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         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2501                                         dir, dot_ldp, dot_dot_ldp);
2502 }
2503
2504 /**
2505  * Create an local agent inode for remote entry
2506  */
2507 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2508                                                   struct osd_device *osd,
2509                                                   struct osd_object *pobj,
2510                                                   const struct lu_fid *fid,
2511                                                   __u32 type,
2512                                                   struct thandle *th)
2513 {
2514         struct osd_thread_info  *info = osd_oti_get(env);
2515         struct inode            *local;
2516         struct osd_thandle      *oh;
2517         int                     rc;
2518         ENTRY;
2519
2520         LASSERT(th);
2521         oh = container_of(th, struct osd_thandle, ot_super);
2522         LASSERT(oh->ot_handle->h_transaction != NULL);
2523
2524         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2525         if (IS_ERR(local)) {
2526                 CERROR("%s: create local error %d\n", osd_name(osd),
2527                        (int)PTR_ERR(local));
2528                 RETURN(local);
2529         }
2530
2531         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2532         unlock_new_inode(local);
2533
2534         /* Set special LMA flag for local agent inode */
2535         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2536         if (rc != 0) {
2537                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2538                        osd_name(osd), PFID(fid), rc);
2539                 RETURN(ERR_PTR(rc));
2540         }
2541
2542         if (!S_ISDIR(type))
2543                 RETURN(local);
2544
2545         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2546                                          lu_object_fid(&pobj->oo_dt.do_lu),
2547                                          fid, oh);
2548         if (rc != 0) {
2549                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2550                         osd_name(osd), PFID(fid), rc);
2551                 RETURN(ERR_PTR(rc));
2552         }
2553
2554         RETURN(local);
2555 }
2556
2557 /**
2558  * Delete local agent inode for remote entry
2559  */
2560 static int osd_delete_local_agent_inode(const struct lu_env *env,
2561                                         struct osd_device *osd,
2562                                         const struct lu_fid *fid,
2563                                         __u32 ino, struct osd_thandle *oh)
2564 {
2565         struct osd_thread_info  *oti = osd_oti_get(env);
2566         struct osd_inode_id     *id = &oti->oti_id;
2567         struct inode            *inode;
2568         ENTRY;
2569
2570         id->oii_ino = le32_to_cpu(ino);
2571         id->oii_gen = OSD_OII_NOGEN;
2572         inode = osd_iget(oti, osd, id);
2573         if (IS_ERR(inode)) {
2574                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2575                        PFID(fid), id->oii_ino, id->oii_gen);
2576                 RETURN(PTR_ERR(inode));
2577         }
2578
2579         clear_nlink(inode);
2580         mark_inode_dirty(inode);
2581         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2582                 osd_name(osd), PFID(fid), inode->i_ino);
2583         iput(inode);
2584         RETURN(0);
2585 }
2586
2587 /**
2588  * OSD layer object create function for interoperability mode (b11826).
2589  * This is mostly similar to osd_object_create(). Only difference being, fid is
2590  * inserted into inode ea here.
2591  *
2592  * \retval   0, on success
2593  * \retval -ve, on error
2594  */
2595 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2596                                 struct lu_attr *attr,
2597                                 struct dt_allocation_hint *hint,
2598                                 struct dt_object_format *dof,
2599                                 struct thandle *th)
2600 {
2601         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2602         struct osd_object       *obj    = osd_dt_obj(dt);
2603         struct osd_thread_info  *info   = osd_oti_get(env);
2604         int                      result;
2605
2606         ENTRY;
2607
2608         if (dt_object_exists(dt))
2609                 RETURN(-EEXIST);
2610
2611         LASSERT(osd_invariant(obj));
2612         LASSERT(!dt_object_remote(dt));
2613         LASSERT(osd_write_locked(env, obj));
2614         LASSERT(th != NULL);
2615
2616         if (unlikely(fid_is_acct(fid)))
2617                 /* Quota files can't be created from the kernel any more,
2618                  * 'tune2fs -O quota' will take care of creating them */
2619                 RETURN(-EPERM);
2620
2621         result = __osd_object_create(info, obj, attr, hint, dof, th);
2622         if (result == 0) {
2623                 if (fid_is_idif(fid) &&
2624                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2625                         struct lu_fid *tfid = &info->oti_fid;
2626                         struct ost_id *oi   = &info->oti_ostid;
2627
2628                         fid_to_ostid(fid, oi);
2629                         ostid_to_fid(tfid, oi, 0);
2630                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2631                                                 LMAC_FID_ON_OST, 0);
2632                 } else {
2633                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2634                                 fid_is_on_ost(info, osd_obj2dev(obj),
2635                                               fid, OI_CHECK_FLD) ?
2636                                 LMAC_FID_ON_OST : 0, 0);
2637                 }
2638         }
2639
2640         if (result == 0)
2641                 result = __osd_oi_insert(env, obj, fid, th);
2642
2643         LASSERT(ergo(result == 0,
2644                      dt_object_exists(dt) && !dt_object_remote(dt)));
2645         LINVRNT(osd_invariant(obj));
2646         RETURN(result);
2647 }
2648
2649 static int osd_declare_object_ref_add(const struct lu_env *env,
2650                                       struct dt_object *dt,
2651                                       struct thandle *handle)
2652 {
2653         struct osd_thandle       *oh;
2654
2655         /* it's possible that object doesn't exist yet */
2656         LASSERT(handle != NULL);
2657
2658         oh = container_of0(handle, struct osd_thandle, ot_super);
2659         LASSERT(oh->ot_handle == NULL);
2660
2661         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2662                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2663
2664         return 0;
2665 }
2666
2667 /*
2668  * Concurrency: @dt is write locked.
