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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd/osd_handler.c
  37  *
  38  * Top-level entry points into osd module
  39  *
  40  * Author: Nikita Danilov <nikita@clusterfs.com>
  41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
  42  */
  43
  44 #define DEBUG_SUBSYSTEM S_OSD
  45
  46 #include <linux/module.h>
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51
  52 /* LUSTRE_VERSION_CODE */
  53 #include <lustre_ver.h>
  54 /* prerequisite for linux/xattr.h */
  55 #include <linux/types.h>
  56 /* prerequisite for linux/xattr.h */
  57 #include <linux/fs.h>
  58 /* XATTR_{REPLACE,CREATE} */
  59 #include <linux/xattr.h>
  60
  61 #include <ldiskfs/ldiskfs.h>
  62 #include <ldiskfs/xattr.h>
  63 #undef ENTRY
  64 /*
  65  * struct OBD_{ALLOC,FREE}*()
  66  * OBD_FAIL_CHECK
  67  */
  68 #include <obd_support.h>
  69 /* struct ptlrpc_thread */
  70 #include <lustre_net.h>
  71 #include <lustre_fid.h>
  72 /* process_config */
  73 #include <lustre_param.h>
  74
  75 #include "osd_internal.h"
  76 #include "osd_dynlocks.h"
  77
  78 /* llo_* api support */
  79 #include <md_object.h>
  80 #include <lustre_quota.h>
  81
  82 #include <lustre_linkea.h>
  83
  84 int ldiskfs_pdo = 1;
  85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  86                 "ldiskfs with parallel directory operations");
  87
  88 int ldiskfs_track_declares_assert;
  89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
  90                 "LBUG during tracking of declares");
  91
  92 /* Slab to allocate dynlocks */
  93 struct kmem_cache *dynlock_cachep;
  94
  95 /* Slab to allocate osd_it_ea */
  96 struct kmem_cache *osd_itea_cachep;
  97
  98 static struct lu_kmem_descr ldiskfs_caches[] = {
  99         {
 100                 .ckd_cache = &dynlock_cachep,
 101                 .ckd_name  = "dynlock_cache",
 102                 .ckd_size  = sizeof(struct dynlock_handle)
 103         },
 104         {
 105                 .ckd_cache = &osd_itea_cachep,
 106                 .ckd_name  = "osd_itea_cache",
 107                 .ckd_size  = sizeof(struct osd_it_ea)
 108         },
 109         {
 110                 .ckd_cache = NULL
 111         }
 112 };
 113
 114 static const char dot[] = ".";
 115 static const char dotdot[] = "..";
 116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
 117
 118 static const struct lu_object_operations      osd_lu_obj_ops;
 119 static const struct dt_object_operations      osd_obj_ops;
 120 static const struct dt_object_operations      osd_obj_ea_ops;
 121 static const struct dt_object_operations      osd_obj_otable_it_ops;
 122 static const struct dt_index_operations       osd_index_iam_ops;
 123 static const struct dt_index_operations       osd_index_ea_ops;
 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
 153 /*
 154  * Concurrency: doesn't matter
 155  */
 156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 157 {
 158         struct osd_thread_info *oti = osd_oti_get(env);
 159         return oti->oti_w_locks > 0 && o->oo_owner == env;
 160 }
 161
 162 /*
 163  * Concurrency: doesn't access mutable data
 164  */
 165 static int osd_root_get(const struct lu_env *env,
 166                         struct dt_device *dev, struct lu_fid *f)
 167 {
 168         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 169         return 0;
 170 }
 171
 172 /*
 173  * OSD object methods.
 174  */
 175
 176 /*
 177  * Concurrency: no concurrent access is possible that early in object
 178  * life-cycle.
 179  */
 180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 181                                           const struct lu_object_header *hdr,
 182                                           struct lu_device *d)
 183 {
 184         struct osd_object *mo;
 185
 186         OBD_ALLOC_PTR(mo);
 187         if (mo != NULL) {
 188                 struct lu_object *l;
 189
 190                 l = &mo->oo_dt.do_lu;
 191                 dt_object_init(&mo->oo_dt, NULL, d);
 192                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 193                 l->lo_ops = &osd_lu_obj_ops;
 194                 init_rwsem(&mo->oo_sem);
 195                 init_rwsem(&mo->oo_ext_idx_sem);
 196                 spin_lock_init(&mo->oo_guard);
 197                 return l;
 198         } else {
 199                 return NULL;
 200         }
 201 }
 202
 203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 204                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 205 {
 206         int rc;
 207
 208         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 209         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 210                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 211         if (rc > 0) {
 212                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
 213                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 214                 rc = 0;
 215                 lustre_lma_swab(lma);
 216                 /* Check LMA compatibility */
 217                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
 218                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
 219                               "for fid = "DFID", ino = %lu\n",
 220                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 221                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 222                               PFID(&lma->lma_self_fid), inode->i_ino);
 223                         rc = -EOPNOTSUPP;
 224                 }
 225         } else if (rc == 0) {
 226                 rc = -ENODATA;
 227         }
 228
 229         return rc;
 230 }
 231
 232 /*
 233  * retrieve object from backend ext fs.
 234  **/
 235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 236                        struct osd_inode_id *id)
 237 {
 238         struct inode *inode = NULL;
 239
 240         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 241         if (IS_ERR(inode)) {
 242                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 243                        id->oii_ino, PTR_ERR(inode));
 244         } else if (id->oii_gen != OSD_OII_NOGEN &&
 245                    inode->i_generation != id->oii_gen) {
 246                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
 247                        "i_generation = %u\n",
 248                        id->oii_ino, id->oii_gen, inode->i_generation);
 249                 iput(inode);
 250                 inode = ERR_PTR(-ESTALE);
 251         } else if (inode->i_nlink == 0) {
 252                 /* due to parallel readdir and unlink,
 253                 * we can have dead inode here. */
 254                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 255                 iput(inode);
 256                 inode = ERR_PTR(-ESTALE);
 257         } else if (is_bad_inode(inode)) {
 258                 CWARN("%.16s: bad inode: ino = %u\n",
 259                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 260                 iput(inode);
 261                 inode = ERR_PTR(-ENOENT);
 262         } else {
 263                 if (id->oii_gen == OSD_OII_NOGEN)
 264                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 265
 266                 /* Do not update file c/mtime in ldiskfs.
 267                  * NB: we don't have any lock to protect this because we don't
 268                  * have reference on osd_object now, but contention with
 269                  * another lookup + attr_set can't happen in the tiny window
 270                  * between if (...) and set S_NOCMTIME. */
 271                 if (!(inode->i_flags & S_NOCMTIME))
 272                         inode->i_flags |= S_NOCMTIME;
 273         }
 274         return inode;
 275 }
 276
 277 static struct inode *
 278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 279              struct osd_inode_id *id, struct lu_fid *fid)
 280 {
 281         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 282         struct inode            *inode;
 283         int                      rc;
 284
 285         inode = osd_iget(info, dev, id);
 286         if (IS_ERR(inode))
 287                 return inode;
 288
 289         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 290         if (rc == 0) {
 291                 *fid = lma->lma_self_fid;
 292         } else if (rc == -ENODATA) {
 293                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 294                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 295                 else
 296                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 297         } else {
 298                 iput(inode);
 299                 inode = ERR_PTR(rc);
 300         }
 301         return inode;
 302 }
 303
 304 /**
 305  * \retval +v: new filter_fid, does not contain self-fid
 306  * \retval 0:  filter_fid_old, contains self-fid
 307  * \retval -v: other failure cases
 308  */
 309 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
 310                  struct dentry *dentry, struct lu_fid *fid)
 311 {
 312         struct filter_fid_old   *ff     = &info->oti_ff;
 313         struct ost_id           *ostid  = &info->oti_ostid;
 314         int                      rc;
 315
 316         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
 317         if (rc == sizeof(*ff)) {
 318                 rc = 0;
 319                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
 320                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
 321                 /* XXX: should use real OST index in the future. LU-3569 */
 322                 ostid_to_fid(fid, ostid, 0);
 323         } else if (rc == sizeof(struct filter_fid)) {
 324                 rc = 1;
 325         } else if (rc >= 0) {
 326                 rc = -EINVAL;
 327         }
 328
 329         return rc;
 330 }
 331
 332 static int osd_lma_self_repair(struct osd_thread_info *info,
 333                                struct osd_device *osd, struct inode *inode,
 334                                const struct lu_fid *fid, __u32 compat)
 335 {
 336         handle_t *jh;
 337         int       rc;
 338
 339         LASSERT(current->journal_info == NULL);
 340
 341         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
 342                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
 343         if (IS_ERR(jh)) {
 344                 rc = PTR_ERR(jh);
 345                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
 346                       osd_name(osd), rc);
 347                 return rc;
 348         }
 349
 350         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
 351         if (rc != 0)
 352                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
 353                       osd_name(osd), rc);
 354         ldiskfs_journal_stop(jh);
 355         return rc;
 356 }
 357
 358 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
 359 {
 360         struct osd_thread_info  *info   = osd_oti_get(env);
 361         struct osd_device       *osd    = osd_obj2dev(obj);
 362         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
 363         struct inode            *inode  = obj->oo_inode;
 364         struct dentry           *dentry = &info->oti_obj_dentry;
 365         struct lu_fid           *fid    = NULL;
 366         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
 367         int                      rc;
 368         ENTRY;
 369
 370         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
 371         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
 372                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
 373         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
 374                 fid = &lma->lma_self_fid;
 375                 rc = osd_get_idif(info, inode, dentry, fid);
 376                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
 377                         /* For the given OST-object, if it has neither LMA nor
 378                          * FID in XATTR_NAME_FID, then the given FID (which is
 379                          * contained in the @obj, from client RPC for locating
 380                          * the OST-object) is trusted. We use it to generate
 381                          * the LMA. */
 382                         osd_lma_self_repair(info, osd, inode, rfid,
 383                                             LMAC_FID_ON_OST);
 384                         RETURN(0);
 385                 }
 386         }
 387
 388         if (unlikely(rc == -ENODATA))
 389                 RETURN(0);
 390
 391         if (rc < 0)
 392                 RETURN(rc);
 393
 394         if (rc > 0) {
 395                 rc = 0;
 396                 lustre_lma_swab(lma);
 397                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
 398                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
 399                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
 400                               "fid = "DFID", ino = %lu\n", osd_name(osd),
 401                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
 402                               PFID(rfid), inode->i_ino);
 403                         rc = -EOPNOTSUPP;
 404                 } else {
 405                         fid = &lma->lma_self_fid;
 406                 }
 407         }
 408
 409         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
 410                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
 411                         struct ost_id   *oi   = &info->oti_ostid;
 412                         struct lu_fid   *fid1 = &info->oti_fid3;
 413                         __u32            idx  = fid_idif_ost_idx(rfid);
 414
 415                         /* For old IDIF, the OST index is not part of the IDIF,
 416                          * Means that different OSTs may have the same IDIFs.
 417                          * Under such case, we need to make some compatible
 418                          * check to make sure to trigger OI scrub properly. */
 419                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
 420                                 /* Given @rfid is new, LMA is old. */
 421                                 fid_to_ostid(fid, oi);
 422                                 ostid_to_fid(fid1, oi, idx);
 423                                 if (lu_fid_eq(fid1, rfid)) {
 424                                         if (osd->od_index_in_idif)
 425                                                 osd_lma_self_repair(info, osd,
 426                                                         inode, rfid,
 427                                                         LMAC_FID_ON_OST);
 428                                         RETURN(0);
 429                                 }
 430                         }
 431                 }
 432
 433                 rc = -EREMCHG;
 434         }
 435
 436         RETURN(rc);
 437 }
 438
 439 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 440                           const struct lu_fid *fid,
 441                           const struct lu_object_conf *conf)
 442 {
 443         struct osd_thread_info *info;
 444         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 445         struct osd_device      *dev;
 446         struct osd_idmap_cache *oic;
 447         struct osd_inode_id    *id;
 448         struct inode           *inode;
 449         struct osd_scrub       *scrub;
 450         struct scrub_file      *sf;
 451         int                     result;
 452         int                     saved  = 0;
 453         bool                    in_oi  = false;
 454         bool                    in_cache = false;
 455         bool                    triggered = false;
 456         ENTRY;
 457
 458         LINVRNT(osd_invariant(obj));
 459         LASSERT(obj->oo_inode == NULL);
 460         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
 461
 462         dev = osd_dev(ldev);
 463         scrub = &dev->od_scrub;
 464         sf = &scrub->os_file;
 465         info = osd_oti_get(env);
 466         LASSERT(info);
 467         oic = &info->oti_cache;
 468
 469         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 470                 RETURN(-ENOENT);
 471
 472         /* For the object is created as locking anchor, or for the object to
 473          * be created on disk. No need to osd_oi_lookup() at here because FID
 474          * shouldn't never be re-used, if it's really a duplicate FID from
 475          * unexpected reason, we should be able to detect it later by calling
 476          * do_create->osd_oi_insert(). */
 477         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 478                 GOTO(out, result = 0);
 479
 480         /* Search order: 1. per-thread cache. */
 481         if (lu_fid_eq(fid, &oic->oic_fid) &&
 482             likely(oic->oic_dev == dev)) {
 483                 id = &oic->oic_lid;
 484                 in_cache = true;
 485                 goto iget;
 486         }
 487
 488         id = &info->oti_id;
 489         if (!list_empty(&scrub->os_inconsistent_items)) {
 490                 /* Search order: 2. OI scrub pending list. */
 491                 result = osd_oii_lookup(dev, fid, id);
 492                 if (result == 0)
 493                         goto iget;
 494         }
 495
 496         /* Search order: 3. OI files. */
 497         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 498         if (result == -ENOENT) {
 499                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
 500                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 501                                       sf->sf_oi_bitmap))
 502                         GOTO(out, result = 0);
 503
 504                 goto trigger;
 505         }
 506
 507         if (result != 0)
 508                 GOTO(out, result);
 509
 510         in_oi = true;
 511
 512 iget:
 513         inode = osd_iget(info, dev, id);
 514         if (IS_ERR(inode)) {
 515                 result = PTR_ERR(inode);
 516                 if (result != -ENOENT && result != -ESTALE)
 517                         GOTO(out, result);
 518
 519                 if (in_cache)
 520                         fid_zero(&oic->oic_fid);
 521
 522                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 523                 if (result != 0)
 524                         GOTO(out, result = (result == -ENOENT ? 0 : result));
 525
 526                 /* The OI mapping is there, but the inode is NOT there.
 527                  * Two possible cases for that:
 528                  *
 529                  * 1) Backup/restore caused the OI invalid.
 530                  * 2) Someone unlinked the object but NOT removed
 531                  *    the OI mapping, such as mount target device
 532                  *    as ldiskfs, and modify something directly.
 533                  *
 534                  * Generally, when the device is mounted, it will
 535                  * auto check whether the system is restored from
 536                  * file-level backup or not. We trust such detect
 537                  * to distinguish the 1st case from the 2nd case. */
 538                 if (!(scrub->os_file.sf_flags & SF_INCONSISTENT))
 539                         GOTO(out, result = 0);
 540
 541 trigger:
 542                 if (unlikely(triggered))
 543                         GOTO(out, result = saved);
 544
 545                 triggered = true;
 546                 if (thread_is_running(&scrub->os_thread)) {
 547                         result = -EINPROGRESS;
 548                 } else if (!dev->od_noscrub) {
 549                         /* Since we do not know the right OI mapping, we have
 550                          * to trigger OI scrub to scan the whole device. */
 551                         result = osd_scrub_start(dev, SS_AUTO_FULL |
 552                                 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
 553                         CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
 554                                "scrub by RPC for "DFID", rc = %d [1]\n",
 555                                osd_name(dev), PFID(fid), result);
 556                         if (result == 0 || result == -EALREADY)
 557                                 result = -EINPROGRESS;
 558                         else
 559                                 result = -EREMCHG;
 560                 }
 561
 562                 /* We still have chance to get the valid inode: for the
 563                  * object which is referenced by remote name entry, the
 564                  * object on the local MDT will be linked under the dir
 565                  * of "/REMOTE_PARENT_DIR" with its FID string as name.
 566                  *
 567                  * We do not know whether the object for the given FID
 568                  * is referenced by some remote name entry or not, and
 569                  * especially for DNE II, a multiple-linked object may
 570                  * have many name entries reside on many MDTs.
 571                  *
 572                  * To simplify the operation, OSD will not distinguish
 573                  * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
 574                  * only happened for the RPC from other MDT during the
 575                  * OI scrub, or for the client side RPC with FID only,
 576                  * such as FID to path, or from old connected client. */
 577                 saved = result;
 578                 result = osd_lookup_in_remote_parent(info, dev, fid, id);
 579                 if (result == 0) {
 580                         in_oi = false;
 581                         goto iget;
 582                 }
 583
 584                 GOTO(out, result = saved);
 585         }
 586
 587         obj->oo_inode = inode;
 588         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 589
 590         result = osd_check_lma(env, obj);
 591         if (result != 0) {
 592                 iput(inode);
 593                 obj->oo_inode = NULL;
 594
 595                 if (result != -EREMCHG)
 596                         GOTO(out, result);
 597
 598                 if (in_cache)
 599                         fid_zero(&oic->oic_fid);
 600
 601                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
 602                 if (result == 0)
 603                         goto trigger;
 604
 605                 if (result != -ENOENT)
 606                         GOTO(out, result);
 607
 608                 if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) &&
 609                     ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
 610                                      sf->sf_oi_bitmap))
 611                         goto trigger;
 612
 613                 GOTO(out, result = 0);
 614         }
 615
 616         obj->oo_compat_dot_created = 1;
 617         obj->oo_compat_dotdot_created = 1;
 618
 619         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 620                 GOTO(out, result = 0);
 621
 622         LASSERT(obj->oo_hl_head == NULL);
 623         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 624         if (obj->oo_hl_head == NULL) {
 625                 obj->oo_inode = NULL;
 626                 iput(inode);
 627                 GOTO(out, result = -ENOMEM);
 628         }
 629         GOTO(out, result = 0);
 630
 631 out:
 632         LINVRNT(osd_invariant(obj));
 633         return result;
 634 }
 635
 636 /*
 637  * Concurrency: shouldn't matter.
 638  */
 639 static void osd_object_init0(struct osd_object *obj)
 640 {
 641         LASSERT(obj->oo_inode != NULL);
 642         obj->oo_dt.do_body_ops = &osd_body_ops;
 643         obj->oo_dt.do_lu.lo_header->loh_attr |=
 644                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 645 }
 646
 647 /*
 648  * Concurrency: no concurrent access is possible that early in object
 649  * life-cycle.
 650  */
 651 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 652                            const struct lu_object_conf *conf)
 653 {
 654         struct osd_object *obj = osd_obj(l);
 655         int result;
 656
 657         LINVRNT(osd_invariant(obj));
 658
 659         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 660                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 661                 l->lo_header->loh_attr |= LOHA_EXISTS;
 662                 return 0;
 663         }
 664
 665         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 666         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 667         if (result == 0 && obj->oo_inode != NULL)
 668                 osd_object_init0(obj);
 669
 670         LINVRNT(osd_invariant(obj));
 671         return result;
 672 }
 673
 674 /*
 675  * Concurrency: no concurrent access is possible that late in object
 676  * life-cycle.
 677  */
 678 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 679 {
 680         struct osd_object *obj = osd_obj(l);
 681
 682         LINVRNT(osd_invariant(obj));
 683
 684         dt_object_fini(&obj->oo_dt);
 685         if (obj->oo_hl_head != NULL)
 686                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 687         OBD_FREE_PTR(obj);
 688 }
 689
 690 /*
 691  * Concurrency: no concurrent access is possible that late in object
 692  * life-cycle.
 693  */
 694 static void osd_index_fini(struct osd_object *o)
 695 {
 696         struct iam_container *bag;
 697
 698         if (o->oo_dir != NULL) {
 699                 bag = &o->oo_dir->od_container;
 700                 if (o->oo_inode != NULL) {
 701                         if (bag->ic_object == o->oo_inode)
 702                                 iam_container_fini(bag);
 703                 }
 704                 OBD_FREE_PTR(o->oo_dir);
 705                 o->oo_dir = NULL;
 706         }
 707 }
 708
 709 /*
 710  * Concurrency: no concurrent access is possible that late in object
 711  * life-cycle (for all existing callers, that is. New callers have to provide
 712  * their own locking.)
 713  */
 714 static int osd_inode_unlinked(const struct inode *inode)
 715 {
 716         return inode->i_nlink == 0;
 717 }
 718
 719 enum {
 720         OSD_TXN_OI_DELETE_CREDITS    = 20,
 721         OSD_TXN_INODE_DELETE_CREDITS = 20
 722 };
 723
 724 /*
 725  * Journal
 726  */
 727
 728 #if OSD_THANDLE_STATS
 729 /**
 730  * Set time when the handle is allocated
 731  */
 732 static void osd_th_alloced(struct osd_thandle *oth)
 733 {
 734         oth->oth_alloced = cfs_time_current();
 735 }
 736
 737 /**
 738  * Set time when the handle started
 739  */
 740 static void osd_th_started(struct osd_thandle *oth)
 741 {
 742         oth->oth_started = cfs_time_current();
 743 }
 744
 745 /**
 746  * Helper function to convert time interval to microseconds packed in
 747  * long int.
 748  */
 749 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 750 {
 751         struct timeval val;
 752
 753         cfs_duration_usec(cfs_time_sub(end, start), &val);
 754         return val.tv_sec * 1000000 + val.tv_usec;
 755 }
 756
 757 /**
 758  * Check whether the we deal with this handle for too long.
 759  */
 760 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 761                                 cfs_time_t alloced, cfs_time_t started,
 762                                 cfs_time_t closed)
 763 {
 764         cfs_time_t now = cfs_time_current();
 765
 766         LASSERT(dev != NULL);
 767
 768         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 769                             interval_to_usec(alloced, started));
 770         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 771                             interval_to_usec(started, closed));
 772         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 773                             interval_to_usec(closed, now));
 774
 775         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 776                 CWARN("transaction handle %p was open for too long: "
 777                       "now "CFS_TIME_T" ,"
 778                       "alloced "CFS_TIME_T" ,"
 779                       "started "CFS_TIME_T" ,"
 780                       "closed "CFS_TIME_T"\n",
 781                       oth, now, alloced, started, closed);
 782                 libcfs_debug_dumpstack(NULL);
 783         }
 784 }
 785
 786 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 787 {                                                                       \
 788         cfs_time_t __closed = cfs_time_current();                       \
 789         cfs_time_t __alloced = oth->oth_alloced;                        \
 790         cfs_time_t __started = oth->oth_started;                        \
 791                                                                         \
 792         expr;                                                           \
 793         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 794 }
 795
 796 #else /* OSD_THANDLE_STATS */
 797
 798 #define osd_th_alloced(h)                  do {} while(0)
 799 #define osd_th_started(h)                  do {} while(0)
 800 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 801
 802 #endif /* OSD_THANDLE_STATS */
 803
 804 /*
 805  * Concurrency: doesn't access mutable data.
 806  */
 807 static int osd_param_is_not_sane(const struct osd_device *dev,
 808                                  const struct thandle *th)
 809 {
 810         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 811
 812         return oh->ot_credits > osd_transaction_size(dev);
 813 }
 814
 815 /*
 816  * Concurrency: shouldn't matter.
