lustre/osd-ldiskfs/osd_handler.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/osd/osd_handler.c
  33  *
  34  * Top-level entry points into osd module
  35  *
  36  * Author: Nikita Danilov <nikita@clusterfs.com>
  37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  38  */
  39
  40 #define DEBUG_SUBSYSTEM S_OSD
  41
  42 #include <linux/kallsyms.h>
  43 #include <linux/module.h>
  44 #include <linux/user_namespace.h>
  45 #include <linux/uidgid.h>
  46
  47 /* prerequisite for linux/xattr.h */
  48 #include <linux/types.h>
  49 /* prerequisite for linux/xattr.h */
  50 #include <linux/fs.h>
  51 /* XATTR_{REPLACE,CREATE} */
  52 #include <linux/xattr.h>
  53
  54 #include <ldiskfs/ldiskfs.h>
  55 #include <ldiskfs/xattr.h>
  56 #include <ldiskfs/ldiskfs_extents.h>
  57 #undef ENTRY
  58 /*
  59  * struct OBD_{ALLOC,FREE}*()
  60  * OBD_FAIL_CHECK
  61  */
  62 #include <obd_support.h>
  63 /* struct ptlrpc_thread */
  64 #include <lustre_net.h>
  65 #include <lustre_fid.h>
  66 /* process_config */
  67 #include <uapi/linux/lustre/lustre_param.h>
  68
  69 #include "osd_internal.h"
  70 #include "osd_dynlocks.h"
  71
  72 /* llo_* api support */
  73 #include <md_object.h>
  74 #include <lustre_quota.h>
  75
  76 #include <lustre_linkea.h>
  77
  78 /* Maximum EA size is limited by LNET_MTU for remote objects */
  79 #define OSD_MAX_EA_SIZE 1048364
  80
  81 int ldiskfs_pdo = 1;
  82 module_param(ldiskfs_pdo, int, 0644);
  83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
  84
  85 int ldiskfs_track_declares_assert;
  86 module_param(ldiskfs_track_declares_assert, int, 0644);
  87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
  88
  89 /* Slab to allocate dynlocks */
  90 struct kmem_cache *dynlock_cachep;
  91
  92 /* Slab to allocate osd_it_ea */
  93 struct kmem_cache *osd_itea_cachep;
  94
  95 static struct lu_kmem_descr ldiskfs_caches[] = {
  96         {
  97                 .ckd_cache = &dynlock_cachep,
  98                 .ckd_name  = "dynlock_cache",
  99                 .ckd_size  = sizeof(struct dynlock_handle)
 100         },
 101         {
 102                 .ckd_cache = &osd_itea_cachep,
 103                 .ckd_name  = "osd_itea_cache",
 104                 .ckd_size  = sizeof(struct osd_it_ea)
 105         },
 106         {
 107                 .ckd_cache = NULL
 108         }
 109 };
 110
 111 static const char dot[] = ".";
 112 static const char dotdot[] = "..";
 113
 114 static const struct lu_object_operations      osd_lu_obj_ops;
 115 static const struct dt_object_operations      osd_obj_ops;
 116 static const struct dt_object_operations      osd_obj_otable_it_ops;
 117 static const struct dt_index_operations       osd_index_iam_ops;
 118 static const struct dt_index_operations       osd_index_ea_ops;
 119
 120 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
 121                           const struct lu_fid *fid);
 122 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
 123                                                 struct osd_device *osd);
 124
 125 int osd_trans_declare_op2rb[] = {
 126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
 127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
 128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
 129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
 130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
 131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
 132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
 133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
 134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
 135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
 136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 137 };
 138
 139 static int osd_has_index(const struct osd_object *obj)
 140 {
 141         return obj->oo_dt.do_index_ops != NULL;
 142 }
 143
 144 static int osd_object_invariant(const struct lu_object *l)
 145 {
 146         return osd_invariant(osd_obj(l));
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
 153 {
 154         struct osd_thread_info *oti = osd_oti_get(env);
 155
 156         return oti->oti_w_locks > 0 && o->oo_owner == env;
 157 }
 158
 159 /*
 160  * Concurrency: doesn't access mutable data
 161  */
 162 static int osd_root_get(const struct lu_env *env,
 163                         struct dt_device *dev, struct lu_fid *f)
 164 {
 165         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 166         return 0;
 167 }
 168
 169 /*
 170  * the following set of functions are used to maintain per-thread
 171  * cache of FID->ino mapping. this mechanism is needed to resolve
 172  * FID to inode at dt_insert() which in turn stores ino in the
 173  * directory entries to keep ldiskfs compatible with ext[34].
 174  * due to locking-originated restrictions we can't lookup ino
 175  * using LU cache (deadlock is possible). lookup using OI is quite
 176  * expensive. so instead we maintain this cache and methods like
 177  * dt_create() fill it. so in the majority of cases dt_insert() is
 178  * able to find needed mapping in lockless manner.
 179  */
 180 static struct osd_idmap_cache *
 181 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
 182              const struct lu_fid *fid)
 183 {
 184         struct osd_thread_info *oti = osd_oti_get(env);
 185         struct osd_idmap_cache *idc = oti->oti_ins_cache;
 186         int i;
 187
 188         for (i = 0; i < oti->oti_ins_cache_used; i++) {
 189                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
 190                         continue;
 191                 if (idc[i].oic_dev != osd)
 192                         continue;
 193
 194                 return idc + i;
 195         }
 196
 197         return NULL;
 198 }
 199
 200 static struct osd_idmap_cache *
 201 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
 202             const struct lu_fid *fid)
 203 {
 204         struct osd_thread_info *oti   = osd_oti_get(env);
 205         struct osd_idmap_cache *idc;
 206         int i;
 207
 208         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
 209                 i = oti->oti_ins_cache_size * 2;
 210                 if (i == 0)
 211                         i = OSD_INS_CACHE_SIZE;
 212                 OBD_ALLOC(idc, sizeof(*idc) * i);
 213                 if (idc == NULL)
 214                         return ERR_PTR(-ENOMEM);
 215                 if (oti->oti_ins_cache != NULL) {
 216                         memcpy(idc, oti->oti_ins_cache,
 217                                oti->oti_ins_cache_used * sizeof(*idc));
 218                         OBD_FREE(oti->oti_ins_cache,
 219                                  oti->oti_ins_cache_used * sizeof(*idc));
 220                 }
 221                 oti->oti_ins_cache = idc;
 222                 oti->oti_ins_cache_size = i;
 223         }
 224
 225         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
 226         idc->oic_fid = *fid;
 227         idc->oic_dev = osd;
 228         idc->oic_lid.oii_ino = 0;
 229         idc->oic_lid.oii_gen = 0;
 230         idc->oic_remote = 0;
 231
 232         return idc;
 233 }
 234
 235 /*
 236  * lookup mapping for the given fid in the cache, initialize a
 237  * new one if not found. the initialization checks whether the
 238  * object is local or remote. for local objects, OI is used to
 239  * learn ino/generation. the function is used when the caller
 240  * has no information about the object, e.g. at dt_insert().
 241  */
 242 static struct osd_idmap_cache *
 243 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
 244                      const struct lu_fid *fid)
 245 {
 246         struct osd_idmap_cache *idc;
 247         int rc;
 248
 249         idc = osd_idc_find(env, osd, fid);
 250         LASSERT(!IS_ERR(idc));
 251         if (idc != NULL)
 252                 return idc;
 253
 254         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 255                osd->od_svname, PFID(fid));
 256
 257         /* new mapping is needed */
 258         idc = osd_idc_add(env, osd, fid);
 259         if (IS_ERR(idc)) {
 260                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 261                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 262                 return idc;
 263         }
 264
 265         /* initialize it */
 266         rc = osd_remote_fid(env, osd, fid);
 267         if (unlikely(rc < 0))
 268                 return ERR_PTR(rc);
 269
 270         if (rc == 0) {
 271                 /* the object is local, lookup in OI */
 272                 /* XXX: probably cheaper to lookup in LU first? */
 273                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
 274                                    &idc->oic_lid, 0);
 275                 if (unlikely(rc < 0)) {
 276                         CERROR("can't lookup: rc = %d\n", rc);
 277                         return ERR_PTR(rc);
 278                 }
 279         } else {
 280                 /* the object is remote */
 281                 idc->oic_remote = 1;
 282         }
 283
 284         return idc;
 285 }
 286
 287 /*
 288  * lookup mapping for given FID and fill it from the given object.
 289  * the object is lolcal by definition.
 290  */
 291 static int osd_idc_find_and_init(const struct lu_env *env,
 292                                  struct osd_device *osd,
 293                                  struct osd_object *obj)
 294 {
 295         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
 296         struct osd_idmap_cache *idc;
 297
 298         idc = osd_idc_find(env, osd, fid);
 299         LASSERT(!IS_ERR(idc));
 300         if (idc != NULL) {
 301                 if (obj->oo_inode == NULL)
 302                         return 0;
 303                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
 304                         LASSERT(idc->oic_lid.oii_ino == 0);
 305                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 306                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 307                 }
 308                 return 0;
 309         }
 310
 311         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
 312                osd->od_svname, PFID(fid));
 313
 314         /* new mapping is needed */
 315         idc = osd_idc_add(env, osd, fid);
 316         if (IS_ERR(idc)) {
 317                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
 318                        osd->od_svname, PFID(fid), PTR_ERR(idc));
 319                 return PTR_ERR(idc);
 320         }
 321
 322         if (obj->oo_inode != NULL) {
 323                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
 324                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
 325         }
 326         return 0;
 327 }
 328
 329 /*
 330  * OSD object methods.
 331  */
 332
 333 /*
 334  * Concurrency: no concurrent access is possible that early in object
 335  * life-cycle.
 336  */
 337 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 338                                           const struct lu_object_header *hdr,
 339                                           struct lu_device *d)
 340 {
 341         struct osd_object *mo;
 342
 343         OBD_ALLOC_PTR(mo);
 344         if (mo != NULL) {
 345                 struct lu_object *l;
 346                 struct lu_object_header *h;
 347                 struct osd_device *o = osd_dev(d);
 348
 349                 l = &mo->oo_dt.do_lu;
 350                 if (unlikely(o->od_in_init)) {
 351                         OBD_ALLOC_PTR(h);
 352                         if (!h) {
 353                                 OBD_FREE_PTR(mo);
 354                                 return NULL;
 355                         }
 356
 357                         lu_object_header_init(h);
 358                         lu_object_init(l, h, d);
 359                         lu_object_add_top(h, l);
 360                         mo->oo_header = h;
 361                 } else {
 362                         dt_object_init(&mo->oo_dt, NULL, d);
 363                         mo->oo_header = NULL;
 364                 }
 365
 366                 mo->oo_dt.do_ops = &osd_obj_ops;
 367                 l->lo_ops = &osd_lu_obj_ops;
 368                 init_rwsem(&mo->oo_sem);
 369                 init_rwsem(&mo->oo_ext_idx_sem);
 370                 spin_lock_init(&mo->oo_guard);
 371                 INIT_LIST_HEAD(&mo->oo_xattr_list);
 372                 return l;
 373         }
 374         return NULL;
 375 }
 376
 377 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 378                 struct dentry *dentry, struct lustre_ost_attrs *loa)
 379 {
 380         int rc;
 381
 382         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 383                              (void *)loa, sizeof(*loa));
 384         if (rc > 0) {
 385                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
 386
 387                 if (rc < sizeof(*lma))
 388                         return -EINVAL;
 389
 390                 rc = 0;
 391                 lustre_loa_swab(loa, true);
 392                 /* Check LMA compatibility */
 393                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 394                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
 395                               "for fid = "DFID", ino = %lu\n",
 396                               osd_ino2name(inode),
 397                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 398                               PFID(&lma->lma_self_fid), inode->i_ino);
 399                         rc = -EOPNOTSUPP;
 400                 }
 401         } else if (rc == 0) {
 402                 rc = -ENODATA;
 403         }
 404
 405         return rc;
 406 }
 407
 408 /*
 409  * retrieve object from backend ext fs.
 410  **/
 411 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 412                        struct osd_inode_id *id)
 413 {
 414         int rc;
 415         struct inode *inode = NULL;
 416
 417         /*
 418          * if we look for an inode withing a running
 419          * transaction, then we risk to deadlock
 420          * osd_dirent_check_repair() breaks this
 421          */
 422          /* LASSERT(current->journal_info == NULL); */
 423
 424         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 425         if (IS_ERR(inode)) {
 426                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 427                        id->oii_ino, PTR_ERR(inode));
 428         } else if (id->oii_gen != OSD_OII_NOGEN &&
 429                    inode->i_generation != id->oii_gen) {
 430                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 431                        "i_generation = %u\n",
 432                        id->oii_ino, id->oii_gen, inode->i_generation);
 433                 iput(inode);
 434                 inode = ERR_PTR(-ESTALE);
 435         } else if (inode->i_nlink == 0) {
 436                 /*
 437                  * due to parallel readdir and unlink,
 438                  * we can have dead inode here.
 439                  */
 440                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 441                 iput(inode);
 442                 inode = ERR_PTR(-ESTALE);
 443         } else if (is_bad_inode(inode)) {
 444                 CWARN("%s: bad inode: ino = %u\n",
 445                       osd_dev2name(dev), id->oii_ino);
 446                 iput(inode);
 447                 inode = ERR_PTR(-ENOENT);
 448         } else if ((rc = osd_attach_jinode(inode))) {
 449                 iput(inode);
 450                 inode = ERR_PTR(rc);
 451         } else {
 452                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 453                 if (id->oii_gen == OSD_OII_NOGEN)
 454                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 455
 456                 /*
 457                  * Do not update file c/mtime in ldiskfs.
 458                  * NB: we don't have any lock to protect this because we don't
 459                  * have reference on osd_object now, but contention with
 460                  * another lookup + attr_set can't happen in the tiny window
 461                  * between if (...) and set S_NOCMTIME.
 462                  */
 463                 if (!(inode->i_flags & S_NOCMTIME))
 464                         inode->i_flags |= S_NOCMTIME;
 465         }
 466         return inode;
 467 }
 468
 469 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
 470                           handle_t *handle, struct dentry *child,
 471                           struct inode *inode, struct htree_lock *hlock)
 472 {
 473         int rc, rc2;
 474
 475         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
 476         if (rc == -ENOBUFS || rc == -ENOSPC) {
 477                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 478                 struct inode *parent = child->d_parent->d_inode;
 479                 struct lu_fid *fid = NULL;
 480
 481                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
 482                 if (!rc2) {
 483                         fid = &loa->loa_lma.lma_self_fid;
 484                 } else if (rc2 == -ENODATA) {
 485                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
 486                                 fid = &info->oti_fid3;
 487                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 488                         } else if (!osd->od_is_ost && osd->od_index == 0) {
 489                                 fid = &info->oti_fid3;
 490                                 lu_igif_build(fid, parent->i_ino,
 491                                               parent->i_generation);
 492                         }
 493                 }
 494
 495                 if (fid != NULL)
 496                         /* below message is checked in sanity.sh test_129 */
 497                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
 498                               osd_name(osd), parent->i_ino, PFID(fid),
 499                               rc == -ENOSPC ? "has reached" : "is approaching");
 500                 else
 501                         /* below message is checked in sanity.sh test_129 */
 502                         CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
 503                               osd_name(osd), parent->i_ino,
 504                               rc == -ENOSPC ? "has reached" : "is approaching");
 505
 506                 /* ignore such error now */
 507                 if (rc == -ENOBUFS)
 508                         rc = 0;
 509         }
 510
 511         return rc;
 512 }
 513
 514
 515 struct inode *
 516 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 517              struct osd_inode_id *id, struct lu_fid *fid)
 518 {
 519         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 520         struct inode *inode;
 521         int rc;
 522
 523         inode = osd_iget(info, dev, id);
 524         if (IS_ERR(inode))
 525                 return inode;
 526
 527         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
 528         if (!rc) {
 529                 *fid = loa->loa_lma.lma_self_fid;
 530         } else if (rc == -ENODATA) {
 531                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 532                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 533                 else
 534                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 535         } else {
 536                 iput(inode);
 537                 inode = ERR_PTR(rc);
 538         }
 539         return inode;
 540 }
 541
 542 static struct inode *osd_iget_check(struct osd_thread_info *info,
 543                                     struct osd_device *dev,
 544                                     const struct lu_fid *fid,
 545                                     struct osd_inode_id *id,
 546                                     bool trusted)
 547 {
 548         struct inode *inode;
 549         int rc = 0;
 550
 551         ENTRY;
 552
 553         /*
 554          * The cached OI mapping is trustable. If we cannot locate the inode
 555          * via the cached OI mapping, then return the failure to the caller
 556          * directly without further OI checking.
 557          */
 558
 559 again:
 560         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
 561         if (IS_ERR(inode)) {
 562                 rc = PTR_ERR(inode);
 563                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
 564                         goto check_oi;
 565
 566                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
 567                        PFID(fid), id->oii_ino, rc);
 568                 GOTO(put, rc);
 569         }
 570
 571         if (is_bad_inode(inode)) {
 572                 rc = -ENOENT;
 573                 if (!trusted)
 574                         goto check_oi;
 575
 576                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
 577                        PFID(fid), id->oii_ino);
 578                 GOTO(put, rc);
 579         }
 580
 581         if (id->oii_gen != OSD_OII_NOGEN &&
 582             inode->i_generation != id->oii_gen) {
 583                 rc = -ESTALE;
 584                 if (!trusted)
 585                         goto check_oi;
 586
 587                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
 588                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
 589                        id->oii_ino, id->oii_gen, inode->i_generation);
 590                 GOTO(put, rc);
 591         }
 592
 593         if (inode->i_nlink == 0) {
 594                 rc = -ENOENT;
 595                 if (!trusted)
 596                         goto check_oi;
 597
 598                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
 599                        PFID(fid), id->oii_ino);
 600                 GOTO(put, rc);
 601         }
 602
 603         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
 604
 605 check_oi:
 606         if (rc != 0) {
 607                 __u32 saved_ino = id->oii_ino;
 608                 __u32 saved_gen = id->oii_gen;
 609
 610                 LASSERT(!trusted);
 611                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
 612
 613                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 614                 /*
 615                  * XXX: There are four possible cases:
 616                  *      1. rc = 0.
 617                  *         Backup/restore caused the OI invalid.
 618                  *      2. rc = 0.
 619                  *         Someone unlinked the object but NOT removed
 620                  *         the OI mapping, such as mount target device
 621                  *         as ldiskfs, and modify something directly.
 622                  *      3. rc = -ENOENT.
 623                  *         Someone just removed the object between the
 624                  *         former oi_lookup and the iget. It is normal.
 625                  *      4. Other failure cases.
 626                  *
 627                  *      Generally, when the device is mounted, it will
 628                  *      auto check whether the system is restored from
 629                  *      file-level backup or not. We trust such detect
 630                  *      to distinguish the 1st case from the 2nd case:
 631                  *      if the OI files are consistent but may contain
 632                  *      stale OI mappings because of case 2, if iget()
 633                  *      returns -ENOENT or -ESTALE, then it should be
 634                  *      the case 2.
 635                  */
 636                 if (rc != 0)
 637                         /*
 638                          * If the OI mapping was in OI file before the
 639                          * osd_iget_check(), but now, it is disappear,
 640                          * then it must be removed by race. That is a
 641                          * normal race case.
 642                          */
 643                         GOTO(put, rc);
 644
 645                 /*
 646                  * It is the OI scrub updated the OI mapping by race.
 647                  * The new OI mapping must be valid.
 648                  */
 649                 if (saved_ino != id->oii_ino ||
 650                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
 651                         if (!IS_ERR(inode))
 652                                 iput(inode);
 653
 654                         trusted = true;
 655                         goto again;
 656                 }
 657
 658                 if (IS_ERR(inode)) {
 659                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
 660                             SF_INCONSISTENT)
 661                                 /*
 662                                  * It still can be the case 2, but we cannot
 663                                  * distinguish it from the case 1. So return
 664                                  * -EREMCHG to block current operation until
 665                                  *  OI scrub rebuilt the OI mappings.
 666                                  */
 667                                 rc = -EREMCHG;
 668                         else
 669                                 rc = -ENOENT;
 670
 671                         GOTO(put, rc);
 672                 }
 673
 674                 if (inode->i_generation == id->oii_gen)
 675                         rc = -ENOENT;
 676                 else
 677                         rc = -EREMCHG;
 678         } else {
 679                 if (id->oii_gen == OSD_OII_NOGEN)
 680                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 681
 682                 /*
 683                  * Do not update file c/mtime in ldiskfs.
 684                  * NB: we don't have any lock to protect this because we don't
 685                  * have reference on osd_object now, but contention with
 686                  * another lookup + attr_set can't happen in the tiny window
 687                  * between if (...) and set S_NOCMTIME.
 688                  */
 689                 if (!(inode->i_flags & S_NOCMTIME))
 690                         inode->i_flags |= S_NOCMTIME;
 691         }
 692
 693         GOTO(put, rc);
 694
 695 put:
 696         if (rc != 0) {
 697                 if (!IS_ERR(inode))
 698                         iput(inode);
 699
 700                 inode = ERR_PTR(rc);
 701         }
 702
 703         return inode;
 704 }
 705
 706 /**
 707  * \retval +v: new filter_fid does not contain self-fid
 708  * \retval 0:  filter_fid_18_23, contains self-fid
 709  * \retval -v: other failure cases
 710  */
 711 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 712                  struct dentry *dentry, struct lu_fid *fid)
 713 {
 714         struct filter_fid *ff = &info->oti_ff;
 715         struct ost_id *ostid = &info->oti_ostid;
 716         int rc;
 717
 718         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 719         if (rc == sizeof(struct filter_fid_18_23)) {
 720                 struct filter_fid_18_23 *ff_old = (void *)ff;
 721
 722                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
 723                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
 724                 /*
 725                  * XXX: use 0 as the index for compatibility, the caller will
 726                  * handle index related issues when necessary.
 727                  */
 728                 if (!rc)
 729                         ostid_to_fid(fid, ostid, 0);
 730         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
 731                 rc = 1;
 732         } else if (rc >= 0) {
 733                 rc = -EINVAL;
 734         }
 735
 736         return rc;
 737 }
 738
 739 static int osd_lma_self_repair(struct osd_thread_info *info,
 740                                struct osd_device *osd, struct inode *inode,
 741                                const struct lu_fid *fid, __u32 compat)
 742 {
 743         handle_t *jh;
 744         int rc;
 745
 746         LASSERT(current->journal_info == NULL);
 747
 748         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 749                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 750         if (IS_ERR(jh)) {
 751                 rc = PTR_ERR(jh);
 752                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 753                       osd_name(osd), rc);
 754                 return rc;
 755         }
 756
 757         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 758         if (rc != 0)
 759                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 760                       osd_name(osd), rc);
 761         ldiskfs_journal_stop(jh);
 762         return rc;
 763 }
 764
 765 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 766 {
 767         struct osd_thread_info *info = osd_oti_get(env);
 768         struct osd_device *osd = osd_obj2dev(obj);
 769         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
 770         struct lustre_mdt_attrs *lma = &loa->loa_lma;
 771         struct inode *inode = obj->oo_inode;
 772         struct dentry *dentry = &info->oti_obj_dentry;
 773         struct lu_fid *fid = NULL;
 774         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
 775         int rc;
 776
 777         ENTRY;
 778
 779         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 780                              (void *)loa, sizeof(*loa));
 781         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 782                 fid = &lma->lma_self_fid;
 783                 rc = osd_get_idif(info, inode, dentry, fid);
 784                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
 785                         /*
 786                          * For the given OST-object, if it has neither LMA nor
 787                          * FID in XATTR_NAME_FID, then the given FID (which is
 788                          * contained in the @obj, from client RPC for locating
 789                          * the OST-object) is trusted. We use it to generate
 790                          * the LMA.
 791                          */
 792                         osd_lma_self_repair(info, osd, inode, rfid,
 793                                             LMAC_FID_ON_OST);
 794                         RETURN(0);
 795                 }
 796         }
 797
 798         if (rc < 0)
 799                 RETURN(rc);
 800
 801         if (rc > 0) {
 802                 rc = 0;
 803                 lustre_lma_swab(lma);
 804                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 805                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
 806                               S_ISREG(inode->i_mode)))) {
 807                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 808                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 809                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 810                               PFID(rfid), inode->i_ino);
 811                         rc = -EOPNOTSUPP;
 812                 } else {
 813                         fid = &lma->lma_self_fid;
 814                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
 815                             osd->od_is_ost)
 816                                 obj->oo_pfid_in_lma = 1;
 817                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
 818                             !osd->od_is_ost)
 819                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
 820                 }
 821         }
 822
 823         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 824                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 825                         struct ost_id   *oi   = &info->oti_ostid;
 826                         struct lu_fid   *fid1 = &info->oti_fid3;
 827                         __u32            idx  = fid_idif_ost_idx(rfid);
 828
 829                         /*
 830                          * For old IDIF, the OST index is not part of the IDIF,
 831                          * Means that different OSTs may have the same IDIFs.
 832                          * Under such case, we need to make some compatible
 833                          * check to make sure to trigger OI scrub properly.
 834                          */
 835                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 836                                 /* Given @rfid is new, LMA is old. */
 837                                 fid_to_ostid(fid, oi);
 838                                 ostid_to_fid(fid1, oi, idx);
 839                                 if (lu_fid_eq(fid1, rfid)) {
 840                                         if (osd->od_index_in_idif)
 841                                                 osd_lma_self_repair(info, osd,
 842                                                         inode, rfid,
 843                                                         LMAC_FID_ON_OST);
 844                                         RETURN(0);
 845                                 }
 846                         }
 847                 }
 848
 849                 rc = -EREMCHG;
 850         }
 851
 852         RETURN(rc);
 853 }
 854
 855 struct osd_check_lmv_buf {
 856         /* please keep it as first member */
 857         struct dir_context ctx;
 858         struct osd_thread_info *oclb_info;
 859         struct osd_device *oclb_dev;
 860         struct osd_idmap_cache *oclb_oic;
 861         int oclb_items;
 862         bool oclb_found;
 863 };
 864
 865 /**
 866  * It is called internally by ->iterate*() to filter out the
 867  * local slave object's FID of the striped directory.
 868  *
 869  * \retval      1 found the local slave's FID
 870  * \retval      0 continue to check next item
 871  * \retval      -ve for failure
 872  */
 873 #ifdef HAVE_FILLDIR_USE_CTX
 874 static int osd_stripe_dir_filldir(struct dir_context *buf,
 875 #else
 876 static int osd_stripe_dir_filldir(void *buf,
 877 #endif
 878                                   const char *name, int namelen,
 879                                   loff_t offset, __u64 ino, unsigned int d_type)
 880 {
 881         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
 882         struct osd_thread_info *oti = oclb->oclb_info;
 883         struct lu_fid *fid = &oti->oti_fid3;
 884         struct osd_inode_id *id = &oti->oti_id3;
 885         struct osd_device *dev = oclb->oclb_dev;
 886         struct osd_idmap_cache *oic = oclb->oclb_oic;
 887         struct inode *inode;
 888
 889         oclb->oclb_items++;
 890
 891         if (name[0] == '.')
 892                 return 0;
 893
 894         fid_zero(fid);
 895         sscanf(name + 1, SFID, RFID(fid));
 896         if (!fid_is_sane(fid))
 897                 return 0;
 898
 899         if (osd_remote_fid(oti->oti_env, dev, fid))
 900                 return 0;
 901
 902         osd_id_gen(id, ino, OSD_OII_NOGEN);
 903         inode = osd_iget(oti, dev, id);
 904         if (IS_ERR(inode))
 905                 return PTR_ERR(inode);
 906
 907         iput(inode);
 908         osd_add_oi_cache(oti, dev, id, fid);
 909         oic->oic_fid = *fid;
 910         oic->oic_lid = *id;
 911         oic->oic_dev = dev;
 912         osd_oii_insert(dev, oic, true);
 913         oclb->oclb_found = true;
 914
 915         return 1;
 916 }
 917
 918 /*
 919  * When lookup item under striped directory, we need to locate the master
 920  * MDT-object of the striped directory firstly, then the client will send
 921  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
 922  * and the item's name. If the system is restored from MDT file level backup,
 923  * then before the OI scrub completely built the OI files, the OI mappings of
 924  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
 925  * not a problem for the master MDT-object. Because when locate the master
 926  * MDT-object, we will do name based lookup (for the striped directory itself)
 927  * firstly, during such process we can setup the correct OI mapping for the
 928  * master MDT-object. But it will be trouble for the slave MDT-object. Because
 929  * the client will not trigger name based lookup on the MDT to locate the slave
 930  * MDT-object before locating item under the striped directory, then when
 931  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
 932  * is invalid and does not know what the right OI mapping is, then the MDT has
 933  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
 934  * the OI file, related OI mapping is unknown yet, please try again later. And
 935  * then client will re-try the RPC again and again until related OI mapping has
 936  * been updated. That is quite inefficient.
 937  *
 938  * To resolve above trouble, we will handle it as the following two cases:
 939  *
 940  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
 941  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
 942  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
 943  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
 944  *    directly. Please check osd_fid_lookup().
 945  *
 946  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
 947  *    Under such case, during lookup the master MDT-object, we will lookup the
 948  *    slave MDT-object via readdir against the master MDT-object, because the
 949  *    slave MDT-objects information are stored as sub-directories with the name
 950  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
 951  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
 952  *    the correct OI mapping for the slave MDT-object.
 953  */
 954 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
 955                          struct inode *inode, struct osd_idmap_cache *oic)
 956 {
 957         struct lu_buf *buf = &oti->oti_big_buf;
 958         struct dentry *dentry = &oti->oti_obj_dentry;
 959         struct file *filp = &oti->oti_file;
 960         const struct file_operations *fops;
 961         struct lmv_mds_md_v1 *lmv1;
 962         struct osd_check_lmv_buf oclb = {
 963                 .ctx.actor = osd_stripe_dir_filldir,
 964                 .oclb_info = oti,
 965                 .oclb_dev = dev,
 966                 .oclb_oic = oic,
 967                 .oclb_found = false,
 968         };
 969         int rc = 0;
 970
 971         ENTRY;
 972
 973 again:
 974         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
 975                              buf->lb_len);
 976         if (rc == -ERANGE) {
 977                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
 978                 if (rc > 0) {
 979                         lu_buf_realloc(buf, rc);
 980                         if (buf->lb_buf == NULL)
 981                                 GOTO(out, rc = -ENOMEM);
 982
 983                         goto again;
 984                 }
 985         }
 986
 987         if (unlikely(rc == 0 || rc == -ENODATA))
 988                 GOTO(out, rc = 0);
 989
 990         if (rc < 0)
 991                 GOTO(out, rc);
 992
 993         if (unlikely(buf->lb_buf == NULL)) {
 994                 lu_buf_realloc(buf, rc);
 995                 if (buf->lb_buf == NULL)
 996                         GOTO(out, rc = -ENOMEM);
 997
 998                 goto again;
 999         }
1000
1001         lmv1 = buf->lb_buf;
1002         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1003                 GOTO(out, rc = 0);
1004
1005         fops = inode->i_fop;
1006         dentry->d_inode = inode;
1007         dentry->d_sb = inode->i_sb;
1008         filp->f_pos = 0;
1009         filp->f_path.dentry = dentry;
1010         filp->f_flags |= O_NOATIME;
1011         filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
1012         filp->f_mapping = inode->i_mapping;
1013         filp->f_op = fops;
1014         filp->private_data = NULL;
1015         filp->f_cred = current_cred();
1016         filp->f_inode = inode;
1017         rc = osd_security_file_alloc(filp);
1018         if (rc)
1019                 goto out;
1020
1021         do {
1022                 oclb.oclb_items = 0;
1023                 rc = iterate_dir(filp, &oclb.ctx);
1024         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1025                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1026         fops->release(inode, filp);
1027
1028 out:
1029         if (rc < 0)
1030                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
1031                        DFID": rc = %d\n", osd_ino2name(inode),
1032                        inode->i_ino, inode->i_generation,
1033                        PFID(&oic->oic_fid), rc);
1034         else
1035                 rc = 0;
1036
1037         RETURN(rc);
1038 }
1039
1040 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1041                           const struct lu_fid *fid,
1042                           const struct lu_object_conf *conf)
1043 {
1044         struct osd_thread_info *info;
1045         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1046         struct osd_device *dev;
1047         struct osd_idmap_cache *oic;
1048         struct osd_inode_id *id;
1049         struct inode *inode = NULL;
1050         struct lustre_scrub *scrub;
1051         struct scrub_file *sf;
1052         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1053         __u32 saved_ino;
1054         __u32 saved_gen;
1055         int result = 0;
1056         int rc1 = 0;
1057         bool remote = false;
1058         bool trusted = true;
1059         bool updated = false;
1060         bool checked = false;
1061
1062         ENTRY;
1063
1064         LINVRNT(osd_invariant(obj));
1065         LASSERT(obj->oo_inode == NULL);
1066         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
1067
1068         dev = osd_dev(ldev);
1069         scrub = &dev->od_scrub.os_scrub;
1070         sf = &scrub->os_file;
1071         info = osd_oti_get(env);
1072         LASSERT(info);
1073         oic = &info->oti_cache;
1074
1075         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1076                 RETURN(-ENOENT);
1077
1078         /*
1079          * For the object is created as locking anchor, or for the object to
1080          * be created on disk. No need to osd_oi_lookup() at here because FID
1081          * shouldn't never be re-used, if it's really a duplicate FID from
1082          * unexpected reason, we should be able to detect it later by calling
1083          * do_create->osd_oi_insert().