2669  */
2670 static int osd_object_ref_add(const struct lu_env *env,
2671                               struct dt_object *dt, struct thandle *th)
2672 {
2673         struct osd_object  *obj = osd_dt_obj(dt);
2674         struct inode       *inode = obj->oo_inode;
2675         struct osd_thandle *oh;
2676         int                 rc = 0;
2677
2678         if (!dt_object_exists(dt))
2679                 return -ENOENT;
2680
2681         LINVRNT(osd_invariant(obj));
2682         LASSERT(!dt_object_remote(dt));
2683         LASSERT(osd_write_locked(env, obj));
2684         LASSERT(th != NULL);
2685
2686         oh = container_of0(th, struct osd_thandle, ot_super);
2687         LASSERT(oh->ot_handle != NULL);
2688
2689         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2690
2691         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2692                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2693         /*
2694          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2695          * (65000) subdirectories by storing "1" in i_nlink if the link count
2696          * would otherwise overflow. Directory tranversal tools understand
2697          * that (st_nlink == 1) indicates that the filesystem dose not track
2698          * hard links count on the directory, and will not abort subdirectory
2699          * scanning early once (st_nlink - 2) subdirs have been found.
2700          *
2701          * This also has to properly handle the case of inodes with nlink == 0
2702          * in case they are being linked into the PENDING directory
2703          */
2704         spin_lock(&obj->oo_guard);
2705         if (unlikely(inode->i_nlink == 0))
2706                 /* inc_nlink from 0 may cause WARN_ON */
2707                 set_nlink(inode, 1);
2708         else {
2709                 ldiskfs_inc_count(oh->ot_handle, inode);
2710                 if (!S_ISDIR(inode->i_mode))
2711                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2712         }
2713         spin_unlock(&obj->oo_guard);
2714
2715         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2716         LINVRNT(osd_invariant(obj));
2717
2718         osd_trans_exec_check(env, th, OSD_OT_REF_ADD);
2719
2720         return rc;
2721 }
2722
2723 static int osd_declare_object_ref_del(const struct lu_env *env,
2724                                       struct dt_object *dt,
2725                                       struct thandle *handle)
2726 {
2727         struct osd_thandle *oh;
2728
2729         LASSERT(!dt_object_remote(dt));
2730         LASSERT(handle != NULL);
2731
2732         oh = container_of0(handle, struct osd_thandle, ot_super);
2733         LASSERT(oh->ot_handle == NULL);
2734
2735         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2736                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2737
2738         return 0;
2739 }
2740
2741 /*
2742  * Concurrency: @dt is write locked.
2743  */
2744 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2745                               struct thandle *th)
2746 {
2747         struct osd_object       *obj = osd_dt_obj(dt);
2748         struct inode            *inode = obj->oo_inode;
2749         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2750         struct osd_thandle      *oh;
2751
2752         if (!dt_object_exists(dt))
2753                 return -ENOENT;
2754
2755         LINVRNT(osd_invariant(obj));
2756         LASSERT(!dt_object_remote(dt));
2757         LASSERT(osd_write_locked(env, obj));
2758         LASSERT(th != NULL);
2759
2760         oh = container_of0(th, struct osd_thandle, ot_super);
2761         LASSERT(oh->ot_handle != NULL);
2762
2763         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2764
2765         spin_lock(&obj->oo_guard);
2766         /* That can be result of upgrade from old Lustre version and
2767          * applied only to local files.  Just skip this ref_del call.
2768          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2769         if (inode->i_nlink == 0) {
2770                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2771                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2772                              ", maybe an upgraded file? (LU-3915)\n",
2773                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2774                 spin_unlock(&obj->oo_guard);
2775                 return 0;
2776         }
2777
2778         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2779                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2780
2781         ldiskfs_dec_count(oh->ot_handle, inode);
2782         spin_unlock(&obj->oo_guard);
2783
2784         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2785         LINVRNT(osd_invariant(obj));
2786
2787         osd_trans_exec_check(env, th, OSD_OT_REF_DEL);
2788
2789         return 0;
2790 }
2791
2792 /*
2793  * Get the 64-bit version for an inode.
2794  */
2795 static int osd_object_version_get(const struct lu_env *env,
2796                                   struct dt_object *dt, dt_obj_version_t *ver)
2797 {
2798         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2799
2800         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2801                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2802         *ver = LDISKFS_I(inode)->i_fs_version;
2803         return 0;
2804 }
2805
2806 /*
2807  * Concurrency: @dt is read locked.
2808  */
2809 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2810                          struct lu_buf *buf, const char *name)
2811 {
2812         struct osd_object      *obj    = osd_dt_obj(dt);
2813         struct inode           *inode  = obj->oo_inode;
2814         struct osd_thread_info *info   = osd_oti_get(env);
2815         struct dentry          *dentry = &info->oti_obj_dentry;
2816
2817         /* version get is not real XATTR but uses xattr API */
2818         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2819                 /* for version we are just using xattr API but change inode
2820                  * field instead */
2821                 if (buf->lb_len == 0)
2822                         return sizeof(dt_obj_version_t);
2823
2824                 if (buf->lb_len < sizeof(dt_obj_version_t))
2825                         return -ERANGE;
2826
2827                 osd_object_version_get(env, dt, buf->lb_buf);
2828
2829                 return sizeof(dt_obj_version_t);
2830         }
2831
2832         if (!dt_object_exists(dt))
2833                 return -ENOENT;
2834
2835         LASSERT(!dt_object_remote(dt));
2836         LASSERT(inode->i_op != NULL);
2837         LASSERT(inode->i_op->getxattr != NULL);
2838
2839         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2840 }
2841
2842
2843 static int osd_declare_xattr_set(const struct lu_env *env,
2844                                  struct dt_object *dt,
2845                                  const struct lu_buf *buf, const char *name,
2846                                  int fl, struct thandle *handle)
2847 {
2848         struct osd_thandle *oh;
2849         int credits;
2850         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2851
2852         LASSERT(handle != NULL);
2853
2854         oh = container_of0(handle, struct osd_thandle, ot_super);
2855         LASSERT(oh->ot_handle == NULL);
2856
2857         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2858                 /* For non-upgrading case, the LMA is set first and
2859                  * usually fit inode. But for upgrade case, the LMA
2860                  * may be in another separated EA block. */
2861                 if (!dt_object_exists(dt))
2862                         credits = 0;
2863                 else if (fl == LU_XATTR_REPLACE)
2864                         credits = 1;
2865                 else
2866                         goto upgrade;
2867         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2868                 credits = 1;
2869         } else {
2870 upgrade:
2871                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2872                 if (buf && buf->lb_len > sb->s_blocksize) {
2873                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2874                                         sb->s_blocksize_bits;
2875                 }
2876                 /*
2877                  * xattr set may involve inode quota change, reserve credits for
2878                  * dquot_initialize()
2879                  */
2880                 credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2881         }
2882
2883         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2884
2885         return 0;
2886 }
2887
2888 /*
2889  * Set the 64-bit version for object
2890  */
2891 static void osd_object_version_set(const struct lu_env *env,
2892                                    struct dt_object *dt,
2893                                    dt_obj_version_t *new_version)
2894 {
2895         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2896
2897         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2898                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2899
2900         LDISKFS_I(inode)->i_fs_version = *new_version;
2901         /** Version is set after all inode operations are finished,
2902          *  so we should mark it dirty here */
2903         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2904 }
2905
2906 /*
2907  * Concurrency: @dt is write locked.