 817  */
 818 static void osd_trans_commit_cb(struct super_block *sb,
 819                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 820 {
 821         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 822         struct thandle     *th  = &oh->ot_super;
 823         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 824         struct dt_txn_commit_cb *dcb, *tmp;
 825
 826         LASSERT(oh->ot_handle == NULL);
 827
 828         if (error)
 829                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 830
 831         dt_txn_hook_commit(th);
 832
 833         /* call per-transaction callbacks if any */
 834         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
 835                                  dcb_linkage) {
 836                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 837                          "commit callback entry: magic=%x name='%s'\n",
 838                          dcb->dcb_magic, dcb->dcb_name);
 839                 list_del_init(&dcb->dcb_linkage);
 840                 dcb->dcb_func(NULL, th, dcb, error);
 841         }
 842
 843         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
 844         lu_device_put(lud);
 845         th->th_dev = NULL;
 846
 847         lu_context_exit(&th->th_ctx);
 848         lu_context_fini(&th->th_ctx);
 849         OBD_FREE_PTR(oh);
 850 }
 851
 852 static struct thandle *osd_trans_create(const struct lu_env *env,
 853                                         struct dt_device *d)
 854 {
 855         struct osd_thread_info  *oti = osd_oti_get(env);
 856         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
 857         struct osd_thandle      *oh;
 858         struct thandle          *th;
 859         ENTRY;
 860
 861         /* on pending IO in this thread should left from prev. request */
 862         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
 863
 864         th = ERR_PTR(-ENOMEM);
 865         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
 866         if (oh != NULL) {
 867                 oh->ot_quota_trans = &oti->oti_quota_trans;
 868                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 869                 th = &oh->ot_super;
 870                 th->th_dev = d;
 871                 th->th_result = 0;
 872                 th->th_tags = LCT_TX_HANDLE;
 873                 oh->ot_credits = 0;
 874                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
 875                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
 876                 osd_th_alloced(oh);
 877
 878                 memset(oti->oti_declare_ops, 0,
 879                        sizeof(oti->oti_declare_ops));
 880                 memset(oti->oti_declare_ops_cred, 0,
 881                        sizeof(oti->oti_declare_ops_cred));
 882                 memset(oti->oti_declare_ops_used, 0,
 883                        sizeof(oti->oti_declare_ops_used));
 884         }
 885         RETURN(th);
 886 }
 887
 888 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
 889 {
 890         struct osd_thread_info  *oti = osd_oti_get(env);
 891         struct osd_thandle      *oh;
 892
 893         oh = container_of0(th, struct osd_thandle, ot_super);
 894         LASSERT(oh != NULL);
 895
 896         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
 897               oti->oti_declare_ops[OSD_OT_CREATE],
 898               oti->oti_declare_ops_cred[OSD_OT_CREATE],
 899               oti->oti_declare_ops_used[OSD_OT_CREATE],
 900               oti->oti_declare_ops[OSD_OT_DESTROY],
 901               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
 902               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
 903         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
 904               oti->oti_declare_ops[OSD_OT_ATTR_SET],
 905               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
 906               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
 907               oti->oti_declare_ops[OSD_OT_XATTR_SET],
 908               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
 909               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
 910         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
 911               oti->oti_declare_ops[OSD_OT_WRITE],
 912               oti->oti_declare_ops_cred[OSD_OT_WRITE],
 913               oti->oti_declare_ops_used[OSD_OT_WRITE],
 914               oti->oti_declare_ops[OSD_OT_PUNCH],
 915               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
 916               oti->oti_declare_ops_used[OSD_OT_PUNCH],
 917               oti->oti_declare_ops[OSD_OT_QUOTA],
 918               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
 919               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
 920         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
 921               oti->oti_declare_ops[OSD_OT_INSERT],
 922               oti->oti_declare_ops_cred[OSD_OT_INSERT],
 923               oti->oti_declare_ops_used[OSD_OT_INSERT],
 924               oti->oti_declare_ops[OSD_OT_DELETE],
 925               oti->oti_declare_ops_cred[OSD_OT_DELETE],
 926               oti->oti_declare_ops_used[OSD_OT_DELETE]);
 927         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
 928               oti->oti_declare_ops[OSD_OT_REF_ADD],
 929               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
 930               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
 931               oti->oti_declare_ops[OSD_OT_REF_DEL],
 932               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
 933               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
 934 }
 935
 936 /*
 937  * Concurrency: shouldn't matter.
 938  */
 939 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 940                            struct thandle *th)
 941 {
 942         struct osd_thread_info *oti = osd_oti_get(env);
 943         struct osd_device  *dev = osd_dt_dev(d);
 944         handle_t           *jh;
 945         struct osd_thandle *oh;
 946         int rc;
 947
 948         ENTRY;
 949
 950         LASSERT(current->journal_info == NULL);
 951
 952         oh = container_of0(th, struct osd_thandle, ot_super);
 953         LASSERT(oh != NULL);
 954         LASSERT(oh->ot_handle == NULL);
 955
 956         rc = dt_txn_hook_start(env, d, th);
 957         if (rc != 0)
 958                 GOTO(out, rc);
 959
 960         if (unlikely(osd_param_is_not_sane(dev, th))) {
 961                 static unsigned long last_printed;
 962                 static int last_credits;
 963
 964                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 965                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 966                       oh->ot_credits,
 967                       osd_journal(dev)->j_max_transaction_buffers);
 968
 969                 osd_trans_dump_creds(env, th);
 970
 971                 if (last_credits != oh->ot_credits &&
 972                     time_after(jiffies, last_printed +
 973                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
 974                         libcfs_debug_dumpstack(NULL);
 975                         last_credits = oh->ot_credits;
 976                         last_printed = jiffies;
 977                 }
 978                 /* XXX Limit the credits to 'max_transaction_buffers', and
 979                  *     let the underlying filesystem to catch the error if
 980                  *     we really need so many credits.
 981                  *
 982                  *     This should be removed when we can calculate the
 983                  *     credits precisely. */
 984                 oh->ot_credits = osd_transaction_size(dev);
 985         }
 986
 987         /*
 988          * XXX temporary stuff. Some abstraction layer should
 989          * be used.
 990          */
 991         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
 992         osd_th_started(oh);
 993         if (!IS_ERR(jh)) {
 994                 oh->ot_handle = jh;
 995                 LASSERT(oti->oti_txns == 0);
 996                 lu_context_init(&th->th_ctx, th->th_tags);
 997                 lu_context_enter(&th->th_ctx);
 998
 999                 lu_device_get(&d->dd_lu_dev);
1000                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1001                               "osd-tx", th);
1002                 oti->oti_txns++;
1003                 rc = 0;
1004         } else {
1005                 rc = PTR_ERR(jh);
1006         }
1007 out:
1008         RETURN(rc);
1009 }
1010
1011 static int osd_seq_exists(const struct lu_env *env,
1012                           struct osd_device *osd, u64 seq)
1013 {
1014         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1015         struct seq_server_site  *ss = osd_seq_site(osd);
1016         int                     rc;
1017         ENTRY;
1018
1019         LASSERT(ss != NULL);
1020         LASSERT(ss->ss_server_fld != NULL);
1021
1022         rc = osd_fld_lookup(env, osd, seq, range);
1023         if (rc != 0) {
1024                 if (rc != -ENOENT)
1025                         CERROR("%s: can't lookup FLD sequence "LPX64
1026                                ": rc = %d\n", osd_name(osd), seq, rc);
1027                 RETURN(0);
1028         }
1029
1030         RETURN(ss->ss_node_id == range->lsr_index);
1031 }
1032
1033 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1034 {
1035         struct dt_txn_commit_cb *dcb;
1036         struct dt_txn_commit_cb *tmp;
1037
1038         /* call per-transaction stop callbacks if any */
1039         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1040                                  dcb_linkage) {
1041                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1042                          "commit callback entry: magic=%x name='%s'\n",
1043                          dcb->dcb_magic, dcb->dcb_name);
1044                 list_del_init(&dcb->dcb_linkage);
1045                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1046         }
1047 }
1048
1049 /*
1050  * Concurrency: shouldn't matter.
1051  */
1052 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1053                           struct thandle *th)
1054 {
1055         int                     rc = 0;
1056         struct osd_thandle     *oh;
1057         struct osd_thread_info *oti = osd_oti_get(env);
1058         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1059         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1060         struct qsd_instance    *qsd = osd->od_quota_slave;
1061         struct lquota_trans    *qtrans;
1062         ENTRY;
1063
1064         oh = container_of0(th, struct osd_thandle, ot_super);
1065
1066         qtrans = oh->ot_quota_trans;
1067         oh->ot_quota_trans = NULL;
1068
1069         if (oh->ot_handle != NULL) {
1070                 handle_t *hdl = oh->ot_handle;
1071
1072                 /*
1073                  * add commit callback
1074                  * notice we don't do this in osd_trans_start()
1075                  * as underlying transaction can change during truncate
1076                  */
1077                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1078                                          &oh->ot_jcb);
1079
1080                 LASSERT(oti->oti_txns == 1);
1081                 oti->oti_txns--;
1082
1083                 rc = dt_txn_hook_stop(env, th);
1084                 if (rc != 0)
1085                         CERROR("%s: failed in transaction hook: rc = %d\n",
1086                                osd_name(osd), rc);
1087
1088                 osd_trans_stop_cb(oh, rc);
1089                 /* hook functions might modify th_sync */
1090                 hdl->h_sync = th->th_sync;
1091
1092                 oh->ot_handle = NULL;
1093                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1094                 if (rc != 0)
1095                         CERROR("%s: failed to stop transaction: rc = %d\n",
1096                                osd_name(osd), rc);
1097         } else {
1098                 osd_trans_stop_cb(oh, th->th_result);
1099                 OBD_FREE_PTR(oh);
1100         }
1101
1102         /* inform the quota slave device that the transaction is stopping */
1103         qsd_op_end(env, qsd, qtrans);
1104
1105         /* as we want IO to journal and data IO be concurrent, we don't block
1106          * awaiting data IO completion in osd_do_bio(), instead we wait here
1107          * once transaction is submitted to the journal. all reqular requests
1108          * don't do direct IO (except read/write), thus this wait_event becomes
1109          * no-op for them.
1110          *
1111          * IMPORTANT: we have to wait till any IO submited by the thread is
1112          * completed otherwise iobuf may be corrupted by different request
1113          */
1114         wait_event(iobuf->dr_wait,
1115                        atomic_read(&iobuf->dr_numreqs) == 0);
1116         osd_fini_iobuf(osd, iobuf);
1117         if (!rc)
1118                 rc = iobuf->dr_error;
1119
1120         RETURN(rc);
1121 }
1122
1123 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1124 {
1125         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1126                                                ot_super);
1127
1128         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1129         LASSERT(&dcb->dcb_func != NULL);
1130         if (dcb->dcb_flags & DCB_TRANS_STOP)
1131                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1132         else
1133                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1134
1135         return 0;
1136 }
1137
1138 /*
1139  * Called just before object is freed. Releases all resources except for
1140  * object itself (that is released by osd_object_free()).
1141  *
1142  * Concurrency: no concurrent access is possible that late in object
1143  * life-cycle.
1144  */
1145 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1146 {
1147         struct osd_object *obj   = osd_obj(l);
1148         struct inode      *inode = obj->oo_inode;
1149
1150         LINVRNT(osd_invariant(obj));
1151
1152         /*
1153          * If object is unlinked remove fid->ino mapping from object index.
1154          */
1155
1156         osd_index_fini(obj);
1157         if (inode != NULL) {
1158                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1159                 qid_t                    uid = i_uid_read(inode);
1160                 qid_t                    gid = i_gid_read(inode);
1161
1162                 iput(inode);
1163                 obj->oo_inode = NULL;
1164
1165                 if (qsd != NULL) {
1166                         struct osd_thread_info  *info = osd_oti_get(env);
1167                         struct lquota_id_info   *qi = &info->oti_qi;
1168
1169                         /* Release granted quota to master if necessary */
1170                         qi->lqi_id.qid_uid = uid;
1171                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1172
1173                         qi->lqi_id.qid_uid = gid;
1174                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1175                 }
1176         }
1177 }
1178
1179 /*
1180  * Concurrency: ->loo_object_release() is called under site spin-lock.
1181  */
1182 static void osd_object_release(const struct lu_env *env,
1183                                struct lu_object *l)
1184 {
1185 }
1186
1187 /*
1188  * Concurrency: shouldn't matter.
1189  */
1190 static int osd_object_print(const struct lu_env *env, void *cookie,
1191                             lu_printer_t p, const struct lu_object *l)
1192 {
1193         struct osd_object *o = osd_obj(l);
1194         struct iam_descr  *d;
1195
1196         if (o->oo_dir != NULL)
1197                 d = o->oo_dir->od_container.ic_descr;
1198         else
1199                 d = NULL;
1200         return (*p)(env, cookie,
1201                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1202                     o, o->oo_inode,
1203                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1204                     o->oo_inode ? o->oo_inode->i_generation : 0,
1205                     d ? d->id_ops->id_name : "plain");
1206 }
1207
1208 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1209
1210 /*
1211  * Concurrency: shouldn't matter.
1212  */
1213 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1214                struct obd_statfs *sfs)
1215 {
1216         struct osd_device  *osd = osd_dt_dev(d);
1217         struct super_block *sb = osd_sb(osd);
1218         struct kstatfs     *ksfs;
1219         int result = 0;
1220
1221         if (unlikely(osd->od_mnt == NULL))
1222                 return -EINPROGRESS;
1223
1224         /* osd_lproc.c call this without env, allocate ksfs for that case */
1225         if (unlikely(env == NULL)) {
1226                 OBD_ALLOC_PTR(ksfs);
1227                 if (ksfs == NULL)
1228                         return -ENOMEM;
1229         } else {
1230                 ksfs = &osd_oti_get(env)->oti_ksfs;
1231         }
1232
1233         spin_lock(&osd->od_osfs_lock);
1234         result = sb->s_op->statfs(sb->s_root, ksfs);
1235         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1236                 statfs_pack(sfs, ksfs);
1237                 if (unlikely(sb->s_flags & MS_RDONLY))
1238                         sfs->os_state = OS_STATE_READONLY;
1239                 if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1240                                               LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1241                         sfs->os_maxbytes = sb->s_maxbytes;
1242                 else
1243                         sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1244         }
1245         spin_unlock(&osd->od_osfs_lock);
1246
1247         if (unlikely(env == NULL))
1248                 OBD_FREE_PTR(ksfs);
1249
1250         /* Reserve a small amount of space for local objects like last_rcvd,
1251          * llog, quota files, ... */
1252         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1253                 sfs->os_bavail = 0;
1254         } else {
1255                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1256                 /** Take out metadata overhead for indirect blocks */
1257                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1258         }
1259
1260         return result;
1261 }
1262
1263 /**
1264  * Estimate space needed for file creations. We assume the largest filename
1265  * which is 2^64 - 1, hence a filename of 20 chars.
1266  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1267  */
1268 #ifdef __LDISKFS_DIR_REC_LEN
1269 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1270 #else
1271 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1272 #endif
1273
1274 /*
1275  * Concurrency: doesn't access mutable data.
1276  */
1277 static void osd_conf_get(const struct lu_env *env,
1278                          const struct dt_device *dev,
1279                          struct dt_device_param *param)
1280 {
1281         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1282         int                ea_overhead;
1283
1284         /*
1285          * XXX should be taken from not-yet-existing fs abstraction layer.
1286          */
1287         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1288         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1289         param->ddp_block_shift  = sb->s_blocksize_bits;
1290         param->ddp_mount_type     = LDD_MT_LDISKFS;
1291         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1292                 param->ddp_maxbytes = sb->s_maxbytes;
1293         else
1294                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1295         /* Overhead estimate should be fairly accurate, so we really take a tiny
1296          * error margin which also avoids fragmenting the filesystem too much */
1297         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1298         /* inode are statically allocated, so per-inode space consumption
1299          * is the space consumed by the directory entry */
1300         param->ddp_inodespace     = PER_OBJ_USAGE;
1301         /* per-fragment overhead to be used by the client code */
1302         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1303         param->ddp_mntopts      = 0;
1304         if (test_opt(sb, XATTR_USER))
1305                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1306         if (test_opt(sb, POSIX_ACL))
1307                 param->ddp_mntopts |= MNTOPT_ACL;
1308
1309         /* LOD might calculate the max stripe count based on max_ea_size,
1310          * so we need take account in the overhead as well,
1311          * xattr_header + magic + xattr_entry_head */
1312         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1313                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1314
1315 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1316         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1317                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1318                                                                 ea_overhead;
1319         else
1320 #endif
1321                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1322 }
1323
1324 /*
1325  * Concurrency: shouldn't matter.
1326  */
1327 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1328 {
1329         int rc;
1330
1331         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1332
1333         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1334
1335         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1336                LUSTRE_OSD_LDISKFS_NAME, rc);
1337
1338         return rc;
1339 }
1340
1341 /**
1342  * Start commit for OSD device.
1343  *
1344  * An implementation of dt_commit_async method for OSD device.
1345  * Asychronously starts underlayng fs sync and thereby a transaction
1346  * commit.
1347  *
1348  * \param env environment
1349  * \param d dt device
1350  *
1351  * \see dt_device_operations
1352  */
1353 static int osd_commit_async(const struct lu_env *env,
1354                             struct dt_device *d)
1355 {
1356         struct super_block *s = osd_sb(osd_dt_dev(d));
1357         ENTRY;
1358
1359         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1360         RETURN(s->s_op->sync_fs(s, 0));
1361 }
1362
1363 /*
1364  * Concurrency: shouldn't matter.
1365  */
1366
1367 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1368 {
1369         struct super_block *sb = osd_sb(osd_dt_dev(d));
1370         struct block_device *dev = sb->s_bdev;
1371 #ifdef HAVE_DEV_SET_RDONLY
1372         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1373         int rc = 0;
1374 #else
1375         int rc = -EOPNOTSUPP;
1376 #endif
1377         ENTRY;
1378
1379 #ifdef HAVE_DEV_SET_RDONLY
1380         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1381
1382         if (jdev && (jdev != dev)) {
1383                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1384                        (long)jdev);
1385                 dev_set_rdonly(jdev);
1386         }
1387         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1388         dev_set_rdonly(dev);
1389 #else
1390         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1391                osd_dt_dev(d)->od_svname, (long)dev, rc);
1392 #endif
1393         RETURN(rc);
1394 }
1395
1396 /**
1397  * Note: we do not count into QUOTA here.
1398  * If we mount with --data_journal we may need more.
1399  */
1400 const int osd_dto_credits_noquota[DTO_NR] = {
1401         /**
1402          * Insert.
1403          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1404          * SINGLEDATA_TRANS_BLOCKS(8)
1405          * XXX Note: maybe iam need more, since iam have more level than
1406          *           EXT3 htree.
1407          */
1408         [DTO_INDEX_INSERT]  = 16,
1409         /**
1410          * Delete
1411          * just modify a single entry, probably merge few within a block
1412          */
1413         [DTO_INDEX_DELETE]  = 1,
1414         /**
1415          * Used for OI scrub
1416          */
1417         [DTO_INDEX_UPDATE]  = 16,
1418         /**
1419          * 4(inode, inode bits, groups, GDT)
1420          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1421          */
1422         [DTO_OBJECT_CREATE] = 4,
1423         /**
1424          * 4(inode, inode bits, groups, GDT)
1425          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1426          */
1427         [DTO_OBJECT_DELETE] = 4,
1428         /**
1429          * Attr set credits (inode)
1430          */
1431         [DTO_ATTR_SET_BASE] = 1,
1432         /**
1433          * Xattr set. The same as xattr of EXT3.
1434          * DATA_TRANS_BLOCKS(14)
1435          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1436          * are also counted in. Do not know why?
1437          */
1438         [DTO_XATTR_SET]     = 14,
1439         /**
1440          * credits for inode change during write.
1441          */
1442         [DTO_WRITE_BASE]    = 3,
1443         /**
1444          * credits for single block write.
1445          */
1446         [DTO_WRITE_BLOCK]   = 14,
1447         /**
1448          * Attr set credits for chown.
1449          * This is extra credits for setattr, and it is null without quota
1450          */
1451         [DTO_ATTR_SET_CHOWN] = 0
1452 };
1453
1454 static const struct dt_device_operations osd_dt_ops = {
1455         .dt_root_get       = osd_root_get,
1456         .dt_statfs         = osd_statfs,
1457         .dt_trans_create   = osd_trans_create,
1458         .dt_trans_start    = osd_trans_start,
1459         .dt_trans_stop     = osd_trans_stop,
1460         .dt_trans_cb_add   = osd_trans_cb_add,
1461         .dt_conf_get       = osd_conf_get,
1462         .dt_sync           = osd_sync,
1463         .dt_ro             = osd_ro,
1464         .dt_commit_async   = osd_commit_async,
1465 };
1466
1467 static void osd_object_read_lock(const struct lu_env *env,
1468                                  struct dt_object *dt, unsigned role)
1469 {
1470         struct osd_object *obj = osd_dt_obj(dt);
1471         struct osd_thread_info *oti = osd_oti_get(env);
1472
1473         LINVRNT(osd_invariant(obj));
1474
1475         LASSERT(obj->oo_owner != env);
1476         down_read_nested(&obj->oo_sem, role);
1477
1478         LASSERT(obj->oo_owner == NULL);
1479         oti->oti_r_locks++;
1480 }
1481
1482 static void osd_object_write_lock(const struct lu_env *env,
1483                                   struct dt_object *dt, unsigned role)
1484 {
1485         struct osd_object *obj = osd_dt_obj(dt);
1486         struct osd_thread_info *oti = osd_oti_get(env);
1487
1488         LINVRNT(osd_invariant(obj));
1489
1490         LASSERT(obj->oo_owner != env);
1491         down_write_nested(&obj->oo_sem, role);
1492
1493         LASSERT(obj->oo_owner == NULL);
1494         obj->oo_owner = env;
1495         oti->oti_w_locks++;
1496 }
1497
1498 static void osd_object_read_unlock(const struct lu_env *env,
1499                                    struct dt_object *dt)
1500 {
1501         struct osd_object *obj = osd_dt_obj(dt);
1502         struct osd_thread_info *oti = osd_oti_get(env);
1503
1504         LINVRNT(osd_invariant(obj));
1505
1506         LASSERT(oti->oti_r_locks > 0);
1507         oti->oti_r_locks--;
1508         up_read(&obj->oo_sem);
1509 }
1510
1511 static void osd_object_write_unlock(const struct lu_env *env,
1512                                     struct dt_object *dt)
1513 {
1514         struct osd_object *obj = osd_dt_obj(dt);
1515         struct osd_thread_info *oti = osd_oti_get(env);
1516
1517         LINVRNT(osd_invariant(obj));
1518
1519         LASSERT(obj->oo_owner == env);
1520         LASSERT(oti->oti_w_locks > 0);
1521         oti->oti_w_locks--;
1522         obj->oo_owner = NULL;
1523         up_write(&obj->oo_sem);
1524 }
1525
1526 static int osd_object_write_locked(const struct lu_env *env,
1527                                    struct dt_object *dt)
1528 {
1529         struct osd_object *obj = osd_dt_obj(dt);
1530
1531         LINVRNT(osd_invariant(obj));
1532
1533         return obj->oo_owner == env;
1534 }
1535
1536 static struct timespec *osd_inode_time(const struct lu_env *env,
1537                                        struct inode *inode, __u64 seconds)
1538 {
1539         struct osd_thread_info  *oti = osd_oti_get(env);
1540         struct timespec         *t   = &oti->oti_time;
1541
1542         t->tv_sec = seconds;
1543         t->tv_nsec = 0;
1544         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1545         return t;
1546 }
1547
1548
1549 static void osd_inode_getattr(const struct lu_env *env,
1550                               struct inode *inode, struct lu_attr *attr)
1551 {
1552         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1553                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1554                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1555                            LA_TYPE;
1556
1557         attr->la_atime   = LTIME_S(inode->i_atime);
1558         attr->la_mtime   = LTIME_S(inode->i_mtime);
1559         attr->la_ctime   = LTIME_S(inode->i_ctime);
1560         attr->la_mode    = inode->i_mode;
1561         attr->la_size    = i_size_read(inode);
1562         attr->la_blocks  = inode->i_blocks;
1563         attr->la_uid     = i_uid_read(inode);
1564         attr->la_gid     = i_gid_read(inode);
1565         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1566         attr->la_nlink   = inode->i_nlink;
1567         attr->la_rdev    = inode->i_rdev;
1568         attr->la_blksize = 1 << inode->i_blkbits;
1569         attr->la_blkbits = inode->i_blkbits;
1570 }
1571
1572 static int osd_attr_get(const struct lu_env *env,
1573                         struct dt_object *dt,
1574                         struct lu_attr *attr)
1575 {
1576         struct osd_object *obj = osd_dt_obj(dt);
1577
1578         if (!dt_object_exists(dt))
1579                 return -ENOENT;
1580
1581         LASSERT(!dt_object_remote(dt));
1582         LINVRNT(osd_invariant(obj));
1583
1584         spin_lock(&obj->oo_guard);
1585         osd_inode_getattr(env, obj->oo_inode, attr);
1586         spin_unlock(&obj->oo_guard);
1587         return 0;
1588 }
1589
1590 static int osd_declare_attr_set(const struct lu_env *env,
1591                                 struct dt_object *dt,
1592                                 const struct lu_attr *attr,
1593                                 struct thandle *handle)
1594 {
1595         struct osd_thandle     *oh;
1596         struct osd_object      *obj;
1597         struct osd_thread_info *info = osd_oti_get(env);
1598         struct lquota_id_info  *qi = &info->oti_qi;
1599         qid_t                   uid;
1600         qid_t                   gid;
1601         long long               bspace;
1602         int                     rc = 0;
1603         bool                    enforce;
1604         ENTRY;
1605
1606         LASSERT(dt != NULL);
1607         LASSERT(handle != NULL);
1608
1609         obj = osd_dt_obj(dt);
1610         LASSERT(osd_invariant(obj));
1611
1612         oh = container_of0(handle, struct osd_thandle, ot_super);
1613         LASSERT(oh->ot_handle == NULL);
1614
1615         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1616                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1617
1618         if (attr == NULL || obj->oo_inode == NULL)
1619                 RETURN(rc);
1620
1621         bspace   = obj->oo_inode->i_blocks;
1622         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1623         bspace   = toqb(bspace);
1624
1625         /* Changing ownership is always preformed by super user, it should not
1626          * fail with EDQUOT.