1084          */
1085         if (conf && conf->loc_flags & LOC_F_NEW)
1086                 GOTO(out, result = 0);
1087
1088         /* Search order: 1. per-thread cache. */
1089         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1090                 id = &oic->oic_lid;
1091                 goto iget;
1092         }
1093
1094         id = &info->oti_id;
1095         if (!list_empty(&scrub->os_inconsistent_items)) {
1096                 /* Search order: 2. OI scrub pending list. */
1097                 result = osd_oii_lookup(dev, fid, id);
1098                 if (!result)
1099                         goto iget;
1100         }
1101
1102         /*
1103          * The OI mapping in the OI file can be updated by the OI scrub
1104          * when we locate the inode via FID. So it may be not trustable.
1105          */
1106         trusted = false;
1107
1108         /* Search order: 3. OI files. */
1109         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1110         if (result == -ENOENT) {
1111                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1112                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1113                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1114                                       sf->sf_oi_bitmap))
1115                         GOTO(out, result = 0);
1116
1117                 goto trigger;
1118         }
1119
1120         /* -ESTALE is returned if inode of OST object doesn't exist */
1121         if (result == -ESTALE &&
1122             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1123                 GOTO(out, result = 0);
1124         }
1125
1126         if (result)
1127                 GOTO(out, result);
1128
1129 iget:
1130         obj->oo_inode = NULL;
1131         /* for later passes through checks, not true on first pass */
1132         if (!IS_ERR_OR_NULL(inode))
1133                 iput(inode);
1134
1135         inode = osd_iget_check(info, dev, fid, id, trusted);
1136         if (!IS_ERR(inode)) {
1137                 obj->oo_inode = inode;
1138                 result = 0;
1139                 if (remote)
1140                         goto trigger;
1141
1142                 goto check_lma;
1143         }
1144
1145         result = PTR_ERR(inode);
1146         if (result == -ENOENT || result == -ESTALE)
1147                 GOTO(out, result = 0);
1148
1149         if (result != -EREMCHG)
1150                 GOTO(out, result);
1151
1152 trigger:
1153         /*
1154          * We still have chance to get the valid inode: for the
1155          * object which is referenced by remote name entry, the
1156          * object on the local MDT will be linked under the dir
1157          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1158          *
1159          * We do not know whether the object for the given FID
1160          * is referenced by some remote name entry or not, and
1161          * especially for DNE II, a multiple-linked object may
1162          * have many name entries reside on many MDTs.
1163          *
1164          * To simplify the operation, OSD will not distinguish
1165          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1166          * only happened for the RPC from other MDT during the
1167          * OI scrub, or for the client side RPC with FID only,
1168          * such as FID to path, or from old connected client.
1169          */
1170         if (!remote) {
1171                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1172                 if (!rc1) {
1173                         remote = true;
1174                         trusted = true;
1175                         flags |= SS_AUTO_PARTIAL;
1176                         flags &= ~SS_AUTO_FULL;
1177                         goto iget;
1178                 }
1179         }
1180
1181         if (thread_is_running(&scrub->os_thread)) {
1182                 if (scrub->os_partial_scan && !scrub->os_in_join)
1183                         goto join;
1184
1185                 if (IS_ERR_OR_NULL(inode) || result)
1186                         GOTO(out, result = -EINPROGRESS);
1187
1188                 LASSERT(remote);
1189                 LASSERT(obj->oo_inode == inode);
1190
1191                 osd_add_oi_cache(info, dev, id, fid);
1192                 osd_oii_insert(dev, oic, true);
1193                 goto found;
1194         }
1195
1196         if (dev->od_auto_scrub_interval == AS_NEVER) {
1197                 if (!remote)
1198                         GOTO(out, result = -EREMCHG);
1199
1200                 LASSERT(!result);
1201                 LASSERT(obj->oo_inode == inode);
1202
1203                 osd_add_oi_cache(info, dev, id, fid);
1204                 goto found;
1205         }
1206
1207 join:
1208         rc1 = osd_scrub_start(env, dev, flags);
1209         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1210                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1211         if (rc1 && rc1 != -EALREADY)
1212                 GOTO(out, result = -EREMCHG);
1213
1214         if (IS_ERR_OR_NULL(inode) || result)
1215                 GOTO(out, result = -EINPROGRESS);
1216
1217         LASSERT(remote);
1218         LASSERT(obj->oo_inode == inode);
1219
1220         osd_add_oi_cache(info, dev, id, fid);
1221         osd_oii_insert(dev, oic, true);
1222         goto found;
1223
1224 check_lma:
1225         checked = true;
1226         if (unlikely(obj->oo_header))
1227                 goto found;
1228
1229         result = osd_check_lma(env, obj);
1230         if (!result)
1231                 goto found;
1232
1233         LASSERTF(id->oii_ino == inode->i_ino &&
1234                  id->oii_gen == inode->i_generation,
1235                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1236                  PFID(fid), id->oii_ino, id->oii_gen,
1237                  inode->i_ino, inode->i_generation);
1238
1239         saved_ino = inode->i_ino;
1240         saved_gen = inode->i_generation;
1241
1242         if (unlikely(result == -ENODATA)) {
1243                 /*
1244                  * If the OI scrub updated the OI mapping by race, it
1245                  * must be valid. Trust the inode that has no LMA EA.
1246                  */
1247                 if (updated)
1248                         goto found;
1249
1250                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1251                 if (!result) {
1252                         /*
1253                          * The OI mapping is still there, the inode is still
1254                          * valid. It is just becaues the inode has no LMA EA.
1255                          */
1256                         if (saved_ino == id->oii_ino &&
1257                             saved_gen == id->oii_gen)
1258                                 goto found;
1259
1260                         /*
1261                          * It is the OI scrub updated the OI mapping by race.
1262                          * The new OI mapping must be valid.
1263                          */
1264                         trusted = true;
1265                         updated = true;
1266                         goto iget;
1267                 }
1268
1269                 /*
1270                  * "result == -ENOENT" means that the OI mappinghas been
1271                  * removed by race, so the inode belongs to other object.
1272                  *
1273                  * Others error can be returned  directly.
1274                  */
1275                 if (result == -ENOENT) {
1276                         LASSERT(trusted);
1277
1278                         obj->oo_inode = NULL;
1279                         result = 0;
1280                 }
1281         }
1282
1283         if (result != -EREMCHG)
1284                 GOTO(out, result);
1285
1286         LASSERT(!updated);
1287
1288         /*
1289          * if two OST objects map to the same inode, and inode mode is
1290          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1291          * reserved by precreate, and not written yet, in this case, don't
1292          * set inode for the object whose FID mismatch, so that it can create
1293          * inode and not block precreate.
1294          */
1295         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1296             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1297                 obj->oo_inode = NULL;
1298                 GOTO(out, result = 0);
1299         }
1300
1301         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1302         /*
1303          * "result == -ENOENT" means the cached OI mapping has been removed
1304          * from the OI file by race, above inode belongs to other object.
1305          */
1306         if (result == -ENOENT) {
1307                 LASSERT(trusted);
1308
1309                 obj->oo_inode = NULL;
1310                 GOTO(out, result = 0);
1311         }
1312
1313         if (result)
1314                 GOTO(out, result);
1315
1316         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1317                 result = -EREMCHG;
1318                 goto trigger;
1319         }
1320
1321         /*
1322          * It is the OI scrub updated the OI mapping by race.
1323          * The new OI mapping must be valid.
1324          */
1325         trusted = true;
1326         updated = true;
1327         goto iget;
1328
1329 found:
1330         if (!checked) {
1331                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1332                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1333
1334                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1335                 if (!result) {
1336                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1337                             dev->od_is_ost)
1338                                 obj->oo_pfid_in_lma = 1;
1339                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1340                             !dev->od_is_ost)
1341                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1342                 } else if (result != -ENODATA) {
1343                         GOTO(out, result);
1344                 }
1345         }
1346
1347         obj->oo_compat_dot_created = 1;
1348         obj->oo_compat_dotdot_created = 1;
1349
1350         if (S_ISDIR(inode->i_mode) &&
1351             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1352                 osd_check_lmv(info, dev, inode, oic);
1353
1354         result = osd_attach_jinode(inode);
1355         if (result)
1356                 GOTO(out, result);
1357
1358         if (!ldiskfs_pdo)
1359                 GOTO(out, result = 0);
1360
1361         LASSERT(!obj->oo_hl_head);
1362         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1363
1364         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1365
1366 out:
1367         if (result || !obj->oo_inode) {
1368                 if (!IS_ERR_OR_NULL(inode))
1369                         iput(inode);
1370
1371                 obj->oo_inode = NULL;
1372                 if (trusted)
1373                         fid_zero(&oic->oic_fid);
1374         }
1375
1376         LINVRNT(osd_invariant(obj));
1377         return result;
1378 }
1379
1380 /*
1381  * Concurrency: shouldn't matter.
1382  */
1383 static void osd_object_init0(struct osd_object *obj)
1384 {
1385         LASSERT(obj->oo_inode != NULL);
1386         obj->oo_dt.do_body_ops = &osd_body_ops;
1387         obj->oo_dt.do_lu.lo_header->loh_attr |=
1388                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1389 }
1390
1391 /*
1392  * Concurrency: no concurrent access is possible that early in object
1393  * life-cycle.
1394  */
1395 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1396                            const struct lu_object_conf *conf)
1397 {
1398         struct osd_object *obj = osd_obj(l);
1399         int result;
1400
1401         LINVRNT(osd_invariant(obj));
1402
1403         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1404                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1405                 l->lo_header->loh_attr |= LOHA_EXISTS;
1406                 return 0;
1407         }
1408
1409         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1410         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1411         if (result == 0 && obj->oo_inode != NULL) {
1412                 struct osd_thread_info *oti = osd_oti_get(env);
1413                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1414
1415                 osd_object_init0(obj);
1416                 if (unlikely(obj->oo_header))
1417                         return 0;
1418
1419                 result = osd_get_lma(oti, obj->oo_inode,
1420                                      &oti->oti_obj_dentry, loa);
1421                 if (!result) {
1422                         /*
1423                          * Convert LMAI flags to lustre LMA flags
1424                          * and cache it to oo_lma_flags
1425                          */
1426                         obj->oo_lma_flags =
1427                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1428                 } else if (result == -ENODATA) {
1429                         result = 0;
1430                 }
1431         }
1432
1433         LINVRNT(osd_invariant(obj));
1434         return result;
1435 }
1436
1437 /*
1438  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1439  * The left part is for value, binary mode.
1440  */
1441 struct osd_xattr_entry {
1442         struct list_head        oxe_list;
1443         size_t                  oxe_len;
1444         size_t                  oxe_namelen;
1445         bool                    oxe_exist;
1446         struct rcu_head         oxe_rcu;
1447         char                    oxe_buf[0];
1448 };
1449
1450 static int osd_oxc_get(struct osd_object *obj, const char *name,
1451                        struct lu_buf *buf)
1452 {
1453         struct osd_xattr_entry *tmp;
1454         struct osd_xattr_entry *oxe = NULL;
1455         size_t namelen = strlen(name);
1456         int rc;
1457
1458         ENTRY;
1459
1460         rcu_read_lock();
1461         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1462                 if (namelen == tmp->oxe_namelen &&
1463                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1464                         oxe = tmp;
1465                         break;
1466                 }
1467         }
1468
1469         if (oxe == NULL)
1470                 GOTO(out, rc = -ENOENT);
1471
1472         if (!oxe->oxe_exist)
1473                 GOTO(out, rc = -ENODATA);
1474
1475         /* vallen */
1476         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1477         LASSERT(rc > 0);
1478
1479         if (buf->lb_buf == NULL)
1480                 GOTO(out, rc);
1481
1482         if (buf->lb_len < rc)
1483                 GOTO(out, rc = -ERANGE);
1484
1485         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1486         EXIT;
1487 out:
1488         rcu_read_unlock();
1489
1490         return rc;
1491 }
1492
1493 static void osd_oxc_free(struct rcu_head *head)
1494 {
1495         struct osd_xattr_entry *oxe;
1496
1497         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1498         OBD_FREE(oxe, oxe->oxe_len);
1499 }
1500
1501 static void osd_oxc_add(struct osd_object *obj, const char *name,
1502                         const char *buf, int buflen)
1503 {
1504         struct osd_xattr_entry *oxe;
1505         struct osd_xattr_entry *old = NULL;
1506         struct osd_xattr_entry *tmp;
1507         size_t namelen = strlen(name);
1508         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1509
1510         OBD_ALLOC(oxe, len);
1511         if (oxe == NULL)
1512                 return;
1513
1514         INIT_LIST_HEAD(&oxe->oxe_list);
1515         oxe->oxe_len = len;
1516         oxe->oxe_namelen = namelen;
1517         memcpy(oxe->oxe_buf, name, namelen);
1518         if (buflen > 0) {
1519                 LASSERT(buf != NULL);
1520                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1521                 oxe->oxe_exist = true;
1522         } else {
1523                 oxe->oxe_exist = false;
1524         }
1525
1526         /* this should be rarely called, just remove old and add new */
1527         spin_lock(&obj->oo_guard);
1528         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1529                 if (namelen == tmp->oxe_namelen &&
1530                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1531                         old = tmp;
1532                         break;
1533                 }
1534         }
1535         if (old != NULL) {
1536                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1537                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1538         } else {
1539                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1540         }
1541         spin_unlock(&obj->oo_guard);
1542 }
1543
1544 static void osd_oxc_del(struct osd_object *obj, const char *name)
1545 {
1546         struct osd_xattr_entry *oxe;
1547         size_t namelen = strlen(name);
1548
1549         spin_lock(&obj->oo_guard);
1550         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1551                 if (namelen == oxe->oxe_namelen &&
1552                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1553                         list_del_rcu(&oxe->oxe_list);
1554                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1555                         break;
1556                 }
1557         }
1558         spin_unlock(&obj->oo_guard);
1559 }
1560
1561 static void osd_oxc_fini(struct osd_object *obj)
1562 {
1563         struct osd_xattr_entry *oxe, *next;
1564
1565         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1566                 list_del(&oxe->oxe_list);
1567                 OBD_FREE(oxe, oxe->oxe_len);
1568         }
1569 }
1570
1571 /*
1572  * Concurrency: no concurrent access is possible that late in object
1573  * life-cycle.
1574  */
1575 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1576 {
1577         struct osd_object *obj = osd_obj(l);
1578         struct lu_object_header *h = obj->oo_header;
1579
1580         LINVRNT(osd_invariant(obj));
1581
1582         osd_oxc_fini(obj);
1583         dt_object_fini(&obj->oo_dt);
1584         if (obj->oo_hl_head != NULL)
1585                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1586         OBD_FREE_PTR(obj);
1587         if (unlikely(h)) {
1588                 lu_object_header_fini(h);
1589                 OBD_FREE_PTR(h);
1590         }
1591 }
1592
1593 /*
1594  * Concurrency: no concurrent access is possible that late in object
1595  * life-cycle.
1596  */
1597 static void osd_index_fini(struct osd_object *o)
1598 {
1599         struct iam_container *bag;
1600
1601         if (o->oo_dir != NULL) {
1602                 bag = &o->oo_dir->od_container;
1603                 if (o->oo_inode != NULL) {
1604                         if (bag->ic_object == o->oo_inode)
1605                                 iam_container_fini(bag);
1606                 }
1607                 OBD_FREE_PTR(o->oo_dir);
1608                 o->oo_dir = NULL;
1609         }
1610 }
1611
1612 /*
1613  * Concurrency: no concurrent access is possible that late in object
1614  * life-cycle (for all existing callers, that is. New callers have to provide
1615  * their own locking.)
1616  */
1617 static int osd_inode_unlinked(const struct inode *inode)
1618 {
1619         return inode->i_nlink == 0;
1620 }
1621
1622 enum {
1623         OSD_TXN_OI_DELETE_CREDITS    = 20,
1624         OSD_TXN_INODE_DELETE_CREDITS = 20
1625 };
1626
1627 /*
1628  * Journal
1629  */
1630
1631 #if OSD_THANDLE_STATS
1632 /**
1633  * Set time when the handle is allocated
1634  */
1635 static void osd_th_alloced(struct osd_thandle *oth)
1636 {
1637         oth->oth_alloced = ktime_get();
1638 }
1639
1640 /**
1641  * Set time when the handle started
1642  */
1643 static void osd_th_started(struct osd_thandle *oth)
1644 {
1645         oth->oth_started = ktime_get();
1646 }
1647
1648 /**
1649  * Check whether the we deal with this handle for too long.
1650  */
1651 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1652                                 ktime_t alloced, ktime_t started,
1653                                 ktime_t closed)
1654 {
1655         ktime_t now = ktime_get();
1656
1657         LASSERT(dev != NULL);
1658
1659         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1660                             ktime_us_delta(started, alloced));
1661         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1662                             ktime_us_delta(closed, started));
1663         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1664                             ktime_us_delta(now, closed));
1665
1666         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1667                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1668                                 oth, now, alloced, started, closed);
1669                 libcfs_debug_dumpstack(NULL);
1670         }
1671 }
1672
1673 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1674 {                                                                       \
1675         ktime_t __closed = ktime_get();                                 \
1676         ktime_t __alloced = oth->oth_alloced;                           \
1677         ktime_t __started = oth->oth_started;                           \
1678                                                                         \
1679         expr;                                                           \
1680         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1681 }
1682
1683 #else /* OSD_THANDLE_STATS */
1684
1685 #define osd_th_alloced(h)                  do {} while(0)
1686 #define osd_th_started(h)                  do {} while(0)
1687 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1688
1689 #endif /* OSD_THANDLE_STATS */
1690
1691 /*
1692  * Concurrency: doesn't access mutable data.
1693  */
1694 static int osd_param_is_not_sane(const struct osd_device *dev,
1695                                  const struct thandle *th)
1696 {
1697         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1698
1699         return oh->ot_credits > osd_transaction_size(dev);
1700 }
1701
1702 /*
1703  * Concurrency: shouldn't matter.
1704  */
1705 static void osd_trans_commit_cb(struct super_block *sb,
1706                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1707 {
1708         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1709         struct thandle *th = &oh->ot_super;
1710         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1711         struct dt_txn_commit_cb *dcb, *tmp;
1712
1713         LASSERT(oh->ot_handle == NULL);
1714
1715         if (error)
1716                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1717
1718         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1719         /* call per-transaction callbacks if any */
1720         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1721                                  dcb_linkage) {
1722                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1723                          "commit callback entry: magic=%x name='%s'\n",
1724                          dcb->dcb_magic, dcb->dcb_name);
1725                 list_del_init(&dcb->dcb_linkage);
1726                 dcb->dcb_func(NULL, th, dcb, error);
1727         }
1728
1729         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1730         lu_device_put(lud);
1731         th->th_dev = NULL;
1732
1733         OBD_FREE_PTR(oh);
1734 }
1735
1736 static struct thandle *osd_trans_create(const struct lu_env *env,
1737                                         struct dt_device *d)
1738 {
1739         struct osd_thread_info *oti = osd_oti_get(env);
1740         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1741         struct osd_thandle *oh;
1742         struct thandle *th;
1743
1744         ENTRY;
1745
1746         if (d->dd_rdonly) {
1747                 CERROR("%s: someone try to start transaction under "
1748                        "readonly mode, should be disabled.\n",
1749                        osd_name(osd_dt_dev(d)));
1750                 dump_stack();
1751                 RETURN(ERR_PTR(-EROFS));
1752         }
1753
1754         /* on pending IO in this thread should left from prev. request */
1755         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1756
1757         sb_start_write(osd_sb(osd_dt_dev(d)));
1758
1759         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1760         if (!oh) {
1761                 sb_end_write(osd_sb(osd_dt_dev(d)));
1762                 RETURN(ERR_PTR(-ENOMEM));
1763         }
1764
1765         oh->ot_quota_trans = &oti->oti_quota_trans;
1766         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1767         th = &oh->ot_super;
1768         th->th_dev = d;
1769         th->th_result = 0;
1770         oh->ot_credits = 0;
1771         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1772         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1773         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1774         osd_th_alloced(oh);
1775
1776         memset(oti->oti_declare_ops, 0,
1777                sizeof(oti->oti_declare_ops));
1778         memset(oti->oti_declare_ops_cred, 0,
1779                sizeof(oti->oti_declare_ops_cred));
1780         memset(oti->oti_declare_ops_used, 0,
1781                sizeof(oti->oti_declare_ops_used));
1782
1783         oti->oti_ins_cache_depth++;
1784
1785         RETURN(th);
1786 }
1787
1788 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1789 {
1790         struct osd_thread_info *oti = osd_oti_get(env);
1791         struct osd_thandle *oh;
1792
1793         oh = container_of0(th, struct osd_thandle, ot_super);
1794         LASSERT(oh != NULL);
1795
1796         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1797               oti->oti_declare_ops[OSD_OT_CREATE],
1798               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1799               oti->oti_declare_ops_used[OSD_OT_CREATE],
1800               oti->oti_declare_ops[OSD_OT_DESTROY],
1801               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1802               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1803         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1804               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1805               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1806               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1807               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1808               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1809               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1810         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1811               oti->oti_declare_ops[OSD_OT_WRITE],
1812               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1813               oti->oti_declare_ops_used[OSD_OT_WRITE],
1814               oti->oti_declare_ops[OSD_OT_PUNCH],
1815               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1816               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1817               oti->oti_declare_ops[OSD_OT_QUOTA],
1818               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1819               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1820         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1821               oti->oti_declare_ops[OSD_OT_INSERT],
1822               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1823               oti->oti_declare_ops_used[OSD_OT_INSERT],
1824               oti->oti_declare_ops[OSD_OT_DELETE],
1825               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1826               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1827         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1828               oti->oti_declare_ops[OSD_OT_REF_ADD],
1829               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1830               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1831               oti->oti_declare_ops[OSD_OT_REF_DEL],
1832               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1833               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1834 }
1835
1836 /*
1837  * Concurrency: shouldn't matter.
1838  */
1839 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1840                            struct thandle *th)
1841 {
1842         struct osd_thread_info *oti = osd_oti_get(env);
1843         struct osd_device *dev = osd_dt_dev(d);
1844         handle_t *jh;
1845         struct osd_thandle *oh;
1846         int rc;
1847
1848         ENTRY;
1849
1850         LASSERT(current->journal_info == NULL);
1851
1852         oh = container_of0(th, struct osd_thandle, ot_super);
1853         LASSERT(oh != NULL);
1854         LASSERT(oh->ot_handle == NULL);
1855
1856         rc = dt_txn_hook_start(env, d, th);
1857         if (rc != 0)
1858                 GOTO(out, rc);
1859
1860         if (unlikely(osd_param_is_not_sane(dev, th))) {
1861                 static unsigned long last_printed;
1862                 static int last_credits;
1863
1864                 /*
1865                  * don't make noise on a tiny testing systems
1866                  * actual credits misuse will be caught anyway
1867                  */
1868                 if (last_credits != oh->ot_credits &&
1869                     time_after(jiffies, last_printed +
1870                                cfs_time_seconds(60)) &&
1871                     osd_transaction_size(dev) > 512) {
1872                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1873                               oh->ot_credits, osd_transaction_size(dev));
1874                         osd_trans_dump_creds(env, th);
1875                         libcfs_debug_dumpstack(NULL);
1876                         last_credits = oh->ot_credits;
1877                         last_printed = jiffies;
1878                 }
1879                 /*
1880                  * XXX Limit the credits to 'max_transaction_buffers', and
1881                  *     let the underlying filesystem to catch the error if
1882                  *     we really need so many credits.
1883                  *
1884                  *     This should be removed when we can calculate the
1885                  *     credits precisely.
1886                  */
1887                 oh->ot_credits = osd_transaction_size(dev);
1888         } else if (ldiskfs_track_declares_assert != 0) {
1889                 /*
1890                  * reserve few credits to prevent an assertion in JBD
1891                  * our debugging mechanism will be able to detected
1892                  * overuse. this can help to debug single-update
1893                  * transactions
1894                  */
1895                 oh->ot_credits += 10;
1896                 if (unlikely(osd_param_is_not_sane(dev, th)))
1897                         oh->ot_credits = osd_transaction_size(dev);
1898         }
1899
1900         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1901                 GOTO(out, rc = -EIO);
1902
1903         /*
1904          * XXX temporary stuff. Some abstraction layer should
1905          * be used.
1906          */
1907         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1908         osd_th_started(oh);
1909         if (!IS_ERR(jh)) {
1910                 oh->ot_handle = jh;
1911                 LASSERT(oti->oti_txns == 0);
1912
1913                 lu_device_get(&d->dd_lu_dev);
1914                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1915                               "osd-tx", th);
1916                 oti->oti_txns++;
1917                 rc = 0;
1918         } else {
1919                 rc = PTR_ERR(jh);
1920         }
1921 out:
1922         RETURN(rc);
1923 }
1924
1925 static int osd_seq_exists(const struct lu_env *env,
1926                           struct osd_device *osd, u64 seq)
1927 {
1928         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
1929         struct seq_server_site *ss = osd_seq_site(osd);
1930         int rc;
1931
1932         ENTRY;
1933
1934         LASSERT(ss != NULL);
1935         LASSERT(ss->ss_server_fld != NULL);
1936
1937         rc = osd_fld_lookup(env, osd, seq, range);
1938         if (rc != 0) {
1939                 if (rc != -ENOENT)
1940                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1941                                osd_name(osd), seq, rc);
1942                 RETURN(0);
1943         }
1944
1945         RETURN(ss->ss_node_id == range->lsr_index);
1946 }
1947
1948 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1949 {
1950         struct dt_txn_commit_cb *dcb;
1951         struct dt_txn_commit_cb *tmp;
1952
1953         /* call per-transaction stop callbacks if any */
1954         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1955                                  dcb_linkage) {
1956                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1957                          "commit callback entry: magic=%x name='%s'\n",
1958                          dcb->dcb_magic, dcb->dcb_name);
1959                 list_del_init(&dcb->dcb_linkage);
1960                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1961         }
1962 }
1963
1964 /*
1965  * Concurrency: shouldn't matter.
1966  */
1967 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1968                           struct thandle *th)
1969 {
1970         struct osd_thread_info *oti = osd_oti_get(env);
1971         struct osd_thandle *oh;
1972         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1973         struct osd_device *osd = osd_dt_dev(th->th_dev);
1974         struct qsd_instance *qsd = osd_def_qsd(osd);
1975         struct lquota_trans *qtrans;
1976         LIST_HEAD(truncates);
1977         int rc = 0, remove_agents = 0;
1978
1979         ENTRY;
1980
1981         oh = container_of0(th, struct osd_thandle, ot_super);
1982
1983         remove_agents = oh->ot_remove_agents;
1984
1985         qtrans = oh->ot_quota_trans;
1986         oh->ot_quota_trans = NULL;
1987
1988         /* move locks to local list, stop tx, execute truncates */
1989         list_splice(&oh->ot_trunc_locks, &truncates);
1990
1991         if (oh->ot_handle != NULL) {
1992                 int rc2;
1993
1994                 handle_t *hdl = oh->ot_handle;
1995
1996                 /*
1997                  * add commit callback
1998                  * notice we don't do this in osd_trans_start()
1999                  * as underlying transaction can change during truncate
2000                  */
2001                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2002                                              &oh->ot_jcb);
2003
2004                 LASSERT(oti->oti_txns == 1);
2005                 oti->oti_txns--;
2006
2007                 rc = dt_txn_hook_stop(env, th);
2008                 if (rc != 0)
2009                         CERROR("%s: failed in transaction hook: rc = %d\n",
2010                                osd_name(osd), rc);
2011
2012                 osd_trans_stop_cb(oh, rc);
2013                 /* hook functions might modify th_sync */
2014                 hdl->h_sync = th->th_sync;
2015
2016                 oh->ot_handle = NULL;
2017                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2018                 if (rc2 != 0)
2019                         CERROR("%s: failed to stop transaction: rc = %d\n",
2020                                osd_name(osd), rc2);
2021                 if (!rc)
2022                         rc = rc2;
2023
2024                 osd_process_truncates(&truncates);
2025         } else {
2026                 osd_trans_stop_cb(oh, th->th_result);
2027                 OBD_FREE_PTR(oh);
2028         }
2029
2030         osd_trunc_unlock_all(&truncates);
2031
2032         /* inform the quota slave device that the transaction is stopping */
2033         qsd_op_end(env, qsd, qtrans);
2034
2035         /*
2036          * as we want IO to journal and data IO be concurrent, we don't block
2037          * awaiting data IO completion in osd_do_bio(), instead we wait here
2038          * once transaction is submitted to the journal. all reqular requests
2039          * don't do direct IO (except read/write), thus this wait_event becomes
2040          * no-op for them.
2041          *
2042          * IMPORTANT: we have to wait till any IO submited by the thread is
2043          * completed otherwise iobuf may be corrupted by different request
2044          */
2045         wait_event(iobuf->dr_wait,
2046                        atomic_read(&iobuf->dr_numreqs) == 0);
2047         osd_fini_iobuf(osd, iobuf);
2048         if (!rc)
2049                 rc = iobuf->dr_error;
2050
2051         if (unlikely(remove_agents != 0))
2052                 osd_process_scheduled_agent_removals(env, osd);
2053
2054         oti->oti_ins_cache_depth--;
2055         /* reset OI cache for safety */
2056         if (oti->oti_ins_cache_depth == 0)
2057                 oti->oti_ins_cache_used = 0;
2058
2059         sb_end_write(osd_sb(osd));
2060
2061         RETURN(rc);
2062 }
2063
2064 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2065 {
2066         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
2067                                                ot_super);
2068
2069         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2070         LASSERT(&dcb->dcb_func != NULL);
2071         if (dcb->dcb_flags & DCB_TRANS_STOP)
2072                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2073         else
2074                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2075
2076         return 0;
2077 }
2078
2079 /*
2080  * Called just before object is freed. Releases all resources except for
2081  * object itself (that is released by osd_object_free()).
2082  *
2083  * Concurrency: no concurrent access is possible that late in object
2084  * life-cycle.
2085  */
2086 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2087 {
2088         struct osd_object *obj = osd_obj(l);
2089         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2090         struct inode *inode = obj->oo_inode;
2091         __u64 projid;
2092         qid_t uid;
2093         qid_t gid;
2094
2095         LINVRNT(osd_invariant(obj));
2096
2097         /*
2098          * If object is unlinked remove fid->ino mapping from object index.
2099          */
2100
2101         osd_index_fini(obj);
2102
2103         if (!inode)
2104                 return;
2105
2106         uid = i_uid_read(inode);
2107         gid = i_gid_read(inode);
2108         projid = i_projid_read(inode);
2109
2110         obj->oo_inode = NULL;
2111         iput(inode);
2112
2113         /* do not rebalance quota if the caller needs to release memory
2114          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2115          * transaction and risk to deadlock - LU-12178 */
2116         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2117                 return;
2118
2119         if (!obj->oo_header && qsd) {
2120                 struct osd_thread_info *info = osd_oti_get(env);
2121                 struct lquota_id_info *qi = &info->oti_qi;
2122
2123                 /* Release granted quota to master if necessary */
2124                 qi->lqi_id.qid_uid = uid;
2125                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2126
2127                 qi->lqi_id.qid_uid = gid;
2128                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2129
2130                 qi->lqi_id.qid_uid = projid;
2131                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2132         }
2133 }
2134
2135 /*
2136  * Concurrency: ->loo_object_release() is called under site spin-lock.
2137  */
2138 static void osd_object_release(const struct lu_env *env,
2139                                struct lu_object *l)
2140 {
2141         struct osd_object *o = osd_obj(l);
2142
2143         /*
2144          * nobody should be releasing a non-destroyed object with nlink=0
2145          * the API allows this, but ldiskfs doesn't like and then report
2146          * this inode as deleted
2147          */
2148         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2149                   o->oo_inode->i_nlink == 0));
2150 }
2151
2152 /*
2153  * Concurrency: shouldn't matter.
2154  */
2155 static int osd_object_print(const struct lu_env *env, void *cookie,
2156                             lu_printer_t p, const struct lu_object *l)
2157 {
2158         struct osd_object *o = osd_obj(l);
2159         struct iam_descr *d;
2160
2161         if (o->oo_dir != NULL)
2162                 d = o->oo_dir->od_container.ic_descr;
2163         else
2164                 d = NULL;
2165         return (*p)(env, cookie,
2166                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2167                     o, o->oo_inode,
2168                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2169                     o->oo_inode ? o->oo_inode->i_generation : 0,
2170                     d ? d->id_ops->id_name : "plain");
2171 }
2172
2173 /*
2174  * Concurrency: shouldn't matter.
2175  */
2176 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2177                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2178 {
2179         struct osd_device *osd = osd_dt_dev(d);
2180         struct super_block *sb = osd_sb(osd);
2181         struct kstatfs *ksfs;
2182         __u64 reserved;
2183         int result = 0;
2184
2185         if (unlikely(osd->od_mnt == NULL))
2186                 return -EINPROGRESS;
2187
2188         /* osd_lproc.c call this without env, allocate ksfs for that case */
2189         if (unlikely(env == NULL)) {
2190                 OBD_ALLOC_PTR(ksfs);
2191                 if (ksfs == NULL)
2192                         return -ENOMEM;
2193         } else {
2194                 ksfs = &osd_oti_get(env)->oti_ksfs;
2195         }
2196
2197         result = sb->s_op->statfs(sb->s_root, ksfs);
2198         if (result)
2199                 goto out;
2200
2201         statfs_pack(sfs, ksfs);
2202         if (unlikely(sb->s_flags & SB_RDONLY))
2203                 sfs->os_state |= OS_STATE_READONLY;
2204
2205         sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
2206
2207         if (ldiskfs_has_feature_extents(sb))
2208                 sfs->os_maxbytes = sb->s_maxbytes;
2209         else
2210                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2211
2212         /*
2213          * Reserve some space so to avoid fragmenting the filesystem too much.