2908  */
2909 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2910                          const struct lu_buf *buf, const char *name, int fl,
2911                          struct thandle *handle)
2912 {
2913         struct osd_object      *obj      = osd_dt_obj(dt);
2914         struct inode           *inode    = obj->oo_inode;
2915         struct osd_thread_info *info     = osd_oti_get(env);
2916         int                     fs_flags = 0;
2917         int                     rc;
2918         ENTRY;
2919
2920         LASSERT(handle != NULL);
2921
2922         /* version set is not real XATTR */
2923         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2924                 /* for version we are just using xattr API but change inode
2925                  * field instead */
2926                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2927                 osd_object_version_set(env, dt, buf->lb_buf);
2928                 return sizeof(dt_obj_version_t);
2929         }
2930
2931         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2932                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2933
2934         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2935         if (fl & LU_XATTR_REPLACE)
2936                 fs_flags |= XATTR_REPLACE;
2937
2938         if (fl & LU_XATTR_CREATE)
2939                 fs_flags |= XATTR_CREATE;
2940
2941         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2942                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2943                 int                      rc;
2944
2945                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2946                 if (rc != 0)
2947                         RETURN(rc);
2948
2949                 lma->lma_incompat |= LMAI_STRIPED;
2950                 lustre_lma_swab(lma);
2951                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2952                                      sizeof(*lma), XATTR_REPLACE);
2953                 if (rc != 0)
2954                         RETURN(rc);
2955         }
2956
2957         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
2958             strcmp(name, XATTR_NAME_LINK) == 0)
2959                 return -ENOSPC;
2960
2961         rc = __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2962                                fs_flags);
2963         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2964
2965         return rc;
2966 }
2967
2968 /*
2969  * Concurrency: @dt is read locked.
2970  */
2971 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2972                           const struct lu_buf *buf)
2973 {
2974         struct osd_object      *obj    = osd_dt_obj(dt);
2975         struct inode           *inode  = obj->oo_inode;
2976         struct osd_thread_info *info   = osd_oti_get(env);
2977         struct dentry          *dentry = &info->oti_obj_dentry;
2978
2979         if (!dt_object_exists(dt))
2980                 return -ENOENT;
2981
2982         LASSERT(!dt_object_remote(dt));
2983         LASSERT(inode->i_op != NULL);
2984         LASSERT(inode->i_op->listxattr != NULL);
2985
2986         dentry->d_inode = inode;
2987         dentry->d_sb = inode->i_sb;
2988         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2989 }
2990
2991 static int osd_declare_xattr_del(const struct lu_env *env,
2992                                  struct dt_object *dt, const char *name,
2993                                  struct thandle *handle)
2994 {
2995         struct osd_thandle *oh;
2996         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2997
2998         LASSERT(!dt_object_remote(dt));
2999         LASSERT(handle != NULL);
3000
3001         oh = container_of0(handle, struct osd_thandle, ot_super);
3002         LASSERT(oh->ot_handle == NULL);
3003
3004         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3005                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3006         /*
3007          * xattr del may involve inode quota change, reserve credits for
3008          * dquot_initialize()
3009          */
3010         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
3011
3012         return 0;
3013 }
3014
3015 /*
3016  * Concurrency: @dt is write locked.
3017  */
3018 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3019                          const char *name, struct thandle *handle)
3020 {
3021         struct osd_object      *obj    = osd_dt_obj(dt);
3022         struct inode           *inode  = obj->oo_inode;
3023         struct osd_thread_info *info   = osd_oti_get(env);
3024         struct dentry          *dentry = &info->oti_obj_dentry;
3025         int                     rc;
3026
3027         if (!dt_object_exists(dt))
3028                 return -ENOENT;
3029
3030         LASSERT(!dt_object_remote(dt));
3031         LASSERT(inode->i_op != NULL);
3032         LASSERT(inode->i_op->removexattr != NULL);
3033         LASSERT(handle != NULL);
3034
3035         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3036
3037         ll_vfs_dq_init(inode);
3038         dentry->d_inode = inode;
3039         dentry->d_sb = inode->i_sb;
3040         rc = inode->i_op->removexattr(dentry, name);
3041         osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
3042         return rc;
3043 }
3044
3045 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
3046                            __u64 start, __u64 end)
3047 {
3048         struct osd_object       *obj    = osd_dt_obj(dt);
3049         struct inode            *inode  = obj->oo_inode;
3050         struct osd_thread_info  *info   = osd_oti_get(env);
3051         struct dentry           *dentry = &info->oti_obj_dentry;
3052         struct file             *file   = &info->oti_file;
3053         int                     rc;
3054
3055         ENTRY;
3056
3057         dentry->d_inode = inode;
3058         dentry->d_sb = inode->i_sb;
3059         file->f_path.dentry = dentry;
3060         file->f_mapping = inode->i_mapping;
3061         file->f_op = inode->i_fop;
3062         set_file_inode(file, inode);
3063
3064         rc = ll_vfs_fsync_range(file, start, end, 0);
3065
3066         RETURN(rc);
3067 }
3068
3069 /*
3070  * Index operations.
