldto_fini = osd_type_fini,
8184
8185         .ldto_start = osd_type_start,
8186         .ldto_stop  = osd_type_stop,
8187
8188         .ldto_device_alloc = osd_device_alloc,
8189         .ldto_device_free  = osd_device_free,
8190
8191         .ldto_device_init = osd_device_init,
8192         .ldto_device_fini = osd_device_fini
8193 };
8194
8195 static struct lu_device_type osd_device_type = {
8196         .ldt_tags     = LU_DEVICE_DT,
8197         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8198         .ldt_ops      = &osd_device_type_ops,
8199         .ldt_ctx_tags = LCT_LOCAL,
8200 };
8201
8202 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8203 {
8204         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8205         struct super_block *sb = osd_sb(osd);
8206
8207         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8208 }
8209
8210 /*
8211  * lprocfs legacy support.
8212  */
8213 static const struct obd_ops osd_obd_device_ops = {
8214         .o_owner = THIS_MODULE,
8215         .o_connect      = osd_obd_connect,
8216         .o_disconnect   = osd_obd_disconnect,
8217         .o_health_check = osd_health_check,
8218 };
8219
8220 static ssize_t track_declares_assert_show(struct kobject *kobj,
8221                                    struct attribute *attr,
8222                                    char *buf)
8223 {
8224         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8225 }
8226
8227 static ssize_t track_declares_assert_store(struct kobject *kobj,
8228                                            struct attribute *attr,
8229                                            const char *buffer, size_t count)
8230 {
8231         bool track_declares_assert;
8232         int rc;
8233
8234         rc = kstrtobool(buffer, &track_declares_assert);
8235         if (rc)
8236                 return rc;
8237
8238         ldiskfs_track_declares_assert = track_declares_assert;
8239
8240         return count;
8241 }
8242 LUSTRE_RW_ATTR(track_declares_assert);
8243
8244 static int __init osd_init(void)
8245 {
8246         struct kobject *kobj;
8247         int rc;
8248
8249         BUILD_BUG_ON(BH_DXLock >=
8250                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8251 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8252         /* please, try to keep osd_thread_info smaller than a page */
8253         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8254 #endif
8255
8256         osd_oi_mod_init();
8257
8258         rc = lu_kmem_init(ldiskfs_caches);
8259         if (rc)
8260                 return rc;
8261
8262 #ifdef CONFIG_KALLSYMS
8263         priv_security_file_alloc =
8264                 (void *)kallsyms_lookup_name("security_file_alloc");
8265 #endif
8266
8267         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8268                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8269         if (rc) {
8270                 lu_kmem_fini(ldiskfs_caches);
8271                 return rc;
8272         }
8273
8274         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8275         if (kobj) {
8276                 rc = sysfs_create_file(kobj,
8277                                        &lustre_attr_track_declares_assert.attr);
8278                 kobject_put(kobj);
8279                 if (rc) {
8280                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8281                         rc = 0;
8282                 }
8283         }
8284         return rc;
8285 }
8286
8287 static void __exit osd_exit(void)
8288 {
8289         struct kobject *kobj;
8290
8291         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8292         if (kobj) {
8293                 sysfs_remove_file(kobj,
8294                                   &lustre_attr_track_declares_assert.attr);
8295                 kobject_put(kobj);
8296         }
8297         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8298         lu_kmem_fini(ldiskfs_caches);
8299 }
8300
8301 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8302 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8303 MODULE_VERSION(LUSTRE_VERSION_STRING);
8304 MODULE_LICENSE("GPL");
8305
8306 module_init(osd_init);
8307 module_exit(osd_exit);