2214          * Fragmentation not only impacts performance, but can also increase
2215          * metadata overhead significantly, causing grant calculation to be
2216          * wrong.
2217          *
2218          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2219          */
2220         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2221         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2222         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2223                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2224
2225         sfs->os_blocks -= reserved;
2226         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2227         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2228
2229 out:
2230         if (unlikely(env == NULL))
2231                 OBD_FREE_PTR(ksfs);
2232         return result;
2233 }
2234
2235 /**
2236  * Estimate space needed for file creations. We assume the largest filename
2237  * which is 2^64 - 1, hence a filename of 20 chars.
2238  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2239  */
2240 #ifdef __LDISKFS_DIR_REC_LEN
2241 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2242 #else
2243 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2244 #endif
2245
2246 /*
2247  * Concurrency: doesn't access mutable data.
2248  */
2249 static void osd_conf_get(const struct lu_env *env,
2250                          const struct dt_device *dev,
2251                          struct dt_device_param *param)
2252 {
2253         struct osd_device *d = osd_dt_dev(dev);
2254         struct super_block *sb = osd_sb(d);
2255         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2256         const char *name;
2257         int ea_overhead;
2258
2259         /*
2260          * XXX should be taken from not-yet-existing fs abstraction layer.
2261          */
2262         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2263         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2264         param->ddp_symlink_max  = sb->s_blocksize;
2265         param->ddp_mount_type   = LDD_MT_LDISKFS;
2266         if (ldiskfs_has_feature_extents(sb))
2267                 param->ddp_maxbytes = sb->s_maxbytes;
2268         else
2269                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2270         /*
2271          * inode are statically allocated, so per-inode space consumption
2272          * is the space consumed by the directory entry
2273          */
2274         param->ddp_inodespace     = PER_OBJ_USAGE;
2275         /*
2276          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2277          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2278          * maximum length to not under-count the actual number of extents
2279          * needed for writing a file if there are sub-optimal block allocations.
2280          */
2281         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2282         /* worst-case extent insertion metadata overhead */
2283         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2284         param->ddp_mntopts = 0;
2285         if (test_opt(sb, XATTR_USER))
2286                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2287         if (test_opt(sb, POSIX_ACL))
2288                 param->ddp_mntopts |= MNTOPT_ACL;
2289
2290         /*
2291          * LOD might calculate the max stripe count based on max_ea_size,
2292          * so we need take account in the overhead as well,
2293          * xattr_header + magic + xattr_entry_head
2294          */
2295         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2296                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2297
2298 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2299         if (ldiskfs_has_feature_ea_inode(sb))
2300                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2301                                                                 ea_overhead;
2302         else
2303 #endif
2304                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2305
2306         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2307                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2308
2309         /*
2310          * Preferred RPC size for efficient disk IO.  4MB shows good
2311          * all-around performance for ldiskfs, but use bigalloc chunk size
2312          * by default if larger.
2313          */
2314 #if defined(LDISKFS_CLUSTER_SIZE)
2315         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2316                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2317         else
2318 #endif
2319                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2320
2321         param->ddp_t10_cksum_type = 0;
2322         if (bi) {
2323                 unsigned short interval = blk_integrity_interval(bi);
2324                 name = blk_integrity_name(bi);
2325                 /*
2326                  * Expected values:
2327                  * T10-DIF-TYPE1-CRC
2328                  * T10-DIF-TYPE3-CRC
2329                  * T10-DIF-TYPE1-IP
2330                  * T10-DIF-TYPE3-IP
2331                  */
2332                 if (strncmp(name, "T10-DIF-TYPE",
2333                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2334                         /* also skip "1/3-" at end */
2335                         const int type_off = sizeof("T10-DIF-TYPE.");
2336                         char type_number = name[type_off - 2];
2337
2338                         if (interval != 512 && interval != 4096) {
2339                                 CERROR("%s: unsupported T10PI sector size %u\n",
2340                                        d->od_svname, interval);
2341                         } else if (type_number != '1' && type_number != '3') {
2342                                 CERROR("%s: unsupported T10PI type %s\n",
2343                                        d->od_svname, name);
2344                         } else if (strcmp(name + type_off, "CRC") == 0) {
2345                                 d->od_t10_type = type_number == '1' ?
2346                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2347                                 param->ddp_t10_cksum_type = interval == 512 ?
2348                                         OBD_CKSUM_T10CRC512 :
2349                                         OBD_CKSUM_T10CRC4K;
2350                         } else if (strcmp(name + type_off, "IP") == 0) {
2351                                 d->od_t10_type = type_number == '1' ?
2352                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2353                                 param->ddp_t10_cksum_type = interval == 512 ?
2354                                         OBD_CKSUM_T10IP512 :
2355                                         OBD_CKSUM_T10IP4K;
2356                         } else {
2357                                 CERROR("%s: unsupported checksum type of "
2358                                        "T10PI type '%s'",
2359                                        d->od_svname, name);
2360                         }
2361
2362                 } else {
2363                         CERROR("%s: unsupported T10PI type '%s'",
2364                                d->od_svname, name);
2365                 }
2366         }
2367 }
2368
2369 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2370 {
2371         return osd_sb(osd_dt_dev(d));
2372 }
2373
2374 /*
2375  * Concurrency: shouldn't matter.
2376  */
2377 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2378 {
2379         int rc;
2380         struct super_block *s = osd_sb(osd_dt_dev(d));
2381         ENTRY;
2382
2383         down_read(&s->s_umount);
2384         rc = s->s_op->sync_fs(s, 1);
2385         up_read(&s->s_umount);
2386
2387         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2388                rc);
2389
2390         return rc;
2391 }
2392
2393 /**
2394  * Start commit for OSD device.
2395  *
2396  * An implementation of dt_commit_async method for OSD device.
2397  * Asychronously starts underlayng fs sync and thereby a transaction
2398  * commit.
2399  *
2400  * \param env environment
2401  * \param d dt device
2402  *
2403  * \see dt_device_operations
2404  */
2405 static int osd_commit_async(const struct lu_env *env,
2406                             struct dt_device *d)
2407 {
2408         struct super_block *s = osd_sb(osd_dt_dev(d));
2409         int rc;
2410
2411         ENTRY;
2412
2413         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2414         down_read(&s->s_umount);
2415         rc = s->s_op->sync_fs(s, 0);
2416         up_read(&s->s_umount);
2417
2418         RETURN(rc);
2419 }
2420
2421 /* Our own copy of the set readonly functions if present, or NU if not. */
2422 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2423 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2424 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2425 static int (*priv_security_file_alloc)(struct file *file);
2426
2427 int osd_security_file_alloc(struct file *file)
2428 {
2429         if (priv_security_file_alloc)
2430                 return priv_security_file_alloc(file);
2431         return 0;
2432 }
2433
2434 /*
2435  * Concurrency: shouldn't matter.
2436  */
2437 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2438 {
2439         struct super_block *sb = osd_sb(osd_dt_dev(d));
2440         struct block_device *dev = sb->s_bdev;
2441         int rc = -EOPNOTSUPP;
2442
2443         ENTRY;
2444
2445         if (priv_dev_set_rdonly) {
2446                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2447
2448                 rc = 0;
2449                 CERROR("*** setting %s read-only ***\n",
2450                        osd_dt_dev(d)->od_svname);
2451
2452                 if (sb->s_op->freeze_fs) {
2453                         rc = sb->s_op->freeze_fs(sb);
2454                         if (rc)
2455                                 goto out;
2456                 }
2457
2458                 if (jdev && (jdev != dev)) {
2459                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2460                                (long)jdev);
2461                         priv_dev_set_rdonly(jdev);
2462                 }
2463                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2464                 priv_dev_set_rdonly(dev);
2465
2466                 if (sb->s_op->unfreeze_fs)
2467                         sb->s_op->unfreeze_fs(sb);
2468         }
2469
2470 out:
2471         if (rc)
2472                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2473                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2474
2475         RETURN(rc);
2476 }
2477
2478 /**
2479  * Note: we do not count into QUOTA here.
2480  * If we mount with --data_journal we may need more.
2481  */
2482 const int osd_dto_credits_noquota[DTO_NR] = {
2483         /**
2484          * Insert.
2485          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2486          * SINGLEDATA_TRANS_BLOCKS(8)
2487          * XXX Note: maybe iam need more, since iam have more level than
2488          *           EXT3 htree.
2489          */
2490         [DTO_INDEX_INSERT]  = 16,
2491         /**
2492          * Delete
2493          * just modify a single entry, probably merge few within a block
2494          */
2495         [DTO_INDEX_DELETE]  = 1,
2496         /**
2497          * Used for OI scrub
2498          */
2499         [DTO_INDEX_UPDATE]  = 16,
2500         /**
2501          * 4(inode, inode bits, groups, GDT)
2502          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2503          */
2504         [DTO_OBJECT_CREATE] = 4,
2505         /**
2506          * 4(inode, inode bits, groups, GDT)
2507          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2508          */
2509         [DTO_OBJECT_DELETE] = 4,
2510         /**
2511          * Attr set credits (inode)
2512          */
2513         [DTO_ATTR_SET_BASE] = 1,
2514         /**
2515          * Xattr set. The same as xattr of EXT3.
2516          * DATA_TRANS_BLOCKS(14)
2517          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2518          * are also counted in. Do not know why?
2519          */
2520         [DTO_XATTR_SET]     = 14,
2521         /**
2522          * credits for inode change during write.
2523          */
2524         [DTO_WRITE_BASE]    = 3,
2525         /**
2526          * credits for single block write.
2527          */
2528         [DTO_WRITE_BLOCK]   = 14,
2529         /**
2530          * Attr set credits for chown.
2531          * This is extra credits for setattr, and it is null without quota
2532          */
2533         [DTO_ATTR_SET_CHOWN] = 0
2534 };
2535
2536 static const struct dt_device_operations osd_dt_ops = {
2537         .dt_root_get       = osd_root_get,
2538         .dt_statfs         = osd_statfs,
2539         .dt_trans_create   = osd_trans_create,
2540         .dt_trans_start    = osd_trans_start,
2541         .dt_trans_stop     = osd_trans_stop,
2542         .dt_trans_cb_add   = osd_trans_cb_add,
2543         .dt_conf_get       = osd_conf_get,
2544         .dt_mnt_sb_get     = osd_mnt_sb_get,
2545         .dt_sync           = osd_sync,
2546         .dt_ro             = osd_ro,
2547         .dt_commit_async   = osd_commit_async,
2548 };
2549
2550 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2551                           unsigned int role)
2552 {
2553         struct osd_object *obj = osd_dt_obj(dt);
2554         struct osd_thread_info *oti = osd_oti_get(env);
2555
2556         LINVRNT(osd_invariant(obj));
2557
2558         LASSERT(obj->oo_owner != env);
2559         down_read_nested(&obj->oo_sem, role);
2560
2561         LASSERT(obj->oo_owner == NULL);
2562         oti->oti_r_locks++;
2563 }
2564
2565 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2566                            unsigned int role)
2567 {
2568         struct osd_object *obj = osd_dt_obj(dt);
2569         struct osd_thread_info *oti = osd_oti_get(env);
2570
2571         LINVRNT(osd_invariant(obj));
2572
2573         LASSERT(obj->oo_owner != env);
2574         down_write_nested(&obj->oo_sem, role);
2575
2576         LASSERT(obj->oo_owner == NULL);
2577         obj->oo_owner = env;
2578         oti->oti_w_locks++;
2579 }
2580
2581 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2582 {
2583         struct osd_object *obj = osd_dt_obj(dt);
2584         struct osd_thread_info *oti = osd_oti_get(env);
2585
2586         LINVRNT(osd_invariant(obj));
2587
2588         LASSERT(oti->oti_r_locks > 0);
2589         oti->oti_r_locks--;
2590         up_read(&obj->oo_sem);
2591 }
2592
2593 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2594 {
2595         struct osd_object *obj = osd_dt_obj(dt);
2596         struct osd_thread_info *oti = osd_oti_get(env);
2597
2598         LINVRNT(osd_invariant(obj));
2599
2600         LASSERT(obj->oo_owner == env);
2601         LASSERT(oti->oti_w_locks > 0);
2602         oti->oti_w_locks--;
2603         obj->oo_owner = NULL;
2604         up_write(&obj->oo_sem);
2605 }
2606
2607 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2608 {
2609         struct osd_object *obj = osd_dt_obj(dt);
2610
2611         LINVRNT(osd_invariant(obj));
2612
2613         return obj->oo_owner == env;
2614 }
2615
2616 static void osd_inode_getattr(const struct lu_env *env,
2617                               struct inode *inode, struct lu_attr *attr)
2618 {
2619         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2620                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2621                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2622                            LA_BLKSIZE | LA_TYPE;
2623
2624         attr->la_atime = inode->i_atime.tv_sec;
2625         attr->la_mtime = inode->i_mtime.tv_sec;
2626         attr->la_ctime = inode->i_ctime.tv_sec;
2627         attr->la_mode    = inode->i_mode;
2628         attr->la_size    = i_size_read(inode);
2629         attr->la_blocks  = inode->i_blocks;
2630         attr->la_uid     = i_uid_read(inode);
2631         attr->la_gid     = i_gid_read(inode);
2632         attr->la_projid  = i_projid_read(inode);
2633         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2634         attr->la_nlink   = inode->i_nlink;
2635         attr->la_rdev    = inode->i_rdev;
2636         attr->la_blksize = 1 << inode->i_blkbits;
2637         attr->la_blkbits = inode->i_blkbits;
2638         /*
2639          * Ext4 did not transfer inherit flags from raw inode
2640          * to inode flags, and ext4 internally test raw inode
2641          * @i_flags directly. Instead of patching ext4, we do it here.
2642          */
2643         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2644                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2645 }
2646
2647 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2648                         struct lu_attr *attr)
2649 {
2650         struct osd_object *obj = osd_dt_obj(dt);
2651
2652         if (unlikely(!dt_object_exists(dt)))
2653                 return -ENOENT;
2654         if (unlikely(obj->oo_destroyed))
2655                 return -ENOENT;
2656
2657         LASSERT(!dt_object_remote(dt));
2658         LINVRNT(osd_invariant(obj));
2659
2660         spin_lock(&obj->oo_guard);
2661         osd_inode_getattr(env, obj->oo_inode, attr);
2662         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2663                 attr->la_valid |= LA_FLAGS;
2664                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2665         }
2666         spin_unlock(&obj->oo_guard);
2667
2668         return 0;
2669 }
2670
2671 static int osd_declare_attr_qid(const struct lu_env *env,
2672                                 struct osd_object *obj,
2673                                 struct osd_thandle *oh, long long bspace,
2674                                 qid_t old_id, qid_t new_id, bool enforce,
2675                                 unsigned int type, bool ignore_edquot)
2676 {
2677         int rc;
2678         struct osd_thread_info *info = osd_oti_get(env);
2679         struct lquota_id_info  *qi = &info->oti_qi;
2680
2681         qi->lqi_type = type;
2682         /* inode accounting */
2683         qi->lqi_is_blk = false;
2684
2685         /* one more inode for the new id ... */
2686         qi->lqi_id.qid_uid = new_id;
2687         qi->lqi_space      = 1;
2688         /* Reserve credits for the new id */
2689         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2690         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2691                 rc = 0;
2692         if (rc)
2693                 RETURN(rc);
2694
2695         /* and one less inode for the current id */
2696         qi->lqi_id.qid_uid = old_id;
2697         qi->lqi_space = -1;
2698         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2699         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2700                 rc = 0;
2701         if (rc)
2702                 RETURN(rc);
2703
2704         /* block accounting */
2705         qi->lqi_is_blk = true;
2706
2707         /* more blocks for the new id ... */
2708         qi->lqi_id.qid_uid = new_id;
2709         qi->lqi_space      = bspace;
2710         /*
2711          * Credits for the new uid has been reserved, re-use "obj"
2712          * to save credit reservation.
2713          */
2714         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2715         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2716                 rc = 0;
2717         if (rc)
2718                 RETURN(rc);
2719
2720         /* and finally less blocks for the current uid */
2721         qi->lqi_id.qid_uid = old_id;
2722         qi->lqi_space      = -bspace;
2723         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2724         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2725                 rc = 0;
2726
2727         RETURN(rc);
2728 }
2729
2730 static int osd_declare_attr_set(const struct lu_env *env,
2731                                 struct dt_object *dt,
2732                                 const struct lu_attr *attr,
2733                                 struct thandle *handle)
2734 {
2735         struct osd_thandle *oh;
2736         struct osd_object *obj;
2737         qid_t uid;
2738         qid_t gid;
2739         long long bspace;
2740         int rc = 0;
2741         bool enforce;
2742
2743         ENTRY;
2744
2745         LASSERT(dt != NULL);
2746         LASSERT(handle != NULL);
2747
2748         obj = osd_dt_obj(dt);
2749         LASSERT(osd_invariant(obj));
2750
2751         oh = container_of0(handle, struct osd_thandle, ot_super);
2752         LASSERT(oh->ot_handle == NULL);
2753
2754         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2755                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2756
2757         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2758                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2759
2760         if (attr == NULL || obj->oo_inode == NULL)
2761                 RETURN(rc);
2762
2763         bspace   = obj->oo_inode->i_blocks << 9;
2764         bspace   = toqb(bspace);
2765
2766         /*
2767          * Changing ownership is always preformed by super user, it should not
2768          * fail with EDQUOT unless required explicitly.
2769          *
2770          * We still need to call the osd_declare_qid() to calculate the journal
2771          * credits for updating quota accounting files and to trigger quota
2772          * space adjustment once the operation is completed.
2773          */
2774         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2775                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2776
2777                 if (!ignore_edquot)
2778                         CDEBUG(D_QUOTA, "%s: enforce quota on UID %u, GID %u"
2779                                "(the quota space is %lld)\n",
2780                                obj->oo_inode->i_sb->s_id, attr->la_uid,
2781                                attr->la_gid, bspace);
2782
2783                 /* USERQUOTA */
2784                 uid = i_uid_read(obj->oo_inode);
2785                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2786                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2787                                           attr->la_uid, enforce, USRQUOTA,
2788                                           true);
2789                 if (rc)
2790                         RETURN(rc);
2791
2792                 gid = i_gid_read(obj->oo_inode);
2793                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2794                        attr->la_uid, gid, attr->la_gid);
2795                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2796                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2797                                           attr->la_gid, enforce, GRPQUOTA,
2798                                           ignore_edquot);
2799                 if (rc)
2800                         RETURN(rc);
2801
2802         }
2803 #ifdef HAVE_PROJECT_QUOTA
2804         if (attr->la_valid & LA_PROJID) {
2805                 __u32 projid = i_projid_read(obj->oo_inode);
2806
2807                 enforce = (attr->la_valid & LA_PROJID) &&
2808                                         (attr->la_projid != projid);
2809                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2810                                           (qid_t)projid, (qid_t)attr->la_projid,
2811                                           enforce, PRJQUOTA, true);
2812                 if (rc)
2813                         RETURN(rc);
2814         }
2815 #endif
2816         RETURN(rc);
2817 }
2818
2819 static int osd_inode_setattr(const struct lu_env *env,
2820                              struct inode *inode, const struct lu_attr *attr)
2821 {
2822         __u64 bits = attr->la_valid;
2823
2824         /* Only allow set size for regular file */
2825         if (!S_ISREG(inode->i_mode))
2826                 bits &= ~(LA_SIZE | LA_BLOCKS);
2827
2828         if (bits == 0)
2829                 return 0;
2830
2831         if (bits & LA_ATIME)
2832                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2833         if (bits & LA_CTIME)
2834                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2835         if (bits & LA_MTIME)
2836                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2837         if (bits & LA_SIZE) {
2838                 spin_lock(&inode->i_lock);
2839                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2840                 i_size_write(inode, attr->la_size);
2841                 spin_unlock(&inode->i_lock);
2842         }
2843
2844         /*
2845          * OSD should not change "i_blocks" which is used by quota.
2846          * "i_blocks" should be changed by ldiskfs only.
2847          */
2848         if (bits & LA_MODE)
2849                 inode->i_mode = (inode->i_mode & S_IFMT) |
2850                                 (attr->la_mode & ~S_IFMT);
2851         if (bits & LA_UID)
2852                 i_uid_write(inode, attr->la_uid);
2853         if (bits & LA_GID)
2854                 i_gid_write(inode, attr->la_gid);
2855         if (bits & LA_PROJID)
2856                 i_projid_write(inode, attr->la_projid);
2857         if (bits & LA_NLINK)
2858                 set_nlink(inode, attr->la_nlink);
2859         if (bits & LA_RDEV)
2860                 inode->i_rdev = attr->la_rdev;
2861
2862         if (bits & LA_FLAGS) {
2863                 /* always keep S_NOCMTIME */
2864                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2865                                  S_NOCMTIME;
2866                 /*
2867                  * Ext4 did not transfer inherit flags from
2868                  * @inode->i_flags to raw inode i_flags when writing
2869                  * flags, we do it explictly here.
2870                  */
2871                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2872                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2873                 else
2874                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2875         }
2876         return 0;
2877 }
2878
2879 #ifdef HAVE_PROJECT_QUOTA
2880 static int osd_transfer_project(struct inode *inode, __u32 projid)
2881 {
2882         struct super_block *sb = inode->i_sb;
2883         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
2884         int err;
2885         kprojid_t kprojid;
2886         struct ldiskfs_iloc iloc;
2887         struct ldiskfs_inode *raw_inode;
2888         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
2889
2890         if (!ldiskfs_has_feature_project(sb)) {
2891                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
2892                         == LDISKFS_DEF_PROJID);
2893                 if (projid != LDISKFS_DEF_PROJID)
2894                         return -EOPNOTSUPP;
2895                 else
2896                         return 0;
2897         }
2898
2899         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
2900                 return -EOPNOTSUPP;
2901
2902         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
2903         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
2904                 return 0;
2905
2906         err = ldiskfs_get_inode_loc(inode, &iloc);
2907         if (err)
2908                 return err;
2909
2910         raw_inode = ldiskfs_raw_inode(&iloc);
2911         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
2912                 err = -EOVERFLOW;
2913                 brelse(iloc.bh);
2914                 return err;
2915         }
2916         brelse(iloc.bh);
2917
2918         dquot_initialize(inode);
2919         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
2920         if (transfer_to[PRJQUOTA]) {
2921                 err = __dquot_transfer(inode, transfer_to);
2922                 dqput(transfer_to[PRJQUOTA]);
2923                 if (err)
2924                         return err;
2925         }
2926
2927         return err;
2928 }
2929 #endif
2930
2931 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2932 {
2933         int rc;
2934
2935         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2936             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2937                 struct iattr iattr;
2938
2939                 CDEBUG(D_QUOTA,
2940                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
2941                        inode->i_ino, i_uid_read(inode), attr->la_uid,
2942                        i_gid_read(inode), attr->la_gid);
2943
2944                 dquot_initialize(inode);
2945                 iattr.ia_valid = 0;
2946                 if (attr->la_valid & LA_UID)
2947                         iattr.ia_valid |= ATTR_UID;
2948                 if (attr->la_valid & LA_GID)
2949                         iattr.ia_valid |= ATTR_GID;
2950                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2951                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2952
2953                 rc = dquot_transfer(inode, &iattr);
2954                 if (rc) {
2955                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2956                                "enforcement enabled on the ldiskfs "
2957                                "filesystem?\n", inode->i_sb->s_id, rc);
2958                         return rc;
2959                 }
2960         }
2961
2962         /* Handle project id transfer here properly */
2963         if (attr->la_valid & LA_PROJID &&
2964             attr->la_projid != i_projid_read(inode)) {
2965 #ifdef HAVE_PROJECT_QUOTA
2966                 rc = osd_transfer_project(inode, attr->la_projid);
2967 #else
2968                 rc = -ENOTSUPP;
2969 #endif
2970                 if (rc) {
2971                         CERROR("%s: quota transfer failed: rc = %d. Is project "
2972                                "enforcement enabled on the ldiskfs "
2973                                "filesystem?\n", inode->i_sb->s_id, rc);
2974                         return rc;
2975                 }
2976         }
2977         return 0;
2978 }
2979
2980 static int osd_attr_set(const struct lu_env *env,
2981                         struct dt_object *dt,
2982                         const struct lu_attr *attr,
2983                         struct thandle *handle)
2984 {
2985         struct osd_object *obj = osd_dt_obj(dt);
2986         struct inode *inode;
2987         int rc;
2988
2989         if (!dt_object_exists(dt))
2990                 return -ENOENT;
2991
2992         LASSERT(handle != NULL);
2993         LASSERT(!dt_object_remote(dt));
2994         LASSERT(osd_invariant(obj));
2995
2996         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2997
2998         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
2999             !osd_obj2dev(obj)->od_is_ost) {
3000                 struct osd_thread_info *oti = osd_oti_get(env);
3001                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3002                 struct lu_fid *fid1 = &oti->oti_fid;
3003                 struct osd_inode_id *id = &oti->oti_id;
3004                 struct iam_path_descr *ipd;
3005                 struct iam_container *bag;
3006                 struct osd_thandle *oh;
3007                 int rc;
3008
3009                 fid_cpu_to_be(fid1, fid0);
3010                 memset(id, 1, sizeof(*id));
3011                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3012                                   fid0)->oi_dir.od_container;
3013                 ipd = osd_idx_ipd_get(env, bag);
3014                 if (unlikely(ipd == NULL))
3015                         RETURN(-ENOMEM);
3016
3017                 oh = container_of0(handle, struct osd_thandle, ot_super);
3018                 rc = iam_update(oh->ot_handle, bag,
3019                                 (const struct iam_key *)fid1,
3020                                 (const struct iam_rec *)id, ipd);
3021                 osd_ipd_put(env, bag, ipd);
3022                 return(rc > 0 ? 0 : rc);
3023         }
3024
3025         inode = obj->oo_inode;
3026
3027         rc = osd_quota_transfer(inode, attr);
3028         if (rc)
3029                 return rc;
3030
3031         spin_lock(&obj->oo_guard);
3032         rc = osd_inode_setattr(env, inode, attr);
3033         spin_unlock(&obj->oo_guard);
3034         if (rc != 0)
3035                 GOTO(out, rc);
3036
3037         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3038
3039         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3040
3041         if (!(attr->la_valid & LA_FLAGS))
3042                 GOTO(out, rc);
3043
3044         /* Let's check if there are extra flags need to be set into LMA */
3045         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3046                 struct osd_thread_info *info = osd_oti_get(env);
3047                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3048
3049                 LASSERT(!obj->oo_pfid_in_lma);
3050
3051                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
3052                                  &info->oti_ost_attrs);
3053                 if (rc)
3054                         GOTO(out, rc);
3055
3056                 lma->lma_incompat |=
3057                         lustre_to_lma_flags(attr->la_flags);
3058                 lustre_lma_swab(lma);
3059
3060                 osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3061
3062                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
3063                                      lma, sizeof(*lma), XATTR_REPLACE);
3064                 if (rc != 0) {
3065                         struct osd_device *osd = osd_obj2dev(obj);
3066
3067                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
3068                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
3069                               lma->lma_incompat, rc);
3070                 } else {
3071                         obj->oo_lma_flags =
3072                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
3073                 }
3074                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3075         }
3076 out:
3077
3078         return rc;
3079 }
3080
3081 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
3082                                            struct osd_object *obj,
3083                                            const char *name, const int namelen)
3084 {
3085         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
3086 }
3087
3088 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
3089                       umode_t mode, struct dt_allocation_hint *hint,
3090                       struct thandle *th, struct lu_attr *attr)
3091 {
3092         int result;
3093         struct osd_device *osd = osd_obj2dev(obj);
3094         struct osd_thandle *oth;
3095         struct dt_object *parent = NULL;
3096         struct inode *inode;
3097         uid_t owner[2] = {0, 0};
3098
3099         if (attr->la_valid & LA_UID)
3100                 owner[0] = attr->la_uid;
3101         if (attr->la_valid & LA_GID)
3102                 owner[1] = attr->la_gid;
3103
3104         LINVRNT(osd_invariant(obj));
3105         LASSERT(obj->oo_inode == NULL);
3106         LASSERT(obj->oo_hl_head == NULL);
3107
3108         if (S_ISDIR(mode) && ldiskfs_pdo) {
3109                 obj->oo_hl_head =
3110                         ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
3111                 if (obj->oo_hl_head == NULL)
3112                         return -ENOMEM;
3113         }
3114
3115         oth = container_of(th, struct osd_thandle, ot_super);
3116         LASSERT(oth->ot_handle->h_transaction != NULL);
3117
3118         if (hint != NULL && hint->dah_parent != NULL &&
3119             !dt_object_remote(hint->dah_parent))
3120                 parent = hint->dah_parent;
3121
3122         inode = ldiskfs_create_inode(oth->ot_handle,
3123                                      parent ? osd_dt_obj(parent)->oo_inode :
3124                                               osd_sb(osd)->s_root->d_inode,
3125                                      mode, owner);
3126         if (!IS_ERR(inode)) {
3127                 /* Do not update file c/mtime in ldiskfs. */
3128                 inode->i_flags |= S_NOCMTIME;
3129
3130                 /*
3131                  * For new created object, it must be consistent,
3132                  * and it is unnecessary to scrub against it.
3133                  */
3134                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
3135
3136                 obj->oo_inode = inode;
3137                 result = 0;
3138         } else {
3139                 if (obj->oo_hl_head != NULL) {
3140                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
3141                         obj->oo_hl_head = NULL;
3142                 }
3143                 result = PTR_ERR(inode);
3144         }
3145         LINVRNT(osd_invariant(obj));
3146         return result;
3147 }
3148
3149 enum {
3150         OSD_NAME_LEN = 255
3151 };
3152
3153 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
3154                      struct lu_attr *attr,
3155                      struct dt_allocation_hint *hint,
3156                      struct dt_object_format *dof,
3157                      struct thandle *th)
3158 {
3159         int result;
3160         struct osd_thandle *oth;
3161         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
3162
3163         LASSERT(S_ISDIR(attr->la_mode));
3164
3165         oth = container_of(th, struct osd_thandle, ot_super);
3166         LASSERT(oth->ot_handle->h_transaction != NULL);
3167         result = osd_mkfile(info, obj, mode, hint, th, attr);
3168
3169         return result;
3170 }
3171
3172 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
3173                         struct lu_attr *attr,
3174                         struct dt_allocation_hint *hint,
3175                         struct dt_object_format *dof,
3176                         struct thandle *th)
3177 {
3178         int result;
3179         struct osd_thandle *oth;
3180         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
3181
3182         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
3183
3184         LASSERT(S_ISREG(attr->la_mode));
3185
3186         oth = container_of(th, struct osd_thandle, ot_super);
3187         LASSERT(oth->ot_handle->h_transaction != NULL);
3188
3189         result = osd_mkfile(info, obj, mode, hint, th, attr);
3190         if (result == 0) {
3191                 LASSERT(obj->oo_inode != NULL);
3192                 if (feat->dif_flags & DT_IND_VARKEY)
3193                         result = iam_lvar_create(obj->oo_inode,
3194                                                  feat->dif_keysize_max,
3195                                                  feat->dif_ptrsize,
3196                                                  feat->dif_recsize_max,
3197                                                  oth->ot_handle);
3198                 else
3199                         result = iam_lfix_create(obj->oo_inode,
3200                                                  feat->dif_keysize_max,
3201                                                  feat->dif_ptrsize,
3202                                                  feat->dif_recsize_max,
3203                                                  oth->ot_handle);
3204         }
3205         return result;
3206 }
3207
3208 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
3209                      struct lu_attr *attr,
3210                      struct dt_allocation_hint *hint,
3211                      struct dt_object_format *dof,
3212                      struct thandle *th)
3213 {
3214         LASSERT(S_ISREG(attr->la_mode));
3215         return osd_mkfile(info, obj, (attr->la_mode &
3216                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3217                           attr);
3218 }
3219
3220 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
3221                      struct lu_attr *attr,
3222                      struct dt_allocation_hint *hint,
3223                      struct dt_object_format *dof,
3224                      struct thandle *th)
3225 {
3226         LASSERT(S_ISLNK(attr->la_mode));
3227         return osd_mkfile(info, obj, (attr->la_mode &
3228                          (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th,
3229                           attr);
3230 }
3231
3232 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
3233                      struct lu_attr *attr,
3234                      struct dt_allocation_hint *hint,
3235                      struct dt_object_format *dof,
3236                      struct thandle *th)
3237 {
3238         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
3239         int result;
3240
3241         LINVRNT(osd_invariant(obj));
3242         LASSERT(obj->oo_inode == NULL);
3243         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
3244                 S_ISFIFO(mode) || S_ISSOCK(mode));
3245
3246         result = osd_mkfile(info, obj, mode, hint, th, attr);
3247         if (result == 0) {
3248                 LASSERT(obj->oo_inode != NULL);
3249                 /*
3250                  * This inode should be marked dirty for i_rdev.  Currently
3251                  * that is done in the osd_attr_init().