3071  */
3072
3073 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3074                            const struct dt_index_features *feat)
3075 {
3076         struct iam_descr *descr;
3077
3078         if (osd_object_is_root(o))
3079                 return feat == &dt_directory_features;
3080
3081         LASSERT(o->oo_dir != NULL);
3082
3083         descr = o->oo_dir->od_container.ic_descr;
3084         if (feat == &dt_directory_features) {
3085                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3086                         return 1;
3087                 else
3088                         return 0;
3089         } else {
3090                 return
3091                         feat->dif_keysize_min <= descr->id_key_size &&
3092                         descr->id_key_size <= feat->dif_keysize_max &&
3093                         feat->dif_recsize_min <= descr->id_rec_size &&
3094                         descr->id_rec_size <= feat->dif_recsize_max &&
3095                         !(feat->dif_flags & (DT_IND_VARKEY |
3096                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3097                         ergo(feat->dif_flags & DT_IND_UPDATE,
3098                              1 /* XXX check that object (and file system) is
3099                                 * writable */);
3100         }
3101 }
3102
3103 static int osd_iam_container_init(const struct lu_env *env,
3104                                   struct osd_object *obj,
3105                                   struct osd_directory *dir)
3106 {
3107         struct iam_container *bag = &dir->od_container;
3108         int result;
3109
3110         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3111         if (result != 0)
3112                 return result;
3113
3114         result = iam_container_setup(bag);
3115         if (result == 0)
3116                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3117         else
3118                 iam_container_fini(bag);
3119
3120         return result;
3121 }
3122
3123
3124 /*
3125  * Concurrency: no external locking is necessary.
3126  */
3127 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3128                          const struct dt_index_features *feat)
3129 {
3130         int                      result;
3131         int                      skip_iam = 0;
3132         struct osd_object       *obj = osd_dt_obj(dt);
3133
3134         LINVRNT(osd_invariant(obj));
3135
3136         if (osd_object_is_root(obj)) {
3137                 dt->do_index_ops = &osd_index_ea_ops;
3138                 result = 0;
3139         } else if (feat == &dt_directory_features) {
3140                 dt->do_index_ops = &osd_index_ea_ops;
3141                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3142                         result = 0;
3143                 else
3144                         result = -ENOTDIR;
3145                 skip_iam = 1;
3146         } else if (unlikely(feat == &dt_otable_features)) {
3147                 dt->do_index_ops = &osd_otable_ops;
3148                 return 0;
3149         } else if (unlikely(feat == &dt_acct_features)) {
3150                 dt->do_index_ops = &osd_acct_index_ops;
3151                 result = 0;
3152                 skip_iam = 1;
3153         } else if (!osd_has_index(obj)) {
3154                 struct osd_directory *dir;
3155
3156                 OBD_ALLOC_PTR(dir);
3157                 if (dir != NULL) {
3158
3159                         spin_lock(&obj->oo_guard);
3160                         if (obj->oo_dir == NULL)
3161                                 obj->oo_dir = dir;
3162                         else
3163                                 /*
3164                                  * Concurrent thread allocated container data.
3165                                  */
3166                                 OBD_FREE_PTR(dir);
3167                         spin_unlock(&obj->oo_guard);
3168                         /*
3169                          * Now, that we have container data, serialize its
3170                          * initialization.
3171                          */
3172                         down_write(&obj->oo_ext_idx_sem);
3173                         /*
3174                          * recheck under lock.
3175                          */
3176                         if (!osd_has_index(obj))
3177                                 result = osd_iam_container_init(env, obj,
3178                                                                 obj->oo_dir);
3179                         else
3180                                 result = 0;
3181                         up_write(&obj->oo_ext_idx_sem);
3182                 } else {
3183                         result = -ENOMEM;
3184                 }
3185         } else {
3186                 result = 0;
3187         }
3188
3189         if (result == 0 && skip_iam == 0) {
3190                 if (!osd_iam_index_probe(env, obj, feat))
3191                         result = -ENOTDIR;
3192         }
3193         LINVRNT(osd_invariant(obj));
3194
3195         if (result == 0 && feat == &dt_quota_glb_features &&
3196             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3197                 result = osd_quota_migration(env, dt);
3198
3199         return result;
3200 }
3201
3202 static int osd_otable_it_attr_get(const struct lu_env *env,
3203                                  struct dt_object *dt,
3204                                  struct lu_attr *attr)
3205 {
3206         attr->la_valid = 0;
3207         return 0;
3208 }
3209
3210 static const struct dt_object_operations osd_obj_ops = {
3211         .do_read_lock         = osd_object_read_lock,
3212         .do_write_lock        = osd_object_write_lock,
3213         .do_read_unlock       = osd_object_read_unlock,
3214         .do_write_unlock      = osd_object_write_unlock,
3215         .do_write_locked      = osd_object_write_locked,
3216         .do_attr_get          = osd_attr_get,
3217         .do_declare_attr_set  = osd_declare_attr_set,
3218         .do_attr_set          = osd_attr_set,
3219         .do_ah_init           = osd_ah_init,
3220         .do_declare_create    = osd_declare_object_create,
3221         .do_create            = osd_object_create,
3222         .do_declare_destroy   = osd_declare_object_destroy,
3223         .do_destroy           = osd_object_destroy,
3224         .do_index_try         = osd_index_try,
3225         .do_declare_ref_add   = osd_declare_object_ref_add,
3226         .do_ref_add           = osd_object_ref_add,
3227         .do_declare_ref_del   = osd_declare_object_ref_del,
3228         .do_ref_del           = osd_object_ref_del,
3229         .do_xattr_get         = osd_xattr_get,
3230         .do_declare_xattr_set = osd_declare_xattr_set,
3231         .do_xattr_set         = osd_xattr_set,
3232         .do_declare_xattr_del = osd_declare_xattr_del,
3233         .do_xattr_del         = osd_xattr_del,
3234         .do_xattr_list        = osd_xattr_list,
3235         .do_object_sync       = osd_object_sync,
3236 };
3237
3238 /**
3239  * dt_object_operations for interoperability mode
3240  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3241  */