1627          *
1628          * We still need to call the osd_declare_qid() to calculate the journal
1629          * credits for updating quota accounting files and to trigger quota
1630          * space adjustment once the operation is completed.*/
1631         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1632                 /* USERQUOTA */
1633                 uid = i_uid_read(obj->oo_inode);
1634                 qi->lqi_type = USRQUOTA;
1635                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1636                 /* inode accounting */
1637                 qi->lqi_is_blk = false;
1638
1639                 /* one more inode for the new uid ... */
1640                 qi->lqi_id.qid_uid = attr->la_uid;
1641                 qi->lqi_space      = 1;
1642                 /* Reserve credits for the new uid */
1643                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1644                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1645                         rc = 0;
1646                 if (rc)
1647                         RETURN(rc);
1648
1649                 /* and one less inode for the current uid */
1650                 qi->lqi_id.qid_uid = uid;
1651                 qi->lqi_space      = -1;
1652                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1653                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1654                         rc = 0;
1655                 if (rc)
1656                         RETURN(rc);
1657
1658                 /* block accounting */
1659                 qi->lqi_is_blk = true;
1660
1661                 /* more blocks for the new uid ... */
1662                 qi->lqi_id.qid_uid = attr->la_uid;
1663                 qi->lqi_space      = bspace;
1664                 /*
1665                  * Credits for the new uid has been reserved, re-use "obj"
1666                  * to save credit reservation.
1667                  */
1668                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1669                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1670                         rc = 0;
1671                 if (rc)
1672                         RETURN(rc);
1673
1674                 /* and finally less blocks for the current uid */
1675                 qi->lqi_id.qid_uid = uid;
1676                 qi->lqi_space      = -bspace;
1677                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1678                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1679                         rc = 0;
1680                 if (rc)
1681                         RETURN(rc);
1682
1683                 /* GROUP QUOTA */
1684                 gid = i_gid_read(obj->oo_inode);
1685                 qi->lqi_type = GRPQUOTA;
1686                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1687
1688                 /* inode accounting */
1689                 qi->lqi_is_blk = false;
1690
1691                 /* one more inode for the new gid ... */
1692                 qi->lqi_id.qid_gid = attr->la_gid;
1693                 qi->lqi_space      = 1;
1694                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1695                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1696                         rc = 0;
1697                 if (rc)
1698                         RETURN(rc);
1699
1700                 /* and one less inode for the current gid */
1701                 qi->lqi_id.qid_gid = gid;
1702                 qi->lqi_space      = -1;
1703                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1704                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1705                         rc = 0;
1706                 if (rc)
1707                         RETURN(rc);
1708
1709                 /* block accounting */
1710                 qi->lqi_is_blk = true;
1711
1712                 /* more blocks for the new gid ... */
1713                 qi->lqi_id.qid_gid = attr->la_gid;
1714                 qi->lqi_space      = bspace;
1715                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1716                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1717                         rc = 0;
1718                 if (rc)
1719                         RETURN(rc);
1720
1721                 /* and finally less blocks for the current gid */
1722                 qi->lqi_id.qid_gid = gid;
1723                 qi->lqi_space      = -bspace;
1724                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1725                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1726                         rc = 0;
1727                 if (rc)
1728                         RETURN(rc);
1729         }
1730
1731         RETURN(rc);
1732 }
1733
1734 static int osd_inode_setattr(const struct lu_env *env,
1735                              struct inode *inode, const struct lu_attr *attr)
1736 {
1737         __u64 bits = attr->la_valid;
1738
1739         /* Only allow set size for regular file */
1740         if (!S_ISREG(inode->i_mode))
1741                 bits &= ~(LA_SIZE | LA_BLOCKS);
1742
1743         if (bits == 0)
1744                 return 0;
1745
1746         if (bits & LA_ATIME)
1747                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1748         if (bits & LA_CTIME)
1749                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1750         if (bits & LA_MTIME)
1751                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1752         if (bits & LA_SIZE) {
1753                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1754                 i_size_write(inode, attr->la_size);
1755         }
1756
1757 #if 0
1758         /* OSD should not change "i_blocks" which is used by quota.
1759          * "i_blocks" should be changed by ldiskfs only. */
1760         if (bits & LA_BLOCKS)
1761                 inode->i_blocks = attr->la_blocks;
1762 #endif
1763         if (bits & LA_MODE)
1764                 inode->i_mode = (inode->i_mode & S_IFMT) |
1765                                 (attr->la_mode & ~S_IFMT);
1766         if (bits & LA_UID)
1767                 i_uid_write(inode, attr->la_uid);
1768         if (bits & LA_GID)
1769                 i_gid_write(inode, attr->la_gid);
1770         if (bits & LA_NLINK)
1771                 set_nlink(inode, attr->la_nlink);
1772         if (bits & LA_RDEV)
1773                 inode->i_rdev = attr->la_rdev;
1774
1775         if (bits & LA_FLAGS) {
1776                 /* always keep S_NOCMTIME */
1777                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1778                                  S_NOCMTIME;
1779         }
1780         return 0;
1781 }
1782
1783 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1784 {
1785         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1786             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1787                 struct iattr    iattr;
1788                 int             rc;
1789
1790                 ll_vfs_dq_init(inode);
1791                 iattr.ia_valid = 0;
1792                 if (attr->la_valid & LA_UID)
1793                         iattr.ia_valid |= ATTR_UID;
1794                 if (attr->la_valid & LA_GID)
1795                         iattr.ia_valid |= ATTR_GID;
1796                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1797                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1798
1799                 rc = ll_vfs_dq_transfer(inode, &iattr);
1800                 if (rc) {
1801                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1802                                "enforcement enabled on the ldiskfs "
1803                                "filesystem?\n", inode->i_sb->s_id, rc);
1804                         return rc;
1805                 }
1806         }
1807         return 0;
1808 }
1809
1810 static int osd_attr_set(const struct lu_env *env,
1811                         struct dt_object *dt,
1812                         const struct lu_attr *attr,
1813                         struct thandle *handle)
1814 {
1815         struct osd_object *obj = osd_dt_obj(dt);
1816         struct inode      *inode;
1817         int rc;
1818
1819         if (!dt_object_exists(dt))
1820                 return -ENOENT;
1821
1822         LASSERT(handle != NULL);
1823         LASSERT(!dt_object_remote(dt));
1824         LASSERT(osd_invariant(obj));
1825
1826         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1827
1828         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1829                 struct osd_thread_info  *oti  = osd_oti_get(env);
1830                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1831                 struct lu_fid           *fid1 = &oti->oti_fid;
1832                 struct osd_inode_id     *id   = &oti->oti_id;
1833                 struct iam_path_descr   *ipd;
1834                 struct iam_container    *bag;
1835                 struct osd_thandle      *oh;
1836                 int                      rc;
1837
1838                 fid_cpu_to_be(fid1, fid0);
1839                 memset(id, 1, sizeof(*id));
1840                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1841                                   fid0)->oi_dir.od_container;
1842                 ipd = osd_idx_ipd_get(env, bag);
1843                 if (unlikely(ipd == NULL))
1844                         RETURN(-ENOMEM);
1845
1846                 oh = container_of0(handle, struct osd_thandle, ot_super);
1847                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1848                                 (const struct iam_rec *)id, ipd);
1849                 osd_ipd_put(env, bag, ipd);
1850                 return(rc > 0 ? 0 : rc);
1851         }
1852
1853         inode = obj->oo_inode;
1854
1855         rc = osd_quota_transfer(inode, attr);
1856         if (rc)
1857                 return rc;
1858
1859         spin_lock(&obj->oo_guard);
1860         rc = osd_inode_setattr(env, inode, attr);
1861         spin_unlock(&obj->oo_guard);
1862
1863         if (!rc)
1864                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1865
1866         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
1867
1868         return rc;
1869 }
1870
1871 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1872                                            struct osd_object *obj,
1873                                            const char *name, const int namelen)
1874 {
1875         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1876 }
1877
1878 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1879                       umode_t mode, struct dt_allocation_hint *hint,
1880                       struct thandle *th)
1881 {
1882         int result;
1883         struct osd_device  *osd = osd_obj2dev(obj);
1884         struct osd_thandle *oth;
1885         struct dt_object   *parent = NULL;
1886         struct inode       *inode;
1887
1888         LINVRNT(osd_invariant(obj));
1889         LASSERT(obj->oo_inode == NULL);
1890         LASSERT(obj->oo_hl_head == NULL);
1891
1892         if (S_ISDIR(mode) && ldiskfs_pdo) {
1893                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1894                 if (obj->oo_hl_head == NULL)
1895                         return -ENOMEM;
1896         }
1897
1898         oth = container_of(th, struct osd_thandle, ot_super);
1899         LASSERT(oth->ot_handle->h_transaction != NULL);
1900
1901         if (hint != NULL && hint->dah_parent != NULL &&
1902             !dt_object_remote(hint->dah_parent))
1903                 parent = hint->dah_parent;
1904
1905         inode = ldiskfs_create_inode(oth->ot_handle,
1906                                      parent ? osd_dt_obj(parent)->oo_inode :
1907                                               osd_sb(osd)->s_root->d_inode,
1908                                      mode);
1909         if (!IS_ERR(inode)) {
1910                 /* Do not update file c/mtime in ldiskfs. */
1911                 inode->i_flags |= S_NOCMTIME;
1912
1913                 /* For new created object, it must be consistent,
1914                  * and it is unnecessary to scrub against it. */
1915                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1916
1917                 obj->oo_inode = inode;
1918                 result = 0;
1919         } else {
1920                 if (obj->oo_hl_head != NULL) {
1921                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1922                         obj->oo_hl_head = NULL;
1923                 }
1924                 result = PTR_ERR(inode);
1925         }
1926         LINVRNT(osd_invariant(obj));
1927         return result;
1928 }
1929
1930 enum {
1931         OSD_NAME_LEN = 255
1932 };
1933
1934 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1935                      struct lu_attr *attr,
1936                      struct dt_allocation_hint *hint,
1937                      struct dt_object_format *dof,
1938                      struct thandle *th)
1939 {
1940         int result;
1941         struct osd_thandle *oth;
1942         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1943
1944         LASSERT(S_ISDIR(attr->la_mode));
1945
1946         oth = container_of(th, struct osd_thandle, ot_super);
1947         LASSERT(oth->ot_handle->h_transaction != NULL);
1948         result = osd_mkfile(info, obj, mode, hint, th);
1949
1950         return result;
1951 }
1952
1953 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1954                         struct lu_attr *attr,
1955                         struct dt_allocation_hint *hint,
1956                         struct dt_object_format *dof,
1957                         struct thandle *th)
1958 {
1959         int result;
1960         struct osd_thandle *oth;
1961         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1962
1963         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1964
1965         LASSERT(S_ISREG(attr->la_mode));
1966
1967         oth = container_of(th, struct osd_thandle, ot_super);
1968         LASSERT(oth->ot_handle->h_transaction != NULL);
1969
1970         result = osd_mkfile(info, obj, mode, hint, th);
1971         if (result == 0) {
1972                 LASSERT(obj->oo_inode != NULL);
1973                 if (feat->dif_flags & DT_IND_VARKEY)
1974                         result = iam_lvar_create(obj->oo_inode,
1975                                                  feat->dif_keysize_max,
1976                                                  feat->dif_ptrsize,
1977                                                  feat->dif_recsize_max,
1978                                                  oth->ot_handle);
1979                 else
1980                         result = iam_lfix_create(obj->oo_inode,
1981                                                  feat->dif_keysize_max,
1982                                                  feat->dif_ptrsize,
1983                                                  feat->dif_recsize_max,
1984                                                  oth->ot_handle);
1985
1986         }
1987         return result;
1988 }
1989
1990 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1991                      struct lu_attr *attr,
1992                      struct dt_allocation_hint *hint,
1993                      struct dt_object_format *dof,
1994                      struct thandle *th)
1995 {
1996         LASSERT(S_ISREG(attr->la_mode));
1997         return osd_mkfile(info, obj, (attr->la_mode &
1998                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1999 }
2000
2001 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2002                      struct lu_attr *attr,
2003                      struct dt_allocation_hint *hint,
2004                      struct dt_object_format *dof,
2005                      struct thandle *th)
2006 {
2007         LASSERT(S_ISLNK(attr->la_mode));
2008         return osd_mkfile(info, obj, (attr->la_mode &
2009                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2010 }
2011
2012 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2013                      struct lu_attr *attr,
2014                      struct dt_allocation_hint *hint,
2015                      struct dt_object_format *dof,
2016                      struct thandle *th)
2017 {
2018         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2019         int result;
2020
2021         LINVRNT(osd_invariant(obj));
2022         LASSERT(obj->oo_inode == NULL);
2023         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2024                 S_ISFIFO(mode) || S_ISSOCK(mode));
2025
2026         result = osd_mkfile(info, obj, mode, hint, th);
2027         if (result == 0) {
2028                 LASSERT(obj->oo_inode != NULL);
2029                 /*
2030                  * This inode should be marked dirty for i_rdev.  Currently
2031                  * that is done in the osd_attr_init().
2032                  */
2033                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2034                                    attr->la_rdev);
2035         }
2036         LINVRNT(osd_invariant(obj));
2037         return result;
2038 }
2039
2040 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2041                               struct lu_attr *,
2042                               struct dt_allocation_hint *hint,
2043                               struct dt_object_format *dof,
2044                               struct thandle *);
2045
2046 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2047 {
2048         osd_obj_type_f result;
2049
2050         switch (type) {
2051         case DFT_DIR:
2052                 result = osd_mkdir;
2053                 break;
2054         case DFT_REGULAR:
2055                 result = osd_mkreg;
2056                 break;
2057         case DFT_SYM:
2058                 result = osd_mksym;
2059                 break;
2060         case DFT_NODE:
2061                 result = osd_mknod;
2062                 break;
2063         case DFT_INDEX:
2064                 result = osd_mk_index;
2065                 break;
2066
2067         default:
2068                 LBUG();
2069                 break;
2070         }
2071         return result;
2072 }
2073
2074
2075 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2076                         struct dt_object *parent, struct dt_object *child,
2077                         umode_t child_mode)
2078 {
2079         LASSERT(ah);
2080
2081         ah->dah_parent = parent;
2082         ah->dah_mode = child_mode;
2083 }
2084
2085 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2086                           struct lu_attr *attr, struct dt_object_format *dof)
2087 {
2088         struct inode   *inode = obj->oo_inode;
2089         __u64           valid = attr->la_valid;
2090         int             result;
2091
2092         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2093
2094         if (dof->dof_type != DFT_NODE)
2095                 attr->la_valid &= ~LA_RDEV;
2096         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2097                 attr->la_valid &= ~LA_ATIME;
2098         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2099                 attr->la_valid &= ~LA_CTIME;
2100         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2101                 attr->la_valid &= ~LA_MTIME;
2102
2103         result = osd_quota_transfer(inode, attr);
2104         if (result)
2105                 return;
2106
2107         if (attr->la_valid != 0) {
2108                 result = osd_inode_setattr(info->oti_env, inode, attr);
2109                 /*
2110                  * The osd_inode_setattr() should always succeed here.  The
2111                  * only error that could be returned is EDQUOT when we are
2112                  * trying to change the UID or GID of the inode. However, this
2113                  * should not happen since quota enforcement is no longer
2114                  * enabled on ldiskfs (lquota takes care of it).
2115                  */
2116                 LASSERTF(result == 0, "%d\n", result);
2117                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2118         }
2119
2120         attr->la_valid = valid;
2121 }
2122
2123 /**
2124  * Helper function for osd_object_create()
2125  *
2126  * \retval 0, on success
2127  */
2128 static int __osd_object_create(struct osd_thread_info *info,
2129                                struct osd_object *obj, struct lu_attr *attr,
2130                                struct dt_allocation_hint *hint,
2131                                struct dt_object_format *dof,
2132                                struct thandle *th)
2133 {
2134         int     result;
2135         __u32   umask;
2136
2137         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2138
2139         /* we drop umask so that permissions we pass are not affected */
2140         umask = current->fs->umask;
2141         current->fs->umask = 0;
2142
2143         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2144                                                   th);
2145         if (likely(obj->oo_inode != NULL)) {
2146                 LASSERT(obj->oo_inode->i_state & I_NEW);
2147
2148                 /* Unlock the inode before attr initialization to avoid
2149                  * unnecessary dqget operations. LU-6378 */
2150                 unlock_new_inode(obj->oo_inode);
2151         }
2152
2153         if (likely(result == 0)) {
2154                 osd_attr_init(info, obj, attr, dof);
2155                 osd_object_init0(obj);
2156         }
2157
2158         /* restore previous umask value */
2159         current->fs->umask = umask;
2160
2161         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2162
2163         return result;
2164 }
2165
2166 /**
2167  * Helper function for osd_object_create()
2168  *
2169  * \retval 0, on success
2170  */
2171 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2172                            const struct lu_fid *fid, struct thandle *th)
2173 {
2174         struct osd_thread_info *info = osd_oti_get(env);
2175         struct osd_inode_id    *id   = &info->oti_id;
2176         struct osd_device      *osd  = osd_obj2dev(obj);
2177         struct osd_thandle     *oh;
2178         int                     rc;
2179
2180         LASSERT(obj->oo_inode != NULL);
2181
2182         oh = container_of0(th, struct osd_thandle, ot_super);
2183         LASSERT(oh->ot_handle);
2184         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2185
2186         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2187         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2188         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2189
2190         return rc;
2191 }
2192
2193 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2194                    u64 seq, struct lu_seq_range *range)
2195 {
2196         struct seq_server_site  *ss = osd_seq_site(osd);
2197
2198         if (fid_seq_is_idif(seq)) {
2199                 fld_range_set_ost(range);
2200                 range->lsr_index = idif_ost_idx(seq);
2201                 return 0;
2202         }
2203
2204         if (!fid_seq_in_fldb(seq)) {
2205                 fld_range_set_mdt(range);
2206                 if (ss != NULL)
2207                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2208                          * at all */
2209                         range->lsr_index = ss->ss_node_id;
2210                 return 0;
2211         }
2212
2213         LASSERT(ss != NULL);
2214         fld_range_set_any(range);
2215         /* OSD will only do local fld lookup */
2216         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2217 }
2218
2219 /*
2220  * Concurrency: no external locking is necessary.
2221  */
2222 static int osd_declare_object_create(const struct lu_env *env,
2223                                      struct dt_object *dt,
2224                                      struct lu_attr *attr,
2225                                      struct dt_allocation_hint *hint,
2226                                      struct dt_object_format *dof,
2227                                      struct thandle *handle)
2228 {
2229         struct osd_thandle      *oh;
2230         int                      rc;
2231         ENTRY;
2232
2233         LASSERT(handle != NULL);
2234
2235         oh = container_of0(handle, struct osd_thandle, ot_super);
2236         LASSERT(oh->ot_handle == NULL);
2237
2238         /* EA object consumes more credits than regular object: osd_mk_index
2239          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
2240          * leaf_node, could involves the block, block bitmap, groups, GDT
2241          * change for each block, so add 4 * 2 credits in that case. */
2242         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2243                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
2244                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
2245         /* Reuse idle OI block may cause additional one OI block
2246          * to be changed. */
2247         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2248                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2249
2250         if (!attr)
2251                 RETURN(0);
2252
2253         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2254                                    osd_dt_obj(dt), false, NULL, false);
2255         if (rc != 0)
2256                 RETURN(rc);
2257
2258         RETURN(rc);
2259 }
2260
2261 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2262                              struct lu_attr *attr,
2263                              struct dt_allocation_hint *hint,
2264                              struct dt_object_format *dof, struct thandle *th)
2265 {
2266         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2267         struct osd_object       *obj    = osd_dt_obj(dt);
2268         struct osd_thread_info  *info   = osd_oti_get(env);
2269         int result;
2270         ENTRY;
2271
2272         if (dt_object_exists(dt))
2273                 return -EEXIST;
2274
2275         LINVRNT(osd_invariant(obj));
2276         LASSERT(!dt_object_remote(dt));
2277         LASSERT(osd_write_locked(env, obj));
2278         LASSERT(th != NULL);
2279
2280         if (unlikely(fid_is_acct(fid)))
2281                 /* Quota files can't be created from the kernel any more,
2282                  * 'tune2fs -O quota' will take care of creating them */
2283                 RETURN(-EPERM);
2284
2285         result = __osd_object_create(info, obj, attr, hint, dof, th);
2286         if (result == 0)
2287                 result = __osd_oi_insert(env, obj, fid, th);
2288
2289         LASSERT(ergo(result == 0,
2290                 dt_object_exists(dt) && !dt_object_remote(dt)));
2291
2292         LASSERT(osd_invariant(obj));
2293         RETURN(result);
2294 }
2295
2296 /**
2297  * Called to destroy on-disk representation of the object
2298  *
2299  * Concurrency: must be locked
2300  */
2301 static int osd_declare_object_destroy(const struct lu_env *env,
2302                                       struct dt_object *dt,
2303                                       struct thandle *th)
2304 {
2305         struct osd_object  *obj = osd_dt_obj(dt);
2306         struct inode       *inode = obj->oo_inode;
2307         struct osd_thandle *oh;
2308         int                 rc;
2309         ENTRY;
2310
2311         oh = container_of0(th, struct osd_thandle, ot_super);
2312         LASSERT(oh->ot_handle == NULL);
2313         LASSERT(inode);
2314
2315         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2316                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2317         /* Recycle idle OI leaf may cause additional three OI blocks
2318          * to be changed. */
2319         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2320                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2321         /* one less inode */
2322         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2323                                    -1, oh, obj, false, NULL, false);
2324         if (rc)
2325                 RETURN(rc);
2326         /* data to be truncated */
2327         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2328                                    0, oh, obj, true, NULL, false);
2329         RETURN(rc);
2330 }
2331
2332 static int osd_object_destroy(const struct lu_env *env,
2333                               struct dt_object *dt,
2334                               struct thandle *th)
2335 {
2336         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2337         struct osd_object      *obj = osd_dt_obj(dt);
2338         struct inode           *inode = obj->oo_inode;
2339         struct osd_device      *osd = osd_obj2dev(obj);
2340         struct osd_thandle     *oh;
2341         int                     result;
2342         ENTRY;
2343
2344         oh = container_of0(th, struct osd_thandle, ot_super);
2345         LASSERT(oh->ot_handle);
2346         LASSERT(inode);
2347         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2348
2349         if (unlikely(fid_is_acct(fid)))
2350                 RETURN(-EPERM);
2351
2352         if (S_ISDIR(inode->i_mode)) {
2353                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2354                         inode->i_nlink == 2);
2355                 /* it will check/delete the inode from remote parent,
2356                  * how to optimize it? unlink performance impaction XXX */
2357                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2358                 if (result != 0 && result != -ENOENT) {
2359                         CERROR("%s: delete inode "DFID": rc = %d\n",
2360                                osd_name(osd), PFID(fid), result);
2361                 }
2362                 spin_lock(&obj->oo_guard);
2363                 clear_nlink(inode);
2364                 spin_unlock(&obj->oo_guard);
2365                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2366         }
2367
2368         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2369
2370         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2371                                OI_CHECK_FLD);
2372
2373         osd_trans_exec_check(env, th, OSD_OT_DESTROY);
2374         /* XXX: add to ext3 orphan list */
2375         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2376
2377         /* not needed in the cache anymore */
2378         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2379         obj->oo_destroyed = 1;
2380
2381         RETURN(0);
2382 }
2383
2384 /**
2385  * Put the fid into lustre_mdt_attrs, and then place the structure
2386  * inode's ea. This fid should not be altered during the life time
2387  * of the inode.