3252                  */
3253                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
3254                                    attr->la_rdev);
3255         }
3256         LINVRNT(osd_invariant(obj));
3257         return result;
3258 }
3259
3260 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
3261                               struct lu_attr *,
3262                               struct dt_allocation_hint *hint,
3263                               struct dt_object_format *dof,
3264                               struct thandle *);
3265
3266 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
3267 {
3268         osd_obj_type_f result;
3269
3270         switch (type) {
3271         case DFT_DIR:
3272                 result = osd_mkdir;
3273                 break;
3274         case DFT_REGULAR:
3275                 result = osd_mkreg;
3276                 break;
3277         case DFT_SYM:
3278                 result = osd_mksym;
3279                 break;
3280         case DFT_NODE:
3281                 result = osd_mknod;
3282                 break;
3283         case DFT_INDEX:
3284                 result = osd_mk_index;
3285                 break;
3286
3287         default:
3288                 LBUG();
3289                 break;
3290         }
3291         return result;
3292 }
3293
3294
3295 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
3296                         struct dt_object *parent, struct dt_object *child,
3297                         umode_t child_mode)
3298 {
3299         LASSERT(ah);
3300
3301         ah->dah_parent = parent;
3302         ah->dah_mode = child_mode;
3303
3304         if (parent != NULL && !dt_object_remote(parent)) {
3305                 /* will help to find FID->ino at dt_insert("..") */
3306                 struct osd_object *pobj = osd_dt_obj(parent);
3307
3308                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
3309         }
3310 }
3311
3312 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
3313                           struct lu_attr *attr, struct dt_object_format *dof)
3314 {
3315         struct inode *inode = obj->oo_inode;
3316         __u64 valid = attr->la_valid;
3317         int result;
3318
3319         attr->la_valid &= ~(LA_TYPE | LA_MODE);
3320
3321         if (dof->dof_type != DFT_NODE)
3322                 attr->la_valid &= ~LA_RDEV;
3323         if ((valid & LA_ATIME) && (attr->la_atime == inode->i_atime.tv_sec))
3324                 attr->la_valid &= ~LA_ATIME;
3325         if ((valid & LA_CTIME) && (attr->la_ctime == inode->i_ctime.tv_sec))
3326                 attr->la_valid &= ~LA_CTIME;
3327         if ((valid & LA_MTIME) && (attr->la_mtime == inode->i_mtime.tv_sec))
3328                 attr->la_valid &= ~LA_MTIME;
3329
3330         result = osd_quota_transfer(inode, attr);
3331         if (result)
3332                 return;
3333
3334         if (attr->la_valid != 0) {
3335                 result = osd_inode_setattr(info->oti_env, inode, attr);
3336                 /*
3337                  * The osd_inode_setattr() should always succeed here.  The
3338                  * only error that could be returned is EDQUOT when we are
3339                  * trying to change the UID or GID of the inode. However, this
3340                  * should not happen since quota enforcement is no longer
3341                  * enabled on ldiskfs (lquota takes care of it).
3342                  */
3343                 LASSERTF(result == 0, "%d\n", result);
3344                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3345         }
3346
3347         attr->la_valid = valid;
3348 }
3349
3350 /**
3351  * Helper function for osd_create()
3352  *
3353  * \retval 0, on success
3354  */
3355 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3356                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3357                         struct dt_object_format *dof, struct thandle *th)
3358 {
3359         int result;
3360         __u32 umask;
3361
3362         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3363
3364         /* we drop umask so that permissions we pass are not affected */
3365         umask = current->fs->umask;
3366         current->fs->umask = 0;
3367
3368         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3369                                                   th);
3370         if (likely(obj->oo_inode != NULL)) {
3371                 LASSERT(obj->oo_inode->i_state & I_NEW);
3372
3373                 /*
3374                  * Unlock the inode before attr initialization to avoid
3375                  * unnecessary dqget operations. LU-6378
3376                  */
3377                 unlock_new_inode(obj->oo_inode);
3378         }
3379
3380         if (likely(result == 0)) {
3381                 osd_attr_init(info, obj, attr, dof);
3382                 osd_object_init0(obj);
3383         }
3384
3385         /* restore previous umask value */
3386         current->fs->umask = umask;
3387
3388         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3389
3390         return result;
3391 }
3392
3393 /**
3394  * Helper function for osd_create()
3395  *
3396  * \retval 0, on success
3397  */
3398 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3399                            const struct lu_fid *fid, struct thandle *th)
3400 {
3401         struct osd_thread_info *info = osd_oti_get(env);
3402         struct osd_inode_id    *id   = &info->oti_id;
3403         struct osd_device      *osd  = osd_obj2dev(obj);
3404         struct osd_thandle     *oh;
3405         int rc;
3406
3407         LASSERT(obj->oo_inode != NULL);
3408
3409         oh = container_of0(th, struct osd_thandle, ot_super);
3410         LASSERT(oh->ot_handle);
3411         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3412
3413         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3414         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3415                            OI_CHECK_FLD, NULL);
3416         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
3417                 struct lu_fid next_fid = *fid;
3418
3419                 /* insert next object in advance, and map to the same inode */
3420                 next_fid.f_oid++;
3421                 if (next_fid.f_oid != 0) {
3422                         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3423                         osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
3424                                       OI_CHECK_FLD, NULL);
3425                         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3426                 }
3427         }
3428
3429         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3430
3431         return rc;
3432 }
3433
3434 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3435                    u64 seq, struct lu_seq_range *range)
3436 {
3437         struct seq_server_site *ss = osd_seq_site(osd);
3438
3439         if (fid_seq_is_idif(seq)) {
3440                 fld_range_set_ost(range);
3441                 range->lsr_index = idif_ost_idx(seq);
3442                 return 0;
3443         }
3444
3445         if (!fid_seq_in_fldb(seq)) {
3446                 fld_range_set_mdt(range);
3447                 if (ss != NULL)
3448                         /*
3449                          * FIXME: If ss is NULL, it suppose not get lsr_index
3450                          * at all
3451                          */
3452                         range->lsr_index = ss->ss_node_id;
3453                 return 0;
3454         }
3455
3456         LASSERT(ss != NULL);
3457         fld_range_set_any(range);
3458         /* OSD will only do local fld lookup */
3459         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3460 }
3461
3462 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3463                               struct lu_attr *attr,
3464                               struct dt_allocation_hint *hint,
3465                               struct dt_object_format *dof,
3466                               struct thandle *handle)
3467 {
3468         struct osd_thandle *oh;
3469         int rc;
3470
3471         ENTRY;
3472
3473         LASSERT(handle != NULL);
3474
3475         oh = container_of0(handle, struct osd_thandle, ot_super);
3476         LASSERT(oh->ot_handle == NULL);
3477
3478         /*
3479          * EA object consumes more credits than regular object: osd_mk_index
3480          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3481          * leaf_node, could involves the block, block bitmap, groups, GDT
3482          * change for each block, so add 4 * 2 credits in that case.
3483          */
3484         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3485                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3486                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3487         /*
3488          * Reuse idle OI block may cause additional one OI block
3489          * to be changed.
3490          */
3491         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3492                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3493         if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
3494                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3495                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3496
3497         /* will help to find FID->ino mapping at dt_insert() */
3498         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3499                                    osd_dt_obj(dt));
3500         if (rc != 0)
3501                 RETURN(rc);
3502
3503         if (!attr)
3504                 RETURN(0);
3505
3506         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3507                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3508                                    NULL, OSD_QID_INODE);
3509         if (rc != 0)
3510                 RETURN(rc);
3511
3512         RETURN(rc);
3513 }
3514
3515 /**
3516  * Called to destroy on-disk representation of the object
3517  *
3518  * Concurrency: must be locked
3519  */
3520 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3521                                struct thandle *th)
3522 {
3523         struct osd_object *obj = osd_dt_obj(dt);
3524         struct inode *inode = obj->oo_inode;
3525         struct osd_thandle *oh;
3526         int rc;
3527
3528         ENTRY;
3529
3530         if (inode == NULL)
3531                 RETURN(-ENOENT);
3532
3533         oh = container_of0(th, struct osd_thandle, ot_super);
3534         LASSERT(oh->ot_handle == NULL);
3535
3536         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3537                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3538
3539         /* For removing agent entry */
3540         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu))
3541                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
3542
3543         /*
3544          * Recycle idle OI leaf may cause additional three OI blocks
3545          * to be changed.
3546          */
3547         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3548                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3549                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3550         /* one less inode */
3551         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3552                                    i_projid_read(inode), -1, oh, obj, NULL,
3553                                    OSD_QID_INODE);
3554         if (rc)
3555                 RETURN(rc);
3556         /* data to be truncated */
3557         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3558                                    i_projid_read(inode), 0, oh, obj, NULL,
3559                                    OSD_QID_BLK);
3560         if (rc)
3561                 RETURN(rc);
3562
3563         /*
3564          * will help to find FID->ino when this object is being
3565          * added to PENDING
3566          */
3567         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3568
3569         RETURN(rc);
3570 }
3571
3572 static int osd_destroy(const struct lu_env *env, struct dt_object *dt,
3573                        struct thandle *th)
3574 {
3575         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3576         struct osd_object *obj = osd_dt_obj(dt);
3577         struct inode *inode = obj->oo_inode;
3578         struct osd_device *osd = osd_obj2dev(obj);
3579         struct osd_thandle *oh;
3580         int result;
3581
3582         ENTRY;
3583
3584         oh = container_of0(th, struct osd_thandle, ot_super);
3585         LASSERT(oh->ot_handle);
3586         LASSERT(inode);
3587         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
3588
3589         if (unlikely(fid_is_acct(fid)))
3590                 RETURN(-EPERM);
3591
3592         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu)) {
3593                 result = osd_delete_from_remote_parent(env, osd, obj, oh, true);
3594                 if (result != 0)
3595                         CERROR("%s: remove agent entry "DFID": rc = %d\n",
3596                                osd_name(osd), PFID(fid), result);
3597         }
3598
3599         if (S_ISDIR(inode->i_mode)) {
3600                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
3601                         inode->i_nlink == 2);
3602
3603                 spin_lock(&obj->oo_guard);
3604                 clear_nlink(inode);
3605                 spin_unlock(&obj->oo_guard);
3606                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3607         }
3608
3609         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
3610
3611         ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
3612
3613         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3614                 result = osd_oi_delete(osd_oti_get(env), osd, fid,
3615                                        oh->ot_handle, OI_CHECK_FLD);
3616
3617         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
3618         /* XXX: add to ext3 orphan list */
3619         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
3620
3621         /* not needed in the cache anymore */
3622         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
3623         obj->oo_destroyed = 1;
3624
3625         RETURN(0);
3626 }
3627
3628 /**
3629  * Put the fid into lustre_mdt_attrs, and then place the structure
3630  * inode's ea. This fid should not be altered during the life time
3631  * of the inode.
3632  *
3633  * \retval +ve, on success
3634  * \retval -ve, on error
3635  *
3636  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
3637  */
3638 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
3639                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
3640 {
3641         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
3642         struct lustre_mdt_attrs *lma = &loa->loa_lma;
3643         int rc;
3644
3645         ENTRY;
3646
3647         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
3648                 RETURN(0);
3649
3650         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
3651                 rc = -ENOMEM;
3652
3653         lustre_loa_init(loa, fid, compat, incompat);
3654         lustre_loa_swab(loa, false);
3655
3656         /*
3657          * For the OST device with 256 bytes inode size by default,
3658          * the PFID EA will be stored together with LMA EA to avoid
3659          * performance trouble. Otherwise the PFID EA can be stored
3660          * independently. LU-8998
3661          */
3662         if ((compat & LMAC_FID_ON_OST) &&
3663             LDISKFS_INODE_SIZE(inode->i_sb) <= 256)
3664                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3665                                      sizeof(*loa), XATTR_CREATE);
3666         else
3667                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3668                                      sizeof(*lma), XATTR_CREATE);
3669         /*
3670          * LMA may already exist, but we need to check that all the
3671          * desired compat/incompat flags have been added.
3672          */
3673         if (unlikely(rc == -EEXIST)) {
3674                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
3675                                      XATTR_NAME_LMA, (void *)loa, sizeof(*loa));
3676                 if (rc < 0)
3677                         RETURN(rc);
3678
3679                 if (rc < sizeof(*lma))
3680                         RETURN(-EINVAL);
3681
3682                 lustre_loa_swab(loa, true);
3683                 if (lu_fid_eq(fid, &lma->lma_self_fid) &&
3684                     ((compat == 0 && incompat == 0) ||
3685                      (!(~lma->lma_compat & compat) &&
3686                       !(~lma->lma_incompat & incompat))))
3687                         RETURN(0);
3688
3689                 lma->lma_self_fid = *fid;
3690                 lma->lma_compat |= compat;
3691                 lma->lma_incompat |= incompat;
3692                 if (rc == sizeof(*lma)) {
3693                         lustre_lma_swab(lma);
3694                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3695                                              sizeof(*lma), XATTR_REPLACE);
3696                 } else {
3697                         lustre_loa_swab(loa, false);
3698                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa,
3699                                              sizeof(*loa), XATTR_REPLACE);
3700                 }
3701         }
3702
3703         RETURN(rc);
3704 }
3705
3706 /**
3707  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
3708  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
3709  * To have compatilibility with 1.8 ldiskfs driver we need to have
3710  * magic number at start of fid data.
3711  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
3712  * its inmemory API.
3713  */
3714 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
3715                                          const struct lu_fid *fid)
3716 {
3717         if (!fid_is_namespace_visible(fid) ||
3718             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
3719                 param->edp_magic = 0;
3720                 return;
3721         }
3722
3723         param->edp_magic = LDISKFS_LUFID_MAGIC;
3724         param->edp_len =  sizeof(struct lu_fid) + 1;
3725         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
3726 }
3727
3728 /**
3729  * Try to read the fid from inode ea into dt_rec.
3730  *
3731  * \param fid object fid.
3732  *
3733  * \retval 0 on success
3734  */
3735 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
3736                           __u32 ino, struct lu_fid *fid,
3737                           struct osd_inode_id *id)
3738 {
3739         struct osd_thread_info *info  = osd_oti_get(env);
3740         struct inode *inode;
3741
3742         ENTRY;
3743
3744         osd_id_gen(id, ino, OSD_OII_NOGEN);
3745         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
3746         if (IS_ERR(inode))
3747                 RETURN(PTR_ERR(inode));
3748
3749         iput(inode);
3750         RETURN(0);
3751 }
3752
3753 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
3754                                         struct inode *dir,
3755                                         struct inode *parent_dir,
3756                                         const struct lu_fid *dot_fid,
3757                                         const struct lu_fid *dot_dot_fid,
3758                                         struct osd_thandle *oth)
3759 {
3760         struct ldiskfs_dentry_param *dot_ldp;
3761         struct ldiskfs_dentry_param *dot_dot_ldp;
3762         __u32 saved_nlink = dir->i_nlink;
3763         int rc;
3764
3765         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3766         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3767
3768         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3769         dot_ldp->edp_magic = 0;
3770
3771         rc = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3772                                     dir, dot_ldp, dot_dot_ldp);
3773         /*
3774          * The ldiskfs_add_dot_dotdot() may dir->i_nlink as 2, then
3775          * the subseqent ref_add() will increase the dir->i_nlink
3776          * as 3. That is incorrect for new created directory.
3777          *
3778          * It looks like hack, because we want to make the OSD API
3779          * to be order-independent for new created directory object
3780          * between dt_insert(..) and ref_add() operations.
3781          *
3782          * Here, we only restore the in-RAM dir-inode's nlink attr,
3783          * becuase if the nlink attr is not 2, then there will be
3784          * ref_add() called following the dt_insert(..), such call
3785          * will make both the in-RAM and on-disk dir-inode's nlink
3786          * attr to be set as 2. LU-7447
3787          */
3788         set_nlink(dir, saved_nlink);
3789         return rc;
3790 }
3791
3792 /**
3793  * Create an local agent inode for remote entry
3794  */
3795 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
3796                                                   struct osd_device *osd,
3797                                                   struct osd_object *pobj,
3798                                                   const struct lu_fid *fid,
3799                                                   __u32 type,
3800                                                   struct thandle *th)
3801 {
3802         struct osd_thread_info *info = osd_oti_get(env);
3803         struct inode *local;
3804         struct osd_thandle *oh;
3805         int rc;
3806
3807         ENTRY;
3808
3809         LASSERT(th);
3810         oh = container_of(th, struct osd_thandle, ot_super);
3811         LASSERT(oh->ot_handle->h_transaction != NULL);
3812
3813         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type,
3814                                      NULL);
3815         if (IS_ERR(local)) {
3816                 CERROR("%s: create local error %d\n", osd_name(osd),
3817                        (int)PTR_ERR(local));
3818                 RETURN(local);
3819         }
3820
3821         /*
3822          * restore i_gid in case S_ISGID is set, we will inherit S_ISGID and set
3823          * correct gid on remote file, not agent here
3824          */
3825         local->i_gid = current_fsgid();
3826         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
3827
3828         /* e2fsck doesn't like empty symlinks.  Store remote FID as symlink.
3829          * That gives e2fsck something to look at and be happy, and allows
3830          * debugging if we need to determine where this symlink came from.
3831          */
3832         if (S_ISLNK(type)) {
3833                 BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
3834                 rc = snprintf((char *)LDISKFS_I(local)->i_data,
3835                               LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
3836
3837                 i_size_write(local, rc);
3838                 LDISKFS_I(local)->i_disksize = rc;
3839         }
3840         unlock_new_inode(local);
3841
3842         /* Agent inode should not have project ID */
3843 #ifdef  HAVE_PROJECT_QUOTA
3844         if (LDISKFS_I(pobj->oo_inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
3845             i_projid_read(pobj->oo_inode) != 0) {
3846                 rc = osd_transfer_project(local, 0);
3847                 if (rc) {
3848                         CERROR("%s: quota transfer failed: rc = %d. Is project "
3849                                "quota enforcement enabled on the ldiskfs "
3850                                "filesystem?\n", local->i_sb->s_id, rc);
3851                         RETURN(ERR_PTR(rc));
3852                 }
3853         }
3854 #endif
3855         /* Set special LMA flag for local agent inode */
3856         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
3857         if (rc != 0) {
3858                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
3859                        osd_name(osd), PFID(fid), rc);
3860                 RETURN(ERR_PTR(rc));
3861         }
3862
3863         if (!S_ISDIR(type))
3864                 RETURN(local);
3865
3866         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
3867                                          lu_object_fid(&pobj->oo_dt.do_lu),
3868                                          fid, oh);
3869         if (rc != 0) {
3870                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
3871                         osd_name(osd), PFID(fid), rc);
3872                 RETURN(ERR_PTR(rc));
3873         }
3874
3875         RETURN(local);
3876 }
3877
3878 /**
3879  * when direntry is deleted, we have to take care of possible agent inode
3880  * referenced by that. unfortunately we can't do this at that point:
3881  * iget() within a running transaction leads to deadlock and we better do
3882  * not call that every delete declaration to save performance. so we put
3883  * a potention agent inode on a list and process that once the transaction
3884  * is over. Notice it's not any worse in terms of real orphans as regular
3885  * object destroy doesn't put inodes on the on-disk orphan list. this should
3886  * be addressed separately
3887  */
3888 static int osd_schedule_agent_inode_removal(const struct lu_env *env,
3889                                             struct osd_thandle *oh,
3890                                             __u32 ino)
3891 {
3892         struct osd_device *osd = osd_dt_dev(oh->ot_super.th_dev);
3893         struct osd_obj_orphan *oor;
3894
3895         OBD_ALLOC_PTR(oor);
3896         if (oor == NULL)
3897                 return -ENOMEM;
3898
3899         oor->oor_ino = ino;
3900         oor->oor_env = (struct lu_env *)env;
3901         spin_lock(&osd->od_osfs_lock);
3902         list_add(&oor->oor_list, &osd->od_orphan_list);
3903         spin_unlock(&osd->od_osfs_lock);
3904
3905         oh->ot_remove_agents = 1;
3906
3907         return 0;
3908
3909 }
3910
3911 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
3912                                                 struct osd_device *osd)
3913 {
3914         struct osd_thread_info *info = osd_oti_get(env);
3915         struct osd_obj_orphan *oor, *tmp;
3916         struct osd_inode_id id;
3917         LIST_HEAD(list);
3918         struct inode *inode;
3919         struct lu_fid fid;
3920         handle_t *jh;
3921         __u32 ino;
3922
3923         spin_lock(&osd->od_osfs_lock);
3924         list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
3925                 if (oor->oor_env == env)
3926                         list_move(&oor->oor_list, &list);
3927         }
3928         spin_unlock(&osd->od_osfs_lock);
3929
3930         list_for_each_entry_safe(oor, tmp, &list, oor_list) {
3931
3932                 ino = oor->oor_ino;
3933
3934                 list_del(&oor->oor_list);
3935                 OBD_FREE_PTR(oor);
3936
3937                 osd_id_gen(&id, ino, OSD_OII_NOGEN);
3938                 inode = osd_iget_fid(info, osd, &id, &fid);
3939                 if (IS_ERR(inode))
3940                         continue;
3941
3942                 if (!osd_remote_fid(env, osd, &fid)) {
3943                         iput(inode);
3944                         continue;
3945                 }
3946
3947                 jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC, 1);
3948                 clear_nlink(inode);
3949                 mark_inode_dirty(inode);
3950                 ldiskfs_journal_stop(jh);
3951                 iput(inode);
3952         }
3953
3954         return 0;
3955 }
3956
3957 /**
3958  * OSD layer object create function for OST objects (b=11826).
3959  *
3960  * The FID is inserted into inode xattr here.
3961  *
3962  * \retval   0, on success
3963  * \retval -ve, on error
3964  */
3965 static int osd_create(const struct lu_env *env, struct dt_object *dt,
3966                       struct lu_attr *attr, struct dt_allocation_hint *hint,
3967                       struct dt_object_format *dof, struct thandle *th)
3968 {
3969         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3970         struct osd_object *obj = osd_dt_obj(dt);
3971         struct osd_thread_info *info = osd_oti_get(env);
3972         int result, on_ost = 0;
3973
3974         ENTRY;
3975
3976         if (dt_object_exists(dt))
3977                 RETURN(-EEXIST);
3978
3979         LINVRNT(osd_invariant(obj));
3980         LASSERT(!dt_object_remote(dt));
3981         LASSERT(osd_is_write_locked(env, obj));
3982         LASSERT(th != NULL);
3983
3984         if (unlikely(fid_is_acct(fid)))
3985                 /*
3986                  * Quota files can't be created from the kernel any more,
3987                  * 'tune2fs -O quota' will take care of creating them
3988                  */
3989                 RETURN(-EPERM);
3990
3991         result = __osd_create(info, obj, attr, hint, dof, th);
3992         if (result == 0) {
3993                 if (fid_is_idif(fid) &&
3994                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
3995                         struct lu_fid *tfid = &info->oti_fid;
3996                         struct ost_id *oi   = &info->oti_ostid;
3997
3998                         fid_to_ostid(fid, oi);
3999                         ostid_to_fid(tfid, oi, 0);
4000                         on_ost = 1;
4001                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
4002                                                 LMAC_FID_ON_OST, 0);
4003                 } else {
4004                         on_ost = fid_is_on_ost(info, osd_obj2dev(obj),
4005                                                fid, OI_CHECK_FLD);
4006                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
4007                                                 on_ost ? LMAC_FID_ON_OST : 0,
4008                                                 0);
4009                 }
4010                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
4011                         obj->oo_dt.do_body_ops = &osd_body_ops;
4012         }
4013
4014         if (!result && !CFS_FAIL_CHECK(OBD_FAIL_OSD_NO_OI_ENTRY))
4015                 result = __osd_oi_insert(env, obj, fid, th);
4016
4017         /*
4018          * a small optimization - dt_insert() isn't usually applied
4019          * to OST objects, so we don't need to cache OI mapping for
4020          * OST objects
4021          */
4022         if (result == 0 && on_ost == 0) {
4023                 struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4024
4025                 result = osd_idc_find_and_init(env, osd, obj);
4026                 LASSERT(result == 0);
4027         }
4028
4029         LASSERT(ergo(result == 0,
4030                      dt_object_exists(dt) && !dt_object_remote(dt)));
4031         LINVRNT(osd_invariant(obj));
4032         RETURN(result);
4033 }
4034
4035 static int osd_declare_ref_add(const struct lu_env *env, struct dt_object *dt,
4036                                struct thandle *handle)
4037 {
4038         struct osd_thandle *oh;
4039
4040         /* it's possible that object doesn't exist yet */
4041         LASSERT(handle != NULL);
4042
4043         oh = container_of0(handle, struct osd_thandle, ot_super);
4044         LASSERT(oh->ot_handle == NULL);
4045
4046         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
4047                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4048
4049         osd_idc_find_and_init(env, osd_dev(dt->do_lu.lo_dev), osd_dt_obj(dt));
4050
4051         return 0;
4052 }
4053
4054 /*
4055  * Concurrency: @dt is write locked.
4056  */
4057 static int osd_ref_add(const struct lu_env *env, struct dt_object *dt,
4058                        struct thandle *th)
4059 {
4060         struct osd_object *obj = osd_dt_obj(dt);
4061         struct inode *inode = obj->oo_inode;
4062         struct osd_thandle *oh;
4063         int rc = 0;
4064
4065         if (!dt_object_exists(dt) || obj->oo_destroyed)
4066                 return -ENOENT;
4067
4068         LINVRNT(osd_invariant(obj));
4069         LASSERT(!dt_object_remote(dt));
4070         LASSERT(osd_is_write_locked(env, obj));
4071         LASSERT(th != NULL);
4072
4073         oh = container_of0(th, struct osd_thandle, ot_super);
4074         LASSERT(oh->ot_handle != NULL);
4075
4076         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
4077
4078         CDEBUG(D_INODE, DFID" increase nlink %d\n",
4079                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4080         /*
4081          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
4082          * (65000) subdirectories by storing "1" in i_nlink if the link count
4083          * would otherwise overflow. Directory tranversal tools understand
4084          * that (st_nlink == 1) indicates that the filesystem dose not track
4085          * hard links count on the directory, and will not abort subdirectory
4086          * scanning early once (st_nlink - 2) subdirs have been found.
4087          *
4088          * This also has to properly handle the case of inodes with nlink == 0
4089          * in case they are being linked into the PENDING directory
4090          */
4091         spin_lock(&obj->oo_guard);
4092         if (unlikely(inode->i_nlink == 0))
4093                 /* inc_nlink from 0 may cause WARN_ON */
4094                 set_nlink(inode, 1);
4095         else {
4096                 ldiskfs_inc_count(oh->ot_handle, inode);
4097                 if (!S_ISDIR(inode->i_mode))
4098                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
4099         }
4100         spin_unlock(&obj->oo_guard);
4101
4102         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4103         LINVRNT(osd_invariant(obj));
4104
4105         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
4106
4107         return rc;
4108 }
4109
4110 static int osd_declare_ref_del(const struct lu_env *env, struct dt_object *dt,
4111                                struct thandle *handle)
4112 {
4113         struct osd_thandle *oh;
4114
4115         if (!dt_object_exists(dt))
4116                 return -ENOENT;
4117
4118         LASSERT(!dt_object_remote(dt));
4119         LASSERT(handle != NULL);
4120
4121         oh = container_of0(handle, struct osd_thandle, ot_super);
4122         LASSERT(oh->ot_handle == NULL);
4123
4124         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
4125                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
4126
4127         return 0;
4128 }
4129
4130 /*
4131  * Concurrency: @dt is write locked.
4132  */
4133 static int osd_ref_del(const struct lu_env *env, struct dt_object *dt,
4134                        struct thandle *th)
4135 {
4136         struct osd_object *obj = osd_dt_obj(dt);
4137         struct inode *inode = obj->oo_inode;
4138         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
4139         struct osd_thandle *oh;
4140
4141         if (!dt_object_exists(dt))
4142                 return -ENOENT;
4143
4144         LINVRNT(osd_invariant(obj));
4145         LASSERT(!dt_object_remote(dt));
4146         LASSERT(osd_is_write_locked(env, obj));
4147         LASSERT(th != NULL);
4148
4149         oh = container_of0(th, struct osd_thandle, ot_super);
4150         LASSERT(oh->ot_handle != NULL);
4151
4152         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
4153
4154         spin_lock(&obj->oo_guard);
4155         /*
4156          * That can be result of upgrade from old Lustre version and
4157          * applied only to local files.  Just skip this ref_del call.
4158          * ext4_unlink() only treats this as a warning, don't LASSERT here.
4159          */
4160         if (inode->i_nlink == 0) {
4161                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
4162                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
4163                              ", maybe an upgraded file? (LU-3915)\n",
4164                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
4165                 spin_unlock(&obj->oo_guard);
4166                 return 0;
4167         }
4168
4169         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
4170                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
4171
4172         ldiskfs_dec_count(oh->ot_handle, inode);
4173         spin_unlock(&obj->oo_guard);
4174
4175         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4176         LINVRNT(osd_invariant(obj));
4177
4178         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
4179
4180         return 0;
4181 }
4182
4183 /*
4184  * Concurrency: @dt is read locked.
4185  */
4186 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
4187                          struct lu_buf *buf, const char *name)
4188 {
4189         struct osd_object      *obj    = osd_dt_obj(dt);
4190         struct inode           *inode  = obj->oo_inode;
4191         struct osd_thread_info *info   = osd_oti_get(env);
4192         struct dentry          *dentry = &info->oti_obj_dentry;
4193         bool cache_xattr = false;
4194         int rc;
4195
4196         LASSERT(buf);
4197
4198         /* version get is not real XATTR but uses xattr API */
4199         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4200                 dt_obj_version_t *ver = buf->lb_buf;
4201
4202                 /*
4203                  * for version we are just using xattr API but change inode
4204                  * field instead
4205                  */
4206                 if (buf->lb_len == 0)
4207                         return sizeof(dt_obj_version_t);
4208
4209                 if (buf->lb_len < sizeof(dt_obj_version_t))
4210                         return -ERANGE;
4211
4212                 CDEBUG(D_INODE, "Get version %#llx for inode %lu\n",
4213                        LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4214
4215                 *ver = LDISKFS_I(inode)->i_fs_version;
4216
4217                 return sizeof(dt_obj_version_t);
4218         }
4219
4220         if (!dt_object_exists(dt))
4221                 return -ENOENT;
4222
4223         LASSERT(!dt_object_remote(dt));
4224         LASSERT(inode->i_op != NULL);
4225 #ifdef HAVE_IOP_XATTR
4226         LASSERT(inode->i_op->getxattr != NULL);
4227 #endif
4228
4229         if (strcmp(name, XATTR_NAME_LOV) == 0 ||
4230             strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0)
4231                 cache_xattr = true;
4232
4233         if (cache_xattr) {
4234                 rc = osd_oxc_get(obj, name, buf);
4235                 if (rc != -ENOENT)
4236                         return rc;
4237         }
4238
4239         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4240                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4241                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4242                 struct filter_fid *ff;
4243                 struct ost_layout *ol;
4244
4245                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4246                 if (rc)
4247                         return rc;
4248
4249                 LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4250
4251                 rc = sizeof(*ff);
4252                 if (buf->lb_len == 0 || !buf->lb_buf)
4253                         return rc;
4254
4255                 if (buf->lb_len < rc)
4256                         return -ERANGE;
4257
4258                 ff = buf->lb_buf;
4259                 ol = &ff->ff_layout;
4260                 ol->ol_stripe_count = cpu_to_le32(loa->loa_parent_fid.f_ver >>
4261                                                   PFID_STRIPE_IDX_BITS);
4262                 ol->ol_stripe_size = cpu_to_le32(loa->loa_stripe_size);
4263                 loa->loa_parent_fid.f_ver &= PFID_STRIPE_COUNT_MASK;
4264                 fid_cpu_to_le(&ff->ff_parent, &loa->loa_parent_fid);
4265                 if (lma->lma_compat & LMAC_COMP_INFO) {
4266                         ol->ol_comp_start = cpu_to_le64(loa->loa_comp_start);
4267                         ol->ol_comp_end = cpu_to_le64(loa->loa_comp_end);
4268                         ol->ol_comp_id = cpu_to_le32(loa->loa_comp_id);
4269                 } else {
4270                         ol->ol_comp_start = 0;
4271                         ol->ol_comp_end = 0;
4272                         ol->ol_comp_id = 0;
4273                 }
4274         } else {
4275                 rc = __osd_xattr_get(inode, dentry, name,
4276                                      buf->lb_buf, buf->lb_len);
4277         }
4278
4279         if (cache_xattr) {
4280                 if (rc == -ENOENT || rc == -ENODATA)
4281                         osd_oxc_add(obj, name, NULL, 0);
4282                 else if (rc > 0 && buf->lb_buf != NULL)
4283                         osd_oxc_add(obj, name, buf->lb_buf, rc);
4284         }
4285
4286         return rc;
4287 }
4288
4289 static int osd_declare_xattr_set(const struct lu_env *env,
4290                                  struct dt_object *dt,
4291                                  const struct lu_buf *buf, const char *name,
4292                                  int fl, struct thandle *handle)
4293 {
4294         struct osd_thandle *oh;
4295         int credits = 0;
4296         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4297
4298         LASSERT(handle != NULL);
4299
4300         oh = container_of0(handle, struct osd_thandle, ot_super);
4301         LASSERT(oh->ot_handle == NULL);
4302
4303         if (strcmp(name, XATTR_NAME_LMA) == 0) {
4304                 /*
4305                  * For non-upgrading case, the LMA is set first and
4306                  * usually fit inode. But for upgrade case, the LMA
4307                  * may be in another separated EA block.
4308                  */
4309                 if (dt_object_exists(dt)) {
4310                         if (fl == LU_XATTR_REPLACE)
4311                                 credits = 1;
4312                         else
4313                                 goto upgrade;
4314                 }
4315         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4316                 credits = 1;
4317         } else if (strcmp(name, XATTR_NAME_FID) == 0) {
4318                 /* We may need to delete the old PFID EA. */
4319                 credits = LDISKFS_MAXQUOTAS_DEL_BLOCKS(sb);
4320                 if (fl == LU_XATTR_REPLACE)
4321                         credits += 1;
4322                 else
4323                         goto upgrade;
4324         } else {
4325                 /*
4326                  * If some name entry resides on remote MDT, then will create
4327                  * agent entry under remote parent. On the other hand, if the
4328                  * remote entry will be removed, then related agent entry may
4329                  * need to be removed from the remote parent. So there may be
4330                  * kinds of cases, let's declare enough credits. The credits
4331                  * for create agent entry is enough for remove case.