3242 static const struct dt_object_operations osd_obj_ea_ops = {
3243         .do_read_lock         = osd_object_read_lock,
3244         .do_write_lock        = osd_object_write_lock,
3245         .do_read_unlock       = osd_object_read_unlock,
3246         .do_write_unlock      = osd_object_write_unlock,
3247         .do_write_locked      = osd_object_write_locked,
3248         .do_attr_get          = osd_attr_get,
3249         .do_declare_attr_set  = osd_declare_attr_set,
3250         .do_attr_set          = osd_attr_set,
3251         .do_ah_init           = osd_ah_init,
3252         .do_declare_create    = osd_declare_object_create,
3253         .do_create            = osd_object_ea_create,
3254         .do_declare_destroy   = osd_declare_object_destroy,
3255         .do_destroy           = osd_object_destroy,
3256         .do_index_try         = osd_index_try,
3257         .do_declare_ref_add   = osd_declare_object_ref_add,
3258         .do_ref_add           = osd_object_ref_add,
3259         .do_declare_ref_del   = osd_declare_object_ref_del,
3260         .do_ref_del           = osd_object_ref_del,
3261         .do_xattr_get         = osd_xattr_get,
3262         .do_declare_xattr_set = osd_declare_xattr_set,
3263         .do_xattr_set         = osd_xattr_set,
3264         .do_declare_xattr_del = osd_declare_xattr_del,
3265         .do_xattr_del         = osd_xattr_del,
3266         .do_xattr_list        = osd_xattr_list,
3267         .do_object_sync       = osd_object_sync,
3268 };
3269
3270 static const struct dt_object_operations osd_obj_otable_it_ops = {
3271         .do_attr_get    = osd_otable_it_attr_get,
3272         .do_index_try   = osd_index_try,
3273 };
3274
3275 static int osd_index_declare_iam_delete(const struct lu_env *env,
3276                                         struct dt_object *dt,
3277                                         const struct dt_key *key,
3278                                         struct thandle *handle)
3279 {
3280         struct osd_thandle    *oh;
3281
3282         oh = container_of0(handle, struct osd_thandle, ot_super);
3283         LASSERT(oh->ot_handle == NULL);
3284
3285         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3286                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3287
3288         return 0;
3289 }
3290
3291 /**
3292  *      delete a (key, value) pair from index \a dt specified by \a key
3293  *
3294  *      \param  dt      osd index object
3295  *      \param  key     key for index
3296  *      \param  rec     record reference
3297  *      \param  handle  transaction handler
3298  *
3299  *      \retval  0  success
3300  *      \retval -ve   failure
3301  */
3302 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3303                                 const struct dt_key *key,
3304                                 struct thandle *handle)
3305 {
3306         struct osd_thread_info *oti = osd_oti_get(env);
3307         struct osd_object      *obj = osd_dt_obj(dt);
3308         struct osd_thandle     *oh;
3309         struct iam_path_descr  *ipd;
3310         struct iam_container   *bag = &obj->oo_dir->od_container;
3311         int                     rc;
3312         ENTRY;
3313
3314         if (!dt_object_exists(dt))
3315                 RETURN(-ENOENT);
3316
3317         LINVRNT(osd_invariant(obj));
3318         LASSERT(!dt_object_remote(dt));
3319         LASSERT(bag->ic_object == obj->oo_inode);
3320         LASSERT(handle != NULL);
3321
3322         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3323
3324         ipd = osd_idx_ipd_get(env, bag);
3325         if (unlikely(ipd == NULL))
3326                 RETURN(-ENOMEM);
3327
3328         oh = container_of0(handle, struct osd_thandle, ot_super);
3329         LASSERT(oh->ot_handle != NULL);
3330         LASSERT(oh->ot_handle->h_transaction != NULL);
3331
3332         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3333                 /* swab quota uid/gid provided by caller */
3334                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3335                 key = (const struct dt_key *)&oti->oti_quota_id;
3336         }
3337
3338         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3339         osd_ipd_put(env, bag, ipd);
3340         LINVRNT(osd_invariant(obj));
3341         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3342         RETURN(rc);
3343 }
3344
3345 static int osd_index_declare_ea_delete(const struct lu_env *env,
3346                                        struct dt_object *dt,
3347                                        const struct dt_key *key,
3348                                        struct thandle *handle)
3349 {
3350         struct osd_thandle *oh;
3351         struct inode       *inode;
3352         int                 rc;
3353         ENTRY;
3354
3355         LASSERT(!dt_object_remote(dt));
3356         LASSERT(handle != NULL);
3357
3358         oh = container_of0(handle, struct osd_thandle, ot_super);
3359         LASSERT(oh->ot_handle == NULL);
3360
3361         /* due to DNE we may need to remove an agent inode */
3362         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3363                              osd_dto_credits_noquota[DTO_INDEX_DELETE] +
3364                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3365
3366         inode = osd_dt_obj(dt)->oo_inode;
3367         LASSERT(inode);
3368
3369         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3370                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3371         RETURN(rc);
3372 }
3373
3374 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3375                                           struct dt_rec *fid)
3376 {
3377         struct osd_fid_pack *rec;
3378         int                  rc = -ENODATA;
3379
3380         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3381                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3382                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3383                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3384                         rc = -EINVAL;
3385         }
3386         return rc;
3387 }
3388
3389 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3390                           const struct lu_fid *fid)
3391 {
3392         struct seq_server_site  *ss = osd_seq_site(osd);
3393         ENTRY;
3394
3395         /* FID seqs not in FLDB, must be local seq */
3396         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3397                 RETURN(0);
3398
3399         /* If FLD is not being initialized yet, it only happens during the
3400          * initialization, likely during mgs initialization, and we assume
3401          * this is local FID. */
3402         if (ss == NULL || ss->ss_server_fld == NULL)
3403                 RETURN(0);
3404
3405         /* Only check the local FLDB here */
3406         if (osd_seq_exists(env, osd, fid_seq(fid)))
3407                 RETURN(0);
3408
3409         RETURN(1);
3410 }
3411
3412 /**
3413  * Index delete function for interoperability mode (b11826).
3414  * It will remove the directory entry added by osd_index_ea_insert().
3415  * This entry is needed to maintain name->fid mapping.