2388  *
2389  * \retval +ve, on success
2390  * \retval -ve, on error
2391  *
2392  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2393  */
2394 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2395                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2396 {
2397         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2398         int                      rc;
2399         ENTRY;
2400
2401         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2402                 RETURN(0);
2403
2404         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_OST_EA_FID_SET))
2405                 rc = -ENOMEM;
2406
2407         lustre_lma_init(lma, fid, compat, incompat);
2408         lustre_lma_swab(lma);
2409
2410         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2411                              XATTR_CREATE);
2412         /* LMA may already exist, but we need to check that all the
2413          * desired compat/incompat flags have been added. */
2414         if (unlikely(rc == -EEXIST)) {
2415                 if (compat == 0 && incompat == 0)
2416                         RETURN(0);
2417
2418                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2419                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2420                                      LMA_OLD_SIZE);
2421                 if (rc <= 0)
2422                         RETURN(-EINVAL);
2423
2424                 lustre_lma_swab(lma);
2425                 if (!(~lma->lma_compat & compat) &&
2426                     !(~lma->lma_incompat & incompat))
2427                         RETURN(0);
2428
2429                 lma->lma_compat |= compat;
2430                 lma->lma_incompat |= incompat;
2431                 lustre_lma_swab(lma);
2432                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2433                                      sizeof(*lma), XATTR_REPLACE);
2434         }
2435
2436         RETURN(rc);
2437 }
2438
2439 /**
2440  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2441  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2442  * To have compatilibility with 1.8 ldiskfs driver we need to have
2443  * magic number at start of fid data.
2444  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2445  * its inmemory API.
2446  */
2447 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2448                                          const struct lu_fid *fid)
2449 {
2450         if (!fid_is_namespace_visible(fid) ||
2451             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2452                 param->edp_magic = 0;
2453                 return;
2454         }
2455
2456         param->edp_magic = LDISKFS_LUFID_MAGIC;
2457         param->edp_len =  sizeof(struct lu_fid) + 1;
2458         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2459 }
2460
2461 /**
2462  * Try to read the fid from inode ea into dt_rec.
2463  *
2464  * \param fid object fid.
2465  *
2466  * \retval 0 on success
2467  */
2468 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2469                           __u32 ino, struct lu_fid *fid,
2470                           struct osd_inode_id *id)
2471 {
2472         struct osd_thread_info *info  = osd_oti_get(env);
2473         struct inode           *inode;
2474         ENTRY;
2475
2476         osd_id_gen(id, ino, OSD_OII_NOGEN);
2477         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2478         if (IS_ERR(inode))
2479                 RETURN(PTR_ERR(inode));
2480
2481         iput(inode);
2482         RETURN(0);
2483 }
2484
2485 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2486                                         struct inode *dir,
2487                                         struct inode *parent_dir,
2488                                         const struct lu_fid *dot_fid,
2489                                         const struct lu_fid *dot_dot_fid,
2490                                         struct osd_thandle *oth)
2491 {
2492         struct ldiskfs_dentry_param *dot_ldp;
2493         struct ldiskfs_dentry_param *dot_dot_ldp;
2494
2495         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2496         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2497
2498         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2499         dot_ldp->edp_magic = 0;
2500
2501         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2502                                       dir, dot_ldp, dot_dot_ldp);
2503 }
2504
2505 /**
2506  * Create an local agent inode for remote entry
2507  */
2508 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2509                                                   struct osd_device *osd,
2510                                                   struct osd_object *pobj,
2511                                                   const struct lu_fid *fid,
2512                                                   __u32 type,
2513                                                   struct thandle *th)
2514 {
2515         struct osd_thread_info  *info = osd_oti_get(env);
2516         struct inode            *local;
2517         struct osd_thandle      *oh;
2518         int                     rc;
2519         ENTRY;
2520
2521         LASSERT(th);
2522         oh = container_of(th, struct osd_thandle, ot_super);
2523         LASSERT(oh->ot_handle->h_transaction != NULL);
2524
2525         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2526         if (IS_ERR(local)) {
2527                 CERROR("%s: create local error %d\n", osd_name(osd),
2528                        (int)PTR_ERR(local));
2529                 RETURN(local);
2530         }
2531
2532         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2533         unlock_new_inode(local);
2534
2535         /* Set special LMA flag for local agent inode */
2536         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2537         if (rc != 0) {
2538                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2539                        osd_name(osd), PFID(fid), rc);
2540                 RETURN(ERR_PTR(rc));
2541         }
2542
2543         if (!S_ISDIR(type))
2544                 RETURN(local);
2545
2546         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2547                                          lu_object_fid(&pobj->oo_dt.do_lu),
2548                                          fid, oh);
2549         if (rc != 0) {
2550                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2551                         osd_name(osd), PFID(fid), rc);
2552                 RETURN(ERR_PTR(rc));
2553         }
2554
2555         RETURN(local);
2556 }
2557
2558 /**
2559  * Delete local agent inode for remote entry
2560  */
2561 static int osd_delete_local_agent_inode(const struct lu_env *env,
2562                                         struct osd_device *osd,
2563                                         const struct lu_fid *fid,
2564                                         __u32 ino, struct osd_thandle *oh)
2565 {
2566         struct osd_thread_info  *oti = osd_oti_get(env);
2567         struct osd_inode_id     *id = &oti->oti_id;
2568         struct inode            *inode;
2569         ENTRY;
2570
2571         id->oii_ino = le32_to_cpu(ino);
2572         id->oii_gen = OSD_OII_NOGEN;
2573         inode = osd_iget(oti, osd, id);
2574         if (IS_ERR(inode)) {
2575                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2576                        PFID(fid), id->oii_ino, id->oii_gen);
2577                 RETURN(PTR_ERR(inode));
2578         }
2579
2580         clear_nlink(inode);
2581         mark_inode_dirty(inode);
2582         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2583                 osd_name(osd), PFID(fid), inode->i_ino);
2584         iput(inode);
2585         RETURN(0);
2586 }
2587
2588 /**
2589  * OSD layer object create function for interoperability mode (b11826).
2590  * This is mostly similar to osd_object_create(). Only difference being, fid is
2591  * inserted into inode ea here.
2592  *
2593  * \retval   0, on success
2594  * \retval -ve, on error
2595  */
2596 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2597                                 struct lu_attr *attr,
2598                                 struct dt_allocation_hint *hint,
2599                                 struct dt_object_format *dof,
2600                                 struct thandle *th)
2601 {
2602         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2603         struct osd_object       *obj    = osd_dt_obj(dt);
2604         struct osd_thread_info  *info   = osd_oti_get(env);
2605         int                      result;
2606
2607         ENTRY;
2608
2609         if (dt_object_exists(dt))
2610                 RETURN(-EEXIST);
2611
2612         LASSERT(osd_invariant(obj));
2613         LASSERT(!dt_object_remote(dt));
2614         LASSERT(osd_write_locked(env, obj));
2615         LASSERT(th != NULL);
2616
2617         if (unlikely(fid_is_acct(fid)))
2618                 /* Quota files can't be created from the kernel any more,
2619                  * 'tune2fs -O quota' will take care of creating them */
2620                 RETURN(-EPERM);
2621
2622         result = __osd_object_create(info, obj, attr, hint, dof, th);
2623         if (result == 0) {
2624                 if (fid_is_idif(fid) &&
2625                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2626                         struct lu_fid *tfid = &info->oti_fid;
2627                         struct ost_id *oi   = &info->oti_ostid;
2628
2629                         fid_to_ostid(fid, oi);
2630                         ostid_to_fid(tfid, oi, 0);
2631                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2632                                                 LMAC_FID_ON_OST, 0);
2633                 } else {
2634                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2635                                 fid_is_on_ost(info, osd_obj2dev(obj),
2636                                               fid, OI_CHECK_FLD) ?
2637                                 LMAC_FID_ON_OST : 0, 0);
2638                 }
2639         }
2640
2641         if (result == 0)
2642                 result = __osd_oi_insert(env, obj, fid, th);
2643
2644         LASSERT(ergo(result == 0,
2645                      dt_object_exists(dt) && !dt_object_remote(dt)));
2646         LINVRNT(osd_invariant(obj));
2647         RETURN(result);
2648 }
2649
2650 static int osd_declare_object_ref_add(const struct lu_env *env,
2651                                       struct dt_object *dt,
2652                                       struct thandle *handle)
2653 {
2654         struct osd_thandle       *oh;
2655
2656         /* it's possible that object doesn't exist yet */
2657         LASSERT(handle != NULL);
2658
2659         oh = container_of0(handle, struct osd_thandle, ot_super);
2660         LASSERT(oh->ot_handle == NULL);
2661
2662         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2663                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2664
2665         return 0;
2666 }
2667
2668 /*
2669  * Concurrency: @dt is write locked.
2670  */
2671 static int osd_object_ref_add(const struct lu_env *env,
2672                               struct dt_object *dt, struct thandle *th)
2673 {
2674         struct osd_object  *obj = osd_dt_obj(dt);
2675         struct inode       *inode = obj->oo_inode;
2676         struct osd_thandle *oh;
2677         int                 rc = 0;
2678
2679         if (!dt_object_exists(dt))
2680                 return -ENOENT;
2681
2682         LINVRNT(osd_invariant(obj));
2683         LASSERT(!dt_object_remote(dt));
2684         LASSERT(osd_write_locked(env, obj));
2685         LASSERT(th != NULL);
2686
2687         oh = container_of0(th, struct osd_thandle, ot_super);
2688         LASSERT(oh->ot_handle != NULL);
2689
2690         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2691
2692         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2693                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2694         /*
2695          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2696          * (65000) subdirectories by storing "1" in i_nlink if the link count
2697          * would otherwise overflow. Directory tranversal tools understand
2698          * that (st_nlink == 1) indicates that the filesystem dose not track
2699          * hard links count on the directory, and will not abort subdirectory
2700          * scanning early once (st_nlink - 2) subdirs have been found.
2701          *
2702          * This also has to properly handle the case of inodes with nlink == 0
2703          * in case they are being linked into the PENDING directory
2704          */
2705         spin_lock(&obj->oo_guard);
2706         if (unlikely(inode->i_nlink == 0))
2707                 /* inc_nlink from 0 may cause WARN_ON */
2708                 set_nlink(inode, 1);
2709         else {
2710                 ldiskfs_inc_count(oh->ot_handle, inode);
2711                 if (!S_ISDIR(inode->i_mode))
2712                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2713         }
2714         spin_unlock(&obj->oo_guard);
2715
2716         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2717         LINVRNT(osd_invariant(obj));
2718
2719         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2720
2721         return rc;
2722 }
2723
2724 static int osd_declare_object_ref_del(const struct lu_env *env,
2725                                       struct dt_object *dt,
2726                                       struct thandle *handle)
2727 {
2728         struct osd_thandle *oh;
2729
2730         LASSERT(!dt_object_remote(dt));
2731         LASSERT(handle != NULL);
2732
2733         oh = container_of0(handle, struct osd_thandle, ot_super);
2734         LASSERT(oh->ot_handle == NULL);
2735
2736         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2737                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2738
2739         return 0;
2740 }
2741
2742 /*
2743  * Concurrency: @dt is write locked.
2744  */
2745 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2746                               struct thandle *th)
2747 {
2748         struct osd_object       *obj = osd_dt_obj(dt);
2749         struct inode            *inode = obj->oo_inode;
2750         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2751         struct osd_thandle      *oh;
2752
2753         if (!dt_object_exists(dt))
2754                 return -ENOENT;
2755
2756         LINVRNT(osd_invariant(obj));
2757         LASSERT(!dt_object_remote(dt));
2758         LASSERT(osd_write_locked(env, obj));
2759         LASSERT(th != NULL);
2760
2761         oh = container_of0(th, struct osd_thandle, ot_super);
2762         LASSERT(oh->ot_handle != NULL);
2763
2764         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2765
2766         spin_lock(&obj->oo_guard);
2767         /* That can be result of upgrade from old Lustre version and
2768          * applied only to local files.  Just skip this ref_del call.
2769          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2770         if (inode->i_nlink == 0) {
2771                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2772                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2773                              ", maybe an upgraded file? (LU-3915)\n",
2774                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2775                 spin_unlock(&obj->oo_guard);
2776                 return 0;
2777         }
2778
2779         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2780                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2781
2782         ldiskfs_dec_count(oh->ot_handle, inode);
2783         spin_unlock(&obj->oo_guard);
2784
2785         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2786         LINVRNT(osd_invariant(obj));
2787
2788         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
2789
2790         return 0;
2791 }
2792
2793 /*
2794  * Get the 64-bit version for an inode.
2795  */
2796 static int osd_object_version_get(const struct lu_env *env,
2797                                   struct dt_object *dt, dt_obj_version_t *ver)
2798 {
2799         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2800
2801         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2802                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2803         *ver = LDISKFS_I(inode)->i_fs_version;
2804         return 0;
2805 }
2806
2807 /*
2808  * Concurrency: @dt is read locked.
2809  */
2810 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2811                          struct lu_buf *buf, const char *name)
2812 {
2813         struct osd_object      *obj    = osd_dt_obj(dt);
2814         struct inode           *inode  = obj->oo_inode;
2815         struct osd_thread_info *info   = osd_oti_get(env);
2816         struct dentry          *dentry = &info->oti_obj_dentry;
2817
2818         /* version get is not real XATTR but uses xattr API */
2819         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2820                 /* for version we are just using xattr API but change inode
2821                  * field instead */
2822                 if (buf->lb_len == 0)
2823                         return sizeof(dt_obj_version_t);
2824
2825                 if (buf->lb_len < sizeof(dt_obj_version_t))
2826                         return -ERANGE;
2827
2828                 osd_object_version_get(env, dt, buf->lb_buf);
2829
2830                 return sizeof(dt_obj_version_t);
2831         }
2832
2833         if (!dt_object_exists(dt))
2834                 return -ENOENT;
2835
2836         LASSERT(!dt_object_remote(dt));
2837         LASSERT(inode->i_op != NULL);
2838         LASSERT(inode->i_op->getxattr != NULL);
2839
2840         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2841 }
2842
2843
2844 static int osd_declare_xattr_set(const struct lu_env *env,
2845                                  struct dt_object *dt,
2846                                  const struct lu_buf *buf, const char *name,
2847                                  int fl, struct thandle *handle)
2848 {
2849         struct osd_thandle *oh;
2850         int credits;
2851         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2852
2853         LASSERT(handle != NULL);
2854
2855         oh = container_of0(handle, struct osd_thandle, ot_super);
2856         LASSERT(oh->ot_handle == NULL);
2857
2858         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2859                 /* For non-upgrading case, the LMA is set first and
2860                  * usually fit inode. But for upgrade case, the LMA
2861                  * may be in another separated EA block. */
2862                 if (!dt_object_exists(dt))
2863                         credits = 0;
2864                 else if (fl == LU_XATTR_REPLACE)
2865                         credits = 1;
2866                 else
2867                         goto upgrade;
2868         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2869                 credits = 1;
2870         } else {
2871 upgrade:
2872                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2873                 if (buf && buf->lb_len > sb->s_blocksize) {
2874                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2875                                         sb->s_blocksize_bits;
2876                 }
2877                 /*
2878                  * xattr set may involve inode quota change, reserve credits for
2879                  * dquot_initialize()
2880                  */
2881                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2882         }
2883
2884         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2885
2886         return 0;
2887 }
2888
2889 /*
2890  * Set the 64-bit version for object
2891  */
2892 static void osd_object_version_set(const struct lu_env *env,
2893                                    struct dt_object *dt,
2894                                    dt_obj_version_t *new_version)
2895 {
2896         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2897
2898         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2899                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2900
2901         LDISKFS_I(inode)->i_fs_version = *new_version;
2902         /** Version is set after all inode operations are finished,
2903          *  so we should mark it dirty here */
2904         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2905 }
2906
2907 /*
2908  * Concurrency: @dt is write locked.
2909  */
2910 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2911                          const struct lu_buf *buf, const char *name, int fl,
2912                          struct thandle *handle)
2913 {
2914         struct osd_object      *obj      = osd_dt_obj(dt);
2915         struct inode           *inode    = obj->oo_inode;
2916         struct osd_thread_info *info     = osd_oti_get(env);
2917         int                     fs_flags = 0;
2918         int                     rc;
2919         ENTRY;
2920
2921         LASSERT(handle != NULL);
2922
2923         /* version set is not real XATTR */
2924         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2925                 /* for version we are just using xattr API but change inode
2926                  * field instead */
2927                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2928                 osd_object_version_set(env, dt, buf->lb_buf);
2929                 return sizeof(dt_obj_version_t);
2930         }
2931
2932         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2933                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2934
2935         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2936         if (fl & LU_XATTR_REPLACE)
2937                 fs_flags |= XATTR_REPLACE;
2938
2939         if (fl & LU_XATTR_CREATE)
2940                 fs_flags |= XATTR_CREATE;
2941
2942         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2943                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2944                 int                      rc;
2945
2946                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2947                 if (rc != 0)
2948                         RETURN(rc);
2949
2950                 lma->lma_incompat |= LMAI_STRIPED;
2951                 lustre_lma_swab(lma);
2952                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2953                                      sizeof(*lma), XATTR_REPLACE);
2954                 if (rc != 0)
2955                         RETURN(rc);
2956         }
2957
2958         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
2959             strcmp(name, XATTR_NAME_LINK) == 0)
2960                 return -ENOSPC;
2961
2962         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2963                                fs_flags);
2964         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2965
2966         return rc;
2967 }
2968
2969 /*
2970  * Concurrency: @dt is read locked.
2971  */
2972 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2973                           const struct lu_buf *buf)
2974 {
2975         struct osd_object      *obj    = osd_dt_obj(dt);
2976         struct inode           *inode  = obj->oo_inode;
2977         struct osd_thread_info *info   = osd_oti_get(env);
2978         struct dentry          *dentry = &info->oti_obj_dentry;
2979
2980         if (!dt_object_exists(dt))
2981                 return -ENOENT;
2982
2983         LASSERT(!dt_object_remote(dt));
2984         LASSERT(inode->i_op != NULL);
2985         LASSERT(inode->i_op->listxattr != NULL);
2986
2987         dentry->d_inode = inode;
2988         dentry->d_sb = inode->i_sb;
2989         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2990 }
2991
2992 static int osd_declare_xattr_del(const struct lu_env *env,
2993                                  struct dt_object *dt, const char *name,
2994                                  struct thandle *handle)
2995 {
2996         struct osd_thandle *oh;
2997         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2998
2999         LASSERT(!dt_object_remote(dt));
3000         LASSERT(handle != NULL);
3001
3002         oh = container_of0(handle, struct osd_thandle, ot_super);
3003         LASSERT(oh->ot_handle == NULL);
3004
3005         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3006                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3007         /*
3008          * xattr del may involve inode quota change, reserve credits for
3009          * dquot_initialize()
3010          */
3011         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3012
3013         return 0;
3014 }
3015
3016 /*
3017  * Concurrency: @dt is write locked.
3018  */
3019 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3020                          const char *name, struct thandle *handle)
3021 {
3022         struct osd_object      *obj    = osd_dt_obj(dt);
3023         struct inode           *inode  = obj->oo_inode;
3024         struct osd_thread_info *info   = osd_oti_get(env);
3025         struct dentry          *dentry = &info->oti_obj_dentry;
3026         int                     rc;
3027
3028         if (!dt_object_exists(dt))
3029                 return -ENOENT;
3030
3031         LASSERT(!dt_object_remote(dt));
3032         LASSERT(inode->i_op != NULL);
3033         LASSERT(inode->i_op->removexattr != NULL);
3034         LASSERT(handle != NULL);
3035
3036         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3037
3038         ll_vfs_dq_init(inode);
3039         dentry->d_inode = inode;
3040         dentry->d_sb = inode->i_sb;
3041         rc = inode->i_op->removexattr(dentry, name);
3042         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3043         return rc;
3044 }
3045
3046 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3047                            __u64 start, __u64 end)
3048 {
3049         struct osd_object       *obj    = osd_dt_obj(dt);
3050         struct inode            *inode  = obj->oo_inode;
3051         struct osd_thread_info  *info   = osd_oti_get(env);
3052         struct dentry           *dentry = &info->oti_obj_dentry;
3053         struct file             *file   = &info->oti_file;
3054         int                     rc;
3055
3056         ENTRY;
3057
3058         dentry->d_inode = inode;
3059         dentry->d_sb = inode->i_sb;
3060         file->f_path.dentry = dentry;
3061         file->f_mapping = inode->i_mapping;
3062         file->f_op = inode->i_fop;
3063         set_file_inode(file, inode);
3064
3065         rc = ll_vfs_fsync_range(file, start, end, 0);
3066
3067         RETURN(rc);
3068 }
3069
3070 /*
3071  * Index operations.
3072  */
3073
3074 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3075                            const struct dt_index_features *feat)
3076 {
3077         struct iam_descr *descr;
3078
3079         if (osd_object_is_root(o))
3080                 return feat == &dt_directory_features;
3081
3082         LASSERT(o->oo_dir != NULL);
3083
3084         descr = o->oo_dir->od_container.ic_descr;
3085         if (feat == &dt_directory_features) {
3086                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3087                         return 1;
3088                 else
3089                         return 0;
3090         } else {
3091                 return
3092                         feat->dif_keysize_min <= descr->id_key_size &&
3093                         descr->id_key_size <= feat->dif_keysize_max &&
3094                         feat->dif_recsize_min <= descr->id_rec_size &&
3095                         descr->id_rec_size <= feat->dif_recsize_max &&
3096                         !(feat->dif_flags & (DT_IND_VARKEY |
3097                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3098                         ergo(feat->dif_flags & DT_IND_UPDATE,
3099                              1 /* XXX check that object (and file system) is
3100                                 * writable */);
3101         }
3102 }
3103
3104 static int osd_iam_container_init(const struct lu_env *env,
3105                                   struct osd_object *obj,
3106                                   struct osd_directory *dir)
3107 {
3108         struct iam_container *bag = &dir->od_container;
3109         int result;
3110
3111         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3112         if (result != 0)
3113                 return result;
3114
3115         result = iam_container_setup(bag);
3116         if (result == 0)
3117                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3118         else
3119                 iam_container_fini(bag);
3120
3121         return result;
3122 }
3123
3124
3125 /*
3126  * Concurrency: no external locking is necessary.
3127  */
3128 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3129                          const struct dt_index_features *feat)
3130 {
3131         int                      result;
3132         int                      skip_iam = 0;
3133         struct osd_object       *obj = osd_dt_obj(dt);
3134
3135         LINVRNT(osd_invariant(obj));
3136
3137         if (osd_object_is_root(obj)) {
3138                 dt->do_index_ops = &osd_index_ea_ops;
3139                 result = 0;
3140         } else if (feat == &dt_directory_features) {
3141                 dt->do_index_ops = &osd_index_ea_ops;
3142                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3143                         result = 0;
3144                 else
3145                         result = -ENOTDIR;
3146                 skip_iam = 1;
3147         } else if (unlikely(feat == &dt_otable_features)) {
3148                 dt->do_index_ops = &osd_otable_ops;
3149                 return 0;
3150         } else if (unlikely(feat == &dt_acct_features)) {
3151                 dt->do_index_ops = &osd_acct_index_ops;
3152                 result = 0;
3153                 skip_iam = 1;
3154         } else if (!osd_has_index(obj)) {
3155                 struct osd_directory *dir;
3156
3157                 OBD_ALLOC_PTR(dir);
3158                 if (dir != NULL) {
3159
3160                         spin_lock(&obj->oo_guard);
3161                         if (obj->oo_dir == NULL)
3162                                 obj->oo_dir = dir;
3163                         else
3164                                 /*
3165                                  * Concurrent thread allocated container data.
3166                                  */
3167                                 OBD_FREE_PTR(dir);
3168                         spin_unlock(&obj->oo_guard);
3169                         /*
3170                          * Now, that we have container data, serialize its
3171                          * initialization.
3172                          */
3173                         down_write(&obj->oo_ext_idx_sem);
3174                         /*
3175                          * recheck under lock.