4332                  */
4333                 if (strcmp(name, XATTR_NAME_LINK) == 0) {
4334                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4335                         if (dt_object_exists(dt))
4336                                 credits += 1; /* For updating LMA */
4337                 }
4338
4339 upgrade:
4340                 credits += osd_dto_credits_noquota[DTO_XATTR_SET];
4341
4342                 if (buf != NULL) {
4343                         ssize_t buflen;
4344
4345                         if (buf->lb_buf == NULL && dt_object_exists(dt)) {
4346                                 /*
4347                                  * learn xattr size from osd_xattr_get if
4348                                  * attribute has not been read yet
4349                                  */
4350                                 buflen = __osd_xattr_get(
4351                                     osd_dt_obj(dt)->oo_inode,
4352                                     &osd_oti_get(env)->oti_obj_dentry,
4353                                     name, NULL, 0);
4354                                 if (buflen < 0)
4355                                         buflen = 0;
4356                         } else {
4357                                 buflen = buf->lb_len;
4358                         }
4359
4360                         if (buflen > sb->s_blocksize) {
4361                                 credits += osd_calc_bkmap_credits(
4362                                     sb, NULL, 0, -1,
4363                                     (buflen + sb->s_blocksize - 1) >>
4364                                     sb->s_blocksize_bits);
4365                         }
4366                 }
4367                 /*
4368                  * xattr set may involve inode quota change, reserve credits for
4369                  * dquot_initialize()
4370                  */
4371                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4372         }
4373
4374         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
4375
4376         return 0;
4377 }
4378
4379 static int osd_xattr_set_pfid(const struct lu_env *env, struct osd_object *obj,
4380                               const struct lu_buf *buf, int fl,
4381                               struct thandle *handle)
4382 {
4383         struct osd_thread_info *info = osd_oti_get(env);
4384         struct dentry *dentry = &info->oti_obj_dentry;
4385         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4386         struct lustre_mdt_attrs *lma = &loa->loa_lma;
4387         struct inode *inode = obj->oo_inode;
4388         struct filter_fid *ff = buf->lb_buf;
4389         struct ost_layout *ol = &ff->ff_layout;
4390         int flags = XATTR_REPLACE;
4391         int rc;
4392
4393         ENTRY;
4394
4395         if (buf->lb_len != sizeof(*ff) && buf->lb_len != sizeof(struct lu_fid))
4396                 RETURN(-EINVAL);
4397
4398         rc = osd_get_lma(info, inode, dentry, loa);
4399         if (rc == -ENODATA) {
4400                 /* Usually for upgarding from old device */
4401                 lustre_loa_init(loa, lu_object_fid(&obj->oo_dt.do_lu),
4402                                 LMAC_FID_ON_OST, 0);
4403                 flags = XATTR_CREATE;
4404         } else if (rc) {
4405                 RETURN(rc);
4406         }
4407
4408         if (!rc && lma->lma_compat & LMAC_STRIPE_INFO) {
4409                 if ((fl & LU_XATTR_CREATE) && !(fl & LU_XATTR_REPLACE))
4410                         RETURN(-EEXIST);
4411
4412                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256) {
4413                         /* Separate PFID EA from LMA */
4414                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4415                         lustre_lma_swab(lma);
4416                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4417                                              sizeof(*lma), XATTR_REPLACE);
4418                         if (!rc) {
4419                                 obj->oo_pfid_in_lma = 0;
4420                                 rc = LU_XATTR_CREATE;
4421                         }
4422
4423                         RETURN(rc);
4424                 }
4425         } else {
4426                 if (LDISKFS_INODE_SIZE(inode->i_sb) > 256)
4427                         RETURN(fl);
4428
4429                 /*
4430                  * Old client does not send stripe information,
4431                  * then store the PFID EA on disk separatedly.
4432                  */
4433                 if (unlikely(buf->lb_len == sizeof(struct lu_fid) ||
4434                              ol->ol_stripe_size == 0))
4435                         RETURN(fl);
4436
4437                 /* Remove old PFID EA entry firstly. */
4438                 dquot_initialize(inode);
4439                 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
4440                 if (rc == -ENODATA) {
4441                         if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
4442                                 RETURN(rc);
4443                 } else if (rc) {
4444                         RETURN(rc);
4445                 }
4446         }
4447
4448         fid_le_to_cpu(&loa->loa_parent_fid, &ff->ff_parent);
4449         if (likely(ol->ol_stripe_size != 0)) {
4450                 loa->loa_parent_fid.f_ver |= le32_to_cpu(ol->ol_stripe_count) <<
4451                                              PFID_STRIPE_IDX_BITS;
4452                 loa->loa_stripe_size = le32_to_cpu(ol->ol_stripe_size);
4453                 lma->lma_compat |= LMAC_STRIPE_INFO;
4454                 if (ol->ol_comp_id != 0) {
4455                         loa->loa_comp_id = le32_to_cpu(ol->ol_comp_id);
4456                         loa->loa_comp_start = le64_to_cpu(ol->ol_comp_start);
4457                         loa->loa_comp_end = le64_to_cpu(ol->ol_comp_end);
4458                         lma->lma_compat |= LMAC_COMP_INFO;
4459                 }
4460         }
4461
4462         lustre_loa_swab(loa, false);
4463
4464         /* Store the PFID EA inside LMA. */
4465         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, loa, sizeof(*loa),
4466                              flags);
4467         if (!rc)
4468                 obj->oo_pfid_in_lma = 1;
4469
4470         RETURN(rc);
4471 }
4472
4473 /*
4474  * In DNE environment, the object (in spite of regular file or directory)
4475  * and its name entry may reside on different MDTs. Under such case, we will
4476  * create an agent entry on the MDT where the object resides. The agent entry
4477  * references the object locally, that makes the object to be visible to the
4478  * userspace when mounted as 'ldiskfs' directly. Then the userspace tools,
4479  * such as 'tar' can handle the object properly.
4480  *
4481  * We handle the agent entry during set linkEA that is the common interface
4482  * for both regular file and directroy, can handle kinds of cases, such as
4483  * create/link/unlink/rename, and so on.
4484  *
4485  * NOTE: we can NOT do that when ea_{insert,delete} that is only for directory.
4486  *
4487  * XXX: There are two known issues:
4488  * 1. For one object, we will create at most one agent entry even if there
4489  *    may be more than one cross-MDTs hard links on the object. So the local
4490  *    e2fsck may claim that the object's nlink is larger than the name entries
4491  *    that reference such inode. And in further, the e2fsck will fix the nlink
4492  *    attribute to match the local references. Then it will cause the object's
4493  *    nlink attribute to be inconsistent with the global references. it is bad
4494  *    but not fatal. The ref_del() can handle the zero-referenced case. On the
4495  *    other hand, the global namespace LFSCK can repair the object's attribute
4496  *    according to the linkEA.
4497  * 2. There may be too many hard links on the object as to its linkEA overflow,
4498  *    then the linkEA entry for cross-MDTs reference may be discarded. If such
4499  *    case happened, then at this point, we do not know whether there are some
4500  *    cross-MDTs reference. But there are local references, it guarantees that
4501  *    object is visible to userspace when mounted as 'ldiskfs'. That is enough.
4502  */
4503 static int osd_xattr_handle_linkea(const struct lu_env *env,
4504                                    struct osd_device *osd,
4505                                    struct osd_object *obj,
4506                                    const struct lu_buf *buf,
4507                                    struct thandle *handle)
4508 {
4509         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
4510         struct lu_fid *tfid = &osd_oti_get(env)->oti_fid3;
4511         struct linkea_data ldata = { .ld_buf = (struct lu_buf *)buf };
4512         struct lu_name tmpname;
4513         struct osd_thandle *oh;
4514         int rc;
4515         bool remote = false;
4516
4517         ENTRY;
4518
4519         oh = container_of0(handle, struct osd_thandle, ot_super);
4520         LASSERT(oh->ot_handle != NULL);
4521
4522         rc = linkea_init_with_rec(&ldata);
4523         if (!rc) {
4524                 linkea_first_entry(&ldata);
4525                 while (ldata.ld_lee != NULL && !remote) {
4526                         linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen,
4527                                             &tmpname, tfid);
4528                         if (osd_remote_fid(env, osd, tfid) > 0)
4529                                 remote = true;
4530                         else
4531                                 linkea_next_entry(&ldata);
4532                 }
4533         } else if (rc == -ENODATA) {
4534                 rc = 0;
4535         } else {
4536                 RETURN(rc);
4537         }
4538
4539         if (lu_object_has_agent_entry(&obj->oo_dt.do_lu) && !remote) {
4540                 rc = osd_delete_from_remote_parent(env, osd, obj, oh, false);
4541                 if (rc)
4542                         CERROR("%s: failed to remove agent entry for "DFID
4543                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4544         } else if (!lu_object_has_agent_entry(&obj->oo_dt.do_lu) && remote) {
4545                 rc = osd_add_to_remote_parent(env, osd, obj, oh);
4546                 if (rc)
4547                         CERROR("%s: failed to create agent entry for "DFID
4548                                ": rc = %d\n", osd_name(osd), PFID(fid), rc);
4549         }
4550
4551         RETURN(rc);
4552 }
4553
4554 /*
4555  * Concurrency: @dt is write locked.
4556  */
4557 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
4558                          const struct lu_buf *buf, const char *name, int fl,
4559                          struct thandle *handle)
4560 {
4561         struct osd_object *obj = osd_dt_obj(dt);
4562         struct osd_device *osd = osd_obj2dev(obj);
4563         struct inode *inode = obj->oo_inode;
4564         struct osd_thread_info *info = osd_oti_get(env);
4565         int fs_flags = 0;
4566         int len;
4567         int rc;
4568
4569         ENTRY;
4570
4571         LASSERT(handle);
4572         LASSERT(buf);
4573
4574         /* version set is not real XATTR */
4575         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
4576                 dt_obj_version_t *version = buf->lb_buf;
4577
4578                 /*
4579                  * for version we are just using xattr API but change inode
4580                  * field instead
4581                  */
4582                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
4583
4584                 CDEBUG(D_INODE, "Set version %#llx (old %#llx) for inode %lu\n",
4585                        *version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
4586
4587                 LDISKFS_I(inode)->i_fs_version = *version;
4588                 /*
4589                  * Version is set after all inode operations are finished,
4590                  * so we should mark it dirty here
4591                  */
4592                 osd_dirty_inode(inode, I_DIRTY_DATASYNC);
4593
4594                 RETURN(0);
4595         }
4596
4597         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
4598                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
4599
4600         len = buf->lb_len;
4601         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4602
4603         /*
4604          * For the OST device with 256 bytes inode size by default,
4605          * the PFID EA will be stored together with LMA EA to avoid
4606          * performance trouble. Otherwise the PFID EA can be stored
4607          * independently. LU-8998
4608          */
4609         if (strcmp(name, XATTR_NAME_FID) == 0 && osd->od_is_ost &&
4610             (LDISKFS_INODE_SIZE(inode->i_sb) <= 256 || obj->oo_pfid_in_lma)) {
4611                 LASSERT(buf->lb_buf);
4612
4613                 fl = osd_xattr_set_pfid(env, obj, buf, fl, handle);
4614                 if (fl <= 0)
4615                         RETURN(fl);
4616         } else if (strcmp(name, XATTR_NAME_LMV) == 0) {
4617                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
4618                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
4619
4620                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
4621                 if (rc)
4622                         RETURN(rc);
4623
4624                 lma->lma_incompat |= LMAI_STRIPED;
4625                 lustre_lma_swab(lma);
4626                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4627                                      sizeof(*lma), XATTR_REPLACE);
4628                 if (rc != 0)
4629                         RETURN(rc);
4630         } else if (strcmp(name, XATTR_NAME_LINK) == 0) {
4631                 LASSERT(!osd->od_is_ost);
4632
4633                 rc = osd_xattr_handle_linkea(env, osd, obj, buf, handle);
4634                 if (rc)
4635                         RETURN(rc);
4636         }
4637
4638         if (fl & LU_XATTR_REPLACE)
4639                 fs_flags |= XATTR_REPLACE;
4640
4641         if (fl & LU_XATTR_CREATE)
4642                 fs_flags |= XATTR_CREATE;
4643
4644         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, len, fs_flags);
4645         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4646
4647         if (rc == 0 &&
4648             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4649              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4650                 osd_oxc_add(obj, name, buf->lb_buf, buf->lb_len);
4651
4652         return rc;
4653 }
4654
4655 /*
4656  * Concurrency: @dt is read locked.
4657  */
4658 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
4659                           const struct lu_buf *buf)
4660 {
4661         struct osd_object *obj = osd_dt_obj(dt);
4662         struct inode *inode = obj->oo_inode;
4663         struct osd_thread_info *info = osd_oti_get(env);
4664         struct dentry *dentry = &info->oti_obj_dentry;
4665
4666         if (!dt_object_exists(dt))
4667                 return -ENOENT;
4668
4669         LASSERT(!dt_object_remote(dt));
4670         LASSERT(inode->i_op != NULL);
4671         LASSERT(inode->i_op->listxattr != NULL);
4672
4673         dentry->d_inode = inode;
4674         dentry->d_sb = inode->i_sb;
4675         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
4676 }
4677
4678 static int osd_declare_xattr_del(const struct lu_env *env,
4679                                  struct dt_object *dt, const char *name,
4680                                  struct thandle *handle)
4681 {
4682         struct osd_thandle *oh;
4683         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
4684
4685         LASSERT(!dt_object_remote(dt));
4686         LASSERT(handle != NULL);
4687
4688         oh = container_of0(handle, struct osd_thandle, ot_super);
4689         LASSERT(oh->ot_handle == NULL);
4690
4691         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
4692                              osd_dto_credits_noquota[DTO_XATTR_SET]);
4693         /*
4694          * xattr del may involve inode quota change, reserve credits for
4695          * dquot_initialize()
4696          */
4697         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
4698
4699         return 0;
4700 }
4701
4702 /*
4703  * Concurrency: @dt is write locked.
4704  */
4705 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
4706                          const char *name, struct thandle *handle)
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         int rc;
4713
4714         if (!dt_object_exists(dt))
4715                 return -ENOENT;
4716
4717         LASSERT(!dt_object_remote(dt));
4718         LASSERT(inode->i_op != NULL);
4719         LASSERT(handle != NULL);
4720 #ifdef HAVE_IOP_XATTR
4721         LASSERT(inode->i_op->removexattr != NULL);
4722 #endif
4723
4724         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
4725
4726         if (strcmp(name, XATTR_NAME_FID) == 0 && obj->oo_pfid_in_lma) {
4727                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
4728
4729                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
4730                                  &info->oti_ost_attrs);
4731                 if (!rc) {
4732                         LASSERT(lma->lma_compat & LMAC_STRIPE_INFO);
4733
4734                         lma->lma_compat &= ~(LMAC_STRIPE_INFO | LMAC_COMP_INFO);
4735                         lustre_lma_swab(lma);
4736                         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
4737                                              sizeof(*lma), XATTR_REPLACE);
4738                         if (!rc)
4739                                 obj->oo_pfid_in_lma = 0;
4740                 }
4741         } else {
4742                 dquot_initialize(inode);
4743                 dentry->d_inode = inode;
4744                 dentry->d_sb = inode->i_sb;
4745                 rc = osd_removexattr(dentry, inode, name);
4746         }
4747
4748         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
4749
4750         if (rc == 0 &&
4751             (strcmp(name, XATTR_NAME_LOV) == 0 ||
4752              strcmp(name, XATTR_NAME_DEFAULT_LMV) == 0))
4753                 osd_oxc_del(obj, name);
4754
4755         return rc;
4756 }
4757
4758 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
4759                            __u64 start, __u64 end)
4760 {
4761         struct osd_object *obj = osd_dt_obj(dt);
4762         struct inode *inode = obj->oo_inode;
4763         struct osd_thread_info *info = osd_oti_get(env);
4764         struct dentry *dentry = &info->oti_obj_dentry;
4765         struct file *file = &info->oti_file;
4766         int rc;
4767
4768         ENTRY;
4769
4770         dentry->d_inode = inode;
4771         dentry->d_sb = inode->i_sb;
4772         file->f_path.dentry = dentry;
4773         file->f_mapping = inode->i_mapping;
4774         file->f_op = inode->i_fop;
4775         file->f_inode = inode;
4776
4777         rc = vfs_fsync_range(file, start, end, 0);
4778
4779         RETURN(rc);
4780 }
4781
4782 static int osd_invalidate(const struct lu_env *env, struct dt_object *dt)
4783 {
4784         return 0;
4785 }
4786
4787 /*
4788  * Index operations.
4789  */
4790
4791 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
4792                                const struct dt_index_features *feat)
4793 {
4794         struct iam_descr *descr;
4795
4796         if (osd_object_is_root(o))
4797                 return feat == &dt_directory_features;
4798
4799         LASSERT(o->oo_dir != NULL);
4800
4801         descr = o->oo_dir->od_container.ic_descr;
4802         if (feat == &dt_directory_features) {
4803                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
4804                         return 1;
4805                 else
4806                         return 0;
4807         } else {
4808                 return feat->dif_keysize_min <= descr->id_key_size &&
4809                        descr->id_key_size <= feat->dif_keysize_max &&
4810                        feat->dif_recsize_min <= descr->id_rec_size &&
4811                        descr->id_rec_size <= feat->dif_recsize_max &&
4812                        !(feat->dif_flags & (DT_IND_VARKEY |
4813                                             DT_IND_VARREC | DT_IND_NONUNQ)) &&
4814                        ergo(feat->dif_flags & DT_IND_UPDATE,
4815                             1 /* XXX check that object (and fs) is writable */);
4816         }
4817 }
4818
4819 static int osd_iam_container_init(const struct lu_env *env,
4820                                   struct osd_object *obj,
4821                                   struct osd_directory *dir)
4822 {
4823         struct iam_container *bag = &dir->od_container;
4824         int result;
4825
4826         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
4827         if (result != 0)
4828                 return result;
4829
4830         result = iam_container_setup(bag);
4831         if (result == 0)
4832                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
4833         else
4834                 iam_container_fini(bag);
4835
4836         return result;
4837 }
4838
4839
4840 /*
4841  * Concurrency: no external locking is necessary.
4842  */
4843 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
4844                          const struct dt_index_features *feat)
4845 {
4846         int result;
4847         int skip_iam = 0;
4848         struct osd_object *obj = osd_dt_obj(dt);
4849
4850         LINVRNT(osd_invariant(obj));
4851
4852         if (osd_object_is_root(obj)) {
4853                 dt->do_index_ops = &osd_index_ea_ops;
4854                 result = 0;
4855         } else if (feat == &dt_directory_features) {
4856                 dt->do_index_ops = &osd_index_ea_ops;
4857                 if (obj->oo_inode == NULL || S_ISDIR(obj->oo_inode->i_mode))
4858                         result = 0;
4859                 else
4860                         result = -ENOTDIR;
4861                 skip_iam = 1;
4862         } else if (unlikely(feat == &dt_otable_features)) {
4863                 dt->do_index_ops = &osd_otable_ops;
4864                 return 0;
4865         } else if (unlikely(feat == &dt_acct_features)) {
4866                 dt->do_index_ops = &osd_acct_index_ops;
4867                 result = 0;
4868                 skip_iam = 1;
4869         } else if (!osd_has_index(obj)) {
4870                 struct osd_directory *dir;
4871                 struct osd_device *osd = osd_obj2dev(obj);
4872                 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
4873
4874                 OBD_ALLOC_PTR(dir);
4875                 if (dir) {
4876
4877                         spin_lock(&obj->oo_guard);
4878                         if (obj->oo_dir == NULL)
4879                                 obj->oo_dir = dir;
4880                         else
4881                                 /*
4882                                  * Concurrent thread allocated container data.
4883                                  */
4884                                 OBD_FREE_PTR(dir);
4885                         spin_unlock(&obj->oo_guard);
4886                         /*
4887                          * Now, that we have container data, serialize its
4888                          * initialization.
4889                          */
4890                         down_write(&obj->oo_ext_idx_sem);
4891                         /*
4892                          * recheck under lock.
4893                          */
4894
4895                         if (osd_has_index(obj)) {
4896                                 result = 0;
4897                                 goto unlock;
4898                         }
4899
4900                         result = osd_iam_container_init(env, obj, obj->oo_dir);
4901                         if (result || feat == &dt_lfsck_namespace_features ||
4902                             feat == &dt_lfsck_layout_orphan_features ||
4903                             feat == &dt_lfsck_layout_dangling_features)
4904                                 goto unlock;
4905
4906                         result = osd_index_register(osd, fid,
4907                                                     feat->dif_keysize_max,
4908                                                     feat->dif_recsize_max);
4909                         if (result < 0)
4910                                 CWARN("%s: failed to register index "
4911                                       DFID": rc = %d\n",
4912                                       osd_name(osd), PFID(fid), result);
4913                         else if (result > 0)
4914                                 result = 0;
4915                         else
4916                                 CDEBUG(D_LFSCK, "%s: index object "DFID
4917                                        " (%d/%d) registered\n",
4918                                        osd_name(osd), PFID(fid),
4919                                        (int)feat->dif_keysize_max,
4920                                        (int)feat->dif_recsize_max);
4921
4922 unlock:
4923                         up_write(&obj->oo_ext_idx_sem);
4924                 } else {
4925                         result = -ENOMEM;
4926                 }
4927         } else {
4928                 result = 0;
4929         }
4930
4931         if (result == 0 && skip_iam == 0) {
4932                 if (!osd_iam_index_probe(env, obj, feat))
4933                         result = -ENOTDIR;
4934         }
4935         LINVRNT(osd_invariant(obj));
4936
4937         return result;
4938 }
4939
4940 static int osd_otable_it_attr_get(const struct lu_env *env,
4941                                  struct dt_object *dt,
4942                                  struct lu_attr *attr)
4943 {
4944         attr->la_valid = 0;
4945         return 0;
4946 }
4947
4948 static const struct dt_object_operations osd_obj_ops = {
4949         .do_read_lock           = osd_read_lock,
4950         .do_write_lock          = osd_write_lock,
4951         .do_read_unlock         = osd_read_unlock,
4952         .do_write_unlock        = osd_write_unlock,
4953         .do_write_locked        = osd_write_locked,
4954         .do_attr_get            = osd_attr_get,
4955         .do_declare_attr_set    = osd_declare_attr_set,
4956         .do_attr_set            = osd_attr_set,
4957         .do_ah_init             = osd_ah_init,
4958         .do_declare_create      = osd_declare_create,
4959         .do_create              = osd_create,
4960         .do_declare_destroy     = osd_declare_destroy,
4961         .do_destroy             = osd_destroy,
4962         .do_index_try           = osd_index_try,
4963         .do_declare_ref_add     = osd_declare_ref_add,
4964         .do_ref_add             = osd_ref_add,
4965         .do_declare_ref_del     = osd_declare_ref_del,
4966         .do_ref_del             = osd_ref_del,
4967         .do_xattr_get           = osd_xattr_get,
4968         .do_declare_xattr_set   = osd_declare_xattr_set,
4969         .do_xattr_set           = osd_xattr_set,
4970         .do_declare_xattr_del   = osd_declare_xattr_del,
4971         .do_xattr_del           = osd_xattr_del,
4972         .do_xattr_list          = osd_xattr_list,
4973         .do_object_sync         = osd_object_sync,
4974         .do_invalidate          = osd_invalidate,
4975 };
4976
4977 static const struct dt_object_operations osd_obj_otable_it_ops = {
4978         .do_attr_get    = osd_otable_it_attr_get,
4979         .do_index_try   = osd_index_try,
4980 };
4981
4982 static int osd_index_declare_iam_delete(const struct lu_env *env,
4983                                         struct dt_object *dt,
4984                                         const struct dt_key *key,
4985                                         struct thandle *handle)
4986 {
4987         struct osd_thandle *oh;
4988
4989         oh = container_of0(handle, struct osd_thandle, ot_super);
4990         LASSERT(oh->ot_handle == NULL);
4991
4992         /* Recycle  may cause additional three blocks to be changed. */
4993         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
4994                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
4995
4996         return 0;
4997 }
4998
4999 /**
5000  *      delete a (key, value) pair from index \a dt specified by \a key
5001  *
5002  *      \param  dt      osd index object
5003  *      \param  key     key for index
5004  *      \param  rec     record reference
5005  *      \param  handle  transaction handler
5006  *
5007  *      \retval  0  success
5008  *      \retval -ve   failure
5009  */
5010 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
5011                                 const struct dt_key *key,
5012                                 struct thandle *handle)
5013 {
5014         struct osd_thread_info *oti = osd_oti_get(env);
5015         struct osd_object *obj = osd_dt_obj(dt);
5016         struct osd_thandle *oh;
5017         struct iam_path_descr *ipd;
5018         struct iam_container *bag = &obj->oo_dir->od_container;
5019         int rc;
5020
5021         ENTRY;
5022
5023         if (!dt_object_exists(dt))
5024                 RETURN(-ENOENT);
5025
5026         LINVRNT(osd_invariant(obj));
5027         LASSERT(!dt_object_remote(dt));
5028         LASSERT(bag->ic_object == obj->oo_inode);
5029         LASSERT(handle != NULL);
5030
5031         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5032
5033         ipd = osd_idx_ipd_get(env, bag);
5034         if (unlikely(ipd == NULL))
5035                 RETURN(-ENOMEM);
5036
5037         oh = container_of0(handle, struct osd_thandle, ot_super);
5038         LASSERT(oh->ot_handle != NULL);
5039         LASSERT(oh->ot_handle->h_transaction != NULL);
5040
5041         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5042                 /* swab quota uid/gid provided by caller */
5043                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5044                 key = (const struct dt_key *)&oti->oti_quota_id;
5045         }
5046
5047         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
5048         osd_ipd_put(env, bag, ipd);
5049         LINVRNT(osd_invariant(obj));
5050         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5051         RETURN(rc);
5052 }
5053
5054 static int osd_index_declare_ea_delete(const struct lu_env *env,
5055                                        struct dt_object *dt,
5056                                        const struct dt_key *key,
5057                                        struct thandle *handle)
5058 {
5059         struct osd_thandle *oh;
5060         struct inode *inode;
5061         int rc, credits;
5062
5063         ENTRY;
5064
5065         LASSERT(!dt_object_remote(dt));
5066         LASSERT(handle != NULL);
5067
5068         oh = container_of0(handle, struct osd_thandle, ot_super);
5069         LASSERT(oh->ot_handle == NULL);
5070
5071         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE];
5072         osd_trans_declare_op(env, oh, OSD_OT_DELETE, credits);
5073
5074         inode = osd_dt_obj(dt)->oo_inode;
5075         if (inode == NULL)
5076                 RETURN(-ENOENT);
5077
5078         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
5079                                    i_projid_read(inode), 0, oh, osd_dt_obj(dt),
5080                                    NULL, OSD_QID_BLK);
5081         RETURN(rc);
5082 }
5083
5084 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
5085                                           struct dt_rec *fid)
5086 {
5087         struct osd_fid_pack *rec;
5088         int rc = -ENODATA;
5089
5090         if (de->file_type & LDISKFS_DIRENT_LUFID) {
5091                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5092                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
5093                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
5094                         rc = -EINVAL;
5095         }
5096         return rc;
5097 }
5098
5099 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
5100                           const struct lu_fid *fid)
5101 {
5102         struct seq_server_site *ss = osd_seq_site(osd);
5103
5104         ENTRY;
5105
5106         /* FID seqs not in FLDB, must be local seq */
5107         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
5108                 RETURN(0);
5109
5110         /*
5111          * If FLD is not being initialized yet, it only happens during the
5112          * initialization, likely during mgs initialization, and we assume
5113          * this is local FID.
5114          */
5115         if (ss == NULL || ss->ss_server_fld == NULL)
5116                 RETURN(0);
5117
5118         /* Only check the local FLDB here */
5119         if (osd_seq_exists(env, osd, fid_seq(fid)))
5120                 RETURN(0);
5121
5122         RETURN(1);
5123 }
5124
5125 static void osd_take_care_of_agent(const struct lu_env *env,
5126                                    struct osd_device *osd,
5127                                    struct osd_thandle *oh,
5128                                    struct ldiskfs_dir_entry_2 *de)
5129 {
5130         struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
5131         struct osd_idmap_cache *idc;
5132         int rc, schedule = 0;
5133
5134         LASSERT(de != NULL);
5135
5136         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
5137         if (likely(rc == 0)) {
5138                 idc = osd_idc_find_or_init(env, osd, fid);
5139                 if (IS_ERR(idc) || idc->oic_remote)
5140                         schedule = 1;
5141         } else if (rc == -ENODATA) {
5142                 /*
5143                  * can't get FID, postpone to the end of the
5144                  * transaction when iget() is safe
5145                  */
5146                 schedule = 1;
5147         } else {
5148                 CERROR("%s: can't get FID: rc = %d\n", osd_name(osd), rc);
5149         }
5150         if (schedule)
5151                 osd_schedule_agent_inode_removal(env, oh,
5152                                                  le32_to_cpu(de->inode));
5153 }
5154
5155 /**
5156  * Index delete function for interoperability mode (b11826).
5157  * It will remove the directory entry added by osd_index_ea_insert().
5158  * This entry is needed to maintain name->fid mapping.
5159  *
5160  * \param key,  key i.e. file entry to be deleted
5161  *
5162  * \retval   0, on success
5163  * \retval -ve, on error
5164  */
5165 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
5166                                const struct dt_key *key, struct thandle *handle)
5167 {
5168         struct osd_object *obj = osd_dt_obj(dt);
5169         struct inode *dir = obj->oo_inode;
5170         struct dentry *dentry;
5171         struct osd_thandle *oh;
5172         struct ldiskfs_dir_entry_2 *de = NULL;
5173         struct buffer_head *bh;
5174         struct htree_lock *hlock = NULL;
5175         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5176         int rc;
5177
5178         ENTRY;
5179
5180         if (!dt_object_exists(dt))
5181                 RETURN(-ENOENT);
5182
5183         LINVRNT(osd_invariant(obj));
5184         LASSERT(!dt_object_remote(dt));
5185         LASSERT(handle != NULL);
5186
5187         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
5188
5189         oh = container_of(handle, struct osd_thandle, ot_super);
5190         LASSERT(oh->ot_handle != NULL);
5191         LASSERT(oh->ot_handle->h_transaction != NULL);
5192
5193         dquot_initialize(dir);
5194         dentry = osd_child_dentry_get(env, obj,
5195                                       (char *)key, strlen((char *)key));
5196
5197         if (obj->oo_hl_head != NULL) {
5198                 hlock = osd_oti_get(env)->oti_hlock;
5199                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5200                                    dir, LDISKFS_HLOCK_DEL);
5201         } else {
5202                 down_write(&obj->oo_ext_idx_sem);
5203         }
5204
5205         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5206         if (!IS_ERR(bh)) {
5207                 /*
5208                  * If this is not the ".." entry, it might be a remote DNE
5209                  * entry and  we need to check if the FID is for a remote
5210                  * MDT.  If the FID is  not in the directory entry (e.g.
5211                  * upgraded 1.8 filesystem without dirdata enabled) then
5212                  * we need to get the FID from the LMA. For a remote directory
5213                  * there HAS to be an LMA, it cannot be an IGIF inode in this
5214                  * case.
5215                  *
5216                  * Delete the entry before the agent inode in order to
5217                  * simplify error handling.  At worst an error after deleting
5218                  * the entry first might leak the agent inode afterward. The
5219                  * reverse would need filesystem abort in case of error deleting
5220                  * the entry after the agent had been removed, or leave a
5221                  * dangling entry pointing at a random inode.
5222                  */
5223                 if (strcmp((char *)key, dotdot) != 0)
5224                         osd_take_care_of_agent(env, osd, oh, de);
5225                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
5226                 brelse(bh);
5227         } else {
5228                 rc = PTR_ERR(bh);
5229         }
5230         if (hlock != NULL)
5231                 ldiskfs_htree_unlock(hlock);
5232         else
5233                 up_write(&obj->oo_ext_idx_sem);
5234
5235         GOTO(out, rc);
5236 out:
5237         LASSERT(osd_invariant(obj));
5238         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
5239         RETURN(rc);
5240 }
5241
5242 /**
5243  *      Lookup index for \a key and copy record to \a rec.
5244  *
5245  *      \param  dt      osd index object
5246  *      \param  key     key for index
5247  *      \param  rec     record reference
5248  *
5249  *      \retval  +ve  success : exact mach
5250  *      \retval  0    return record with key not greater than \a key
5251  *      \retval -ve   failure
5252  */
5253 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
5254                                 struct dt_rec *rec, const struct dt_key *key)
5255 {
5256         struct osd_object *obj = osd_dt_obj(dt);
5257         struct iam_path_descr *ipd;
5258         struct iam_container *bag = &obj->oo_dir->od_container;
5259         struct osd_thread_info *oti = osd_oti_get(env);
5260         struct iam_iterator *it = &oti->oti_idx_it;
5261         struct iam_rec *iam_rec;
5262         int rc;
5263
5264         ENTRY;
5265
5266         if (!dt_object_exists(dt))
5267                 RETURN(-ENOENT);
5268
5269         LASSERT(osd_invariant(obj));
5270         LASSERT(!dt_object_remote(dt));
5271         LASSERT(bag->ic_object == obj->oo_inode);
5272
5273         ipd = osd_idx_ipd_get(env, bag);
5274         if (IS_ERR(ipd))
5275                 RETURN(-ENOMEM);
5276
5277         /* got ipd now we can start iterator. */
5278         iam_it_init(it, bag, 0, ipd);
5279
5280         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5281                 /* swab quota uid/gid provided by caller */
5282                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5283                 key = (const struct dt_key *)&oti->oti_quota_id;
5284         }
5285
5286         rc = iam_it_get(it, (struct iam_key *)key);
5287         if (rc >= 0) {
5288                 if (S_ISDIR(obj->oo_inode->i_mode))
5289                         iam_rec = (struct iam_rec *)oti->oti_ldp;
5290                 else
5291                         iam_rec = (struct iam_rec *)rec;
5292
5293                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
5294
5295                 if (S_ISDIR(obj->oo_inode->i_mode))
5296                         osd_fid_unpack((struct lu_fid *)rec,
5297                                        (struct osd_fid_pack *)iam_rec);
5298                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
5299                         osd_quota_unpack(obj, rec);
5300         }
5301
5302         iam_it_put(it);
5303         iam_it_fini(it);
5304         osd_ipd_put(env, bag, ipd);
5305
5306         LINVRNT(osd_invariant(obj));
5307
5308         RETURN(rc);
5309 }
5310
5311 static int osd_index_declare_iam_insert(const struct lu_env *env,
5312                                         struct dt_object *dt,
5313                                         const struct dt_rec *rec,
5314                                         const struct dt_key *key,
5315                                         struct thandle *handle)
5316 {
5317         struct osd_thandle *oh;
5318
5319         LASSERT(handle != NULL);
5320
5321         oh = container_of0(handle, struct osd_thandle, ot_super);
5322         LASSERT(oh->ot_handle == NULL);
5323
5324         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
5325                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
5326
5327         return 0;
5328 }
5329
5330 /**
5331  *      Inserts (key, value) pair in \a dt index object.