3416  *
3417  * \param key,  key i.e. file entry to be deleted
3418  *
3419  * \retval   0, on success
3420  * \retval -ve, on error
3421  */
3422 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3423                                const struct dt_key *key, struct thandle *handle)
3424 {
3425         struct osd_object          *obj = osd_dt_obj(dt);
3426         struct inode               *dir = obj->oo_inode;
3427         struct dentry              *dentry;
3428         struct osd_thandle         *oh;
3429         struct ldiskfs_dir_entry_2 *de = NULL;
3430         struct buffer_head         *bh;
3431         struct htree_lock          *hlock = NULL;
3432         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3433         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3434         int                        rc;
3435         ENTRY;
3436
3437         if (!dt_object_exists(dt))
3438                 RETURN(-ENOENT);
3439
3440         LINVRNT(osd_invariant(obj));
3441         LASSERT(!dt_object_remote(dt));
3442         LASSERT(handle != NULL);
3443
3444         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3445
3446         oh = container_of(handle, struct osd_thandle, ot_super);
3447         LASSERT(oh->ot_handle != NULL);
3448         LASSERT(oh->ot_handle->h_transaction != NULL);
3449
3450         ll_vfs_dq_init(dir);
3451         dentry = osd_child_dentry_get(env, obj,
3452                                       (char *)key, strlen((char *)key));
3453
3454         if (obj->oo_hl_head != NULL) {
3455                 hlock = osd_oti_get(env)->oti_hlock;
3456                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3457                                    dir, LDISKFS_HLOCK_DEL);
3458         } else {
3459                 down_write(&obj->oo_ext_idx_sem);
3460         }
3461
3462         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3463         if (bh) {
3464                 __u32 ino = 0;
3465
3466                 /* If this is not the ".." entry, it might be a remote DNE
3467                  * entry and  we need to check if the FID is for a remote
3468                  * MDT.  If the FID is  not in the directory entry (e.g.
3469                  * upgraded 1.8 filesystem without dirdata enabled) then
3470                  * we need to get the FID from the LMA. For a remote directory
3471                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3472                  * case.
3473                  *
3474                  * Delete the entry before the agent inode in order to
3475                  * simplify error handling.  At worst an error after deleting
3476                  * the entry first might leak the agent inode afterward. The
3477                  * reverse would need filesystem abort in case of error deleting
3478                  * the entry after the agent had been removed, or leave a
3479                  * dangling entry pointing at a random inode. */
3480                 if (strcmp((char *)key, dotdot) != 0) {
3481                         LASSERT(de != NULL);
3482                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3483                         /* If Fid is not in dentry, try to get it from LMA */
3484                         if (rc == -ENODATA) {
3485                                 struct osd_inode_id *id;
3486                                 struct inode *inode;
3487
3488                                 /* Before trying to get fid from the inode,
3489                                  * check whether the inode is valid.
3490                                  *
3491                                  * If the inode has been deleted, do not go
3492                                  * ahead to do osd_ea_fid_get, which will set
3493                                  * the inode to bad inode, which might cause
3494                                  * the inode to be deleted uncorrectly */
3495                                 inode = ldiskfs_iget(osd_sb(osd),
3496                                                      le32_to_cpu(de->inode));
3497                                 if (IS_ERR(inode)) {
3498                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3499                                                "error.\n", osd_name(osd),
3500                                                PFID(fid));
3501                                         rc = PTR_ERR(inode);
3502                                 } else {
3503                                         if (likely(inode->i_nlink != 0)) {
3504                                                 id = &osd_oti_get(env)->oti_id;
3505                                                 rc = osd_ea_fid_get(env, obj,
3506                                                         le32_to_cpu(de->inode),
3507                                                                     fid, id);
3508                                         } else {
3509                                                 CDEBUG(D_INFO, "%s: %u "DFID
3510                                                        "deleted.\n",
3511                                                        osd_name(osd),
3512                                                        le32_to_cpu(de->inode),
3513                                                        PFID(fid));
3514                                                 rc = -ESTALE;
3515                                         }
3516                                         iput(inode);
3517                                 }
3518                         }
3519                         if (rc == 0 &&
3520                             unlikely(osd_remote_fid(env, osd, fid)))
3521                                 /* Need to delete agent inode */
3522                                 ino = le32_to_cpu(de->inode);
3523                 }
3524                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3525                 brelse(bh);
3526                 if (rc == 0 && unlikely(ino != 0)) {
3527                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3528                                                           oh);
3529                         if (rc != 0)
3530                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3531                                        osd_name(osd), PFID(fid), rc);
3532                 }
3533         } else {
3534                 rc = -ENOENT;
3535         }
3536         if (hlock != NULL)
3537                 ldiskfs_htree_unlock(hlock);
3538         else
3539                 up_write(&obj->oo_ext_idx_sem);
3540
3541         if (rc != 0)
3542                 GOTO(out, rc);
3543
3544         /* For inode on the remote MDT, .. will point to
3545          * /Agent directory, Check whether it needs to delete
3546          * from agent directory */
3547         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3548                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3549                                                    oh);
3550                 if (rc != 0 && rc != -ENOENT) {
3551                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3552                                osd_name(osd), PFID(fid), rc);
3553                 }
3554
3555                 if (rc == -ENOENT)
3556                         rc = 0;
3557
3558                 GOTO(out, rc);
3559         }
3560 out:
3561
3562         LASSERT(osd_invariant(obj));
3563         osd_trans_exec_check(env, handle, OSD_OT_DELETE);
3564         RETURN(rc);
3565 }
3566
3567 /**
3568  *      Lookup index for \a key and copy record to \a rec.