3176                          */
3177                         if (!osd_has_index(obj))
3178                                 result = osd_iam_container_init(env, obj,
3179                                                                 obj->oo_dir);
3180                         else
3181                                 result = 0;
3182                         up_write(&obj->oo_ext_idx_sem);
3183                 } else {
3184                         result = -ENOMEM;
3185                 }
3186         } else {
3187                 result = 0;
3188         }
3189
3190         if (result == 0 && skip_iam == 0) {
3191                 if (!osd_iam_index_probe(env, obj, feat))
3192                         result = -ENOTDIR;
3193         }
3194         LINVRNT(osd_invariant(obj));
3195
3196         if (result == 0 && feat == &dt_quota_glb_features &&
3197             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3198                 result = osd_quota_migration(env, dt);
3199
3200         return result;
3201 }
3202
3203 static int osd_otable_it_attr_get(const struct lu_env *env,
3204                                  struct dt_object *dt,
3205                                  struct lu_attr *attr)
3206 {
3207         attr->la_valid = 0;
3208         return 0;
3209 }
3210
3211 static const struct dt_object_operations osd_obj_ops = {
3212         .do_read_lock         = osd_object_read_lock,
3213         .do_write_lock        = osd_object_write_lock,
3214         .do_read_unlock       = osd_object_read_unlock,
3215         .do_write_unlock      = osd_object_write_unlock,
3216         .do_write_locked      = osd_object_write_locked,
3217         .do_attr_get          = osd_attr_get,
3218         .do_declare_attr_set  = osd_declare_attr_set,
3219         .do_attr_set          = osd_attr_set,
3220         .do_ah_init           = osd_ah_init,
3221         .do_declare_create    = osd_declare_object_create,
3222         .do_create            = osd_object_create,
3223         .do_declare_destroy   = osd_declare_object_destroy,
3224         .do_destroy           = osd_object_destroy,
3225         .do_index_try         = osd_index_try,
3226         .do_declare_ref_add   = osd_declare_object_ref_add,
3227         .do_ref_add           = osd_object_ref_add,
3228         .do_declare_ref_del   = osd_declare_object_ref_del,
3229         .do_ref_del           = osd_object_ref_del,
3230         .do_xattr_get         = osd_xattr_get,
3231         .do_declare_xattr_set = osd_declare_xattr_set,
3232         .do_xattr_set         = osd_xattr_set,
3233         .do_declare_xattr_del = osd_declare_xattr_del,
3234         .do_xattr_del         = osd_xattr_del,
3235         .do_xattr_list        = osd_xattr_list,
3236         .do_object_sync       = osd_object_sync,
3237 };
3238
3239 /**
3240  * dt_object_operations for interoperability mode
3241  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3242  */
3243 static const struct dt_object_operations osd_obj_ea_ops = {
3244         .do_read_lock         = osd_object_read_lock,
3245         .do_write_lock        = osd_object_write_lock,
3246         .do_read_unlock       = osd_object_read_unlock,
3247         .do_write_unlock      = osd_object_write_unlock,
3248         .do_write_locked      = osd_object_write_locked,
3249         .do_attr_get          = osd_attr_get,
3250         .do_declare_attr_set  = osd_declare_attr_set,
3251         .do_attr_set          = osd_attr_set,
3252         .do_ah_init           = osd_ah_init,
3253         .do_declare_create    = osd_declare_object_create,
3254         .do_create            = osd_object_ea_create,
3255         .do_declare_destroy   = osd_declare_object_destroy,
3256         .do_destroy           = osd_object_destroy,
3257         .do_index_try         = osd_index_try,
3258         .do_declare_ref_add   = osd_declare_object_ref_add,
3259         .do_ref_add           = osd_object_ref_add,
3260         .do_declare_ref_del   = osd_declare_object_ref_del,
3261         .do_ref_del           = osd_object_ref_del,
3262         .do_xattr_get         = osd_xattr_get,
3263         .do_declare_xattr_set = osd_declare_xattr_set,
3264         .do_xattr_set         = osd_xattr_set,
3265         .do_declare_xattr_del = osd_declare_xattr_del,
3266         .do_xattr_del         = osd_xattr_del,
3267         .do_xattr_list        = osd_xattr_list,
3268         .do_object_sync       = osd_object_sync,
3269 };
3270
3271 static const struct dt_object_operations osd_obj_otable_it_ops = {
3272         .do_attr_get    = osd_otable_it_attr_get,
3273         .do_index_try   = osd_index_try,
3274 };
3275
3276 static int osd_index_declare_iam_delete(const struct lu_env *env,
3277                                         struct dt_object *dt,
3278                                         const struct dt_key *key,
3279                                         struct thandle *handle)
3280 {
3281         struct osd_thandle    *oh;
3282
3283         oh = container_of0(handle, struct osd_thandle, ot_super);
3284         LASSERT(oh->ot_handle == NULL);
3285
3286         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3287                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3288
3289         return 0;
3290 }
3291
3292 /**
3293  *      delete a (key, value) pair from index \a dt specified by \a key
3294  *
3295  *      \param  dt      osd index object
3296  *      \param  key     key for index
3297  *      \param  rec     record reference
3298  *      \param  handle  transaction handler
3299  *
3300  *      \retval  0  success
3301  *      \retval -ve   failure
3302  */
3303 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3304                                 const struct dt_key *key,
3305                                 struct thandle *handle)
3306 {
3307         struct osd_thread_info *oti = osd_oti_get(env);
3308         struct osd_object      *obj = osd_dt_obj(dt);
3309         struct osd_thandle     *oh;
3310         struct iam_path_descr  *ipd;
3311         struct iam_container   *bag = &obj->oo_dir->od_container;
3312         int                     rc;
3313         ENTRY;
3314
3315         if (!dt_object_exists(dt))
3316                 RETURN(-ENOENT);
3317
3318         LINVRNT(osd_invariant(obj));
3319         LASSERT(!dt_object_remote(dt));
3320         LASSERT(bag->ic_object == obj->oo_inode);
3321         LASSERT(handle != NULL);
3322
3323         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3324
3325         ipd = osd_idx_ipd_get(env, bag);
3326         if (unlikely(ipd == NULL))
3327                 RETURN(-ENOMEM);
3328
3329         oh = container_of0(handle, struct osd_thandle, ot_super);
3330         LASSERT(oh->ot_handle != NULL);
3331         LASSERT(oh->ot_handle->h_transaction != NULL);
3332
3333         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3334                 /* swab quota uid/gid provided by caller */
3335                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3336                 key = (const struct dt_key *)&oti->oti_quota_id;
3337         }
3338
3339         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3340         osd_ipd_put(env, bag, ipd);
3341         LINVRNT(osd_invariant(obj));
3342         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3343         RETURN(rc);
3344 }
3345
3346 static int osd_index_declare_ea_delete(const struct lu_env *env,
3347                                        struct dt_object *dt,
3348                                        const struct dt_key *key,
3349                                        struct thandle *handle)
3350 {
3351         struct osd_thandle *oh;
3352         struct inode       *inode;
3353         int                 rc;
3354         ENTRY;
3355
3356         LASSERT(!dt_object_remote(dt));
3357         LASSERT(handle != NULL);
3358
3359         oh = container_of0(handle, struct osd_thandle, ot_super);
3360         LASSERT(oh->ot_handle == NULL);
3361
3362         /* due to DNE we may need to remove an agent inode */
3363         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3364                              osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3365                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3366
3367         inode = osd_dt_obj(dt)->oo_inode;
3368         LASSERT(inode);
3369
3370         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3371                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3372         RETURN(rc);
3373 }
3374
3375 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3376                                           struct dt_rec *fid)
3377 {
3378         struct osd_fid_pack *rec;
3379         int                  rc = -ENODATA;
3380
3381         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3382                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3383                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3384                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3385                         rc = -EINVAL;
3386         }
3387         return rc;
3388 }
3389
3390 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3391                           const struct lu_fid *fid)
3392 {
3393         struct seq_server_site  *ss = osd_seq_site(osd);
3394         ENTRY;
3395
3396         /* FID seqs not in FLDB, must be local seq */
3397         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3398                 RETURN(0);
3399
3400         /* If FLD is not being initialized yet, it only happens during the
3401          * initialization, likely during mgs initialization, and we assume
3402          * this is local FID. */
3403         if (ss == NULL || ss->ss_server_fld == NULL)
3404                 RETURN(0);
3405
3406         /* Only check the local FLDB here */
3407         if (osd_seq_exists(env, osd, fid_seq(fid)))
3408                 RETURN(0);
3409
3410         RETURN(1);
3411 }
3412
3413 /**
3414  * Index delete function for interoperability mode (b11826).
3415  * It will remove the directory entry added by osd_index_ea_insert().
3416  * This entry is needed to maintain name->fid mapping.
3417  *
3418  * \param key,  key i.e. file entry to be deleted
3419  *
3420  * \retval   0, on success
3421  * \retval -ve, on error
3422  */
3423 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3424                                const struct dt_key *key, struct thandle *handle)
3425 {
3426         struct osd_object          *obj = osd_dt_obj(dt);
3427         struct inode               *dir = obj->oo_inode;
3428         struct dentry              *dentry;
3429         struct osd_thandle         *oh;
3430         struct ldiskfs_dir_entry_2 *de = NULL;
3431         struct buffer_head         *bh;
3432         struct htree_lock          *hlock = NULL;
3433         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3434         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3435         int                        rc;
3436         ENTRY;
3437
3438         if (!dt_object_exists(dt))
3439                 RETURN(-ENOENT);
3440
3441         LINVRNT(osd_invariant(obj));
3442         LASSERT(!dt_object_remote(dt));
3443         LASSERT(handle != NULL);
3444
3445         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3446
3447         oh = container_of(handle, struct osd_thandle, ot_super);
3448         LASSERT(oh->ot_handle != NULL);
3449         LASSERT(oh->ot_handle->h_transaction != NULL);
3450
3451         ll_vfs_dq_init(dir);
3452         dentry = osd_child_dentry_get(env, obj,
3453                                       (char *)key, strlen((char *)key));
3454
3455         if (obj->oo_hl_head != NULL) {
3456                 hlock = osd_oti_get(env)->oti_hlock;
3457                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3458                                    dir, LDISKFS_HLOCK_DEL);
3459         } else {
3460                 down_write(&obj->oo_ext_idx_sem);
3461         }
3462
3463         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3464         if (bh) {
3465                 __u32 ino = 0;
3466
3467                 /* If this is not the ".." entry, it might be a remote DNE
3468                  * entry and  we need to check if the FID is for a remote
3469                  * MDT.  If the FID is  not in the directory entry (e.g.
3470                  * upgraded 1.8 filesystem without dirdata enabled) then
3471                  * we need to get the FID from the LMA. For a remote directory
3472                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3473                  * case.
3474                  *
3475                  * Delete the entry before the agent inode in order to
3476                  * simplify error handling.  At worst an error after deleting
3477                  * the entry first might leak the agent inode afterward. The
3478                  * reverse would need filesystem abort in case of error deleting
3479                  * the entry after the agent had been removed, or leave a
3480                  * dangling entry pointing at a random inode. */
3481                 if (strcmp((char *)key, dotdot) != 0) {
3482                         LASSERT(de != NULL);
3483                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3484                         /* If Fid is not in dentry, try to get it from LMA */
3485                         if (rc == -ENODATA) {
3486                                 struct osd_inode_id *id;
3487                                 struct inode *inode;
3488
3489                                 /* Before trying to get fid from the inode,
3490                                  * check whether the inode is valid.
3491                                  *
3492                                  * If the inode has been deleted, do not go
3493                                  * ahead to do osd_ea_fid_get, which will set
3494                                  * the inode to bad inode, which might cause
3495                                  * the inode to be deleted uncorrectly */
3496                                 inode = ldiskfs_iget(osd_sb(osd),
3497                                                      le32_to_cpu(de->inode));
3498                                 if (IS_ERR(inode)) {
3499                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3500                                                "error.\n", osd_name(osd),
3501                                                PFID(fid));
3502                                         rc = PTR_ERR(inode);
3503                                 } else {
3504                                         if (likely(inode->i_nlink != 0)) {
3505                                                 id = &osd_oti_get(env)->oti_id;
3506                                                 rc = osd_ea_fid_get(env, obj,
3507                                                         le32_to_cpu(de->inode),
3508                                                                     fid, id);
3509                                         } else {
3510                                                 CDEBUG(D_INFO, "%s: %u "DFID
3511                                                        "deleted.\n",
3512                                                        osd_name(osd),
3513                                                        le32_to_cpu(de->inode),
3514                                                        PFID(fid));
3515                                                 rc = -ESTALE;
3516                                         }
3517                                         iput(inode);
3518                                 }
3519                         }
3520                         if (rc == 0 &&
3521                             unlikely(osd_remote_fid(env, osd, fid)))
3522                                 /* Need to delete agent inode */
3523                                 ino = le32_to_cpu(de->inode);
3524                 }
3525                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3526                 brelse(bh);
3527                 if (rc == 0 && unlikely(ino != 0)) {
3528                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3529                                                           oh);
3530                         if (rc != 0)
3531                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3532                                        osd_name(osd), PFID(fid), rc);
3533                 }
3534         } else {
3535                 rc = -ENOENT;
3536         }
3537         if (hlock != NULL)
3538                 ldiskfs_htree_unlock(hlock);
3539         else
3540                 up_write(&obj->oo_ext_idx_sem);
3541
3542         if (rc != 0)
3543                 GOTO(out, rc);
3544
3545         /* For inode on the remote MDT, .. will point to
3546          * /Agent directory, Check whether it needs to delete
3547          * from agent directory */
3548         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3549                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3550                                                    oh);
3551                 if (rc != 0 && rc != -ENOENT) {
3552                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3553                                osd_name(osd), PFID(fid), rc);
3554                 }
3555
3556                 if (rc == -ENOENT)
3557                         rc = 0;
3558
3559                 GOTO(out, rc);
3560         }
3561 out:
3562
3563         LASSERT(osd_invariant(obj));
3564         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3565         RETURN(rc);
3566 }
3567
3568 /**
3569  *      Lookup index for \a key and copy record to \a rec.
3570  *
3571  *      \param  dt      osd index object
3572  *      \param  key     key for index
3573  *      \param  rec     record reference
3574  *
3575  *      \retval  +ve  success : exact mach
3576  *      \retval  0    return record with key not greater than \a key
3577  *      \retval -ve   failure
3578  */
3579 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3580                                 struct dt_rec *rec, const struct dt_key *key)
3581 {
3582         struct osd_object      *obj = osd_dt_obj(dt);
3583         struct iam_path_descr  *ipd;
3584         struct iam_container   *bag = &obj->oo_dir->od_container;
3585         struct osd_thread_info *oti = osd_oti_get(env);
3586         struct iam_iterator    *it = &oti->oti_idx_it;
3587         struct iam_rec         *iam_rec;
3588         int                     rc;
3589         ENTRY;
3590
3591         if (!dt_object_exists(dt))
3592                 RETURN(-ENOENT);
3593
3594         LASSERT(osd_invariant(obj));
3595         LASSERT(!dt_object_remote(dt));
3596         LASSERT(bag->ic_object == obj->oo_inode);
3597
3598         ipd = osd_idx_ipd_get(env, bag);
3599         if (IS_ERR(ipd))
3600                 RETURN(-ENOMEM);
3601
3602         /* got ipd now we can start iterator. */
3603         iam_it_init(it, bag, 0, ipd);
3604
3605         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3606                 /* swab quota uid/gid provided by caller */
3607                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3608                 key = (const struct dt_key *)&oti->oti_quota_id;
3609         }
3610
3611         rc = iam_it_get(it, (struct iam_key *)key);
3612         if (rc >= 0) {
3613                 if (S_ISDIR(obj->oo_inode->i_mode))
3614                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3615                 else
3616                         iam_rec = (struct iam_rec *) rec;
3617
3618                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3619
3620                 if (S_ISDIR(obj->oo_inode->i_mode))
3621                         osd_fid_unpack((struct lu_fid *) rec,
3622                                        (struct osd_fid_pack *)iam_rec);
3623                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3624                         osd_quota_unpack(obj, rec);
3625         }
3626
3627         iam_it_put(it);
3628         iam_it_fini(it);
3629         osd_ipd_put(env, bag, ipd);
3630
3631         LINVRNT(osd_invariant(obj));
3632
3633         RETURN(rc);
3634 }
3635
3636 static int osd_index_declare_iam_insert(const struct lu_env *env,
3637                                         struct dt_object *dt,
3638                                         const struct dt_rec *rec,
3639                                         const struct dt_key *key,
3640                                         struct thandle *handle)
3641 {
3642         struct osd_thandle *oh;
3643
3644         LASSERT(handle != NULL);
3645
3646         oh = container_of0(handle, struct osd_thandle, ot_super);
3647         LASSERT(oh->ot_handle == NULL);
3648
3649         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3650                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3651
3652         return 0;
3653 }
3654
3655 /**
3656  *      Inserts (key, value) pair in \a dt index object.
3657  *
3658  *      \param  dt      osd index object
3659  *      \param  key     key for index
3660  *      \param  rec     record reference
3661  *      \param  th      transaction handler
3662  *
3663  *      \retval  0  success
3664  *      \retval -ve failure
3665  */
3666 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3667                                 const struct dt_rec *rec,
3668                                 const struct dt_key *key, struct thandle *th,
3669                                 int ignore_quota)
3670 {
3671         struct osd_object     *obj = osd_dt_obj(dt);
3672         struct iam_path_descr *ipd;
3673         struct osd_thandle    *oh;
3674         struct iam_container  *bag;
3675         struct osd_thread_info *oti = osd_oti_get(env);
3676         struct iam_rec         *iam_rec;
3677         int                     rc;
3678         ENTRY;
3679
3680         if (!dt_object_exists(dt))
3681                 RETURN(-ENOENT);
3682
3683         LINVRNT(osd_invariant(obj));
3684         LASSERT(!dt_object_remote(dt));
3685
3686         bag = &obj->oo_dir->od_container;
3687         LASSERT(bag->ic_object == obj->oo_inode);
3688         LASSERT(th != NULL);
3689
3690         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3691
3692         ipd = osd_idx_ipd_get(env, bag);
3693         if (unlikely(ipd == NULL))
3694                 RETURN(-ENOMEM);
3695
3696         oh = container_of0(th, struct osd_thandle, ot_super);
3697         LASSERT(oh->ot_handle != NULL);
3698         LASSERT(oh->ot_handle->h_transaction != NULL);
3699         if (S_ISDIR(obj->oo_inode->i_mode)) {
3700                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3701                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3702         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3703                 /* pack quota uid/gid */
3704                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3705                 key = (const struct dt_key *)&oti->oti_quota_id;
3706                 /* pack quota record */
3707                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3708                 iam_rec = (struct iam_rec *)rec;
3709         } else {
3710                 iam_rec = (struct iam_rec *)rec;
3711         }
3712
3713         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3714                         iam_rec, ipd);
3715         osd_ipd_put(env, bag, ipd);
3716         LINVRNT(osd_invariant(obj));
3717         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3718         RETURN(rc);
3719 }
3720
3721 /**
3722  * Calls ldiskfs_add_entry() to add directory entry
3723  * into the directory. This is required for
3724  * interoperability mode (b11826)
3725  *
3726  * \retval   0, on success
3727  * \retval -ve, on error
3728  */
3729 static int __osd_ea_add_rec(struct osd_thread_info *info,
3730                             struct osd_object *pobj, struct inode  *cinode,
3731                             const char *name, const struct lu_fid *fid,
3732                             struct htree_lock *hlock, struct thandle *th)
3733 {
3734         struct ldiskfs_dentry_param *ldp;
3735         struct dentry               *child;
3736         struct osd_thandle          *oth;
3737         int                          rc;
3738
3739         oth = container_of(th, struct osd_thandle, ot_super);
3740         LASSERT(oth->ot_handle != NULL);
3741         LASSERT(oth->ot_handle->h_transaction != NULL);
3742         LASSERT(pobj->oo_inode);
3743
3744         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3745         if (unlikely(pobj->oo_inode ==
3746                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3747                 ldp->edp_magic = 0;
3748         else
3749                 osd_get_ldiskfs_dirent_param(ldp, fid);
3750         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3751         child->d_fsdata = (void *)ldp;
3752         ll_vfs_dq_init(pobj->oo_inode);
3753         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3754         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3755                 struct ldiskfs_dir_entry_2      *de;
3756                 struct buffer_head              *bh;
3757                 int                              rc1;
3758
3759                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3760                                             NULL, hlock);
3761                 if (bh != NULL) {
3762                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3763                                                                bh);
3764                         if (rc1 == 0) {
3765                                 if (S_ISDIR(cinode->i_mode))
3766                                         de->file_type = LDISKFS_DIRENT_LUFID |
3767                                                         LDISKFS_FT_REG_FILE;
3768                                 else
3769                                         de->file_type = LDISKFS_DIRENT_LUFID |
3770                                                         LDISKFS_FT_DIR;
3771                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3772                                                               NULL, bh);
3773                                 brelse(bh);
3774                         }
3775                 }
3776         }
3777
3778         RETURN(rc);
3779 }
3780
3781 /**
3782  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3783  * into the directory.Also sets flags into osd object to
3784  * indicate dot and dotdot are created. This is required for
3785  * interoperability mode (b11826)
3786  *
3787  * \param dir   directory for dot and dotdot fixup.
3788  * \param obj   child object for linking
3789  *
3790  * \retval   0, on success
3791  * \retval -ve, on error
3792  */
3793 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3794                               struct osd_object *dir,
3795                               struct inode *parent_dir, const char *name,
3796                               const struct lu_fid *dot_fid,
3797                               const struct lu_fid *dot_dot_fid,
3798                               struct thandle *th)
3799 {
3800         struct inode                *inode = dir->oo_inode;
3801         struct osd_thandle          *oth;
3802         int result = 0;
3803
3804         oth = container_of(th, struct osd_thandle, ot_super);
3805         LASSERT(oth->ot_handle->h_transaction != NULL);
3806         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3807
3808         if (strcmp(name, dot) == 0) {
3809                 if (dir->oo_compat_dot_created) {
3810                         result = -EEXIST;
3811                 } else {
3812                         LASSERT(inode == parent_dir);
3813                         dir->oo_compat_dot_created = 1;
3814                         result = 0;
3815                 }
3816         } else if (strcmp(name, dotdot) == 0) {
3817                 if (!dir->oo_compat_dot_created)
3818                         return -EINVAL;
3819                 /* in case of rename, dotdot is already created */
3820                 if (dir->oo_compat_dotdot_created) {
3821                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3822                                                 dot_dot_fid, NULL, th);
3823                 }
3824
3825                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
3826                         struct lu_fid tfid = *dot_dot_fid;
3827
3828                         tfid.f_oid--;
3829                         result = osd_add_dot_dotdot_internal(info,
3830                                         dir->oo_inode, parent_dir, dot_fid,
3831                                         &tfid, oth);
3832                 } else {
3833                         result = osd_add_dot_dotdot_internal(info,
3834                                         dir->oo_inode, parent_dir, dot_fid,
3835                                         dot_dot_fid, oth);
3836                 }
3837
3838                 if (result == 0)
3839                         dir->oo_compat_dotdot_created = 1;
3840         }
3841
3842         return result;
3843 }
3844
3845
3846 /**
3847  * It will call the appropriate osd_add* function and return the
3848  * value, return by respective functions.
3849  */
3850 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3851                           struct inode *cinode, const char *name,
3852                           const struct lu_fid *fid, struct thandle *th)
3853 {
3854         struct osd_thread_info *info   = osd_oti_get(env);
3855         struct htree_lock      *hlock;
3856         int                     rc;
3857
3858         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3859
3860         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3861                                                    name[2] =='\0'))) {
3862                 if (hlock != NULL) {
3863                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3864                                            pobj->oo_inode, 0);
3865                 } else {
3866                         down_write(&pobj->oo_ext_idx_sem);
3867                 }
3868
3869                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3870                                         lu_object_fid(&pobj->oo_dt.do_lu),
3871                                         fid, th);
3872         } else {
3873                 if (hlock != NULL) {
3874                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3875                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3876                 } else {
3877                         down_write(&pobj->oo_ext_idx_sem);
3878                 }
3879
3880                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3881                         struct lu_fid *tfid = &info->oti_fid;
3882
3883                         *tfid = *fid;
3884                         tfid->f_ver = ~0;
3885                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3886                                               tfid, hlock, th);
3887                 } else {
3888                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3889                                               hlock, th);
3890                 }
3891         }
3892         if (hlock != NULL)
3893                 ldiskfs_htree_unlock(hlock);
3894         else
3895                 up_write(&pobj->oo_ext_idx_sem);
3896
3897         return rc;
3898 }
3899
3900 static void
3901 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3902                       struct osd_idmap_cache *oic)
3903 {
3904         struct osd_scrub    *scrub = &dev->od_scrub;
3905         struct lu_fid       *fid   = &oic->oic_fid;
3906         struct osd_inode_id *id    = &oti->oti_id;
3907         int                  once  = 0;
3908         int                  rc;
3909         ENTRY;
3910
3911         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3912                 RETURN_EXIT;
3913
3914         if (scrub->os_pos_current > id->oii_ino)
3915                 RETURN_EXIT;
3916
3917 again:
3918         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
3919         if (rc != 0 && rc != -ENOENT)
3920                 RETURN_EXIT;
3921
3922         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3923                 RETURN_EXIT;
3924
3925         if (thread_is_running(&scrub->os_thread)) {
3926                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3927                 /* There is race condition between osd_oi_lookup and OI scrub.