5332  *
5333  *      \param  dt      osd index object
5334  *      \param  key     key for index
5335  *      \param  rec     record reference
5336  *      \param  th      transaction handler
5337  *
5338  *      \retval  0  success
5339  *      \retval -ve failure
5340  */
5341 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
5342                                 const struct dt_rec *rec,
5343                                 const struct dt_key *key, struct thandle *th)
5344 {
5345         struct osd_object *obj = osd_dt_obj(dt);
5346         struct iam_path_descr *ipd;
5347         struct osd_thandle *oh;
5348         struct iam_container *bag;
5349         struct osd_thread_info *oti = osd_oti_get(env);
5350         struct iam_rec *iam_rec;
5351         int rc;
5352
5353         ENTRY;
5354
5355         if (!dt_object_exists(dt))
5356                 RETURN(-ENOENT);
5357
5358         LINVRNT(osd_invariant(obj));
5359         LASSERT(!dt_object_remote(dt));
5360
5361         bag = &obj->oo_dir->od_container;
5362         LASSERT(bag->ic_object == obj->oo_inode);
5363         LASSERT(th != NULL);
5364
5365         osd_trans_exec_op(env, th, OSD_OT_INSERT);
5366
5367         ipd = osd_idx_ipd_get(env, bag);
5368         if (unlikely(ipd == NULL))
5369                 RETURN(-ENOMEM);
5370
5371         oh = container_of0(th, struct osd_thandle, ot_super);
5372         LASSERT(oh->ot_handle != NULL);
5373         LASSERT(oh->ot_handle->h_transaction != NULL);
5374         if (S_ISDIR(obj->oo_inode->i_mode)) {
5375                 iam_rec = (struct iam_rec *)oti->oti_ldp;
5376                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec,
5377                              &oti->oti_fid);
5378         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
5379                 /* pack quota uid/gid */
5380                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
5381                 key = (const struct dt_key *)&oti->oti_quota_id;
5382                 /* pack quota record */
5383                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
5384                 iam_rec = (struct iam_rec *)rec;
5385         } else {
5386                 iam_rec = (struct iam_rec *)rec;
5387         }
5388
5389         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
5390                         iam_rec, ipd);
5391         osd_ipd_put(env, bag, ipd);
5392         LINVRNT(osd_invariant(obj));
5393         osd_trans_exec_check(env, th, OSD_OT_INSERT);
5394         RETURN(rc);
5395 }
5396
5397 /**
5398  * Calls ldiskfs_add_entry() to add directory entry
5399  * into the directory. This is required for
5400  * interoperability mode (b11826)
5401  *
5402  * \retval   0, on success
5403  * \retval -ve, on error
5404  */
5405 static int __osd_ea_add_rec(struct osd_thread_info *info,
5406                             struct osd_object *pobj, struct inode  *cinode,
5407                             const char *name, const struct lu_fid *fid,
5408                             struct htree_lock *hlock, struct thandle *th)
5409 {
5410         struct ldiskfs_dentry_param *ldp;
5411         struct dentry *child;
5412         struct osd_thandle *oth;
5413         int rc;
5414
5415         oth = container_of(th, struct osd_thandle, ot_super);
5416         LASSERT(oth->ot_handle != NULL);
5417         LASSERT(oth->ot_handle->h_transaction != NULL);
5418         LASSERT(pobj->oo_inode);
5419
5420         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
5421         if (unlikely(pobj->oo_inode ==
5422                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
5423                 ldp->edp_magic = 0;
5424         else
5425                 osd_get_ldiskfs_dirent_param(ldp, fid);
5426         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
5427         child->d_fsdata = (void *)ldp;
5428         dquot_initialize(pobj->oo_inode);
5429         rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
5430                                    child, cinode, hlock);
5431         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
5432                 struct ldiskfs_dir_entry_2      *de;
5433                 struct buffer_head              *bh;
5434                 int                              rc1;
5435
5436                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
5437                                             NULL, hlock);
5438                 if (!IS_ERR(bh)) {
5439                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
5440                                                                bh);
5441                         if (rc1 == 0) {
5442                                 if (S_ISDIR(cinode->i_mode))
5443                                         de->file_type = LDISKFS_DIRENT_LUFID |
5444                                                         LDISKFS_FT_REG_FILE;
5445                                 else
5446                                         de->file_type = LDISKFS_DIRENT_LUFID |
5447                                                         LDISKFS_FT_DIR;
5448                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
5449                                                               NULL, bh);
5450                         }
5451                         brelse(bh);
5452                 }
5453         }
5454
5455         RETURN(rc);
5456 }
5457
5458 /**
5459  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
5460  * into the directory.Also sets flags into osd object to
5461  * indicate dot and dotdot are created. This is required for
5462  * interoperability mode (b11826)
5463  *
5464  * \param dir   directory for dot and dotdot fixup.
5465  * \param obj   child object for linking
5466  *
5467  * \retval   0, on success
5468  * \retval -ve, on error
5469  */
5470 static int osd_add_dot_dotdot(struct osd_thread_info *info,
5471                               struct osd_object *dir,
5472                               struct inode *parent_dir, const char *name,
5473                               const struct lu_fid *dot_fid,
5474                               const struct lu_fid *dot_dot_fid,
5475                               struct thandle *th)
5476 {
5477         struct inode *inode = dir->oo_inode;
5478         struct osd_thandle *oth;
5479         int result = 0;
5480
5481         oth = container_of(th, struct osd_thandle, ot_super);
5482         LASSERT(oth->ot_handle->h_transaction != NULL);
5483         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
5484
5485         if (strcmp(name, dot) == 0) {
5486                 if (dir->oo_compat_dot_created) {
5487                         result = -EEXIST;
5488                 } else {
5489                         LASSERT(inode->i_ino == parent_dir->i_ino);
5490                         dir->oo_compat_dot_created = 1;
5491                         result = 0;
5492                 }
5493         } else if (strcmp(name, dotdot) == 0) {
5494                 if (!dir->oo_compat_dot_created)
5495                         return -EINVAL;
5496                 /* in case of rename, dotdot is already created */
5497                 if (dir->oo_compat_dotdot_created) {
5498                         return __osd_ea_add_rec(info, dir, parent_dir, name,
5499                                                 dot_dot_fid, NULL, th);
5500                 }
5501
5502                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
5503                         struct lu_fid tfid = *dot_dot_fid;
5504
5505                         tfid.f_oid--;
5506                         result = osd_add_dot_dotdot_internal(info,
5507                                         dir->oo_inode, parent_dir, dot_fid,
5508                                         &tfid, oth);
5509                 } else {
5510                         result = osd_add_dot_dotdot_internal(info,
5511                                         dir->oo_inode, parent_dir, dot_fid,
5512                                         dot_dot_fid, oth);
5513                 }
5514
5515                 if (result == 0)
5516                         dir->oo_compat_dotdot_created = 1;
5517         }
5518
5519         return result;
5520 }
5521
5522
5523 /**
5524  * It will call the appropriate osd_add* function and return the
5525  * value, return by respective functions.
5526  */
5527 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
5528                           struct inode *cinode, const char *name,
5529                           const struct lu_fid *fid, struct thandle *th)
5530 {
5531         struct osd_thread_info *info = osd_oti_get(env);
5532         struct htree_lock *hlock;
5533         int rc;
5534
5535         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
5536
5537         if (name[0] == '.' && (name[1] == '\0' ||
5538                                (name[1] == '.' && name[2] == '\0'))) {
5539                 if (hlock != NULL) {
5540                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5541                                            pobj->oo_inode, 0);
5542                 } else {
5543                         down_write(&pobj->oo_ext_idx_sem);
5544                 }
5545
5546                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
5547                                         lu_object_fid(&pobj->oo_dt.do_lu),
5548                                         fid, th);
5549         } else {
5550                 if (hlock != NULL) {
5551                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
5552                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
5553                 } else {
5554                         down_write(&pobj->oo_ext_idx_sem);
5555                 }
5556
5557                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
5558                         struct lu_fid *tfid = &info->oti_fid;
5559
5560                         *tfid = *fid;
5561                         tfid->f_ver = ~0;
5562                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
5563                                               tfid, hlock, th);
5564                 } else {
5565                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
5566                                               hlock, th);
5567                 }
5568         }
5569         if (hlock != NULL)
5570                 ldiskfs_htree_unlock(hlock);
5571         else
5572                 up_write(&pobj->oo_ext_idx_sem);
5573
5574         return rc;
5575 }
5576
5577 static int
5578 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
5579                       struct osd_idmap_cache *oic)
5580 {
5581         struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
5582         struct lu_fid *fid = &oic->oic_fid;
5583         struct osd_inode_id *id = &oic->oic_lid;
5584         struct inode *inode = NULL;
5585         int once = 0;
5586         bool insert;
5587         int rc;
5588
5589         ENTRY;
5590
5591         if (!fid_is_norm(fid) && !fid_is_igif(fid))
5592                 RETURN(0);
5593
5594         if (thread_is_running(&scrub->os_thread) &&
5595             scrub->os_pos_current > id->oii_ino)
5596                 RETURN(0);
5597
5598         if (dev->od_auto_scrub_interval == AS_NEVER ||
5599             ktime_get_real_seconds() <
5600             scrub->os_file.sf_time_last_complete + dev->od_auto_scrub_interval)
5601                 RETURN(0);
5602
5603 again:
5604         rc = osd_oi_lookup(oti, dev, fid, &oti->oti_id, 0);
5605         if (rc == -ENOENT) {
5606                 __u32 gen = id->oii_gen;
5607
5608                 insert = true;
5609                 if (inode != NULL)
5610                         goto trigger;
5611
5612                 inode = osd_iget(oti, dev, id);
5613                 /* The inode has been removed (by race maybe). */
5614                 if (IS_ERR(inode)) {
5615                         rc = PTR_ERR(inode);
5616
5617                         RETURN(rc == -ESTALE ? -ENOENT : rc);
5618                 }
5619
5620                 /* The OI mapping is lost. */
5621                 if (gen != OSD_OII_NOGEN)
5622                         goto trigger;
5623
5624                 /*
5625                  * The inode may has been reused by others, we do not know,
5626                  * leave it to be handled by subsequent osd_fid_lookup().
5627                  */
5628                 GOTO(out, rc = 0);
5629         } else if (rc || osd_id_eq(id, &oti->oti_id)) {
5630                 GOTO(out, rc);
5631         }
5632
5633         insert = false;
5634
5635 trigger:
5636         if (thread_is_running(&scrub->os_thread)) {
5637                 if (inode == NULL) {
5638                         inode = osd_iget(oti, dev, id);
5639                         /* The inode has been removed (by race maybe). */
5640                         if (IS_ERR(inode)) {
5641                                 rc = PTR_ERR(inode);
5642
5643                                 RETURN(rc == -ESTALE ? -ENOENT : rc);
5644                         }
5645                 }
5646
5647                 rc = osd_oii_insert(dev, oic, insert);
5648                 /*
5649                  * There is race condition between osd_oi_lookup and OI scrub.
5650                  * The OI scrub finished just after osd_oi_lookup() failure.
5651                  * Under such case, it is unnecessary to trigger OI scrub again,
5652                  * but try to call osd_oi_lookup() again.
5653                  */
5654                 if (unlikely(rc == -EAGAIN))
5655                         goto again;
5656
5657                 if (!S_ISDIR(inode->i_mode))
5658                         rc = 0;
5659                 else
5660                         rc = osd_check_lmv(oti, dev, inode, oic);
5661
5662                 GOTO(out, rc);
5663         }
5664
5665         if (dev->od_auto_scrub_interval != AS_NEVER && ++once == 1) {
5666                 rc = osd_scrub_start(oti->oti_env, dev, SS_AUTO_PARTIAL |
5667                                      SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
5668                 CDEBUG(D_LFSCK | D_CONSOLE | D_WARNING,
5669                        "%s: trigger partial OI scrub for RPC inconsistency "
5670                        "checking FID "DFID": rc = %d\n",
5671                        osd_dev2name(dev), PFID(fid), rc);
5672                 if (rc == 0 || rc == -EALREADY)
5673                         goto again;
5674         }
5675
5676         GOTO(out, rc);
5677
5678 out:
5679         if (inode)
5680                 iput(inode);
5681
5682         RETURN(rc);
5683 }
5684
5685 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
5686                                struct osd_device *dev,
5687                                struct osd_idmap_cache *oic,
5688                                struct lu_fid *fid, __u32 ino)
5689 {
5690         struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
5691         struct inode *inode;
5692         int rc;
5693
5694         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
5695         inode = osd_iget(oti, dev, &oic->oic_lid);
5696         if (IS_ERR(inode)) {
5697                 fid_zero(&oic->oic_fid);
5698                 return PTR_ERR(inode);
5699         }
5700
5701         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, loa);
5702         iput(inode);
5703         if (rc != 0)
5704                 fid_zero(&oic->oic_fid);
5705         else
5706                 *fid = oic->oic_fid = loa->loa_lma.lma_self_fid;
5707         return rc;
5708 }
5709
5710 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
5711                       struct osd_inode_id *id, const struct lu_fid *fid)
5712 {
5713         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
5714                id->oii_ino, id->oii_gen, info);
5715         info->oti_cache.oic_lid = *id;
5716         info->oti_cache.oic_fid = *fid;
5717         info->oti_cache.oic_dev = osd;
5718 }
5719
5720 /**
5721  * Get parent FID from the linkEA.
5722  *
5723  * For a directory which parent resides on remote MDT, to satisfy the
5724  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
5725  * the other hand, to make the lookup(..) on the directory can return
5726  * the real parent FID, we append the real parent FID after its ".."
5727  * name entry in the /REMOTE_PARENT_DIR.
5728  *
5729  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
5730  * backup. So after the restore, we cannot get the right parent FID from
5731  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
5732  * we have stored the real parent FID in the directory object's linkEA,
5733  * we can parse the linkEA for the real parent FID.
5734  *
5735  * \param[in] env       pointer to the thread context
5736  * \param[in] obj       pointer to the object to be handled
5737  * \param[out]fid       pointer to the buffer to hold the parent FID
5738  *
5739  * \retval              0 for getting the real parent FID successfully
5740  * \retval              negative error number on failure
5741  */
5742 static int osd_get_pfid_from_linkea(const struct lu_env *env,
5743                                     struct osd_object *obj,
5744                                     struct lu_fid *fid)
5745 {
5746         struct osd_thread_info *oti = osd_oti_get(env);
5747         struct lu_buf *buf = &oti->oti_big_buf;
5748         struct dentry *dentry = &oti->oti_obj_dentry;
5749         struct inode *inode = obj->oo_inode;
5750         struct linkea_data ldata = { NULL };
5751         int rc;
5752
5753         ENTRY;
5754
5755         fid_zero(fid);
5756         if (!S_ISDIR(inode->i_mode))
5757                 RETURN(-EIO);
5758
5759 again:
5760         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5761                              buf->lb_buf, buf->lb_len);
5762         if (rc == -ERANGE) {
5763                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5764                                      NULL, 0);
5765                 if (rc > 0) {
5766                         lu_buf_realloc(buf, rc);
5767                         if (buf->lb_buf == NULL)
5768                                 RETURN(-ENOMEM);
5769
5770                         goto again;
5771                 }
5772         }
5773
5774         if (unlikely(rc == 0))
5775                 RETURN(-ENODATA);
5776
5777         if (rc < 0)
5778                 RETURN(rc);
5779
5780         if (unlikely(buf->lb_buf == NULL)) {
5781                 lu_buf_realloc(buf, rc);
5782                 if (buf->lb_buf == NULL)
5783                         RETURN(-ENOMEM);
5784
5785                 goto again;
5786         }
5787
5788         ldata.ld_buf = buf;
5789         rc = linkea_init_with_rec(&ldata);
5790         if (!rc) {
5791                 linkea_first_entry(&ldata);
5792                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
5793         }
5794
5795         RETURN(rc);
5796 }
5797
5798 static int osd_verify_ent_by_linkea(const struct lu_env *env,
5799                                     struct inode *inode,
5800                                     const struct lu_fid *pfid,
5801                                     const char *name, const int namelen)
5802 {
5803         struct osd_thread_info *oti = osd_oti_get(env);
5804         struct lu_buf *buf = &oti->oti_big_buf;
5805         struct dentry *dentry = &oti->oti_obj_dentry;
5806         struct linkea_data ldata = { NULL };
5807         struct lu_name cname = { .ln_name = name,
5808                                  .ln_namelen = namelen };
5809         int rc;
5810
5811         ENTRY;
5812
5813 again:
5814         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
5815                              buf->lb_buf, buf->lb_len);
5816         if (rc == -ERANGE)
5817                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK, NULL, 0);
5818
5819         if (rc < 0)
5820                 RETURN(rc);
5821
5822         if (unlikely(rc == 0))
5823                 RETURN(-ENODATA);
5824
5825         if (buf->lb_len < rc) {
5826                 lu_buf_realloc(buf, rc);
5827                 if (buf->lb_buf == NULL)
5828                         RETURN(-ENOMEM);
5829
5830                 goto again;
5831         }
5832
5833         ldata.ld_buf = buf;
5834         rc = linkea_init_with_rec(&ldata);
5835         if (!rc)
5836                 rc = linkea_links_find(&ldata, &cname, pfid);
5837
5838         RETURN(rc);
5839 }
5840
5841 /**
5842  * Calls ->lookup() to find dentry. From dentry get inode and
5843  * read inode's ea to get fid. This is required for  interoperability
5844  * mode (b11826)
5845  *
5846  * \retval   0, on success
5847  * \retval -ve, on error
5848  */
5849 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
5850                              struct dt_rec *rec, const struct dt_key *key)
5851 {
5852         struct inode *dir = obj->oo_inode;
5853         struct dentry *dentry;
5854         struct ldiskfs_dir_entry_2 *de;
5855         struct buffer_head *bh;
5856         struct lu_fid *fid = (struct lu_fid *)rec;
5857         struct htree_lock *hlock = NULL;
5858         int ino;
5859         int rc;
5860
5861         ENTRY;
5862
5863         LASSERT(dir->i_op != NULL);
5864         LASSERT(dir->i_op->lookup != NULL);
5865
5866         dentry = osd_child_dentry_get(env, obj,
5867                                       (char *)key, strlen((char *)key));
5868
5869         if (obj->oo_hl_head != NULL) {
5870                 hlock = osd_oti_get(env)->oti_hlock;
5871                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
5872                                    dir, LDISKFS_HLOCK_LOOKUP);
5873         } else {
5874                 down_read(&obj->oo_ext_idx_sem);
5875         }
5876
5877         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5878         if (!IS_ERR(bh)) {
5879                 struct osd_thread_info *oti = osd_oti_get(env);
5880                 struct osd_inode_id *id = &oti->oti_id;
5881                 struct osd_idmap_cache *oic = &oti->oti_cache;
5882                 struct osd_device *dev = osd_obj2dev(obj);
5883
5884                 ino = le32_to_cpu(de->inode);
5885                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
5886                         brelse(bh);
5887                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
5888                         GOTO(out, rc);
5889                 }
5890
5891                 rc = osd_get_fid_from_dentry(de, rec);
5892
5893                 /* done with de, release bh */
5894                 brelse(bh);
5895                 if (rc != 0) {
5896                         if (unlikely(is_remote_parent_ino(dev, ino))) {
5897                                 const char *name = (const char *)key;
5898
5899                                 /*
5900                                  * If the parent is on remote MDT, and there
5901                                  * is no FID-in-dirent, then we have to get
5902                                  * the parent FID from the linkEA.
5903                                  */
5904                                 if (likely(strlen(name) == 2 &&
5905                                            name[0] == '.' && name[1] == '.'))
5906                                         rc = osd_get_pfid_from_linkea(env, obj,
5907                                                                       fid);
5908                         } else {
5909                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5910                         }
5911                 } else {
5912                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5913                 }
5914
5915                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
5916                         fid_zero(&oic->oic_fid);
5917
5918                         GOTO(out, rc);
5919                 }
5920
5921                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
5922                 rc = osd_consistency_check(oti, dev, oic);
5923                 if (rc == -ENOENT)
5924                         fid_zero(&oic->oic_fid);
5925                 else
5926                         /* Other error should not affect lookup result. */
5927                         rc = 0;
5928         } else {
5929                 rc = PTR_ERR(bh);
5930         }
5931
5932         GOTO(out, rc);
5933
5934 out:
5935         if (hlock != NULL)
5936                 ldiskfs_htree_unlock(hlock);
5937         else
5938                 up_read(&obj->oo_ext_idx_sem);
5939         return rc;
5940 }
5941
5942 static int osd_index_declare_ea_insert(const struct lu_env *env,
5943                                        struct dt_object *dt,
5944                                        const struct dt_rec *rec,
5945                                        const struct dt_key *key,
5946                                        struct thandle *handle)
5947 {
5948         struct osd_thandle *oh;
5949         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
5950         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
5951         const struct lu_fid *fid = rec1->rec_fid;
5952         int credits, rc = 0;
5953         struct osd_idmap_cache *idc;
5954
5955         ENTRY;
5956
5957         LASSERT(!dt_object_remote(dt));
5958         LASSERT(handle != NULL);
5959         LASSERT(fid != NULL);
5960         LASSERT(rec1->rec_type != 0);
5961
5962         oh = container_of0(handle, struct osd_thandle, ot_super);
5963         LASSERT(oh->ot_handle == NULL);
5964
5965         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
5966
5967         /*
5968          * we can't call iget() while a transactions is running
5969          * (this can lead to a deadlock), but we need to know
5970          * inum and object type. so we find this information at
5971          * declaration and cache in per-thread info
5972          */
5973         idc = osd_idc_find_or_init(env, osd, fid);
5974         if (IS_ERR(idc))
5975                 RETURN(PTR_ERR(idc));
5976         if (idc->oic_remote) {
5977                 /*
5978                  * a reference to remote inode is represented by an
5979                  * agent inode which we have to create
5980                  */
5981                 credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
5982                 credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
5983         }
5984
5985         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
5986
5987         if (osd_dt_obj(dt)->oo_inode != NULL) {
5988                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
5989
5990                 /*
5991                  * We ignore block quota on meta pool (MDTs), so needn't
5992                  * calculate how many blocks will be consumed by this index
5993                  * insert
5994                  */
5995                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
5996                                            i_gid_read(inode),
5997                                            i_projid_read(inode), 0,
5998                                            oh, osd_dt_obj(dt), NULL,
5999                                            OSD_QID_BLK);
6000                 if (rc)
6001                         RETURN(rc);
6002
6003 #ifdef HAVE_PROJECT_QUOTA
6004                 /*
6005                  * Reserve credits for local agent inode to transfer
6006                  * to 0, quota enforcement is ignored in this case.
6007                  */
6008                 if (idc->oic_remote &&
6009                     LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL &&
6010                     i_projid_read(inode) != 0)
6011                         rc = osd_declare_attr_qid(env, osd_dt_obj(dt), oh,
6012                                                   0, i_projid_read(inode),
6013                                                   0, false, PRJQUOTA, true);
6014 #endif
6015         }
6016
6017         RETURN(rc);
6018 }
6019
6020 /**
6021  * Index add function for interoperability mode (b11826).
6022  * It will add the directory entry.This entry is needed to
6023  * maintain name->fid mapping.
6024  *
6025  * \param key it is key i.e. file entry to be inserted
6026  * \param rec it is value of given key i.e. fid
6027  *
6028  * \retval   0, on success
6029  * \retval -ve, on error
6030  */
6031 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
6032                                const struct dt_rec *rec,
6033                                const struct dt_key *key, struct thandle *th)
6034 {
6035         struct osd_object *obj = osd_dt_obj(dt);
6036         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
6037         struct dt_insert_rec *rec1 = (struct dt_insert_rec *)rec;
6038         const struct lu_fid *fid = rec1->rec_fid;
6039         const char *name = (const char *)key;
6040         struct osd_thread_info *oti = osd_oti_get(env);
6041         struct inode *child_inode = NULL;
6042         struct osd_idmap_cache *idc;
6043         int rc;
6044
6045         ENTRY;
6046
6047         if (!dt_object_exists(dt))
6048                 RETURN(-ENOENT);
6049
6050         LASSERT(osd_invariant(obj));
6051         LASSERT(!dt_object_remote(dt));
6052         LASSERT(th != NULL);
6053
6054         osd_trans_exec_op(env, th, OSD_OT_INSERT);
6055
6056         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
6057
6058         idc = osd_idc_find(env, osd, fid);
6059         if (unlikely(idc == NULL)) {
6060                 idc = osd_idc_find_or_init(env, osd, fid);
6061                 if (IS_ERR(idc)) {
6062                         /*
6063                          * this dt_insert() wasn't declared properly, so
6064                          * FID is missing in OI cache. we better do not
6065                          * lookup FID in FLDB/OI and don't risk to deadlock,
6066                          * but in some special cases (lfsck testing, etc)
6067                          * it's much simpler than fixing a caller.
6068                          *
6069                          * normally this error should be placed after the first
6070                          * find, but migrate may attach source stripes to
6071                          * target, which doesn't create stripes.
6072                          */
6073                         CERROR("%s: "DFID" wasn't declared for insert\n",
6074                                osd_name(osd), PFID(fid));
6075                         dump_stack();
6076                         RETURN(PTR_ERR(idc));
6077                 }
6078         }
6079
6080         if (idc->oic_remote) {
6081                 /* Insert remote entry */
6082                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
6083                         child_inode =
6084                         igrab(osd->od_mdt_map->omm_remote_parent->d_inode);
6085                 } else {
6086                         child_inode = osd_create_local_agent_inode(env, osd,
6087                                         obj, fid, rec1->rec_type & S_IFMT, th);
6088                         if (IS_ERR(child_inode))
6089                                 RETURN(PTR_ERR(child_inode));
6090                 }
6091         } else {
6092                 /* Insert local entry */
6093                 if (unlikely(idc->oic_lid.oii_ino == 0)) {
6094                         /* for a reason OI cache wasn't filled properly */
6095                         CERROR("%s: OIC for "DFID" isn't filled\n",
6096                                osd_name(osd), PFID(fid));
6097                         RETURN(-EINVAL);
6098                 }
6099                 child_inode = oti->oti_inode;
6100                 if (unlikely(child_inode == NULL)) {
6101                         struct ldiskfs_inode_info *lii;
6102
6103                         OBD_ALLOC_PTR(lii);
6104                         if (lii == NULL)
6105                                 RETURN(-ENOMEM);
6106                         child_inode = oti->oti_inode = &lii->vfs_inode;
6107                 }
6108                 child_inode->i_sb = osd_sb(osd);
6109                 child_inode->i_ino = idc->oic_lid.oii_ino;
6110                 child_inode->i_mode = rec1->rec_type & S_IFMT;
6111         }
6112
6113         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
6114
6115         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
6116                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
6117
6118         if (child_inode && child_inode != oti->oti_inode)
6119                 iput(child_inode);
6120         LASSERT(osd_invariant(obj));
6121         osd_trans_exec_check(env, th, OSD_OT_INSERT);
6122         RETURN(rc);
6123 }
6124
6125 /**
6126  *  Initialize osd Iterator for given osd index object.
6127  *
6128  *  \param  dt      osd index object
6129  */
6130
6131 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
6132                                      struct dt_object *dt,
6133                                      __u32 unused)
6134 {
6135         struct osd_it_iam *it;
6136         struct osd_object *obj = osd_dt_obj(dt);
6137         struct lu_object *lo = &dt->do_lu;
6138         struct iam_path_descr *ipd;
6139         struct iam_container *bag = &obj->oo_dir->od_container;
6140
6141         if (!dt_object_exists(dt))
6142                 return ERR_PTR(-ENOENT);
6143
6144         OBD_ALLOC_PTR(it);
6145         if (it == NULL)
6146                 return ERR_PTR(-ENOMEM);
6147
6148         ipd = osd_it_ipd_get(env, bag);
6149         if (likely(ipd != NULL)) {
6150                 it->oi_obj = obj;
6151                 it->oi_ipd = ipd;
6152                 lu_object_get(lo);
6153                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
6154                 return (struct dt_it *)it;
6155         } else {
6156                 OBD_FREE_PTR(it);
6157                 return ERR_PTR(-ENOMEM);
6158         }
6159 }
6160
6161 /**
6162  * free given Iterator.
6163  */
6164 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
6165 {
6166         struct osd_it_iam *it  = (struct osd_it_iam *)di;
6167         struct osd_object *obj = it->oi_obj;
6168
6169         iam_it_fini(&it->oi_it);
6170         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
6171         osd_object_put(env, obj);
6172         OBD_FREE_PTR(it);
6173 }
6174
6175 /**
6176  *  Move Iterator to record specified by \a key
6177  *
6178  *  \param  di      osd iterator
6179  *  \param  key     key for index
6180  *
6181  *  \retval +ve  di points to record with least key not larger than key
6182  *  \retval  0   di points to exact matched key
6183  *  \retval -ve  failure
6184  */
6185
6186 static int osd_it_iam_get(const struct lu_env *env,
6187                           struct dt_it *di, const struct dt_key *key)
6188 {
6189         struct osd_thread_info *oti = osd_oti_get(env);
6190         struct osd_it_iam *it = (struct osd_it_iam *)di;
6191
6192         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6193                 /* swab quota uid/gid */
6194                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
6195                 key = (struct dt_key *)&oti->oti_quota_id;
6196         }
6197
6198         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
6199 }
6200
6201 /**
6202  *  Release Iterator
6203  *
6204  *  \param  di      osd iterator
6205  */
6206 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
6207 {
6208         struct osd_it_iam *it = (struct osd_it_iam *)di;
6209
6210         iam_it_put(&it->oi_it);
6211 }
6212
6213 /**
6214  *  Move iterator by one record
6215  *
6216  *  \param  di      osd iterator
6217  *
6218  *  \retval +1   end of container reached
6219  *  \retval  0   success
6220  *  \retval -ve  failure
6221  */
6222
6223 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
6224 {
6225         struct osd_it_iam *it = (struct osd_it_iam *)di;
6226
6227         return iam_it_next(&it->oi_it);
6228 }
6229
6230 /**
6231  * Return pointer to the key under iterator.
6232  */
6233
6234 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
6235                                      const struct dt_it *di)
6236 {
6237         struct osd_thread_info *oti = osd_oti_get(env);
6238         struct osd_it_iam *it = (struct osd_it_iam *)di;
6239         struct osd_object *obj = it->oi_obj;
6240         struct dt_key *key;
6241
6242         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
6243
6244         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
6245                 /* swab quota uid/gid */
6246                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
6247                 key = (struct dt_key *)&oti->oti_quota_id;
6248         }
6249
6250         return key;
6251 }
6252
6253 /**
6254  * Return size of key under iterator (in bytes)
6255  */
6256
6257 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
6258 {
6259         struct osd_it_iam *it = (struct osd_it_iam *)di;
6260
6261         return iam_it_key_size(&it->oi_it);
6262 }
6263
6264 static inline void
6265 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
6266 {
6267         /* check if file type is required */
6268         if (ent->lde_attrs & LUDA_TYPE) {
6269                 struct luda_type *lt;
6270                 int align = sizeof(*lt) - 1;
6271
6272                 len = (len + align) & ~align;
6273                 lt = (struct luda_type *)(ent->lde_name + len);
6274                 lt->lt_type = cpu_to_le16(DTTOIF(type));
6275         }
6276
6277         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
6278 }
6279
6280 /**
6281  * build lu direct from backend fs dirent.
6282  */
6283
6284 static inline void
6285 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
6286                    char *name, __u16 namelen, __u16 type, __u32 attr)
6287 {
6288         ent->lde_attrs = attr | LUDA_FID;
6289         fid_cpu_to_le(&ent->lde_fid, fid);
6290
6291         ent->lde_hash = cpu_to_le64(offset);
6292         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
6293
6294         strncpy(ent->lde_name, name, namelen);
6295         ent->lde_name[namelen] = '\0';
6296         ent->lde_namelen = cpu_to_le16(namelen);
6297
6298         /* append lustre attributes */
6299         osd_it_append_attrs(ent, namelen, type);
6300 }
6301
6302 /**
6303  * Return pointer to the record under iterator.