3569  *
3570  *      \param  dt      osd index object
3571  *      \param  key     key for index
3572  *      \param  rec     record reference
3573  *
3574  *      \retval  +ve  success : exact mach
3575  *      \retval  0    return record with key not greater than \a key
3576  *      \retval -ve   failure
3577  */
3578 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3579                                 struct dt_rec *rec, const struct dt_key *key)
3580 {
3581         struct osd_object      *obj = osd_dt_obj(dt);
3582         struct iam_path_descr  *ipd;
3583         struct iam_container   *bag = &obj->oo_dir->od_container;
3584         struct osd_thread_info *oti = osd_oti_get(env);
3585         struct iam_iterator    *it = &oti->oti_idx_it;
3586         struct iam_rec         *iam_rec;
3587         int                     rc;
3588         ENTRY;
3589
3590         if (!dt_object_exists(dt))
3591                 RETURN(-ENOENT);
3592
3593         LASSERT(osd_invariant(obj));
3594         LASSERT(!dt_object_remote(dt));
3595         LASSERT(bag->ic_object == obj->oo_inode);
3596
3597         ipd = osd_idx_ipd_get(env, bag);
3598         if (IS_ERR(ipd))
3599                 RETURN(-ENOMEM);
3600
3601         /* got ipd now we can start iterator. */
3602         iam_it_init(it, bag, 0, ipd);
3603
3604         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3605                 /* swab quota uid/gid provided by caller */
3606                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3607                 key = (const struct dt_key *)&oti->oti_quota_id;
3608         }
3609
3610         rc = iam_it_get(it, (struct iam_key *)key);
3611         if (rc >= 0) {
3612                 if (S_ISDIR(obj->oo_inode->i_mode))
3613                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3614                 else
3615                         iam_rec = (struct iam_rec *) rec;
3616
3617                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3618
3619                 if (S_ISDIR(obj->oo_inode->i_mode))
3620                         osd_fid_unpack((struct lu_fid *) rec,
3621                                        (struct osd_fid_pack *)iam_rec);
3622                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3623                         osd_quota_unpack(obj, rec);
3624         }
3625
3626         iam_it_put(it);
3627         iam_it_fini(it);
3628         osd_ipd_put(env, bag, ipd);
3629
3630         LINVRNT(osd_invariant(obj));
3631
3632         RETURN(rc);
3633 }
3634
3635 static int osd_index_declare_iam_insert(const struct lu_env *env,
3636                                         struct dt_object *dt,
3637                                         const struct dt_rec *rec,
3638                                         const struct dt_key *key,
3639                                         struct thandle *handle)
3640 {
3641         struct osd_thandle *oh;
3642
3643         LASSERT(handle != NULL);
3644
3645         oh = container_of0(handle, struct osd_thandle, ot_super);
3646         LASSERT(oh->ot_handle == NULL);
3647
3648         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3649                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3650
3651         return 0;
3652 }
3653
3654 /**
3655  *      Inserts (key, value) pair in \a dt index object.
3656  *
3657  *      \param  dt      osd index object
3658  *      \param  key     key for index
3659  *      \param  rec     record reference
3660  *      \param  th      transaction handler
3661  *
3662  *      \retval  0  success
3663  *      \retval -ve failure
3664  */
3665 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3666                                 const struct dt_rec *rec,
3667                                 const struct dt_key *key, struct thandle *th,
3668                                 int ignore_quota)
3669 {
3670         struct osd_object     *obj = osd_dt_obj(dt);
3671         struct iam_path_descr *ipd;
3672         struct osd_thandle    *oh;
3673         struct iam_container  *bag;
3674         struct osd_thread_info *oti = osd_oti_get(env);
3675         struct iam_rec         *iam_rec;
3676         int                     rc;
3677         ENTRY;
3678
3679         if (!dt_object_exists(dt))
3680                 RETURN(-ENOENT);
3681
3682         LINVRNT(osd_invariant(obj));
3683         LASSERT(!dt_object_remote(dt));
3684
3685         bag = &obj->oo_dir->od_container;
3686         LASSERT(bag->ic_object == obj->oo_inode);
3687         LASSERT(th != NULL);
3688
3689         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3690
3691         ipd = osd_idx_ipd_get(env, bag);
3692         if (unlikely(ipd == NULL))
3693                 RETURN(-ENOMEM);
3694
3695         oh = container_of0(th, struct osd_thandle, ot_super);
3696         LASSERT(oh->ot_handle != NULL);
3697         LASSERT(oh->ot_handle->h_transaction != NULL);
3698         if (S_ISDIR(obj->oo_inode->i_mode)) {
3699                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3700                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3701         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3702                 /* pack quota uid/gid */
3703                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3704                 key = (const struct dt_key *)&oti->oti_quota_id;
3705                 /* pack quota record */
3706                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3707                 iam_rec = (struct iam_rec *)rec;
3708         } else {
3709                 iam_rec = (struct iam_rec *)rec;
3710         }
3711
3712         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3713                         iam_rec, ipd);
3714         osd_ipd_put(env, bag, ipd);
3715         LINVRNT(osd_invariant(obj));
3716         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3717         RETURN(rc);
3718 }
3719
3720 /**
3721  * Calls ldiskfs_add_entry() to add directory entry
3722  * into the directory. This is required for
3723  * interoperability mode (b11826)
3724  *
3725  * \retval   0, on success
3726  * \retval -ve, on error
3727  */
3728 static int __osd_ea_add_rec(struct osd_thread_info *info,
3729                             struct osd_object *pobj, struct inode  *cinode,
3730                             const char *name, const struct lu_fid *fid,
3731                             struct htree_lock *hlock, struct thandle *th)
3732 {
3733         struct ldiskfs_dentry_param *ldp;
3734         struct dentry               *child;
3735         struct osd_thandle          *oth;
3736         int                          rc;
3737
3738         oth = container_of(th, struct osd_thandle, ot_super);
3739         LASSERT(oth->ot_handle != NULL);
3740         LASSERT(oth->ot_handle->h_transaction != NULL);
3741         LASSERT(pobj->oo_inode);
3742
3743         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3744         if (unlikely(pobj->oo_inode ==
3745                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3746                 ldp->edp_magic = 0;
3747         else
3748                 osd_get_ldiskfs_dirent_param(ldp, fid);
3749         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3750         child->d_fsdata = (void *)ldp;
3751         ll_vfs_dq_init(pobj->oo_inode);
3752         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3753         if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
3754                 struct ldiskfs_dir_entry_2      *de;
3755                 struct buffer_head              *bh;
3756                 int                              rc1;
3757
3758                 bh = osd_ldiskfs_find_entry(pobj->oo_inode, &child->d_name, &de,
3759                                             NULL, hlock);
3760                 if (bh != NULL) {
3761                         rc1 = ldiskfs_journal_get_write_access(oth->ot_handle,
3762                                                                bh);
3763                         if (rc1 == 0) {
3764                                 if (S_ISDIR(cinode->i_mode))
3765                                         de->file_type = LDISKFS_DIRENT_LUFID |
3766                                                         LDISKFS_FT_REG_FILE;
3767                                 else
3768                                         de->file_type = LDISKFS_DIRENT_LUFID |
3769                                                         LDISKFS_FT_DIR;
3770                                 ldiskfs_handle_dirty_metadata(oth->ot_handle,
3771                                                               NULL, bh);
3772                                 brelse(bh);
3773                         }
3774                 }
3775         }
3776
3777         RETURN(rc);
3778 }
3779
3780 /**
3781  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3782  * into the directory.Also sets flags into osd object to
3783  * indicate dot and dotdot are created. This is required for
3784  * interoperability mode (b11826)
3785  *
3786  * \param dir   directory for dot and dotdot fixup.