3928                  * The OI scrub finished just after osd_oi_lookup() failure.
3929                  * Under such case, it is unnecessary to trigger OI scrub again,
3930                  * but try to call osd_oi_lookup() again. */
3931                 if (unlikely(rc == -EAGAIN))
3932                         goto again;
3933
3934                 RETURN_EXIT;
3935         }
3936
3937         if (!dev->od_noscrub && ++once == 1) {
3938                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
3939                                      SS_CLEAR_FAILOUT);
3940                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
3941                        "for "DFID", rc = %d [2]\n",
3942                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3943                        PFID(fid), rc);
3944                 if (rc == 0 || rc == -EALREADY)
3945                         goto again;
3946         }
3947
3948         EXIT;
3949 }
3950
3951 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3952                                struct osd_device *dev,
3953                                struct osd_idmap_cache *oic,
3954                                struct lu_fid *fid, __u32 ino)
3955 {
3956         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3957         struct inode            *inode;
3958         int                      rc;
3959
3960         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3961         inode = osd_iget(oti, dev, &oic->oic_lid);
3962         if (IS_ERR(inode)) {
3963                 fid_zero(&oic->oic_fid);
3964                 return PTR_ERR(inode);
3965         }
3966
3967         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3968         iput(inode);
3969         if (rc != 0)
3970                 fid_zero(&oic->oic_fid);
3971         else
3972                 *fid = oic->oic_fid = lma->lma_self_fid;
3973         return rc;
3974 }
3975
3976 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
3977                       struct osd_inode_id *id, const struct lu_fid *fid)
3978 {
3979         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
3980                id->oii_ino, id->oii_gen, info);
3981         info->oti_cache.oic_lid = *id;
3982         info->oti_cache.oic_fid = *fid;
3983         info->oti_cache.oic_dev = osd;
3984 }
3985
3986 /**
3987  * Get parent FID from the linkEA.
3988  *
3989  * For a directory which parent resides on remote MDT, to satisfy the
3990  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
3991  * the other hand, to make the lookup(..) on the directory can return
3992  * the real parent FID, we append the real parent FID after its ".."
3993  * name entry in the /REMOTE_PARENT_DIR.
3994  *
3995  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
3996  * backup. So after the restore, we cannot get the right parent FID from
3997  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
3998  * we have stored the real parent FID in the directory object's linkEA,
3999  * we can parse the linkEA for the real parent FID.
4000  *
4001  * \param[in] env       pointer to the thread context
4002  * \param[in] obj       pointer to the object to be handled
4003  * \param[out]fid       pointer to the buffer to hold the parent FID
4004  *
4005  * \retval              0 for getting the real parent FID successfully
4006  * \retval              negative error number on failure
4007  */
4008 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4009                                     struct osd_object *obj,
4010                                     struct lu_fid *fid)
4011 {
4012         struct osd_thread_info  *oti    = osd_oti_get(env);
4013         struct lu_buf           *buf    = &oti->oti_big_buf;
4014         struct dentry           *dentry = &oti->oti_obj_dentry;
4015         struct inode            *inode  = obj->oo_inode;
4016         struct linkea_data       ldata  = { NULL };
4017         int                      rc;
4018         ENTRY;
4019
4020         fid_zero(fid);
4021         if (!S_ISDIR(inode->i_mode))
4022                 RETURN(-EIO);
4023
4024 again:
4025         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4026                              buf->lb_buf, buf->lb_len);
4027         if (rc == -ERANGE) {
4028                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4029                                      NULL, 0);
4030                 if (rc > 0) {
4031                         lu_buf_realloc(buf, rc);
4032                         if (buf->lb_buf == NULL)
4033                                 RETURN(-ENOMEM);
4034
4035                         goto again;
4036                 }
4037         }
4038
4039         if (unlikely(rc == 0))
4040                 RETURN(-ENODATA);
4041
4042         if (rc < 0)
4043                 RETURN(rc);
4044
4045         if (unlikely(buf->lb_buf == NULL)) {
4046                 lu_buf_realloc(buf, rc);
4047                 if (buf->lb_buf == NULL)
4048                         RETURN(-ENOMEM);
4049
4050                 goto again;
4051         }
4052
4053         ldata.ld_buf = buf;
4054         rc = linkea_init(&ldata);
4055         if (rc == 0) {
4056                 linkea_first_entry(&ldata);
4057                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4058         }
4059
4060         RETURN(rc);
4061 }
4062
4063 /**
4064  * Calls ->lookup() to find dentry. From dentry get inode and
4065  * read inode's ea to get fid. This is required for  interoperability
4066  * mode (b11826)
4067  *
4068  * \retval   0, on success
4069  * \retval -ve, on error
4070  */
4071 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4072                              struct dt_rec *rec, const struct dt_key *key)
4073 {
4074         struct inode                    *dir    = obj->oo_inode;
4075         struct dentry                   *dentry;
4076         struct ldiskfs_dir_entry_2      *de;
4077         struct buffer_head              *bh;
4078         struct lu_fid                   *fid = (struct lu_fid *) rec;
4079         struct htree_lock               *hlock = NULL;
4080         int                             ino;
4081         int                             rc;
4082         ENTRY;
4083
4084         LASSERT(dir->i_op != NULL);
4085         LASSERT(dir->i_op->lookup != NULL);
4086
4087         dentry = osd_child_dentry_get(env, obj,
4088                                       (char *)key, strlen((char *)key));
4089
4090         if (obj->oo_hl_head != NULL) {
4091                 hlock = osd_oti_get(env)->oti_hlock;
4092                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4093                                    dir, LDISKFS_HLOCK_LOOKUP);
4094         } else {
4095                 down_read(&obj->oo_ext_idx_sem);
4096         }
4097
4098         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4099         if (bh) {
4100                 struct osd_thread_info *oti = osd_oti_get(env);
4101                 struct osd_inode_id *id = &oti->oti_id;
4102                 struct osd_idmap_cache *oic = &oti->oti_cache;
4103                 struct osd_device *dev = osd_obj2dev(obj);
4104
4105                 ino = le32_to_cpu(de->inode);
4106                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4107                         brelse(bh);
4108                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4109                         GOTO(out, rc);
4110                 }
4111
4112                 rc = osd_get_fid_from_dentry(de, rec);
4113
4114                 /* done with de, release bh */
4115                 brelse(bh);
4116                 if (rc != 0) {
4117                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
4118                                 const char *name = (const char *)key;
4119
4120                                 /* If the parent is on remote MDT, and there
4121                                  * is no FID-in-dirent, then we have to get
4122                                  * the parent FID from the linkEA.  */
4123                                 if (likely(strlen(name) == 2 &&
4124                                            name[0] == '.' && name[1] == '.'))
4125                                         rc = osd_get_pfid_from_linkea(env, obj,
4126                                                                       fid);
4127                         } else {
4128                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4129                         }
4130                 } else {
4131                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4132                 }
4133
4134                 if (rc != 0) {
4135                         fid_zero(&oic->oic_fid);
4136                         GOTO(out, rc);
4137                 }
4138
4139                 if (osd_remote_fid(env, dev, fid))
4140                         GOTO(out, rc = 0);
4141
4142                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4143                 osd_consistency_check(oti, dev, oic);
4144         } else {
4145                 rc = -ENOENT;
4146         }
4147
4148         GOTO(out, rc);
4149
4150 out:
4151         if (hlock != NULL)
4152                 ldiskfs_htree_unlock(hlock);
4153         else
4154                 up_read(&obj->oo_ext_idx_sem);
4155         return rc;
4156 }
4157
4158 /**
4159  * Find the osd object for given fid.
4160  *
4161  * \param fid need to find the osd object having this fid
4162  *
4163  * \retval osd_object on success
4164  * \retval        -ve on error
4165  */
4166 static struct osd_object *osd_object_find(const struct lu_env *env,
4167                                           struct dt_object *dt,
4168                                           const struct lu_fid *fid)
4169 {
4170         struct lu_device  *ludev = dt->do_lu.lo_dev;
4171         struct osd_object *child = NULL;
4172         struct lu_object  *luch;
4173         struct lu_object  *lo;
4174
4175         /*
4176          * at this point topdev might not exist yet
4177          * (i.e. MGS is preparing profiles). so we can
4178          * not rely on topdev and instead lookup with
4179          * our device passed as topdev. this can't work
4180          * if the object isn't cached yet (as osd doesn't
4181          * allocate lu_header). IOW, the object must be
4182          * in the cache, otherwise lu_object_alloc() crashes
4183          * -bzzz
4184          */
4185         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4186                                  ludev : ludev->ld_site->ls_top_dev,
4187                                  fid, NULL);
4188         if (!IS_ERR(luch)) {
4189                 if (lu_object_exists(luch)) {
4190                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4191                         if (lo != NULL)
4192                                 child = osd_obj(lo);
4193                         else
4194                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4195                                                 "lu_object can't be located"
4196                                                 DFID"\n", PFID(fid));
4197
4198                         if (child == NULL) {
4199                                 lu_object_put(env, luch);
4200                                 CERROR("Unable to get osd_object\n");
4201                                 child = ERR_PTR(-ENOENT);
4202                         }
4203                 } else {
4204                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4205                                         "lu_object does not exists "DFID"\n",
4206                                         PFID(fid));
4207                         lu_object_put(env, luch);
4208                         child = ERR_PTR(-ENOENT);
4209                 }
4210         } else {
4211                 child = ERR_CAST(luch);
4212         }
4213
4214         return child;
4215 }
4216
4217 /**
4218  * Put the osd object once done with it.
4219  *
4220  * \param obj osd object that needs to be put
4221  */
4222 static inline void osd_object_put(const struct lu_env *env,
4223                                   struct osd_object *obj)
4224 {
4225         lu_object_put(env, &obj->oo_dt.do_lu);
4226 }
4227
4228 static int osd_index_declare_ea_insert(const struct lu_env *env,
4229                                        struct dt_object *dt,
4230                                        const struct dt_rec *rec,
4231                                        const struct dt_key *key,
4232                                        struct thandle *handle)
4233 {
4234         struct osd_thandle      *oh;
4235         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4236         struct lu_fid           *fid = (struct lu_fid *)rec;
4237         int                      credits, rc = 0;
4238         ENTRY;
4239
4240         LASSERT(!dt_object_remote(dt));
4241         LASSERT(handle != NULL);
4242
4243         oh = container_of0(handle, struct osd_thandle, ot_super);
4244         LASSERT(oh->ot_handle == NULL);
4245
4246         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4247         if (fid != NULL) {
4248                 rc = osd_remote_fid(env, osd, fid);
4249                 if (unlikely(rc < 0))
4250                         RETURN(rc);
4251                 if (rc > 0) {
4252                         /* a reference to remote inode is represented by an
4253                          * agent inode which we have to create */
4254                         credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4255                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4256                 }
4257                 rc = 0;
4258         }
4259
4260         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4261
4262         if (osd_dt_obj(dt)->oo_inode != NULL) {
4263                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4264
4265                 /* We ignore block quota on meta pool (MDTs), so needn't
4266                  * calculate how many blocks will be consumed by this index
4267                  * insert */
4268                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4269                                            i_gid_read(inode), 0, oh,
4270                                            osd_dt_obj(dt), true, NULL, false);
4271         }
4272
4273         RETURN(rc);
4274 }
4275
4276 /**
4277  * Index add function for interoperability mode (b11826).
4278  * It will add the directory entry.This entry is needed to
4279  * maintain name->fid mapping.
4280  *
4281  * \param key it is key i.e. file entry to be inserted
4282  * \param rec it is value of given key i.e. fid
4283  *
4284  * \retval   0, on success
4285  * \retval -ve, on error
4286  */
4287 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4288                                const struct dt_rec *rec,
4289                                const struct dt_key *key, struct thandle *th,
4290                                int ignore_quota)
4291 {
4292         struct osd_object       *obj = osd_dt_obj(dt);
4293         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4294         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4295         const struct lu_fid     *fid    = rec1->rec_fid;
4296         const char              *name = (const char *)key;
4297         struct osd_thread_info  *oti   = osd_oti_get(env);
4298         struct osd_inode_id     *id    = &oti->oti_id;
4299         struct inode            *child_inode = NULL;
4300         struct osd_object       *child = NULL;
4301         int                     rc;
4302         ENTRY;
4303
4304         if (!dt_object_exists(dt))
4305                 RETURN(-ENOENT);
4306
4307         LASSERT(osd_invariant(obj));
4308         LASSERT(!dt_object_remote(dt));
4309         LASSERT(th != NULL);
4310
4311         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4312
4313         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4314
4315         rc = osd_remote_fid(env, osd, fid);
4316         if (rc < 0) {
4317                 CERROR("%s: Can not find object "DFID" rc %d\n",
4318                        osd_name(osd), PFID(fid), rc);
4319                 RETURN(rc);
4320         }
4321
4322         if (rc == 1) {
4323                 /* Insert remote entry */
4324                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4325                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4326                         struct osd_thandle      *oh;
4327
4328                         /* If parent on remote MDT, we need put this object
4329                          * under AGENT */
4330                         oh = container_of(th, typeof(*oh), ot_super);
4331                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4332                         if (rc != 0) {
4333                                 CERROR("%s: add "DFID" error: rc = %d\n",
4334                                        osd_name(osd),
4335                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4336                                 RETURN(rc);
4337                         }
4338
4339                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4340                 } else {
4341                         child_inode = osd_create_local_agent_inode(env, osd,
4342                                         obj, fid, rec1->rec_type & S_IFMT, th);
4343                         if (IS_ERR(child_inode))
4344                                 RETURN(PTR_ERR(child_inode));
4345                 }
4346         } else {
4347                 /* Insert local entry */
4348                 child = osd_object_find(env, dt, fid);
4349                 if (IS_ERR(child)) {
4350                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4351                                osd_name(osd), PFID(fid),
4352                                id->oii_ino, id->oii_gen,
4353                                (int)PTR_ERR(child));
4354                         RETURN(PTR_ERR(child));
4355                 }
4356                 child_inode = igrab(child->oo_inode);
4357         }
4358
4359         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4360
4361         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4362                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4363
4364         iput(child_inode);
4365         if (child != NULL)
4366                 osd_object_put(env, child);
4367         LASSERT(osd_invariant(obj));
4368         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4369         RETURN(rc);
4370 }
4371
4372 /**
4373  *  Initialize osd Iterator for given osd index object.
4374  *
4375  *  \param  dt      osd index object
4376  */
4377
4378 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4379                                      struct dt_object *dt,
4380                                      __u32 unused)
4381 {
4382         struct osd_it_iam      *it;
4383         struct osd_object      *obj = osd_dt_obj(dt);
4384         struct lu_object       *lo  = &dt->do_lu;
4385         struct iam_path_descr  *ipd;
4386         struct iam_container   *bag = &obj->oo_dir->od_container;
4387
4388         if (!dt_object_exists(dt))
4389                 return ERR_PTR(-ENOENT);
4390
4391         OBD_ALLOC_PTR(it);
4392         if (it == NULL)
4393                 return ERR_PTR(-ENOMEM);
4394
4395         ipd = osd_it_ipd_get(env, bag);
4396         if (likely(ipd != NULL)) {
4397                 it->oi_obj = obj;
4398                 it->oi_ipd = ipd;
4399                 lu_object_get(lo);
4400                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4401                 return (struct dt_it *)it;
4402         } else {
4403                 OBD_FREE_PTR(it);
4404                 return ERR_PTR(-ENOMEM);
4405         }
4406 }
4407
4408 /**
4409  * free given Iterator.
4410  */
4411
4412 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4413 {
4414         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4415         struct osd_object       *obj = it->oi_obj;
4416
4417         iam_it_fini(&it->oi_it);
4418         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4419         lu_object_put(env, &obj->oo_dt.do_lu);
4420         OBD_FREE_PTR(it);
4421 }
4422
4423 /**
4424  *  Move Iterator to record specified by \a key
4425  *
4426  *  \param  di      osd iterator
4427  *  \param  key     key for index
4428  *
4429  *  \retval +ve  di points to record with least key not larger than key
4430  *  \retval  0   di points to exact matched key
4431  *  \retval -ve  failure
4432  */
4433
4434 static int osd_it_iam_get(const struct lu_env *env,
4435                           struct dt_it *di, const struct dt_key *key)
4436 {
4437         struct osd_thread_info  *oti = osd_oti_get(env);
4438         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4439
4440         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4441                 /* swab quota uid/gid */
4442                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4443                 key = (struct dt_key *)&oti->oti_quota_id;
4444         }
4445
4446         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4447 }
4448
4449 /**
4450  *  Release Iterator
4451  *
4452  *  \param  di      osd iterator
4453  */
4454 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4455 {
4456         struct osd_it_iam *it = (struct osd_it_iam *)di;
4457
4458         iam_it_put(&it->oi_it);
4459 }
4460
4461 /**
4462  *  Move iterator by one record
4463  *
4464  *  \param  di      osd iterator
4465  *
4466  *  \retval +1   end of container reached
4467  *  \retval  0   success
4468  *  \retval -ve  failure
4469  */
4470
4471 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4472 {
4473         struct osd_it_iam *it = (struct osd_it_iam *)di;
4474
4475         return iam_it_next(&it->oi_it);
4476 }
4477
4478 /**
4479  * Return pointer to the key under iterator.
4480  */
4481
4482 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4483                                  const struct dt_it *di)
4484 {
4485         struct osd_thread_info *oti = osd_oti_get(env);
4486         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4487         struct osd_object      *obj = it->oi_obj;
4488         struct dt_key          *key;
4489
4490         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4491
4492         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4493                 /* swab quota uid/gid */
4494                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4495                 key = (struct dt_key *)&oti->oti_quota_id;
4496         }
4497
4498         return key;
4499 }
4500
4501 /**
4502  * Return size of key under iterator (in bytes)
4503  */
4504
4505 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4506 {
4507         struct osd_it_iam *it = (struct osd_it_iam *)di;
4508
4509         return iam_it_key_size(&it->oi_it);
4510 }
4511
4512 static inline void
4513 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4514 {
4515         /* check if file type is required */
4516         if (ent->lde_attrs & LUDA_TYPE) {
4517                 struct luda_type *lt;
4518                 int align = sizeof(*lt) - 1;
4519
4520                 len = (len + align) & ~align;
4521                 lt = (struct luda_type *)(ent->lde_name + len);
4522                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4523         }
4524
4525         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4526 }
4527
4528 /**
4529  * build lu direct from backend fs dirent.
4530  */
4531
4532 static inline void
4533 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4534                    char *name, __u16 namelen, __u16 type, __u32 attr)
4535 {
4536         ent->lde_attrs = attr | LUDA_FID;
4537         fid_cpu_to_le(&ent->lde_fid, fid);
4538
4539         ent->lde_hash = cpu_to_le64(offset);
4540         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4541
4542         strncpy(ent->lde_name, name, namelen);
4543         ent->lde_name[namelen] = '\0';
4544         ent->lde_namelen = cpu_to_le16(namelen);
4545
4546         /* append lustre attributes */
4547         osd_it_append_attrs(ent, namelen, type);
4548 }
4549
4550 /**
4551  * Return pointer to the record under iterator.
4552  */
4553 static int osd_it_iam_rec(const struct lu_env *env,
4554                           const struct dt_it *di,
4555                           struct dt_rec *dtrec, __u32 attr)
4556 {
4557         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4558         struct osd_thread_info *info = osd_oti_get(env);
4559         ENTRY;
4560
4561         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4562                 const struct osd_fid_pack *rec;
4563                 struct lu_fid             *fid = &info->oti_fid;
4564                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4565                 char                      *name;
4566                 int                        namelen;
4567                 __u64                      hash;
4568                 int                        rc;
4569
4570                 name = (char *)iam_it_key_get(&it->oi_it);
4571                 if (IS_ERR(name))
4572                         RETURN(PTR_ERR(name));
4573
4574                 namelen = iam_it_key_size(&it->oi_it);
4575
4576                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4577                 if (IS_ERR(rec))
4578                         RETURN(PTR_ERR(rec));
4579
4580                 rc = osd_fid_unpack(fid, rec);
4581                 if (rc)
4582                         RETURN(rc);
4583
4584                 hash = iam_it_store(&it->oi_it);
4585
4586                 /* IAM does not store object type in IAM index (dir) */
4587                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4588                                    0, LUDA_FID);
4589         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4590                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4591                            (struct iam_rec *)dtrec);
4592                 osd_quota_unpack(it->oi_obj, dtrec);
4593         } else {
4594                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4595                            (struct iam_rec *)dtrec);
4596         }
4597
4598         RETURN(0);
4599 }
4600
4601 /**
4602  * Returns cookie for current Iterator position.
4603  */
4604 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4605 {
4606         struct osd_it_iam *it = (struct osd_it_iam *)di;
4607
4608         return iam_it_store(&it->oi_it);
4609 }
4610
4611 /**
4612  * Restore iterator from cookie.
4613  *
4614  * \param  di      osd iterator
4615  * \param  hash    Iterator location cookie
4616  *
4617  * \retval +ve  di points to record with least key not larger than key.
4618  * \retval  0   di points to exact matched key
4619  * \retval -ve  failure
4620  */
4621
4622 static int osd_it_iam_load(const struct lu_env *env,
4623                            const struct dt_it *di, __u64 hash)
4624 {
4625         struct osd_it_iam *it = (struct osd_it_iam *)di;
4626
4627         return iam_it_load(&it->oi_it, hash);
4628 }
4629
4630 static const struct dt_index_operations osd_index_iam_ops = {
4631         .dio_lookup         = osd_index_iam_lookup,
4632         .dio_declare_insert = osd_index_declare_iam_insert,
4633         .dio_insert         = osd_index_iam_insert,
4634         .dio_declare_delete = osd_index_declare_iam_delete,
4635         .dio_delete         = osd_index_iam_delete,
4636         .dio_it     = {
4637                 .init     = osd_it_iam_init,
4638                 .fini     = osd_it_iam_fini,
4639                 .get      = osd_it_iam_get,
4640                 .put      = osd_it_iam_put,
4641                 .next     = osd_it_iam_next,
4642                 .key      = osd_it_iam_key,
4643                 .key_size = osd_it_iam_key_size,
4644                 .rec      = osd_it_iam_rec,
4645                 .store    = osd_it_iam_store,
4646                 .load     = osd_it_iam_load
4647         }
4648 };
4649
4650
4651 /**
4652  * Creates or initializes iterator context.
4653  *
4654  * \retval struct osd_it_ea, iterator structure on success
4655  *
4656  */
4657 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4658                                     struct dt_object *dt,
4659                                     __u32 attr)
4660 {
4661         struct osd_object       *obj  = osd_dt_obj(dt);
4662         struct osd_thread_info  *info = osd_oti_get(env);
4663         struct osd_it_ea        *oie;
4664         struct file             *file;
4665         struct lu_object        *lo   = &dt->do_lu;
4666         struct dentry           *obj_dentry;
4667         ENTRY;
4668
4669         if (!dt_object_exists(dt) || obj->oo_destroyed)
4670                 RETURN(ERR_PTR(-ENOENT));
4671
4672         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4673         if (oie == NULL)
4674                 RETURN(ERR_PTR(-ENOMEM));
4675         obj_dentry = &oie->oie_dentry;
4676
4677         obj_dentry->d_inode = obj->oo_inode;
4678         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4679         obj_dentry->d_name.hash = 0;
4680
4681         oie->oie_rd_dirent       = 0;
4682         oie->oie_it_dirent       = 0;
4683         oie->oie_dirent          = NULL;
4684         if (unlikely(!info->oti_it_ea_buf_used)) {
4685                 oie->oie_buf = info->oti_it_ea_buf;
4686                 info->oti_it_ea_buf_used = 1;
4687         } else {
4688                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4689                 if (oie->oie_buf == NULL)
4690                         RETURN(ERR_PTR(-ENOMEM));
4691         }
4692         oie->oie_obj             = obj;
4693
4694         file = &oie->oie_file;
4695
4696         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4697         if (attr & LUDA_64BITHASH)
4698                 file->f_mode    = FMODE_64BITHASH;
4699         else
4700                 file->f_mode    = FMODE_32BITHASH;
4701         file->f_path.dentry     = obj_dentry;
4702         file->f_mapping         = obj->oo_inode->i_mapping;
4703         file->f_op              = obj->oo_inode->i_fop;
4704         set_file_inode(file, obj->oo_inode);
4705
4706         lu_object_get(lo);
4707         RETURN((struct dt_it *) oie);
4708 }
4709
4710 /**
4711  * Destroy or finishes iterator context.