6304  */
6305 static int osd_it_iam_rec(const struct lu_env *env,
6306                           const struct dt_it *di,
6307                           struct dt_rec *dtrec, __u32 attr)
6308 {
6309         struct osd_it_iam *it = (struct osd_it_iam *)di;
6310         struct osd_thread_info *info = osd_oti_get(env);
6311
6312         ENTRY;
6313
6314         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
6315                 const struct osd_fid_pack *rec;
6316                 struct lu_fid *fid = &info->oti_fid;
6317                 struct lu_dirent *lde = (struct lu_dirent *)dtrec;
6318                 char *name;
6319                 int namelen;
6320                 __u64 hash;
6321                 int rc;
6322
6323                 name = (char *)iam_it_key_get(&it->oi_it);
6324                 if (IS_ERR(name))
6325                         RETURN(PTR_ERR(name));
6326
6327                 namelen = iam_it_key_size(&it->oi_it);
6328
6329                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
6330                 if (IS_ERR(rec))
6331                         RETURN(PTR_ERR(rec));
6332
6333                 rc = osd_fid_unpack(fid, rec);
6334                 if (rc)
6335                         RETURN(rc);
6336
6337                 hash = iam_it_store(&it->oi_it);
6338
6339                 /* IAM does not store object type in IAM index (dir) */
6340                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
6341                                    0, LUDA_FID);
6342         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
6343                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6344                            (struct iam_rec *)dtrec);
6345                 osd_quota_unpack(it->oi_obj, dtrec);
6346         } else {
6347                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
6348                            (struct iam_rec *)dtrec);
6349         }
6350
6351         RETURN(0);
6352 }
6353
6354 /**
6355  * Returns cookie for current Iterator position.
6356  */
6357 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
6358 {
6359         struct osd_it_iam *it = (struct osd_it_iam *)di;
6360
6361         return iam_it_store(&it->oi_it);
6362 }
6363
6364 /**
6365  * Restore iterator from cookie.
6366  *
6367  * \param  di      osd iterator
6368  * \param  hash    Iterator location cookie
6369  *
6370  * \retval +ve  di points to record with least key not larger than key.
6371  * \retval  0   di points to exact matched key
6372  * \retval -ve  failure
6373  */
6374
6375 static int osd_it_iam_load(const struct lu_env *env,
6376                            const struct dt_it *di, __u64 hash)
6377 {
6378         struct osd_it_iam *it = (struct osd_it_iam *)di;
6379
6380         return iam_it_load(&it->oi_it, hash);
6381 }
6382
6383 static const struct dt_index_operations osd_index_iam_ops = {
6384         .dio_lookup         = osd_index_iam_lookup,
6385         .dio_declare_insert = osd_index_declare_iam_insert,
6386         .dio_insert         = osd_index_iam_insert,
6387         .dio_declare_delete = osd_index_declare_iam_delete,
6388         .dio_delete         = osd_index_iam_delete,
6389         .dio_it     = {
6390                 .init     = osd_it_iam_init,
6391                 .fini     = osd_it_iam_fini,
6392                 .get      = osd_it_iam_get,
6393                 .put      = osd_it_iam_put,
6394                 .next     = osd_it_iam_next,
6395                 .key      = osd_it_iam_key,
6396                 .key_size = osd_it_iam_key_size,
6397                 .rec      = osd_it_iam_rec,
6398                 .store    = osd_it_iam_store,
6399                 .load     = osd_it_iam_load
6400         }
6401 };
6402
6403
6404 /**
6405  * Creates or initializes iterator context.
6406  *
6407  * \retval struct osd_it_ea, iterator structure on success
6408  *
6409  */
6410 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
6411                                     struct dt_object *dt,
6412                                     __u32 attr)
6413 {
6414         struct osd_object *obj = osd_dt_obj(dt);
6415         struct osd_thread_info *info = osd_oti_get(env);
6416         struct osd_it_ea *oie;
6417         struct file *file;
6418         struct lu_object *lo = &dt->do_lu;
6419         struct dentry *obj_dentry;
6420
6421         ENTRY;
6422
6423         if (!dt_object_exists(dt) || obj->oo_destroyed)
6424                 RETURN(ERR_PTR(-ENOENT));
6425
6426         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
6427         if (oie == NULL)
6428                 RETURN(ERR_PTR(-ENOMEM));
6429         obj_dentry = &oie->oie_dentry;
6430
6431         obj_dentry->d_inode = obj->oo_inode;
6432         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
6433         obj_dentry->d_name.hash = 0;
6434
6435         oie->oie_rd_dirent       = 0;
6436         oie->oie_it_dirent       = 0;
6437         oie->oie_dirent          = NULL;
6438         if (unlikely(!info->oti_it_ea_buf_used)) {
6439                 oie->oie_buf = info->oti_it_ea_buf;
6440                 info->oti_it_ea_buf_used = 1;
6441         } else {
6442                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6443                 if (oie->oie_buf == NULL)
6444                         RETURN(ERR_PTR(-ENOMEM));
6445         }
6446         oie->oie_obj = obj;
6447
6448         file = &oie->oie_file;
6449
6450         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
6451         if (attr & LUDA_64BITHASH)
6452                 file->f_mode    = FMODE_64BITHASH;
6453         else
6454                 file->f_mode    = FMODE_32BITHASH;
6455         file->f_path.dentry     = obj_dentry;
6456         file->f_mapping         = obj->oo_inode->i_mapping;
6457         file->f_op              = obj->oo_inode->i_fop;
6458         file->f_inode = obj->oo_inode;
6459
6460         lu_object_get(lo);
6461         RETURN((struct dt_it *)oie);
6462 }
6463
6464 /**
6465  * Destroy or finishes iterator context.
6466  *
6467  * \param di iterator structure to be destroyed
6468  */
6469 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
6470 {
6471         struct osd_thread_info *info = osd_oti_get(env);
6472         struct osd_it_ea *oie = (struct osd_it_ea *)di;
6473         struct osd_object *obj = oie->oie_obj;
6474         struct inode *inode = obj->oo_inode;
6475
6476         ENTRY;
6477         oie->oie_file.f_op->release(inode, &oie->oie_file);
6478         osd_object_put(env, obj);
6479         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
6480                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
6481         else
6482                 info->oti_it_ea_buf_used = 0;
6483         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
6484         EXIT;
6485 }
6486
6487 /**
6488  * It position the iterator at given key, so that next lookup continues from
6489  * that key Or it is similar to dio_it->load() but based on a key,
6490  * rather than file position.
6491  *
6492  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
6493  * to the beginning.
6494  *
6495  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
6496  */
6497 static int osd_it_ea_get(const struct lu_env *env,
6498                          struct dt_it *di, const struct dt_key *key)
6499 {
6500         struct osd_it_ea *it = (struct osd_it_ea *)di;
6501
6502         ENTRY;
6503         LASSERT(((const char *)key)[0] == '\0');
6504         it->oie_file.f_pos = 0;
6505         it->oie_rd_dirent = 0;
6506         it->oie_it_dirent = 0;
6507         it->oie_dirent = NULL;
6508
6509         RETURN(+1);
6510 }
6511
6512 /**
6513  * Does nothing
6514  */
6515 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
6516 {
6517 }
6518
6519 struct osd_filldir_cbs {
6520         struct dir_context ctx;
6521         struct osd_it_ea  *it;
6522 };
6523 /**
6524  * It is called internally by ->iterate*(). It fills the
6525  * iterator's in-memory data structure with required
6526  * information i.e. name, namelen, rec_size etc.
6527  *
6528  * \param buf in which information to be filled in.
6529  * \param name name of the file in given dir
6530  *
6531  * \retval 0 on success
6532  * \retval 1 on buffer full
6533  */
6534 #ifdef HAVE_FILLDIR_USE_CTX
6535 static int osd_ldiskfs_filldir(struct dir_context *buf,
6536 #else
6537 static int osd_ldiskfs_filldir(void *buf,
6538 #endif
6539                                const char *name, int namelen,
6540                                loff_t offset, __u64 ino, unsigned int d_type)
6541 {
6542         struct osd_it_ea *it = ((struct osd_filldir_cbs *)buf)->it;
6543         struct osd_object *obj = it->oie_obj;
6544         struct osd_it_ea_dirent *ent = it->oie_dirent;
6545         struct lu_fid *fid = &ent->oied_fid;
6546         struct osd_fid_pack *rec;
6547
6548         ENTRY;
6549
6550 /* this should never happen */
6551         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
6552                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
6553                 RETURN(-EIO);
6554         }
6555
6556         if ((void *)ent - it->oie_buf + sizeof(*ent) + namelen >
6557             OSD_IT_EA_BUFSIZE)
6558                 RETURN(1);
6559
6560         /* "." is just the object itself. */
6561         if (namelen == 1 && name[0] == '.') {
6562                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6563         } else if (d_type & LDISKFS_DIRENT_LUFID) {
6564                 rec = (struct osd_fid_pack *)(name + namelen + 1);
6565                 if (osd_fid_unpack(fid, rec) != 0)
6566                         fid_zero(fid);
6567         } else {
6568                 fid_zero(fid);
6569         }
6570         d_type &= ~LDISKFS_DIRENT_LUFID;
6571
6572         /* NOT export local root. */
6573         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
6574                 ino = obj->oo_inode->i_ino;
6575                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
6576         }
6577
6578         ent->oied_ino     = ino;
6579         ent->oied_off     = offset;
6580         ent->oied_namelen = namelen;
6581         ent->oied_type    = d_type;
6582
6583         memcpy(ent->oied_name, name, namelen);
6584
6585         it->oie_rd_dirent++;
6586         it->oie_dirent = (void *)ent + cfs_size_round(sizeof(*ent) + namelen);
6587         RETURN(0);
6588 }
6589
6590 /**
6591  * Calls ->iterate*() to load a directory entry at a time
6592  * and stored it in iterator's in-memory data structure.
6593  *
6594  * \param di iterator's in memory structure
6595  *
6596  * \retval   0 on success
6597  * \retval -ve on error
6598  * \retval +1 reach the end of entry
6599  */
6600 static int osd_ldiskfs_it_fill(const struct lu_env *env,
6601                                const struct dt_it *di)
6602 {
6603         struct osd_it_ea *it = (struct osd_it_ea *)di;
6604         struct osd_object *obj = it->oie_obj;
6605         struct inode *inode = obj->oo_inode;
6606         struct htree_lock *hlock = NULL;
6607         struct file *filp = &it->oie_file;
6608         int rc = 0;
6609         struct osd_filldir_cbs buf = {
6610                 .ctx.actor = osd_ldiskfs_filldir,
6611                 .it = it
6612         };
6613
6614         ENTRY;
6615         it->oie_dirent = it->oie_buf;
6616         it->oie_rd_dirent = 0;
6617
6618         if (obj->oo_hl_head != NULL) {
6619                 hlock = osd_oti_get(env)->oti_hlock;
6620                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
6621                                    inode, LDISKFS_HLOCK_READDIR);
6622         } else {
6623                 down_read(&obj->oo_ext_idx_sem);
6624         }
6625
6626         filp->f_cred = current_cred();
6627         rc = osd_security_file_alloc(filp);
6628         if (rc)
6629                 RETURN(rc);
6630
6631         filp->f_flags |= O_NOATIME;
6632         filp->f_mode |= FMODE_NONOTIFY;
6633         rc = iterate_dir(filp, &buf.ctx);
6634         if (rc)
6635                 RETURN(rc);
6636
6637         if (hlock != NULL)
6638                 ldiskfs_htree_unlock(hlock);
6639         else
6640                 up_read(&obj->oo_ext_idx_sem);
6641
6642         if (it->oie_rd_dirent == 0) {
6643                 /*
6644                  * If it does not get any dirent, it means it has been reached
6645                  * to the end of the dir
6646                  */
6647                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
6648                 if (rc == 0)
6649                         rc = 1;
6650         } else {
6651                 it->oie_dirent = it->oie_buf;
6652                 it->oie_it_dirent = 1;
6653         }
6654
6655         RETURN(rc);
6656 }
6657
6658 /**
6659  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
6660  * to load a directory entry at a time and stored it in
6661  * iterator's in-memory data structure.
6662  *
6663  * \param di iterator's in memory structure
6664  *
6665  * \retval +ve iterator reached to end
6666  * \retval   0 iterator not reached to end
6667  * \retval -ve on error
6668  */
6669 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
6670 {
6671         struct osd_it_ea *it = (struct osd_it_ea *)di;
6672         int rc;
6673
6674         ENTRY;
6675
6676         if (it->oie_it_dirent < it->oie_rd_dirent) {
6677                 it->oie_dirent =
6678                         (void *)it->oie_dirent +
6679                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
6680                                        it->oie_dirent->oied_namelen);
6681                 it->oie_it_dirent++;
6682                 rc = 0;
6683         } else {
6684                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
6685                         rc = 1;
6686                 else
6687                         rc = osd_ldiskfs_it_fill(env, di);
6688         }
6689
6690         RETURN(rc);
6691 }
6692
6693 /**
6694  * Returns the key at current position from iterator's in memory structure.
6695  *
6696  * \param di iterator's in memory structure
6697  *
6698  * \retval key i.e. struct dt_key on success
6699  */
6700 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
6701                                     const struct dt_it *di)
6702 {
6703         struct osd_it_ea *it = (struct osd_it_ea *)di;
6704
6705         return (struct dt_key *)it->oie_dirent->oied_name;
6706 }
6707
6708 /**
6709  * Returns key's size at current position from iterator's in memory structure.
6710  *
6711  * \param di iterator's in memory structure
6712  *
6713  * \retval key_size i.e. struct dt_key on success
6714  */
6715 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
6716 {
6717         struct osd_it_ea *it = (struct osd_it_ea *)di;
6718
6719         return it->oie_dirent->oied_namelen;
6720 }
6721
6722 #if defined LDISKFS_DIR_ENTRY_LEN && defined LDISKFS_DIR_ENTRY_LEN_
6723 #undef LDISKFS_DIR_REC_LEN
6724 #define LDISKFS_DIR_REC_LEN(de)         LDISKFS_DIR_ENTRY_LEN_((de))
6725 #endif
6726
6727 static inline bool osd_dotdot_has_space(struct ldiskfs_dir_entry_2 *de)
6728 {
6729         if (LDISKFS_DIR_REC_LEN(de) >=
6730             __LDISKFS_DIR_REC_LEN(2 + 1 + sizeof(struct osd_fid_pack)))
6731                 return true;
6732
6733         return false;
6734 }
6735
6736 static inline bool
6737 osd_dirent_has_space(struct ldiskfs_dir_entry_2 *de, __u16 namelen,
6738                      unsigned int blocksize, bool dotdot)
6739 {
6740         if (dotdot)
6741                 return osd_dotdot_has_space(de);
6742
6743         if (ldiskfs_rec_len_from_disk(de->rec_len, blocksize) >=
6744             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
6745                 return true;
6746
6747         return false;
6748 }
6749
6750 static int
6751 osd_dirent_reinsert(const struct lu_env *env, struct osd_device *dev,
6752                     handle_t *jh, struct dentry *dentry,
6753                     const struct lu_fid *fid, struct buffer_head *bh,
6754                     struct ldiskfs_dir_entry_2 *de, struct htree_lock *hlock,
6755                     bool dotdot)
6756 {
6757         struct inode *dir = dentry->d_parent->d_inode;
6758         struct inode *inode = dentry->d_inode;
6759         struct osd_fid_pack *rec;
6760         struct ldiskfs_dentry_param *ldp;
6761         int namelen = dentry->d_name.len;
6762         int rc;
6763         struct osd_thread_info *info = osd_oti_get(env);
6764
6765         ENTRY;
6766
6767         if (!ldiskfs_has_feature_dirdata(inode->i_sb))
6768                 RETURN(0);
6769
6770         /* There is enough space to hold the FID-in-dirent. */
6771         if (osd_dirent_has_space(de, namelen, dir->i_sb->s_blocksize, dotdot)) {
6772                 rc = ldiskfs_journal_get_write_access(jh, bh);
6773                 if (rc != 0)
6774                         RETURN(rc);
6775
6776                 de->name[namelen] = 0;
6777                 rec = (struct osd_fid_pack *)(de->name + namelen + 1);
6778                 rec->fp_len = sizeof(struct lu_fid) + 1;
6779                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
6780                 de->file_type |= LDISKFS_DIRENT_LUFID;
6781                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
6782
6783                 RETURN(rc);
6784         }
6785
6786         LASSERT(!dotdot);
6787
6788         rc = ldiskfs_delete_entry(jh, dir, de, bh);
6789         if (rc != 0)
6790                 RETURN(rc);
6791
6792         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
6793         osd_get_ldiskfs_dirent_param(ldp, fid);
6794         dentry->d_fsdata = (void *)ldp;
6795         dquot_initialize(dir);
6796         rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
6797         /*
6798          * It is too bad, we cannot reinsert the name entry back.
6799          * That means we lose it!
6800          */
6801         if (rc != 0)
6802                 CDEBUG(D_LFSCK, "%s: fail to reinsert the dirent, "
6803                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
6804                        osd_ino2name(inode),
6805                        dir->i_ino, dir->i_generation, namelen,
6806                        dentry->d_name.name, PFID(fid), rc);
6807
6808         RETURN(rc);
6809 }
6810
6811 static int
6812 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
6813                         struct osd_it_ea *it, struct lu_fid *fid,
6814                         struct osd_inode_id *id, __u32 *attr)
6815 {
6816         struct osd_thread_info *info = osd_oti_get(env);
6817         struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
6818         struct osd_device *dev = osd_obj2dev(obj);
6819         struct super_block *sb = osd_sb(dev);
6820         const char *devname = osd_name(dev);
6821         struct osd_it_ea_dirent *ent = it->oie_dirent;
6822         struct inode *dir = obj->oo_inode;
6823         struct htree_lock *hlock = NULL;
6824         struct buffer_head *bh = NULL;
6825         handle_t *jh = NULL;
6826         struct ldiskfs_dir_entry_2 *de;
6827         struct dentry *dentry;
6828         struct inode *inode;
6829         const struct lu_fid *pfid = lu_object_fid(&obj->oo_dt.do_lu);
6830         int credits;
6831         int rc;
6832         bool dotdot = false;
6833         bool dirty = false;
6834
6835         ENTRY;
6836
6837         if (ent->oied_name[0] == '.') {
6838                 if (ent->oied_namelen == 1)
6839                         RETURN(0);
6840
6841                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
6842                         dotdot = true;
6843         }
6844
6845         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
6846         inode = osd_iget(info, dev, id);
6847         if (IS_ERR(inode)) {
6848                 rc = PTR_ERR(inode);
6849                 if (rc == -ENOENT || rc == -ESTALE) {
6850                         /*
6851                          * Maybe dangling name entry, or
6852                          * corrupted directory entry.
6853                          */
6854                         *attr |= LUDA_UNKNOWN;
6855                         rc = 0;
6856                 } else {
6857                         CDEBUG(D_LFSCK, "%s: fail to iget() for dirent "
6858                                "check_repair, dir = %lu/%u, name = %.*s, "
6859                                "ino = %llu, rc = %d\n",
6860                                devname, dir->i_ino, dir->i_generation,
6861                                ent->oied_namelen, ent->oied_name,
6862                                ent->oied_ino, rc);
6863                 }
6864
6865                 RETURN(rc);
6866         }
6867
6868         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
6869                                            ent->oied_namelen);
6870         rc = osd_get_lma(info, inode, dentry, &info->oti_ost_attrs);
6871         if (rc == -ENODATA || !fid_is_sane(&lma->lma_self_fid))
6872                 lma = NULL;
6873         else if (rc != 0)
6874                 GOTO(out, rc);
6875
6876         /*
6877          * We need to ensure that the name entry is still valid.
6878          * Because it may be removed or renamed by other already.
6879          *
6880          * The unlink or rename operation will start journal before PDO lock,
6881          * so to avoid deadlock, here we need to start journal handle before
6882          * related PDO lock also. But because we do not know whether there
6883          * will be something to be repaired before PDO lock, we just start
6884          * journal without conditions.
6885          *
6886          * We may need to remove the name entry firstly, then insert back.
6887          * One credit is for user quota file update.
6888          * One credit is for group quota file update.
6889          * Two credits are for dirty inode.
6890          */
6891         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
6892                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
6893
6894         if (dev->od_dirent_journal != 0) {
6895
6896 again:
6897                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
6898                 if (IS_ERR(jh)) {
6899                         rc = PTR_ERR(jh);
6900                         CDEBUG(D_LFSCK, "%s: fail to start trans for dirent "
6901                                "check_repair, dir = %lu/%u, credits = %d, "
6902                                "name = %.*s, ino = %llu: rc = %d\n",
6903                                devname, dir->i_ino, dir->i_generation, credits,
6904                                ent->oied_namelen, ent->oied_name,
6905                                ent->oied_ino, rc);
6906
6907                         GOTO(out_inode, rc);
6908                 }
6909
6910                 if (obj->oo_hl_head != NULL) {
6911                         hlock = osd_oti_get(env)->oti_hlock;
6912                         /*
6913                          * "0" means exclusive lock for the whole directory.
6914                          * We need to prevent others access such name entry
6915                          * during the delete + insert. Neither HLOCK_ADD nor
6916                          * HLOCK_DEL cannot guarantee the atomicity.
6917                          */
6918                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
6919                 } else {
6920                         down_write(&obj->oo_ext_idx_sem);
6921                 }
6922         } else {
6923                 if (obj->oo_hl_head != NULL) {
6924                         hlock = osd_oti_get(env)->oti_hlock;
6925                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
6926                                            LDISKFS_HLOCK_LOOKUP);
6927                 } else {
6928                         down_read(&obj->oo_ext_idx_sem);
6929                 }
6930         }
6931
6932         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
6933         if (IS_ERR(bh) || le32_to_cpu(de->inode) != inode->i_ino) {
6934                 *attr |= LUDA_IGNORE;
6935
6936                 GOTO(out, rc = 0);
6937         }
6938
6939         /*
6940          * For dotdot entry, if there is not enough space to hold the
6941          * FID-in-dirent, just keep them there. It only happens when the
6942          * device upgraded from 1.8 or restored from MDT file-level backup.
6943          * For the whole directory, only dotdot entry have no FID-in-dirent
6944          * and needs to get FID from LMA when readdir, it will not affect the
6945          * performance much.
6946          */
6947         if (dotdot && !osd_dotdot_has_space(de)) {
6948                 *attr |= LUDA_UNKNOWN;
6949
6950                 GOTO(out, rc = 0);
6951         }
6952
6953         if (lma != NULL) {
6954                 if (lu_fid_eq(fid, &lma->lma_self_fid))
6955                         GOTO(out, rc = 0);
6956
6957                 if (unlikely(lma->lma_compat & LMAC_NOT_IN_OI)) {
6958                         struct lu_fid *tfid = &lma->lma_self_fid;
6959
6960                         if (likely(dotdot &&
6961                                    fid_seq(tfid) == FID_SEQ_LOCAL_FILE &&
6962                                    fid_oid(tfid) == REMOTE_PARENT_DIR_OID)) {
6963                                 /*
6964                                  * It must be REMOTE_PARENT_DIR and as the
6965                                  * 'dotdot' entry of remote directory
6966                                  */
6967                                 *attr |= LUDA_IGNORE;
6968                         } else {
6969                                 CDEBUG(D_LFSCK, "%s: expect remote agent "
6970                                        "parent directory, but got %.*s under "
6971                                        "dir = %lu/%u with the FID "DFID"\n",
6972                                        devname, ent->oied_namelen,
6973                                        ent->oied_name, dir->i_ino,
6974                                        dir->i_generation, PFID(tfid));
6975
6976                                 *attr |= LUDA_UNKNOWN;
6977                         }
6978
6979                         GOTO(out, rc = 0);
6980                 }
6981         }
6982
6983         if (!fid_is_zero(fid)) {
6984                 rc = osd_verify_ent_by_linkea(env, inode, pfid, ent->oied_name,
6985                                               ent->oied_namelen);
6986                 if (rc == -ENOENT ||
6987                     (rc == -ENODATA &&
6988                      !(dev->od_scrub.os_scrub.os_file.sf_flags & SF_UPGRADE))) {
6989                         /*
6990                          * linkEA does not recognize the dirent entry,
6991                          * it may because the dirent entry corruption
6992                          * and points to other's inode.
6993                          */
6994                         CDEBUG(D_LFSCK, "%s: the target inode does not "
6995                                "recognize the dirent, dir = %lu/%u, "
6996                                " name = %.*s, ino = %llu, "
6997                                DFID": rc = %d\n", devname, dir->i_ino,
6998                                dir->i_generation, ent->oied_namelen,
6999                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7000                         *attr |= LUDA_UNKNOWN;
7001
7002                         GOTO(out, rc = 0);
7003                 }
7004
7005                 if (rc && rc != -ENODATA) {
7006                         CDEBUG(D_LFSCK, "%s: fail to verify FID in the dirent, "
7007                                "dir = %lu/%u, name = %.*s, ino = %llu, "
7008                                DFID": rc = %d\n", devname, dir->i_ino,
7009                                dir->i_generation, ent->oied_namelen,
7010                                ent->oied_name, ent->oied_ino, PFID(fid), rc);
7011                         *attr |= LUDA_UNKNOWN;
7012
7013                         GOTO(out, rc = 0);
7014                 }
7015         }
7016
7017         if (lma != NULL) {
7018                 /*
7019                  * linkEA recognizes the dirent entry, the FID-in-LMA is
7020                  * valid, trusted, in spite of fid_is_sane(fid) or not.
7021                  */
7022                 if (*attr & LUDA_VERIFY_DRYRUN) {
7023                         *fid = lma->lma_self_fid;
7024                         *attr |= LUDA_REPAIR;
7025
7026                         GOTO(out, rc = 0);
7027                 }
7028
7029                 if (jh == NULL) {
7030                         brelse(bh);
7031                         dev->od_dirent_journal = 1;
7032                         if (hlock != NULL) {
7033                                 ldiskfs_htree_unlock(hlock);
7034                                 hlock = NULL;
7035                         } else {
7036                                 up_read(&obj->oo_ext_idx_sem);
7037                         }
7038
7039                         goto again;
7040                 }
7041
7042                 *fid = lma->lma_self_fid;
7043                 dirty = true;
7044                 /* Update or append the FID-in-dirent. */
7045                 rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7046                                          bh, de, hlock, dotdot);
7047                 if (rc == 0)
7048                         *attr |= LUDA_REPAIR;
7049                 else
7050                         CDEBUG(D_LFSCK, "%s: fail to re-insert FID after "
7051                                "the dirent, dir = %lu/%u, name = %.*s, "
7052                                "ino = %llu, "DFID": rc = %d\n",
7053                                devname, dir->i_ino, dir->i_generation,
7054                                ent->oied_namelen, ent->oied_name,
7055                                ent->oied_ino, PFID(fid), rc);
7056         } else {
7057                 /* lma is NULL, trust the FID-in-dirent if it is valid. */
7058                 if (*attr & LUDA_VERIFY_DRYRUN) {
7059                         if (fid_is_sane(fid)) {
7060                                 *attr |= LUDA_REPAIR;
7061                         } else if (dev->od_index == 0) {
7062                                 lu_igif_build(fid, inode->i_ino,
7063                                               inode->i_generation);
7064                                 *attr |= LUDA_UPGRADE;
7065                         }
7066
7067                         GOTO(out, rc = 0);
7068                 }
7069
7070                 if (jh == NULL) {
7071                         brelse(bh);
7072                         dev->od_dirent_journal = 1;
7073                         if (hlock != NULL) {
7074                                 ldiskfs_htree_unlock(hlock);
7075                                 hlock = NULL;
7076                         } else {
7077                                 up_read(&obj->oo_ext_idx_sem);
7078                         }
7079
7080                         goto again;
7081                 }
7082
7083                 dirty = true;
7084                 if (unlikely(fid_is_sane(fid))) {
7085                         /*
7086                          * FID-in-dirent exists, but FID-in-LMA is lost.
7087                          * Trust the FID-in-dirent, and add FID-in-LMA.
7088                          */
7089                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
7090                         if (rc == 0)
7091                                 *attr |= LUDA_REPAIR;
7092                         else
7093                                 CDEBUG(D_LFSCK, "%s: fail to set LMA for "
7094                                        "update dirent, dir = %lu/%u, "
7095                                        "name = %.*s, ino = %llu, "
7096                                        DFID": rc = %d\n",
7097                                        devname, dir->i_ino, dir->i_generation,
7098                                        ent->oied_namelen, ent->oied_name,
7099                                        ent->oied_ino, PFID(fid), rc);
7100                 } else if (dev->od_index == 0) {
7101                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
7102                         /*
7103                          * It is probably IGIF object. Only aappend the
7104                          * FID-in-dirent. OI scrub will process FID-in-LMA.
7105                          */
7106                         rc = osd_dirent_reinsert(env, dev, jh, dentry, fid,
7107                                                  bh, de, hlock, dotdot);
7108                         if (rc == 0)
7109                                 *attr |= LUDA_UPGRADE;
7110                         else
7111                                 CDEBUG(D_LFSCK, "%s: fail to append IGIF "
7112                                        "after the dirent, dir = %lu/%u, "
7113                                        "name = %.*s, ino = %llu, "
7114                                        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                 }
7119         }
7120
7121         GOTO(out, rc);
7122
7123 out:
7124         if (!IS_ERR(bh))
7125                 brelse(bh);
7126         if (hlock != NULL) {
7127                 ldiskfs_htree_unlock(hlock);
7128         } else {
7129                 if (dev->od_dirent_journal != 0)
7130                         up_write(&obj->oo_ext_idx_sem);
7131                 else
7132                         up_read(&obj->oo_ext_idx_sem);
7133         }
7134
7135         if (jh != NULL)
7136                 ldiskfs_journal_stop(jh);
7137
7138 out_inode:
7139         iput(inode);
7140         if (rc >= 0 && !dirty)
7141                 dev->od_dirent_journal = 0;
7142
7143         return rc;
7144 }
7145
7146 /**
7147  * Returns the value at current position from iterator's in memory structure.
7148  *
7149  * \param di struct osd_it_ea, iterator's in memory structure
7150  * \param attr attr requested for dirent.
7151  * \param lde lustre dirent
7152  *
7153  * \retval   0 no error and \param lde has correct lustre dirent.
7154  * \retval -ve on error
7155  */
7156 static inline int osd_it_ea_rec(const struct lu_env *env,
7157                                 const struct dt_it *di,
7158                                 struct dt_rec *dtrec, __u32 attr)
7159 {
7160         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
7161         struct osd_object      *obj   = it->oie_obj;
7162         struct osd_device      *dev   = osd_obj2dev(obj);
7163         struct osd_thread_info *oti   = osd_oti_get(env);
7164         struct osd_inode_id    *id    = &oti->oti_id;
7165         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
7166         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
7167         __u32 ino = it->oie_dirent->oied_ino;
7168         int rc = 0;
7169
7170         ENTRY;
7171
7172         LASSERT(!is_remote_parent_ino(dev, obj->oo_inode->i_ino));
7173
7174         if (attr & LUDA_VERIFY) {
7175                 if (unlikely(is_remote_parent_ino(dev, ino))) {
7176                         attr |= LUDA_IGNORE;
7177                         /*
7178                          * If the parent is on remote MDT, and there
7179                          * is no FID-in-dirent, then we have to get
7180                          * the parent FID from the linkEA.
7181                          */
7182                         if (!fid_is_sane(fid) &&
7183                             it->oie_dirent->oied_namelen == 2 &&
7184                             it->oie_dirent->oied_name[0] == '.' &&
7185                             it->oie_dirent->oied_name[1] == '.')
7186                                 osd_get_pfid_from_linkea(env, obj, fid);
7187                 } else {
7188                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
7189                                                      &attr);
7190                 }
7191
7192                 if (!fid_is_sane(fid))
7193                         attr |= LUDA_UNKNOWN;
7194         } else {
7195                 attr &= ~LU_DIRENT_ATTRS_MASK;
7196                 if (!fid_is_sane(fid)) {
7197                         bool is_dotdot = false;
7198
7199                         if (it->oie_dirent->oied_namelen == 2 &&
7200                             it->oie_dirent->oied_name[0] == '.' &&
7201                             it->oie_dirent->oied_name[1] == '.')
7202                                 is_dotdot = true;
7203                         /*
7204                          * If the parent is on remote MDT, and there
7205                          * is no FID-in-dirent, then we have to get
7206                          * the parent FID from the linkEA.
7207                          */
7208                         if (is_remote_parent_ino(dev, ino) && is_dotdot) {
7209                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
7210                         } else {
7211                                 if (is_dotdot == false &&
7212                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
7213                                         RETURN(-ENOENT);
7214
7215                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
7216                         }
7217                 } else {
7218                         osd_id_gen(id, ino, OSD_OII_NOGEN);
7219                 }
7220         }
7221
7222         /* Pack the entry anyway, at least the offset is right. */
7223         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
7224                            it->oie_dirent->oied_name,
7225                            it->oie_dirent->oied_namelen,
7226                            it->oie_dirent->oied_type, attr);
7227
7228         if (rc < 0)
7229                 RETURN(rc);
7230
7231         if (osd_remote_fid(env, dev, fid))
7232                 RETURN(0);
7233
7234         if (likely(!(attr & (LUDA_IGNORE | LUDA_UNKNOWN)) && rc == 0))
7235                 osd_add_oi_cache(oti, dev, id, fid);
7236
7237         RETURN(rc > 0 ? 0 : rc);
7238 }
7239
7240 /**
7241  * Returns the record size size at current position.
7242  *
7243  * This function will return record(lu_dirent) size in bytes.
7244  *
7245  * \param[in] env       execution environment
7246  * \param[in] di        iterator's in memory structure
7247  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
7248  *                      calculate the lu_dirent size.
7249  *
7250  * \retval      record size(in bytes & in memory) of the current lu_dirent
7251  *              entry.