4712  *
4713  * \param di iterator structure to be destroyed
4714  */
4715 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4716 {
4717         struct osd_thread_info  *info = osd_oti_get(env);
4718         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
4719         struct osd_object       *obj    = oie->oie_obj;
4720         struct inode            *inode  = obj->oo_inode;
4721
4722         ENTRY;
4723         oie->oie_file.f_op->release(inode, &oie->oie_file);
4724         lu_object_put(env, &obj->oo_dt.do_lu);
4725         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4726                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4727         else
4728                 info->oti_it_ea_buf_used = 0;
4729         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4730         EXIT;
4731 }
4732
4733 /**
4734  * It position the iterator at given key, so that next lookup continues from
4735  * that key Or it is similar to dio_it->load() but based on a key,
4736  * rather than file position.
4737  *
4738  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4739  * to the beginning.
4740  *
4741  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4742  */
4743 static int osd_it_ea_get(const struct lu_env *env,
4744                          struct dt_it *di, const struct dt_key *key)
4745 {
4746         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4747
4748         ENTRY;
4749         LASSERT(((const char *)key)[0] == '\0');
4750         it->oie_file.f_pos      = 0;
4751         it->oie_rd_dirent       = 0;
4752         it->oie_it_dirent       = 0;
4753         it->oie_dirent          = NULL;
4754
4755         RETURN(+1);
4756 }
4757
4758 /**
4759  * Does nothing
4760  */
4761 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4762 {
4763 }
4764
4765 struct osd_filldir_cbs {
4766 #ifdef HAVE_DIR_CONTEXT
4767         struct dir_context ctx;
4768 #endif
4769         struct osd_it_ea  *it;
4770 };
4771 /**
4772  * It is called internally by ->readdir(). It fills the
4773  * iterator's in-memory data structure with required
4774  * information i.e. name, namelen, rec_size etc.
4775  *
4776  * \param buf in which information to be filled in.
4777  * \param name name of the file in given dir
4778  *
4779  * \retval 0 on success
4780  * \retval 1 on buffer full
4781  */
4782 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4783                                loff_t offset, __u64 ino,
4784                                unsigned d_type)
4785 {
4786         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4787         struct osd_object       *obj  = it->oie_obj;
4788         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4789         struct lu_fid           *fid  = &ent->oied_fid;
4790         struct osd_fid_pack     *rec;
4791         ENTRY;
4792
4793         /* this should never happen */
4794         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4795                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4796                 RETURN(-EIO);
4797         }
4798
4799         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4800             OSD_IT_EA_BUFSIZE)
4801                 RETURN(1);
4802
4803         /* "." is just the object itself. */
4804         if (namelen == 1 && name[0] == '.') {
4805                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4806         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4807                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4808                 if (osd_fid_unpack(fid, rec) != 0)
4809                         fid_zero(fid);
4810         } else {
4811                 fid_zero(fid);
4812         }
4813         d_type &= ~LDISKFS_DIRENT_LUFID;
4814
4815         /* NOT export local root. */
4816         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4817                 ino = obj->oo_inode->i_ino;
4818                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4819         }
4820
4821         ent->oied_ino     = ino;
4822         ent->oied_off     = offset;
4823         ent->oied_namelen = namelen;
4824         ent->oied_type    = d_type;
4825
4826         memcpy(ent->oied_name, name, namelen);
4827
4828         it->oie_rd_dirent++;
4829         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4830         RETURN(0);
4831 }
4832
4833 /**
4834  * Calls ->readdir() to load a directory entry at a time
4835  * and stored it in iterator's in-memory data structure.
4836  *
4837  * \param di iterator's in memory structure
4838  *
4839  * \retval   0 on success
4840  * \retval -ve on error
4841  * \retval +1 reach the end of entry
4842  */
4843 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4844                                const struct dt_it *di)
4845 {
4846         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4847         struct osd_object  *obj   = it->oie_obj;
4848         struct inode       *inode = obj->oo_inode;
4849         struct htree_lock  *hlock = NULL;
4850         struct file        *filp  = &it->oie_file;
4851         int                 rc = 0;
4852         struct osd_filldir_cbs buf = {
4853 #ifdef HAVE_DIR_CONTEXT
4854                 .ctx.actor = osd_ldiskfs_filldir,
4855 #endif
4856                 .it = it
4857         };
4858
4859         ENTRY;
4860         it->oie_dirent = it->oie_buf;
4861         it->oie_rd_dirent = 0;
4862
4863         if (obj->oo_hl_head != NULL) {
4864                 hlock = osd_oti_get(env)->oti_hlock;
4865                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4866                                    inode, LDISKFS_HLOCK_READDIR);
4867         } else {
4868                 down_read(&obj->oo_ext_idx_sem);
4869         }
4870
4871 #ifdef HAVE_DIR_CONTEXT
4872         buf.ctx.pos = filp->f_pos;
4873         rc = inode->i_fop->iterate(filp, &buf.ctx);
4874         filp->f_pos = buf.ctx.pos;
4875 #else
4876         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4877 #endif
4878
4879         if (hlock != NULL)
4880                 ldiskfs_htree_unlock(hlock);
4881         else
4882                 up_read(&obj->oo_ext_idx_sem);
4883
4884         if (it->oie_rd_dirent == 0) {
4885                 /*If it does not get any dirent, it means it has been reached
4886                  *to the end of the dir */
4887                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4888                 if (rc == 0)
4889                         rc = 1;
4890         } else {
4891                 it->oie_dirent = it->oie_buf;
4892                 it->oie_it_dirent = 1;
4893         }
4894
4895         RETURN(rc);
4896 }
4897
4898 /**
4899  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4900  * to load a directory entry at a time and stored it in
4901  * iterator's in-memory data structure.
4902  *
4903  * \param di iterator's in memory structure
4904  *
4905  * \retval +ve iterator reached to end
4906  * \retval   0 iterator not reached to end
4907  * \retval -ve on error
4908  */
4909 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4910 {
4911         struct osd_it_ea *it = (struct osd_it_ea *)di;
4912         int rc;
4913
4914         ENTRY;
4915
4916         if (it->oie_it_dirent < it->oie_rd_dirent) {
4917                 it->oie_dirent =
4918                         (void *) it->oie_dirent +
4919                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4920                                        it->oie_dirent->oied_namelen);
4921                 it->oie_it_dirent++;
4922                 RETURN(0);
4923         } else {
4924                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4925                         rc = +1;
4926                 else
4927                         rc = osd_ldiskfs_it_fill(env, di);
4928         }
4929
4930         RETURN(rc);
4931 }
4932
4933 /**
4934  * Returns the key at current position from iterator's in memory structure.
4935  *
4936  * \param di iterator's in memory structure
4937  *
4938  * \retval key i.e. struct dt_key on success
4939  */
4940 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4941                                     const struct dt_it *di)
4942 {
4943         struct osd_it_ea *it = (struct osd_it_ea *)di;
4944
4945         return (struct dt_key *)it->oie_dirent->oied_name;
4946 }
4947
4948 /**
4949  * Returns the key's size at current position from iterator's in memory structure.
4950  *
4951  * \param di iterator's in memory structure
4952  *
4953  * \retval key_size i.e. struct dt_key on success
4954  */
4955 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4956 {
4957         struct osd_it_ea *it = (struct osd_it_ea *)di;
4958
4959         return it->oie_dirent->oied_namelen;
4960 }
4961
4962 static int
4963 osd_dirent_update(handle_t *jh, struct super_block *sb,
4964                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4965                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4966 {
4967         struct osd_fid_pack *rec;
4968         int                  rc;
4969         ENTRY;
4970
4971         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4972         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4973
4974         rc = ldiskfs_journal_get_write_access(jh, bh);
4975         if (rc != 0)
4976                 RETURN(rc);
4977
4978         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4979         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4980         rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
4981
4982         RETURN(rc);
4983 }
4984
4985 static inline int
4986 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4987 {
4988         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4989             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4990                 return 1;
4991         else
4992                 return 0;
4993 }
4994
4995 static inline int
4996 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
4997 {
4998         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
4999                  "dot_dotdot = %d\n", dot_dotdot);
5000
5001         if (LDISKFS_DIR_REC_LEN(de) >=
5002             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5003                 return 1;
5004         else
5005                 return 0;
5006 }
5007
5008 static int
5009 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5010                     struct inode *dir, struct inode *inode,
5011                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5012                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5013                     struct htree_lock *hlock)
5014 {
5015         struct dentry               *dentry;
5016         struct osd_fid_pack         *rec;
5017         struct ldiskfs_dentry_param *ldp;
5018         int                          rc;
5019         ENTRY;
5020
5021         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5022                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5023                 RETURN(0);
5024
5025         /* There is enough space to hold the FID-in-dirent. */
5026         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5027                                  dir->i_sb->s_blocksize)) {
5028                 rc = ldiskfs_journal_get_write_access(jh, bh);
5029                 if (rc != 0)
5030                         RETURN(rc);
5031
5032                 de->name[de->name_len] = 0;
5033                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5034                 rec->fp_len = sizeof(struct lu_fid) + 1;
5035                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5036                 de->file_type |= LDISKFS_DIRENT_LUFID;
5037
5038                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5039
5040                 RETURN(rc);
5041         }
5042
5043         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5044         if (rc != 0)
5045                 RETURN(rc);
5046
5047         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5048                                            ent->oied_namelen);
5049         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5050         osd_get_ldiskfs_dirent_param(ldp, fid);
5051         dentry->d_fsdata = (void *)ldp;
5052         ll_vfs_dq_init(dir);
5053         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5054         /* It is too bad, we cannot reinsert the name entry back.
5055          * That means we lose it! */
5056         if (rc != 0)
5057                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5058                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5059                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5060                        dir->i_ino, dir->i_generation,
5061                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5062
5063         RETURN(rc);
5064 }
5065
5066 static int
5067 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5068                         struct osd_it_ea *it, struct lu_fid *fid,
5069                         struct osd_inode_id *id, __u32 *attr)
5070 {
5071         struct osd_thread_info     *info        = osd_oti_get(env);
5072         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5073         struct osd_device          *dev         = osd_obj2dev(obj);
5074         struct super_block         *sb          = osd_sb(dev);
5075         const char                 *devname     =
5076                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5077         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5078         struct inode               *dir         = obj->oo_inode;
5079         struct htree_lock          *hlock       = NULL;
5080         struct buffer_head         *bh          = NULL;
5081         handle_t                   *jh          = NULL;
5082         struct ldiskfs_dir_entry_2 *de;
5083         struct dentry              *dentry;
5084         struct inode               *inode;
5085         int                         credits;
5086         int                         rc;
5087         int                         dot_dotdot  = 0;
5088         bool                        dirty       = false;
5089         ENTRY;
5090
5091         if (ent->oied_name[0] == '.') {
5092                 if (ent->oied_namelen == 1)
5093                         dot_dotdot = 1;
5094                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5095                         dot_dotdot = 2;
5096         }
5097
5098         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5099                                       ent->oied_namelen);
5100
5101         /* We need to ensure that the name entry is still valid.
5102          * Because it may be removed or renamed by other already.
5103          *
5104          * The unlink or rename operation will start journal before PDO lock,
5105          * so to avoid deadlock, here we need to start journal handle before
5106          * related PDO lock also. But because we do not know whether there
5107          * will be something to be repaired before PDO lock, we just start
5108          * journal without conditions.
5109          *
5110          * We may need to remove the name entry firstly, then insert back.
5111          * One credit is for user quota file update.
5112          * One credit is for group quota file update.
5113          * Two credits are for dirty inode. */
5114         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5115                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5116
5117 again:
5118         if (dev->od_dirent_journal != 0) {
5119                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5120                 if (IS_ERR(jh)) {
5121                         rc = PTR_ERR(jh);
5122                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5123                                "check_repair, dir = %lu/%u, credits = %d, "
5124                                "name = %.*s: rc = %d\n",
5125                                devname, dir->i_ino, dir->i_generation, credits,
5126                                ent->oied_namelen, ent->oied_name, rc);
5127                         RETURN(rc);
5128                 }
5129
5130                 if (obj->oo_hl_head != NULL) {
5131                         hlock = osd_oti_get(env)->oti_hlock;
5132                         /* "0" means exclusive lock for the whole directory.
5133                          * We need to prevent others access such name entry
5134                          * during the delete + insert. Neither HLOCK_ADD nor
5135                          * HLOCK_DEL cannot guarantee the atomicity. */
5136                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5137                 } else {
5138                         down_write(&obj->oo_ext_idx_sem);
5139                 }
5140         } else {
5141                 if (obj->oo_hl_head != NULL) {
5142                         hlock = osd_oti_get(env)->oti_hlock;
5143                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5144                                            LDISKFS_HLOCK_LOOKUP);
5145                 } else {
5146                         down_read(&obj->oo_ext_idx_sem);
5147                 }
5148         }
5149
5150         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5151         /* For dot/dotdot entry, if there is not enough space to hold the
5152          * FID-in-dirent, just keep them there. It only happens when the
5153          * device upgraded from 1.8 or restored from MDT file-level backup.
5154          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5155          * and needs to get FID from LMA when readdir, it will not affect the
5156          * performance much. */
5157         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5158             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5159                 *attr |= LUDA_IGNORE;
5160                 GOTO(out_journal, rc = 0);
5161         }
5162
5163         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5164         inode = osd_iget(info, dev, id);
5165         if (IS_ERR(inode)) {
5166                 rc = PTR_ERR(inode);
5167                 if (rc == -ENOENT || rc == -ESTALE)
5168                         rc = 1;
5169                 else
5170                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5171                                "check_repair, dir = %lu/%u, name = %.*s: "
5172                                "rc = %d\n",
5173                                devname, dir->i_ino, dir->i_generation,
5174                                ent->oied_namelen, ent->oied_name, rc);
5175
5176                 GOTO(out_journal, rc);
5177         }
5178
5179         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5180         if (rc == 0) {
5181                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5182
5183                 if (fid_is_sane(fid)) {
5184                         /* FID-in-dirent is valid. */
5185                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5186                                 GOTO(out_inode, rc = 0);
5187
5188                         /* Do not repair under dryrun mode. */
5189                         if (*attr & LUDA_VERIFY_DRYRUN) {
5190                                 *attr |= LUDA_REPAIR;
5191                                 GOTO(out_inode, rc = 0);
5192                         }
5193
5194                         if (dev->od_dirent_journal == 0) {
5195                                 iput(inode);
5196                                 brelse(bh);
5197                                 if (hlock != NULL)
5198                                         ldiskfs_htree_unlock(hlock);
5199                                 else
5200                                         up_read(&obj->oo_ext_idx_sem);
5201                                 dev->od_dirent_journal = 1;
5202                                 goto again;
5203                         }
5204
5205                         *fid = lma->lma_self_fid;
5206                         dirty = true;
5207                         /* Update the FID-in-dirent. */
5208                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5209                         if (rc == 0)
5210                                 *attr |= LUDA_REPAIR;
5211                         else
5212                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5213                                        "in the dirent, dir = %lu/%u, "
5214                                        "name = %.*s, "DFID": rc = %d\n",
5215                                        devname, dir->i_ino, dir->i_generation,
5216                                        ent->oied_namelen, ent->oied_name,
5217                                        PFID(fid), rc);
5218                 } else {
5219                         /* Do not repair under dryrun mode. */
5220                         if (*attr & LUDA_VERIFY_DRYRUN) {
5221                                 *fid = lma->lma_self_fid;
5222                                 *attr |= LUDA_REPAIR;
5223                                 GOTO(out_inode, rc = 0);
5224                         }
5225
5226                         if (dev->od_dirent_journal == 0) {
5227                                 iput(inode);
5228                                 brelse(bh);
5229                                 if (hlock != NULL)
5230                                         ldiskfs_htree_unlock(hlock);
5231                                 else
5232                                         up_read(&obj->oo_ext_idx_sem);
5233                                 dev->od_dirent_journal = 1;
5234                                 goto again;
5235                         }
5236
5237                         *fid = lma->lma_self_fid;
5238                         dirty = true;
5239                         /* Append the FID-in-dirent. */
5240                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5241                                                  fid, bh, de, hlock);
5242                         if (rc == 0)
5243                                 *attr |= LUDA_REPAIR;
5244                         else
5245                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5246                                        "after the dirent, dir = %lu/%u, "
5247                                        "name = %.*s, "DFID": rc = %d\n",
5248                                        devname, dir->i_ino, dir->i_generation,
5249                                        ent->oied_namelen, ent->oied_name,
5250                                        PFID(fid), rc);
5251                 }
5252         } else if (rc == -ENODATA) {
5253                 /* Do not repair under dryrun mode. */
5254                 if (*attr & LUDA_VERIFY_DRYRUN) {
5255                         if (fid_is_sane(fid)) {
5256                                 *attr |= LUDA_REPAIR;
5257                         } else {
5258                                 lu_igif_build(fid, inode->i_ino,
5259                                               inode->i_generation);
5260                                 *attr |= LUDA_UPGRADE;
5261                         }
5262                         GOTO(out_inode, rc = 0);
5263                 }
5264
5265                 if (dev->od_dirent_journal == 0) {
5266                         iput(inode);
5267                         brelse(bh);
5268                         if (hlock != NULL)
5269                                 ldiskfs_htree_unlock(hlock);
5270                         else
5271                                 up_read(&obj->oo_ext_idx_sem);
5272                         dev->od_dirent_journal = 1;
5273                         goto again;
5274                 }
5275
5276                 dirty = true;
5277                 if (unlikely(fid_is_sane(fid))) {
5278                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5279                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5280                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5281                         if (rc == 0)
5282                                 *attr |= LUDA_REPAIR;
5283                         else
5284                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5285                                        "update dirent, dir = %lu/%u, "
5286                                        "name = %.*s, "DFID": rc = %d\n",
5287                                        devname, dir->i_ino, dir->i_generation,
5288                                        ent->oied_namelen, ent->oied_name,
5289                                        PFID(fid), rc);
5290                 } else {
5291                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5292                         /* It is probably IGIF object. Only aappend the
5293                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5294                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5295                                                  fid, bh, de, hlock);
5296                         if (rc == 0)
5297                                 *attr |= LUDA_UPGRADE;
5298                         else
5299                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5300                                        "after the dirent, dir = %lu/%u, "
5301                                        "name = %.*s, "DFID": rc = %d\n",
5302                                        devname, dir->i_ino, dir->i_generation,
5303                                        ent->oied_namelen, ent->oied_name,
5304                                        PFID(fid), rc);
5305                 }
5306         }
5307
5308         GOTO(out_inode, rc);
5309
5310 out_inode:
5311         iput(inode);
5312
5313 out_journal:
5314         brelse(bh);
5315         if (hlock != NULL) {
5316                 ldiskfs_htree_unlock(hlock);
5317         } else {
5318                 if (dev->od_dirent_journal != 0)
5319                         up_write(&obj->oo_ext_idx_sem);
5320                 else
5321                         up_read(&obj->oo_ext_idx_sem);
5322         }
5323         if (jh != NULL)
5324                 ldiskfs_journal_stop(jh);
5325         if (rc >= 0 && !dirty)
5326                 dev->od_dirent_journal = 0;
5327         return rc;
5328 }
5329
5330 /**
5331  * Returns the value at current position from iterator's in memory structure.
5332  *
5333  * \param di struct osd_it_ea, iterator's in memory structure
5334  * \param attr attr requested for dirent.
5335  * \param lde lustre dirent
5336  *
5337  * \retval   0 no error and \param lde has correct lustre dirent.
5338  * \retval -ve on error
5339  */
5340 static inline int osd_it_ea_rec(const struct lu_env *env,
5341                                 const struct dt_it *di,
5342                                 struct dt_rec *dtrec, __u32 attr)
5343 {
5344         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5345         struct osd_object      *obj   = it->oie_obj;
5346         struct osd_device      *dev   = osd_obj2dev(obj);
5347         struct osd_thread_info *oti   = osd_oti_get(env);
5348         struct osd_inode_id    *id    = &oti->oti_id;
5349         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5350         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5351         __u32                   ino   = it->oie_dirent->oied_ino;
5352         int                     rc    = 0;
5353         ENTRY;
5354
5355         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5356
5357         if (attr & LUDA_VERIFY) {
5358                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5359                         attr |= LUDA_IGNORE;
5360                         /* If the parent is on remote MDT, and there
5361                          * is no FID-in-dirent, then we have to get
5362                          * the parent FID from the linkEA.  */
5363                         if (!fid_is_sane(fid) &&
5364                             it->oie_dirent->oied_namelen == 2 &&
5365                             it->oie_dirent->oied_name[0] == '.' &&
5366                             it->oie_dirent->oied_name[1] == '.')
5367                                 osd_get_pfid_from_linkea(env, obj, fid);
5368                 } else {
5369                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5370                                                      &attr);
5371                 }
5372         } else {
5373                 attr &= ~LU_DIRENT_ATTRS_MASK;
5374                 if (!fid_is_sane(fid)) {
5375                         bool is_dotdot = false;
5376                         if (it->oie_dirent->oied_namelen == 2 &&
5377                             it->oie_dirent->oied_name[0] == '.' &&
5378                             it->oie_dirent->oied_name[1] == '.')
5379                                 is_dotdot = true;
5380                         /* If the parent is on remote MDT, and there
5381                          * is no FID-in-dirent, then we have to get
5382                          * the parent FID from the linkEA.  */
5383                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5384                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5385                         } else {
5386                                 if (is_dotdot == false &&
5387                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5388                                         RETURN(-ENOENT);
5389
5390                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5391                         }
5392                 } else {
5393                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5394                 }
5395         }
5396
5397         /* Pack the entry anyway, at least the offset is right. */
5398         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5399                            it->oie_dirent->oied_name,
5400                            it->oie_dirent->oied_namelen,
5401                            it->oie_dirent->oied_type, attr);
5402
5403         if (rc < 0)
5404                 RETURN(rc);
5405
5406         if (osd_remote_fid(env, dev, fid))
5407                 RETURN(0);
5408
5409         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5410                 osd_add_oi_cache(oti, dev, id, fid);
5411
5412         RETURN(rc > 0 ? 0 : rc);
5413 }
5414
5415 /**
5416  * Returns the record size size at current position.
5417  *
5418  * This function will return record(lu_dirent) size in bytes.
5419  *
5420  * \param[in] env       execution environment
5421  * \param[in] di        iterator's in memory structure
5422  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5423  *                      calculate the lu_dirent size.
5424  *
5425  * \retval      record size(in bytes & in memory) of the current lu_dirent
5426  *              entry.
5427  */
5428 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5429                               __u32 attr)
5430 {
5431         struct osd_it_ea *it = (struct osd_it_ea *)di;
5432
5433         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5434 }
5435
5436 /**
5437  * Returns a cookie for current position of the iterator head, so that
5438  * user can use this cookie to load/start the iterator next time.
5439  *
5440  * \param di iterator's in memory structure
5441  *
5442  * \retval cookie for current position, on success
5443  */
5444 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5445 {
5446         struct osd_it_ea *it = (struct osd_it_ea *)di;
5447
5448         return it->oie_dirent->oied_off;
5449 }
5450
5451 /**
5452  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5453  * to load a directory entry at a time and stored it i inn,
5454  * in iterator's in-memory data structure.
5455  *
5456  * \param di struct osd_it_ea, iterator's in memory structure
5457  *
5458  * \retval +ve on success
5459  * \retval -ve on error
5460  */
5461 static int osd_it_ea_load(const struct lu_env *env,
5462                           const struct dt_it *di, __u64 hash)
5463 {
5464         struct osd_it_ea *it = (struct osd_it_ea *)di;
5465         int rc;
5466
5467         ENTRY;
5468         it->oie_file.f_pos = hash;
5469
5470         rc =  osd_ldiskfs_it_fill(env, di);
5471         if (rc > 0)
5472                 rc = -ENODATA;
5473
5474         if (rc == 0)
5475                 rc = +1;
5476
5477         RETURN(rc);
5478 }
5479
5480 /**
5481  * Index lookup function for interoperability mode (b11826).
5482  *
5483  * \param key,  key i.e. file name to be searched
5484  *
5485  * \retval +ve, on success
5486  * \retval -ve, on error
5487  */
5488 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5489                                struct dt_rec *rec, const struct dt_key *key)
5490 {
5491         struct osd_object *obj = osd_dt_obj(dt);
5492         int rc = 0;
5493
5494         ENTRY;
5495
5496         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5497         LINVRNT(osd_invariant(obj));
5498
5499         rc = osd_ea_lookup_rec(env, obj, rec, key);
5500         if (rc == 0)
5501                 rc = +1;
5502         RETURN(rc);
5503 }
5504
5505 /**
5506  * Index and Iterator operations for interoperability
5507  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5508  */
5509 static const struct dt_index_operations osd_index_ea_ops = {
5510         .dio_lookup         = osd_index_ea_lookup,
5511         .dio_declare_insert = osd_index_declare_ea_insert,
5512         .dio_insert         = osd_index_ea_insert,
5513         .dio_declare_delete = osd_index_declare_ea_delete,
5514         .dio_delete         = osd_index_ea_delete,
5515         .dio_it     = {
5516                 .init     = osd_it_ea_init,
5517                 .fini     = osd_it_ea_fini,
5518                 .get      = osd_it_ea_get,
5519                 .put      = osd_it_ea_put,
5520                 .next     = osd_it_ea_next,
5521                 .key      = osd_it_ea_key,
5522                 .key_size = osd_it_ea_key_size,
5523                 .rec      = osd_it_ea_rec,
5524                 .rec_size = osd_it_ea_rec_size,
5525                 .store    = osd_it_ea_store,
5526                 .load     = osd_it_ea_load
5527         }
5528 };
5529
5530 static void *osd_key_init(const struct lu_context *ctx,
5531                           struct lu_context_key *key)
5532 {
5533         struct osd_thread_info *info;
5534
5535         OBD_ALLOC_PTR(info);
5536         if (info == NULL)
5537                 return ERR_PTR(-ENOMEM);
5538
5539         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5540         if (info->oti_it_ea_buf == NULL)
5541                 goto out_free_info;
5542
5543         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5544
5545         info->oti_hlock = ldiskfs_htree_lock_alloc();
5546         if (info->oti_hlock == NULL)
5547                 goto out_free_ea;
5548
5549         return info;
5550
5551  out_free_ea:
5552         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5553  out_free_info:
5554         OBD_FREE_PTR(info);
5555         return ERR_PTR(-ENOMEM);
5556 }
5557
5558 static void osd_key_fini(const struct lu_context *ctx,
5559                          struct lu_context_key *key, void* data)
5560 {
5561         struct osd_thread_info *info = data;
5562
5563         if (info->oti_inode != NULL)
5564                 OBD_FREE_PTR(info->oti_inode);
5565         if (info->oti_hlock != NULL)
5566                 ldiskfs_htree_lock_free(info->oti_hlock);
5567         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5568         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5569         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5570         lu_buf_free(&info->oti_big_buf);
5571         OBD_FREE_PTR(info);
5572 }
5573
5574 static void osd_key_exit(const struct lu_context *ctx,
5575                          struct lu_context_key *key, void *data)
5576 {
5577         struct osd_thread_info *info = data;
5578
5579         LASSERT(info->oti_r_locks == 0);
5580         LASSERT(info->oti_w_locks == 0);
5581         LASSERT(info->oti_txns    == 0);
5582 }
5583
5584 /* type constructor/destructor: osd_type_init, osd_type_fini */
5585 LU_TYPE_INIT_FINI(osd, &osd_key);
5586
5587 struct lu_context_key osd_key = {
5588         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5589         .lct_init = osd_key_init,
5590         .lct_fini = osd_key_fini,
5591         .lct_exit = osd_key_exit
5592 };
5593
5594
5595 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5596                            const char *name, struct lu_device *next)
5597 {
5598         struct osd_device *osd = osd_dev(d);
5599
5600         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5601             >= sizeof(osd->od_svname))
5602                 return -E2BIG;
5603         return osd_procfs_init(osd, name);
5604 }
5605
5606 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5607 {
5608         struct seq_server_site  *ss = osd_seq_site(osd);
5609         int                     rc;
5610         ENTRY;
5611
5612         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5613                 RETURN(0);
5614
5615         if (unlikely(ss == NULL))
5616                 RETURN(-ENODEV);
5617
5618         OBD_ALLOC_PTR(osd->od_cl_seq);
5619         if (osd->od_cl_seq == NULL)
5620                 RETURN(-ENOMEM);
5621
5622         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5623                              osd->od_svname, ss->ss_server_seq);
5624         if (rc != 0) {
5625                 OBD_FREE_PTR(osd->od_cl_seq);
5626                 osd->od_cl_seq = NULL;
5627                 RETURN(rc);
5628         }
5629
5630         if (ss->ss_node_id == 0) {
5631                 /* If the OSD on the sequence controller(MDT0), then allocate
5632                  * sequence here, otherwise allocate sequence after connected
5633                  * to MDT0 (see mdt_register_lwp_callback()). */
5634                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
5635                                    &osd->od_cl_seq->lcs_space, env);
5636         }
5637
5638         RETURN(rc);
5639 }
5640
5641 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5642 {
5643         if (osd->od_cl_seq == NULL)
5644                 return;
5645
5646         seq_client_fini(osd->od_cl_seq);
5647         OBD_FREE_PTR(osd->od_cl_seq);
5648         osd->od_cl_seq = NULL;
5649 }
5650
5651 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5652 {
5653         ENTRY;
5654
5655         /* shutdown quota slave instance associated with the device */
5656         if (o->od_quota_slave != NULL) {
5657                 qsd_fini(env, o->od_quota_slave);
5658                 o->od_quota_slave = NULL;
5659         }
5660
5661         osd_fid_fini(env, o);
5662
5663         RETURN(0);
5664 }
5665
5666 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5667 {
5668         ENTRY;
5669
5670         if (o->od_mnt != NULL) {
5671                 shrink_dcache_sb(osd_sb(o));
5672                 osd_sync(env, &o->od_dt_dev);
5673
5674                 mntput(o->od_mnt);
5675                 o->od_mnt = NULL;
5676         }
5677
5678         EXIT;
5679 }
5680
5681 static int osd_mount(const struct lu_env *env,
5682                      struct osd_device *o, struct lustre_cfg *cfg)
5683 {
5684         const char              *name  = lustre_cfg_string(cfg, 0);
5685         const char              *dev  = lustre_cfg_string(cfg, 1);
5686         const char              *opts;
5687         unsigned long            page, s_flags, lmd_flags = 0;
5688         struct page             *__page;
5689         struct file_system_type *type;
5690         char                    *options = NULL;
5691         char                    *str;
5692         struct osd_thread_info  *info = osd_oti_get(env);
5693         struct lu_fid           *fid = &info->oti_fid;
5694         struct inode            *inode;
5695         int                      rc = 0, force_over_128tb = 0;
5696         ENTRY;
5697
5698         if (o->od_mnt != NULL)
5699                 RETURN(0);
5700
5701         if (strlen(dev) >= sizeof(o->od_mntdev))
5702                 RETURN(-E2BIG);
5703         strcpy(o->od_mntdev, dev);
5704
5705         str = lustre_cfg_string(cfg, 2);
5706         s_flags = simple_strtoul(str, NULL, 0);
5707         str = strstr(str, ":");
5708         if (str)
5709                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5710         opts = lustre_cfg_string(cfg, 3);
5711 #ifdef __BIG_ENDIAN
5712         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5713                 CERROR("%s: device %s extents feature is not guaranteed to "
5714                        "work on big-endian systems. Use \"bigendian_extents\" "
5715                        "mount option to override.\n", name, dev);
5716                 RETURN(-EINVAL);
5717         }
5718 #endif
5719         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5720                 force_over_128tb = 1;
5721
5722         __page = alloc_page(GFP_IOFS);
5723         if (__page == NULL)
5724                 GOTO(out, rc = -ENOMEM);
5725         page = (unsigned long)page_address(__page);
5726         options = (char *)page;
5727         *options = '\0';
5728         if (opts != NULL) {
5729                 /* strip out the options for back compatiblity */
5730                 static char *sout[] = {
5731                         "mballoc",
5732                         "iopen",
5733                         "noiopen",
5734                         "iopen_nopriv",
5735                         "extents",
5736                         "noextents",
5737                         /* strip out option we processed in osd */
5738                         "bigendian_extents",
5739                         "force_over_128tb",
5740                         NULL
5741                 };
5742                 strcat(options, opts);
5743                 for (rc = 0, str = options; sout[rc]; ) {
5744                         char *op = strstr(str, sout[rc]);
5745                         if (op == NULL) {
5746                                 rc++;
5747                                 str = options;
5748                                 continue;
5749                         }
5750                         if (op == options || *(op - 1) == ',') {
5751                                 str = op + strlen(sout[rc]);
5752                                 if (*str == ',' || *str == '\0') {
5753                                         *str == ',' ? str++ : str;
5754                                         memmove(op, str, strlen(str) + 1);
5755                                 }
5756                         }
5757                         for (str = op; *str != ',' && *str != '\0'; str++)
5758                                 ;
5759                 }
5760         } else {
5761                 strncat(options, "user_xattr,acl", 14);
5762         }
5763
5764         /* Glom up mount options */
5765         if (*options != '\0')
5766                 strcat(options, ",");
5767         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5768
5769         type = get_fs_type("ldiskfs");
5770         if (!type) {
5771                 CERROR("%s: cannot find ldiskfs module\n", name);
5772                 GOTO(out, rc = -ENODEV);
5773         }
5774
5775         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5776         module_put(type->owner);
5777
5778         if (IS_ERR(o->od_mnt)) {
5779                 rc = PTR_ERR(o->od_mnt);
5780                 o->od_mnt = NULL;
5781                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5782                 GOTO(out, rc);
5783         }
5784
5785         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
5786             force_over_128tb == 0) {
5787                 CERROR("%s: device %s LDISKFS does not support filesystems "
5788                        "greater than 128TB and can cause data corruption. "
5789                        "Use \"force_over_128tb\" mount option to override.\n",
5790                        name, dev);
5791                 GOTO(out, rc = -EINVAL);
5792         }
5793
5794 #ifdef HAVE_DEV_SET_RDONLY
5795         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5796                 CERROR("%s: underlying device %s is marked as read-only. "
5797                        "Setup failed\n", name, dev);
5798                 GOTO(out_mnt, rc = -EROFS);
5799         }
5800 #endif
5801
5802         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5803                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5804                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5805                 GOTO(out_mnt, rc = -EINVAL);
5806         }
5807
5808 #ifdef LDISKFS_MOUNT_DIRDATA
5809         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5810                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5811                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5812         else if (!o->od_is_ost)
5813                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5814                       "enabling the dirdata feature. If you do not want to "
5815                       "downgrade to Lustre-1.x again, you can enable it via "
5816                       "'tune2fs -O dirdata device'\n", name, dev);
5817 #endif
5818         inode = osd_sb(o)->s_root->d_inode;
5819         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5820         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5821         if (rc != 0) {
5822                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5823                 GOTO(out_mnt, rc);
5824         }
5825
5826         if (lmd_flags & LMD_FLG_NOSCRUB)
5827                 o->od_noscrub = 1;
5828
5829         GOTO(out, rc = 0);
5830
5831 out_mnt:
5832         mntput(o->od_mnt);
5833         o->od_mnt = NULL;
5834
5835 out:
5836         if (__page)
5837                 __free_page(__page);
5838
5839         return rc;
5840 }
5841
5842 static struct lu_device *osd_device_fini(const struct lu_env *env,
5843                                          struct lu_device *d)
5844 {
5845         struct osd_device *o = osd_dev(d);
5846         ENTRY;
5847
5848         osd_shutdown(env, o);
5849         osd_procfs_fini(o);
5850         osd_scrub_cleanup(env, o);
5851         osd_obj_map_fini(o);
5852         osd_umount(env, o);
5853
5854         RETURN(NULL);
5855 }
5856
5857 static int osd_device_init0(const struct lu_env *env,
5858                             struct osd_device *o,
5859                             struct lustre_cfg *cfg)
5860 {
5861         struct lu_device        *l = osd2lu_dev(o);
5862         struct osd_thread_info *info;
5863         int                     rc;
5864         int                     cplen = 0;
5865
5866         /* if the module was re-loaded, env can loose its keys */
5867         rc = lu_env_refill((struct lu_env *) env);
5868         if (rc)
5869                 GOTO(out, rc);
5870         info = osd_oti_get(env);
5871         LASSERT(info);
5872
5873         l->ld_ops = &osd_lu_ops;
5874         o->od_dt_dev.dd_ops = &osd_dt_ops;
5875
5876         spin_lock_init(&o->od_osfs_lock);
5877         mutex_init(&o->od_otable_mutex);
5878
5879         o->od_read_cache = 1;
5880         o->od_writethrough_cache = 1;
5881         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5882
5883         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5884                         sizeof(o->od_svname));
5885         if (cplen >= sizeof(o->od_svname)) {
5886                 rc = -E2BIG;
5887                 GOTO(out, rc);
5888         }
5889
5890         o->od_index = -1; /* -1 means index is invalid */
5891         rc = server_name2index(o->od_svname, &o->od_index, NULL);
5892         if (rc == LDD_F_SV_TYPE_OST)
5893                 o->od_is_ost = 1;
5894
5895         o->od_full_scrub_ratio = OFSR_DEFAULT;
5896         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
5897         rc = osd_mount(env, o, cfg);
5898         if (rc != 0)
5899                 GOTO(out, rc);
5900
5901         rc = osd_obj_map_init(env, o);
5902         if (rc != 0)
5903                 GOTO(out_mnt, rc);
5904
5905         rc = lu_site_init(&o->od_site, l);
5906         if (rc != 0)
5907                 GOTO(out_compat, rc);
5908         o->od_site.ls_bottom_dev = l;
5909
5910         rc = lu_site_init_finish(&o->od_site);
5911         if (rc != 0)
5912                 GOTO(out_site, rc);
5913
5914         INIT_LIST_HEAD(&o->od_ios_list);
5915         /* setup scrub, including OI files initialization */
5916         rc = osd_scrub_setup(env, o);
5917         if (rc < 0)
5918                 GOTO(out_site, rc);
5919
5920         rc = osd_procfs_init(o, o->od_svname);
5921         if (rc != 0) {
5922                 CERROR("%s: can't initialize procfs: rc = %d\n",
5923                        o->od_svname, rc);
5924                 GOTO(out_scrub, rc);
5925         }
5926
5927         LASSERT(l->ld_site->ls_linkage.next != NULL);
5928         LASSERT(l->ld_site->ls_linkage.prev != NULL);
5929
5930         /* initialize quota slave instance */
5931         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5932                                      o->od_proc_entry);
5933         if (IS_ERR(o->od_quota_slave)) {
5934                 rc = PTR_ERR(o->od_quota_slave);
5935                 o->od_quota_slave = NULL;
5936                 GOTO(out_procfs, rc);
5937         }
5938
5939         RETURN(0);
5940
5941 out_procfs:
5942         osd_procfs_fini(o);
5943 out_scrub:
5944         osd_scrub_cleanup(env, o);
5945 out_site:
5946         lu_site_fini(&o->od_site);
5947 out_compat:
5948         osd_obj_map_fini(o);
5949 out_mnt:
5950         osd_umount(env, o);
5951 out:
5952         return rc;
5953 }
5954
5955 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5956                                           struct lu_device_type *t,
5957                                           struct lustre_cfg *cfg)
5958 {
5959         struct osd_device *o;
5960         int                rc;
5961
5962         OBD_ALLOC_PTR(o);
5963         if (o == NULL)
5964                 return ERR_PTR(-ENOMEM);
5965
5966         rc = dt_device_init(&o->od_dt_dev, t);
5967         if (rc == 0) {
5968                 /* Because the ctx might be revived in dt_device_init,
5969                  * refill the env here */
5970                 lu_env_refill((struct lu_env *)env);
5971                 rc = osd_device_init0(env, o, cfg);
5972                 if (rc)
5973                         dt_device_fini(&o->od_dt_dev);
5974         }
5975
5976         if (unlikely(rc != 0))
5977                 OBD_FREE_PTR(o);
5978
5979         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5980 }
5981
5982 static struct lu_device *osd_device_free(const struct lu_env *env,
5983                                          struct lu_device *d)
5984 {
5985         struct osd_device *o = osd_dev(d);
5986         ENTRY;
5987
5988         /* XXX: make osd top device in order to release reference */
5989         d->ld_site->ls_top_dev = d;
5990         lu_site_purge(env, d->ld_site, -1);
5991         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5992                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5993                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5994         }
5995         lu_site_fini(&o->od_site);
5996         dt_device_fini(&o->od_dt_dev);
5997         OBD_FREE_PTR(o);
5998         RETURN(NULL);
5999 }
6000
6001 static int osd_process_config(const struct lu_env *env,
6002                               struct lu_device *d, struct lustre_cfg *cfg)
6003 {
6004         struct osd_device               *o = osd_dev(d);
6005         int                             rc;
6006         ENTRY;
6007
6008         switch (cfg->lcfg_command) {
6009         case LCFG_SETUP:
6010                 rc = osd_mount(env, o, cfg);
6011                 break;
6012         case LCFG_CLEANUP:
6013                 lu_dev_del_linkage(d->ld_site, d);
6014                 rc = osd_shutdown(env, o);
6015                 break;
6016         case LCFG_PARAM:
6017                 LASSERT(&o->od_dt_dev);
6018                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6019                                               cfg, &o->od_dt_dev);
6020                 if (rc > 0 || rc == -ENOSYS)
6021                         rc = class_process_proc_param(PARAM_OST,
6022                                                       lprocfs_osd_obd_vars,
6023                                                       cfg, &o->od_dt_dev);
6024                 break;
6025         default:
6026                 rc = -ENOSYS;
6027         }
6028
6029         RETURN(rc);
6030 }
6031
6032 static int osd_recovery_complete(const struct lu_env *env,
6033                                  struct lu_device *d)
6034 {
6035         struct osd_device       *osd = osd_dev(d);
6036         int                      rc = 0;
6037         ENTRY;
6038
6039         if (osd->od_quota_slave == NULL)
6040                 RETURN(0);
6041
6042         /* start qsd instance on recovery completion, this notifies the quota
6043          * slave code that we are about to process new requests now */
6044         rc = qsd_start(env, osd->od_quota_slave);
6045         RETURN(rc);
6046 }
6047
6048 /*
6049  * we use exports to track all osd users
6050  */
6051 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6052                            struct obd_device *obd, struct obd_uuid *cluuid,
6053                            struct obd_connect_data *data, void *localdata)
6054 {
6055         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6056         struct lustre_handle  conn;
6057         int                   rc;
6058         ENTRY;
6059
6060         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6061
6062         rc = class_connect(&conn, obd, cluuid);
6063         if (rc)
6064                 RETURN(rc);
6065
6066         *exp = class_conn2export(&conn);
6067
6068         spin_lock(&osd->od_osfs_lock);
6069         osd->od_connects++;
6070         spin_unlock(&osd->od_osfs_lock);
6071
6072         RETURN(0);
6073 }
6074
6075 /*
6076  * once last export (we don't count self-export) disappeared
6077  * osd can be released
6078  */
6079 static int osd_obd_disconnect(struct obd_export *exp)
6080 {
6081         struct obd_device *obd = exp->exp_obd;
6082         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6083         int                rc, release = 0;
6084         ENTRY;
6085
6086         /* Only disconnect the underlying layers on the final disconnect. */
6087         spin_lock(&osd->od_osfs_lock);
6088         osd->od_connects--;
6089         if (osd->od_connects == 0)
6090                 release = 1;
6091         spin_unlock(&osd->od_osfs_lock);
6092
6093         rc = class_disconnect(exp); /* bz 9811 */
6094
6095         if (rc == 0 && release)
6096                 class_manual_cleanup(obd);
6097         RETURN(rc);
6098 }
6099
6100 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6101                        struct lu_device *dev)
6102 {
6103         struct osd_device       *osd    = osd_dev(dev);
6104         struct lr_server_data   *lsd    =
6105                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6106         int                      result = 0;
6107         ENTRY;
6108
6109         if (osd->od_quota_slave != NULL) {
6110                 /* set up quota slave objects */
6111                 result = qsd_prepare(env, osd->od_quota_slave);
6112                 if (result != 0)
6113                         RETURN(result);
6114         }
6115
6116         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6117 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6118                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6119                         osd->od_index_in_idif = 1;
6120                 } else {
6121                         osd->od_index_in_idif = 0;
6122                         result = osd_register_proc_index_in_idif(osd);
6123                         if (result != 0)
6124                                 RETURN(result);
6125                 }
6126 #else
6127                 osd->od_index_in_idif = 1;
6128 #endif
6129         }
6130
6131         result = osd_fid_init(env, osd);
6132
6133         RETURN(result);
6134 }
6135
6136 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6137                          struct lu_fid *fid, struct md_op_data *op_data)
6138 {
6139         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6140
6141         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6142 }
6143
6144 static const struct lu_object_operations osd_lu_obj_ops = {
6145         .loo_object_init      = osd_object_init,
6146         .loo_object_delete    = osd_object_delete,
6147         .loo_object_release   = osd_object_release,
6148         .loo_object_free      = osd_object_free,
6149         .loo_object_print     = osd_object_print,
6150         .loo_object_invariant = osd_object_invariant
6151 };
6152
6153 const struct lu_device_operations osd_lu_ops = {
6154         .ldo_object_alloc      = osd_object_alloc,
6155         .ldo_process_config    = osd_process_config,
6156         .ldo_recovery_complete = osd_recovery_complete,
6157         .ldo_prepare           = osd_prepare,
6158 };
6159
6160 static const struct lu_device_type_operations osd_device_type_ops = {
6161         .ldto_init = osd_type_init,
6162         .ldto_fini = osd_type_fini,
6163
6164         .ldto_start = osd_type_start,
6165         .ldto_stop  = osd_type_stop,
6166
6167         .ldto_device_alloc = osd_device_alloc,
6168         .ldto_device_free  = osd_device_free,
6169
6170         .ldto_device_init    = osd_device_init,
6171         .ldto_device_fini    = osd_device_fini
6172 };
6173
6174 static struct lu_device_type osd_device_type = {
6175         .ldt_tags     = LU_DEVICE_DT,
6176         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6177         .ldt_ops      = &osd_device_type_ops,
6178         .ldt_ctx_tags = LCT_LOCAL,
6179 };
6180
6181 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6182 {
6183         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6184         struct super_block *sb = osd_sb(osd);
6185
6186         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6187 }
6188
6189 /*
6190  * lprocfs legacy support.
6191  */
6192 static struct obd_ops osd_obd_device_ops = {
6193         .o_owner = THIS_MODULE,
6194         .o_connect      = osd_obd_connect,
6195         .o_disconnect   = osd_obd_disconnect,
6196         .o_fid_alloc    = osd_fid_alloc,
6197         .o_health_check = osd_health_check,
6198 };
6199
6200 static int __init osd_mod_init(void)
6201 {
6202         int rc;
6203
6204 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6205         /* please, try to keep osd_thread_info smaller than a page */
6206         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6207 #endif
6208
6209         osd_oi_mod_init();
6210
6211         rc = lu_kmem_init(ldiskfs_caches);
6212         if (rc)
6213                 return rc;
6214
6215         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6216                                  lprocfs_osd_module_vars,
6217                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6218         if (rc)
6219                 lu_kmem_fini(ldiskfs_caches);
6220         return rc;
6221 }
6222
6223 static void __exit osd_mod_exit(void)
6224 {
6225         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6226         lu_kmem_fini(ldiskfs_caches);
6227 }
6228
6229 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6230 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6231 MODULE_VERSION(LUSTRE_VERSION_STRING);
6232 MODULE_LICENSE("GPL");
6233
6234 module_init(osd_mod_init);
6235 module_exit(osd_mod_exit);