7252  */
7253 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
7254                               __u32 attr)
7255 {
7256         struct osd_it_ea *it = (struct osd_it_ea *)di;
7257
7258         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
7259 }
7260
7261 /**
7262  * Returns a cookie for current position of the iterator head, so that
7263  * user can use this cookie to load/start the iterator next time.
7264  *
7265  * \param di iterator's in memory structure
7266  *
7267  * \retval cookie for current position, on success
7268  */
7269 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
7270 {
7271         struct osd_it_ea *it = (struct osd_it_ea *)di;
7272
7273         return it->oie_dirent->oied_off;
7274 }
7275
7276 /**
7277  * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
7278  * to load a directory entry at a time and stored it i inn,
7279  * in iterator's in-memory data structure.
7280  *
7281  * \param di struct osd_it_ea, iterator's in memory structure
7282  *
7283  * \retval +ve on success
7284  * \retval -ve on error
7285  */
7286 static int osd_it_ea_load(const struct lu_env *env,
7287                           const struct dt_it *di, __u64 hash)
7288 {
7289         struct osd_it_ea *it = (struct osd_it_ea *)di;
7290         int rc;
7291
7292         ENTRY;
7293         it->oie_file.f_pos = hash;
7294
7295         rc =  osd_ldiskfs_it_fill(env, di);
7296         if (rc > 0)
7297                 rc = -ENODATA;
7298
7299         if (rc == 0)
7300                 rc = 1;
7301
7302         RETURN(rc);
7303 }
7304
7305 /**
7306  * Index lookup function for interoperability mode (b11826).
7307  *
7308  * \param key,  key i.e. file name to be searched
7309  *
7310  * \retval +ve, on success
7311  * \retval -ve, on error
7312  */
7313 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
7314                                struct dt_rec *rec, const struct dt_key *key)
7315 {
7316         struct osd_object *obj = osd_dt_obj(dt);
7317         int rc = 0;
7318
7319         ENTRY;
7320
7321         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
7322         LINVRNT(osd_invariant(obj));
7323
7324         rc = osd_ea_lookup_rec(env, obj, rec, key);
7325         if (rc == 0)
7326                 rc = 1;
7327         RETURN(rc);
7328 }
7329
7330 /**
7331  * Index and Iterator operations for interoperability
7332  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
7333  */
7334 static const struct dt_index_operations osd_index_ea_ops = {
7335         .dio_lookup         = osd_index_ea_lookup,
7336         .dio_declare_insert = osd_index_declare_ea_insert,
7337         .dio_insert         = osd_index_ea_insert,
7338         .dio_declare_delete = osd_index_declare_ea_delete,
7339         .dio_delete         = osd_index_ea_delete,
7340         .dio_it     = {
7341                 .init     = osd_it_ea_init,
7342                 .fini     = osd_it_ea_fini,
7343                 .get      = osd_it_ea_get,
7344                 .put      = osd_it_ea_put,
7345                 .next     = osd_it_ea_next,
7346                 .key      = osd_it_ea_key,
7347                 .key_size = osd_it_ea_key_size,
7348                 .rec      = osd_it_ea_rec,
7349                 .rec_size = osd_it_ea_rec_size,
7350                 .store    = osd_it_ea_store,
7351                 .load     = osd_it_ea_load
7352         }
7353 };
7354
7355 static void *osd_key_init(const struct lu_context *ctx,
7356                           struct lu_context_key *key)
7357 {
7358         struct osd_thread_info *info;
7359
7360         OBD_ALLOC_PTR(info);
7361         if (info == NULL)
7362                 return ERR_PTR(-ENOMEM);
7363
7364         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7365         if (info->oti_it_ea_buf == NULL)
7366                 goto out_free_info;
7367
7368         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
7369
7370         info->oti_hlock = ldiskfs_htree_lock_alloc();
7371         if (info->oti_hlock == NULL)
7372                 goto out_free_ea;
7373
7374         return info;
7375
7376 out_free_ea:
7377         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7378 out_free_info:
7379         OBD_FREE_PTR(info);
7380         return ERR_PTR(-ENOMEM);
7381 }
7382
7383 static void osd_key_fini(const struct lu_context *ctx,
7384                          struct lu_context_key *key, void *data)
7385 {
7386         struct osd_thread_info *info = data;
7387         struct ldiskfs_inode_info *lli = LDISKFS_I(info->oti_inode);
7388         struct osd_idmap_cache *idc = info->oti_ins_cache;
7389
7390         if (info->oti_dio_pages) {
7391                 int i;
7392                 for (i = 0; i < PTLRPC_MAX_BRW_PAGES; i++) {
7393                         if (info->oti_dio_pages[i])
7394                                 __free_page(info->oti_dio_pages[i]);
7395                 }
7396                 OBD_FREE(info->oti_dio_pages,
7397                          sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
7398         }
7399
7400         if (info->oti_inode != NULL)
7401                 OBD_FREE_PTR(lli);
7402         if (info->oti_hlock != NULL)
7403                 ldiskfs_htree_lock_free(info->oti_hlock);
7404         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
7405         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
7406         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
7407         lu_buf_free(&info->oti_iobuf.dr_lnb_buf);
7408         lu_buf_free(&info->oti_big_buf);
7409         if (idc != NULL) {
7410                 LASSERT(info->oti_ins_cache_size > 0);
7411                 OBD_FREE(idc, sizeof(*idc) * info->oti_ins_cache_size);
7412                 info->oti_ins_cache = NULL;
7413                 info->oti_ins_cache_size = 0;
7414         }
7415         OBD_FREE_PTR(info);
7416 }
7417
7418 static void osd_key_exit(const struct lu_context *ctx,
7419                          struct lu_context_key *key, void *data)
7420 {
7421         struct osd_thread_info *info = data;
7422
7423         LASSERT(info->oti_r_locks == 0);
7424         LASSERT(info->oti_w_locks == 0);
7425         LASSERT(info->oti_txns    == 0);
7426 }
7427
7428 /* type constructor/destructor: osd_type_init, osd_type_fini */
7429 LU_TYPE_INIT_FINI(osd, &osd_key);
7430
7431 struct lu_context_key osd_key = {
7432         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
7433         .lct_init = osd_key_init,
7434         .lct_fini = osd_key_fini,
7435         .lct_exit = osd_key_exit
7436 };
7437
7438
7439 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
7440                            const char *name, struct lu_device *next)
7441 {
7442         struct osd_device *osd = osd_dev(d);
7443
7444         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname)) >=
7445             sizeof(osd->od_svname))
7446                 return -E2BIG;
7447         return osd_procfs_init(osd, name);
7448 }
7449
7450 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
7451 {
7452         struct seq_server_site *ss = osd_seq_site(osd);
7453         int rc;
7454
7455         ENTRY;
7456
7457         if (osd->od_is_ost || osd->od_cl_seq != NULL)
7458                 RETURN(0);
7459
7460         if (unlikely(ss == NULL))
7461                 RETURN(-ENODEV);
7462
7463         OBD_ALLOC_PTR(osd->od_cl_seq);
7464         if (osd->od_cl_seq == NULL)
7465                 RETURN(-ENOMEM);
7466
7467         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
7468                              osd->od_svname, ss->ss_server_seq);
7469         if (rc != 0) {
7470                 OBD_FREE_PTR(osd->od_cl_seq);
7471                 osd->od_cl_seq = NULL;
7472                 RETURN(rc);
7473         }
7474
7475         if (ss->ss_node_id == 0) {
7476                 /*
7477                  * If the OSD on the sequence controller(MDT0), then allocate
7478                  * sequence here, otherwise allocate sequence after connected
7479                  * to MDT0 (see mdt_register_lwp_callback()).
7480                  */
7481                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
7482                                    &osd->od_cl_seq->lcs_space, env);
7483         }
7484
7485         RETURN(rc);
7486 }
7487
7488 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
7489 {
7490         if (osd->od_cl_seq == NULL)
7491                 return;
7492
7493         seq_client_fini(osd->od_cl_seq);
7494         OBD_FREE_PTR(osd->od_cl_seq);
7495         osd->od_cl_seq = NULL;
7496 }
7497
7498 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
7499 {
7500         ENTRY;
7501
7502         /* shutdown quota slave instance associated with the device */
7503         if (o->od_quota_slave_md != NULL) {
7504                 struct qsd_instance *qsd = o->od_quota_slave_md;
7505
7506                 o->od_quota_slave_md = NULL;
7507                 qsd_fini(env, qsd);
7508         }
7509
7510         if (o->od_quota_slave_dt != NULL) {
7511                 struct qsd_instance *qsd = o->od_quota_slave_dt;
7512
7513                 o->od_quota_slave_dt = NULL;
7514                 qsd_fini(env, qsd);
7515         }
7516
7517         osd_fid_fini(env, o);
7518         osd_scrub_cleanup(env, o);
7519
7520         RETURN(0);
7521 }
7522
7523 static void osd_umount(const struct lu_env *env, struct osd_device *o)
7524 {
7525         ENTRY;
7526
7527         if (o->od_mnt != NULL) {
7528                 shrink_dcache_sb(osd_sb(o));
7529                 osd_sync(env, &o->od_dt_dev);
7530
7531                 mntput(o->od_mnt);
7532                 o->od_mnt = NULL;
7533         }
7534
7535         EXIT;
7536 }
7537
7538 static int osd_mount(const struct lu_env *env,
7539                      struct osd_device *o, struct lustre_cfg *cfg)
7540 {
7541         const char *name = lustre_cfg_string(cfg, 0);
7542         const char *dev = lustre_cfg_string(cfg, 1);
7543         const char *opts;
7544         unsigned long page, s_flags, lmd_flags = 0;
7545         struct page *__page;
7546         struct file_system_type *type;
7547         char *options = NULL;
7548         char *str;
7549         struct osd_thread_info *info = osd_oti_get(env);
7550         struct lu_fid *fid = &info->oti_fid;
7551         struct inode *inode;
7552         int rc = 0, force_over_512tb = 0;
7553
7554         ENTRY;
7555
7556         if (o->od_mnt != NULL)
7557                 RETURN(0);
7558
7559         if (strlen(dev) >= sizeof(o->od_mntdev))
7560                 RETURN(-E2BIG);
7561         strcpy(o->od_mntdev, dev);
7562
7563         str = lustre_cfg_string(cfg, 2);
7564         s_flags = simple_strtoul(str, NULL, 0);
7565         str = strstr(str, ":");
7566         if (str)
7567                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
7568         opts = lustre_cfg_string(cfg, 3);
7569 #ifdef __BIG_ENDIAN
7570         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
7571                 CERROR("%s: device %s extents feature is not guaranteed to "
7572                        "work on big-endian systems. Use \"bigendian_extents\" "
7573                        "mount option to override.\n", name, dev);
7574                 RETURN(-EINVAL);
7575         }
7576 #endif
7577 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
7578         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL) {
7579                 CWARN("force_over_128tb option is deprecated. "
7580                       "Filesystems less than 512TB can be created without any "
7581                       "force options. Use force_over_512tb option for "
7582                       "filesystems greater than 512TB.\n");
7583         }
7584 #endif
7585 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 1, 53, 0)
7586         if (opts != NULL && strstr(opts, "force_over_256tb") != NULL) {
7587                 CWARN("force_over_256tb option is deprecated. "
7588                       "Filesystems less than 512TB can be created without any "
7589                       "force options. Use force_over_512tb option for "
7590                       "filesystems greater than 512TB.\n");
7591         }
7592 #endif
7593
7594         if (opts != NULL && strstr(opts, "force_over_512tb") != NULL)
7595                 force_over_512tb = 1;
7596
7597         __page = alloc_page(GFP_KERNEL);
7598         if (__page == NULL)
7599                 GOTO(out, rc = -ENOMEM);
7600         page = (unsigned long)page_address(__page);
7601         options = (char *)page;
7602         *options = '\0';
7603         if (opts != NULL) {
7604                 /* strip out the options for back compatiblity */
7605                 static char *sout[] = {
7606                         "mballoc",
7607                         "iopen",
7608                         "noiopen",
7609                         "iopen_nopriv",
7610                         "extents",
7611                         "noextents",
7612                         /* strip out option we processed in osd */
7613                         "bigendian_extents",
7614                         "force_over_128tb",
7615                         "force_over_256tb",
7616                         "force_over_512tb",
7617                         NULL
7618                 };
7619                 strncat(options, opts, PAGE_SIZE);
7620                 for (rc = 0, str = options; sout[rc]; ) {
7621                         char *op = strstr(str, sout[rc]);
7622
7623                         if (op == NULL) {
7624                                 rc++;
7625                                 str = options;
7626                                 continue;
7627                         }
7628                         if (op == options || *(op - 1) == ',') {
7629                                 str = op + strlen(sout[rc]);
7630                                 if (*str == ',' || *str == '\0') {
7631                                         *str == ',' ? str++ : str;
7632                                         memmove(op, str, strlen(str) + 1);
7633                                 }
7634                         }
7635                         for (str = op; *str != ',' && *str != '\0'; str++)
7636                                 ;
7637                 }
7638         } else {
7639                 strncat(options, "user_xattr,acl", PAGE_SIZE);
7640         }
7641
7642         /* Glom up mount options */
7643         if (*options != '\0')
7644                 strncat(options, ",", PAGE_SIZE);
7645         strncat(options, "no_mbcache,nodelalloc", PAGE_SIZE);
7646
7647         type = get_fs_type("ldiskfs");
7648         if (!type) {
7649                 CERROR("%s: cannot find ldiskfs module\n", name);
7650                 GOTO(out, rc = -ENODEV);
7651         }
7652
7653         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
7654         module_put(type->owner);
7655
7656         if (IS_ERR(o->od_mnt)) {
7657                 rc = PTR_ERR(o->od_mnt);
7658                 o->od_mnt = NULL;
7659                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
7660                 GOTO(out, rc);
7661         }
7662
7663         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) <<
7664                                  osd_sb(o)->s_blocksize_bits > 512ULL << 40 &&
7665                                  force_over_512tb == 0) {
7666                 CERROR("%s: device %s LDISKFS does not support filesystems "
7667                        "greater than 512TB and can cause data corruption. "
7668                        "Use \"force_over_512tb\" mount option to override.\n",
7669                        name, dev);
7670                 GOTO(out_mnt, rc = -EINVAL);
7671         }
7672
7673         if (lmd_flags & LMD_FLG_DEV_RDONLY) {
7674                 if (priv_dev_set_rdonly) {
7675                         priv_dev_set_rdonly(osd_sb(o)->s_bdev);
7676                         o->od_dt_dev.dd_rdonly = 1;
7677                         LCONSOLE_WARN("%s: set dev_rdonly on this device\n",
7678                                       name);
7679                 } else {
7680                         LCONSOLE_WARN("%s: not support dev_rdonly on this device",
7681                                       name);
7682
7683                         GOTO(out_mnt, rc = -EOPNOTSUPP);
7684                 }
7685         } else if (priv_dev_check_rdonly &&
7686                    priv_dev_check_rdonly(osd_sb(o)->s_bdev)) {
7687                 CERROR("%s: underlying device %s is marked as "
7688                        "read-only. Setup failed\n", name, dev);
7689
7690                 GOTO(out_mnt, rc = -EROFS);
7691         }
7692
7693         if (!ldiskfs_has_feature_journal(o->od_mnt->mnt_sb)) {
7694                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
7695                 GOTO(out_mnt, rc = -EINVAL);
7696         }
7697
7698 #ifdef LDISKFS_MOUNT_DIRDATA
7699         if (ldiskfs_has_feature_dirdata(o->od_mnt->mnt_sb))
7700                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
7701         else if (strstr(name, "MDT")) /* don't complain for MGT or OSTs */
7702                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
7703                       "enabling the dirdata feature. If you do not want to "
7704                       "downgrade to Lustre-1.x again, you can enable it via "
7705                       "'tune2fs -O dirdata device'\n", name, dev);
7706 #endif
7707         inode = osd_sb(o)->s_root->d_inode;
7708         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
7709         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
7710         if (rc != 0) {
7711                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
7712                 GOTO(out_mnt, rc);
7713         }
7714
7715         if (lmd_flags & LMD_FLG_NOSCRUB)
7716                 o->od_auto_scrub_interval = AS_NEVER;
7717
7718         if (blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev))) {
7719                 /* do not use pagecache with flash-backed storage */
7720                 o->od_writethrough_cache = 0;
7721                 o->od_read_cache = 0;
7722         }
7723
7724         GOTO(out, rc = 0);
7725
7726 out_mnt:
7727         mntput(o->od_mnt);
7728         o->od_mnt = NULL;
7729
7730 out:
7731         if (__page)
7732                 __free_page(__page);
7733
7734         return rc;
7735 }
7736
7737 static struct lu_device *osd_device_fini(const struct lu_env *env,
7738                                          struct lu_device *d)
7739 {
7740         struct osd_device *o = osd_dev(d);
7741
7742         ENTRY;
7743
7744         osd_index_backup(env, o, false);
7745         osd_shutdown(env, o);
7746         osd_procfs_fini(o);
7747         osd_obj_map_fini(o);
7748         osd_umount(env, o);
7749
7750         RETURN(NULL);
7751 }
7752
7753 static int osd_device_init0(const struct lu_env *env,
7754                             struct osd_device *o,
7755                             struct lustre_cfg *cfg)
7756 {
7757         struct lu_device *l = osd2lu_dev(o);
7758         struct osd_thread_info *info;
7759         int rc;
7760         int cplen = 0;
7761
7762         /* if the module was re-loaded, env can loose its keys */
7763         rc = lu_env_refill((struct lu_env *)env);
7764         if (rc)
7765                 GOTO(out, rc);
7766         info = osd_oti_get(env);
7767         LASSERT(info);
7768
7769         l->ld_ops = &osd_lu_ops;
7770         o->od_dt_dev.dd_ops = &osd_dt_ops;
7771
7772         spin_lock_init(&o->od_osfs_lock);
7773         mutex_init(&o->od_otable_mutex);
7774         INIT_LIST_HEAD(&o->od_orphan_list);
7775         INIT_LIST_HEAD(&o->od_index_backup_list);
7776         INIT_LIST_HEAD(&o->od_index_restore_list);
7777         spin_lock_init(&o->od_lock);
7778         o->od_index_backup_policy = LIBP_NONE;
7779         o->od_t10_type = 0;
7780
7781         o->od_read_cache = 1;
7782         o->od_writethrough_cache = 1;
7783         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
7784         o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
7785         o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
7786         o->od_auto_scrub_interval = AS_DEFAULT;
7787
7788         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
7789                         sizeof(o->od_svname));
7790         if (cplen >= sizeof(o->od_svname)) {
7791                 rc = -E2BIG;
7792                 GOTO(out, rc);
7793         }
7794
7795         o->od_index_backup_stop = 0;
7796         o->od_index = -1; /* -1 means index is invalid */
7797         rc = server_name2index(o->od_svname, &o->od_index, NULL);
7798         if (rc == LDD_F_SV_TYPE_OST)
7799                 o->od_is_ost = 1;
7800
7801         o->od_full_scrub_ratio = OFSR_DEFAULT;
7802         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
7803         rc = osd_mount(env, o, cfg);
7804         if (rc != 0)
7805                 GOTO(out, rc);
7806
7807         /* Can only check block device after mount */
7808         o->od_nonrotational =
7809                 blk_queue_nonrot(bdev_get_queue(osd_sb(o)->s_bdev));
7810
7811         rc = osd_obj_map_init(env, o);
7812         if (rc != 0)
7813                 GOTO(out_mnt, rc);
7814
7815         rc = lu_site_init(&o->od_site, l);
7816         if (rc != 0)
7817                 GOTO(out_compat, rc);
7818         o->od_site.ls_bottom_dev = l;
7819
7820         rc = lu_site_init_finish(&o->od_site);
7821         if (rc != 0)
7822                 GOTO(out_site, rc);
7823
7824         INIT_LIST_HEAD(&o->od_ios_list);
7825         /* setup scrub, including OI files initialization */
7826         o->od_in_init = 1;
7827         rc = osd_scrub_setup(env, o);
7828         o->od_in_init = 0;
7829         if (rc < 0)
7830                 GOTO(out_site, rc);
7831
7832         rc = osd_procfs_init(o, o->od_svname);
7833         if (rc != 0) {
7834                 CERROR("%s: can't initialize procfs: rc = %d\n",
7835                        o->od_svname, rc);
7836                 GOTO(out_scrub, rc);
7837         }
7838
7839         LASSERT(l->ld_site->ls_linkage.next != NULL);
7840         LASSERT(l->ld_site->ls_linkage.prev != NULL);
7841
7842         /* initialize quota slave instance */
7843         /* currently it's no need to prepare qsd_instance_md for OST */
7844         if (!o->od_is_ost) {
7845                 o->od_quota_slave_md = qsd_init(env, o->od_svname,
7846                                                 &o->od_dt_dev,
7847                                                 o->od_proc_entry, true);
7848                 if (IS_ERR(o->od_quota_slave_md)) {
7849                         rc = PTR_ERR(o->od_quota_slave_md);
7850                         o->od_quota_slave_md = NULL;
7851                         GOTO(out_procfs, rc);
7852                 }
7853         }
7854
7855         o->od_quota_slave_dt = qsd_init(env, o->od_svname, &o->od_dt_dev,
7856                                         o->od_proc_entry, false);
7857
7858         if (IS_ERR(o->od_quota_slave_dt)) {
7859                 if (o->od_quota_slave_md != NULL) {
7860                         qsd_fini(env, o->od_quota_slave_md);
7861                         o->od_quota_slave_md = NULL;
7862                 }
7863
7864                 rc = PTR_ERR(o->od_quota_slave_dt);
7865                 o->od_quota_slave_dt = NULL;
7866                 GOTO(out_procfs, rc);
7867         }
7868
7869         RETURN(0);
7870
7871 out_procfs:
7872         osd_procfs_fini(o);
7873 out_scrub:
7874         osd_scrub_cleanup(env, o);
7875 out_site:
7876         lu_site_fini(&o->od_site);
7877 out_compat:
7878         osd_obj_map_fini(o);
7879 out_mnt:
7880         osd_umount(env, o);
7881 out:
7882         return rc;
7883 }
7884
7885 static struct lu_device *osd_device_alloc(const struct lu_env *env,
7886                                           struct lu_device_type *t,
7887                                           struct lustre_cfg *cfg)
7888 {
7889         struct osd_device *o;
7890         int rc;
7891
7892         OBD_ALLOC_PTR(o);
7893         if (o == NULL)
7894                 return ERR_PTR(-ENOMEM);
7895
7896         rc = dt_device_init(&o->od_dt_dev, t);
7897         if (rc == 0) {
7898                 /*
7899                  * Because the ctx might be revived in dt_device_init,
7900                  * refill the env here
7901                  */
7902                 lu_env_refill((struct lu_env *)env);
7903                 rc = osd_device_init0(env, o, cfg);
7904                 if (rc)
7905                         dt_device_fini(&o->od_dt_dev);
7906         }
7907
7908         if (unlikely(rc != 0))
7909                 OBD_FREE_PTR(o);
7910
7911         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
7912 }
7913
7914 static struct lu_device *osd_device_free(const struct lu_env *env,
7915                                          struct lu_device *d)
7916 {
7917         struct osd_device *o = osd_dev(d);
7918
7919         ENTRY;
7920
7921         /* XXX: make osd top device in order to release reference */
7922         d->ld_site->ls_top_dev = d;
7923         lu_site_purge(env, d->ld_site, -1);
7924         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
7925                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
7926                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
7927         }
7928         lu_site_fini(&o->od_site);
7929         dt_device_fini(&o->od_dt_dev);
7930         OBD_FREE_PTR(o);
7931         RETURN(NULL);
7932 }
7933
7934 static int osd_process_config(const struct lu_env *env,
7935                               struct lu_device *d, struct lustre_cfg *cfg)
7936 {
7937         struct osd_device *o = osd_dev(d);
7938         ssize_t count;
7939         int rc;
7940
7941         ENTRY;
7942
7943         switch (cfg->lcfg_command) {
7944         case LCFG_SETUP:
7945                 rc = osd_mount(env, o, cfg);
7946                 break;
7947         case LCFG_CLEANUP:
7948                 /*
7949                  * For the case LCFG_PRE_CLEANUP is not called in advance,
7950                  * that may happend if hit failure during mount process.
7951                  */
7952                 osd_index_backup(env, o, false);
7953                 lu_dev_del_linkage(d->ld_site, d);
7954                 rc = osd_shutdown(env, o);
7955                 break;
7956         case LCFG_PARAM:
7957                 LASSERT(&o->od_dt_dev);
7958                 count  = class_modify_config(cfg, PARAM_OSD,
7959                                              &o->od_dt_dev.dd_kobj);
7960                 if (count < 0)
7961                         count = class_modify_config(cfg, PARAM_OST,
7962                                                     &o->od_dt_dev.dd_kobj);
7963                 rc = count > 0 ? 0 : count;
7964                 break;
7965         case LCFG_PRE_CLEANUP:
7966                 osd_scrub_stop(o);
7967                 osd_index_backup(env, o,
7968                                  o->od_index_backup_policy != LIBP_NONE);
7969                 rc = 0;
7970                 break;
7971         default:
7972                 rc = -ENOSYS;
7973         }
7974
7975         RETURN(rc);
7976 }
7977
7978 static int osd_recovery_complete(const struct lu_env *env,
7979                                  struct lu_device *d)
7980 {
7981         struct osd_device *osd = osd_dev(d);
7982         int rc = 0;
7983
7984         ENTRY;
7985
7986         if (osd->od_quota_slave_md == NULL && osd->od_quota_slave_dt == NULL)
7987                 RETURN(0);
7988
7989         /*
7990          * start qsd instance on recovery completion, this notifies the quota
7991          * slave code that we are about to process new requests now
7992          */
7993         rc = qsd_start(env, osd->od_quota_slave_dt);
7994         if (rc == 0 && osd->od_quota_slave_md != NULL)
7995                 rc = qsd_start(env, osd->od_quota_slave_md);
7996
7997         RETURN(rc);
7998 }
7999
8000 /*
8001  * we use exports to track all osd users
8002  */
8003 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
8004                            struct obd_device *obd, struct obd_uuid *cluuid,
8005                            struct obd_connect_data *data, void *localdata)
8006 {
8007         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8008         struct lustre_handle conn;
8009         int rc;
8010
8011         ENTRY;
8012
8013         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
8014
8015         rc = class_connect(&conn, obd, cluuid);
8016         if (rc)
8017                 RETURN(rc);
8018
8019         *exp = class_conn2export(&conn);
8020
8021         spin_lock(&osd->od_osfs_lock);
8022         osd->od_connects++;
8023         spin_unlock(&osd->od_osfs_lock);
8024
8025         RETURN(0);
8026 }
8027
8028 /*
8029  * once last export (we don't count self-export) disappeared
8030  * osd can be released
8031  */
8032 static int osd_obd_disconnect(struct obd_export *exp)
8033 {
8034         struct obd_device *obd = exp->exp_obd;
8035         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8036         int rc, release = 0;
8037
8038         ENTRY;
8039
8040         /* Only disconnect the underlying layers on the final disconnect. */
8041         spin_lock(&osd->od_osfs_lock);
8042         osd->od_connects--;
8043         if (osd->od_connects == 0)
8044                 release = 1;
8045         spin_unlock(&osd->od_osfs_lock);
8046
8047         rc = class_disconnect(exp); /* bz 9811 */
8048
8049         if (rc == 0 && release)
8050                 class_manual_cleanup(obd);
8051         RETURN(rc);
8052 }
8053
8054 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
8055                        struct lu_device *dev)
8056 {
8057         struct osd_device *osd = osd_dev(dev);
8058         struct lr_server_data *lsd =
8059                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
8060         int result = 0;
8061
8062         ENTRY;
8063
8064         if (osd->od_quota_slave_md != NULL) {
8065                 /* set up quota slave objects for inode */
8066                 result = qsd_prepare(env, osd->od_quota_slave_md);
8067                 if (result != 0)
8068                         RETURN(result);
8069         }
8070
8071         if (osd->od_quota_slave_dt != NULL) {
8072                 /* set up quota slave objects for block */
8073                 result = qsd_prepare(env, osd->od_quota_slave_dt);
8074                 if (result != 0)
8075                         RETURN(result);
8076         }
8077
8078
8079         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
8080 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
8081                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
8082                         osd->od_index_in_idif = 1;
8083                 } else {
8084                         osd->od_index_in_idif = 0;
8085                         result = osd_register_proc_index_in_idif(osd);
8086                         if (result != 0)
8087                                 RETURN(result);
8088                 }
8089 #else
8090                 osd->od_index_in_idif = 1;
8091 #endif
8092         }
8093
8094         result = osd_fid_init(env, osd);
8095
8096         RETURN(result);
8097 }
8098
8099 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
8100                          struct lu_fid *fid, struct md_op_data *op_data)
8101 {
8102         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
8103
8104         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
8105 }
8106
8107 static const struct lu_object_operations osd_lu_obj_ops = {
8108         .loo_object_init      = osd_object_init,
8109         .loo_object_delete    = osd_object_delete,
8110         .loo_object_release   = osd_object_release,
8111         .loo_object_free      = osd_object_free,
8112         .loo_object_print     = osd_object_print,
8113         .loo_object_invariant = osd_object_invariant
8114 };
8115
8116 const struct lu_device_operations osd_lu_ops = {
8117         .ldo_object_alloc      = osd_object_alloc,
8118         .ldo_process_config    = osd_process_config,
8119         .ldo_recovery_complete = osd_recovery_complete,
8120         .ldo_prepare           = osd_prepare,
8121 };
8122
8123 static const struct lu_device_type_operations osd_device_type_ops = {
8124         .ldto_init = osd_type_init,
8125         .ldto_fini = osd_type_fini,
8126
8127         .ldto_start = osd_type_start,
8128         .ldto_stop  = osd_type_stop,
8129
8130         .ldto_device_alloc = osd_device_alloc,
8131         .ldto_device_free  = osd_device_free,
8132
8133         .ldto_device_init = osd_device_init,
8134         .ldto_device_fini = osd_device_fini
8135 };
8136
8137 static struct lu_device_type osd_device_type = {
8138         .ldt_tags     = LU_DEVICE_DT,
8139         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
8140         .ldt_ops      = &osd_device_type_ops,
8141         .ldt_ctx_tags = LCT_LOCAL,
8142 };
8143
8144 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
8145 {
8146         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
8147         struct super_block *sb = osd_sb(osd);
8148
8149         return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
8150 }
8151
8152 /*
8153  * lprocfs legacy support.
8154  */
8155 static const struct obd_ops osd_obd_device_ops = {
8156         .o_owner = THIS_MODULE,
8157         .o_connect      = osd_obd_connect,
8158         .o_disconnect   = osd_obd_disconnect,
8159         .o_fid_alloc    = osd_fid_alloc,
8160         .o_health_check = osd_health_check,
8161 };
8162
8163 static ssize_t track_declares_assert_show(struct kobject *kobj,
8164                                    struct attribute *attr,
8165                                    char *buf)
8166 {
8167         return sprintf(buf, "%d\n", ldiskfs_track_declares_assert);
8168 }
8169
8170 static ssize_t track_declares_assert_store(struct kobject *kobj,
8171                                            struct attribute *attr,
8172                                            const char *buffer, size_t count)
8173 {
8174         bool track_declares_assert;
8175         int rc;
8176
8177         rc = kstrtobool(buffer, &track_declares_assert);
8178         if (rc)
8179                 return rc;
8180
8181         ldiskfs_track_declares_assert = track_declares_assert;
8182
8183         return count;
8184 }
8185 LUSTRE_RW_ATTR(track_declares_assert);
8186
8187 static int __init osd_init(void)
8188 {
8189         struct kobject *kobj;
8190         int rc;
8191
8192         BUILD_BUG_ON(BH_DXLock >=
8193                      sizeof(((struct buffer_head *)0)->b_state) * 8);
8194 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
8195         /* please, try to keep osd_thread_info smaller than a page */
8196         BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
8197 #endif
8198
8199         osd_oi_mod_init();
8200
8201         rc = lu_kmem_init(ldiskfs_caches);
8202         if (rc)
8203                 return rc;
8204
8205 #ifdef CONFIG_KALLSYMS
8206         priv_security_file_alloc =
8207                 (void *)kallsyms_lookup_name("security_file_alloc");
8208         priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
8209         priv_dev_check_rdonly =
8210                 (void *)kallsyms_lookup_name("dev_check_rdonly");
8211 #endif
8212
8213         rc = class_register_type(&osd_obd_device_ops, NULL, true, NULL,
8214                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
8215         if (rc) {
8216                 lu_kmem_fini(ldiskfs_caches);
8217                 return rc;
8218         }
8219
8220         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8221         if (kobj) {
8222                 rc = sysfs_create_file(kobj,
8223                                        &lustre_attr_track_declares_assert.attr);
8224                 kobject_put(kobj);
8225                 if (rc) {
8226                         CWARN("osd-ldiskfs: track_declares_assert failed to register with sysfs\n");
8227                         rc = 0;
8228                 }
8229         }
8230         return rc;
8231 }
8232
8233 static void __exit osd_exit(void)
8234 {
8235         struct kobject *kobj;
8236
8237         kobj = kset_find_obj(lustre_kset, LUSTRE_OSD_LDISKFS_NAME);
8238         if (kobj) {
8239                 sysfs_remove_file(kobj,
8240                                   &lustre_attr_track_declares_assert.attr);
8241                 kobject_put(kobj);
8242         }
8243         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
8244         lu_kmem_fini(ldiskfs_caches);
8245 }
8246
8247 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
8248 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
8249 MODULE_VERSION(LUSTRE_VERSION_STRING);
8250 MODULE_LICENSE("GPL");
8251
8252 module_init(osd_init);
8253 module_exit(osd_exit);