3787  * \param obj   child object for linking
3788  *
3789  * \retval   0, on success
3790  * \retval -ve, on error
3791  */
3792 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3793                               struct osd_object *dir,
3794                               struct inode *parent_dir, const char *name,
3795                               const struct lu_fid *dot_fid,
3796                               const struct lu_fid *dot_dot_fid,
3797                               struct thandle *th)
3798 {
3799         struct inode                *inode = dir->oo_inode;
3800         struct osd_thandle          *oth;
3801         int result = 0;
3802
3803         oth = container_of(th, struct osd_thandle, ot_super);
3804         LASSERT(oth->ot_handle->h_transaction != NULL);
3805         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3806
3807         if (strcmp(name, dot) == 0) {
3808                 if (dir->oo_compat_dot_created) {
3809                         result = -EEXIST;
3810                 } else {
3811                         LASSERT(inode == parent_dir);
3812                         dir->oo_compat_dot_created = 1;
3813                         result = 0;
3814                 }
3815         } else if (strcmp(name, dotdot) == 0) {
3816                 if (!dir->oo_compat_dot_created)
3817                         return -EINVAL;
3818                 /* in case of rename, dotdot is already created */
3819                 if (dir->oo_compat_dotdot_created) {
3820                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3821                                                 dot_dot_fid, NULL, th);
3822                 }
3823
3824                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_PARENT)) {
3825                         struct lu_fid tfid = *dot_dot_fid;
3826
3827                         tfid.f_oid--;
3828                         result = osd_add_dot_dotdot_internal(info,
3829                                         dir->oo_inode, parent_dir, dot_fid,
3830                                         &tfid, oth);
3831                 } else {
3832                         result = osd_add_dot_dotdot_internal(info,
3833                                         dir->oo_inode, parent_dir, dot_fid,
3834                                         dot_dot_fid, oth);
3835                 }
3836
3837                 if (result == 0)
3838                         dir->oo_compat_dotdot_created = 1;
3839         }
3840
3841         /* ldiskfs_init_new_dir() doesn't call ldiskfs_mark_inode_dirty()
3842          * this seem as an optimization as usually it's called
3843          * later to refresh mtime of the parent. Lustre does not
3844          * update mtime in few cases (e.g. PENDING, .lustre)
3845          * we still need to transfer i_size/etc to the buffer cache */
3846         ldiskfs_mark_inode_dirty(oth->ot_handle, dir->oo_inode);
3847
3848         return result;
3849 }
3850
3851
3852 /**
3853  * It will call the appropriate osd_add* function and return the
3854  * value, return by respective functions.
3855  */
3856 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3857                           struct inode *cinode, const char *name,
3858                           const struct lu_fid *fid, struct thandle *th)
3859 {
3860         struct osd_thread_info *info   = osd_oti_get(env);
3861         struct htree_lock      *hlock;
3862         int                     rc;
3863
3864         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3865
3866         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3867                                                    name[2] =='\0'))) {
3868                 if (hlock != NULL) {
3869                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3870                                            pobj->oo_inode, 0);
3871                 } else {
3872                         down_write(&pobj->oo_ext_idx_sem);
3873                 }
3874
3875                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3876                                         lu_object_fid(&pobj->oo_dt.do_lu),
3877                                         fid, th);
3878         } else {
3879                 if (hlock != NULL) {
3880                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3881                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3882                 } else {
3883                         down_write(&pobj->oo_ext_idx_sem);
3884                 }
3885
3886                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3887                         struct lu_fid *tfid = &info->oti_fid;
3888
3889                         *tfid = *fid;
3890                         tfid->f_ver = ~0;
3891                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3892                                               tfid, hlock, th);
3893                 } else {
3894                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3895                                               hlock, th);
3896                 }
3897         }
3898         if (hlock != NULL)
3899                 ldiskfs_htree_unlock(hlock);
3900         else
3901                 up_write(&pobj->oo_ext_idx_sem);
3902
3903         return rc;
3904 }
3905
3906 static void
3907 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3908                       struct osd_idmap_cache *oic)
3909 {
3910         struct osd_scrub    *scrub = &dev->od_scrub;
3911         struct lu_fid       *fid   = &oic->oic_fid;
3912         struct osd_inode_id *id    = &oti->oti_id;
3913         int                  once  = 0;
3914         int                  rc;
3915         ENTRY;
3916
3917         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3918                 RETURN_EXIT;
3919
3920         if (scrub->os_pos_current > id->oii_ino)
3921                 RETURN_EXIT;
3922
3923 again:
3924         rc = osd_oi_lookup(oti, dev, fid, id, OI_CHECK_FLD);
3925         if (rc != 0 && rc != -ENOENT)
3926                 RETURN_EXIT;
3927
3928         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3929                 RETURN_EXIT;
3930
3931         if (thread_is_running(&scrub->os_thread)) {
3932                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3933                 /* There is race condition between osd_oi_lookup and OI scrub.
3934                  * The OI scrub finished just after osd_oi_lookup() failure.
3935                  * Under such case, it is unnecessary to trigger OI scrub again,
3936                  * but try to call osd_oi_lookup() again. */
3937                 if (unlikely(rc == -EAGAIN))
3938                         goto again;
3939
3940                 RETURN_EXIT;
3941         }
3942
3943         if (!dev->od_noscrub && ++once == 1) {
3944                 rc = osd_scrub_start(dev, SS_AUTO_PARTIAL | SS_CLEAR_DRYRUN |
3945                                      SS_CLEAR_FAILOUT);
3946                 CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI scrub by RPC "
3947                        "for "DFID", rc = %d [2]\n",
3948                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3949                        PFID(fid), rc);
3950                 if (rc == 0 || rc == -EALREADY)
3951                         goto again;
3952         }
3953
3954         EXIT;
3955 }
3956
3957 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3958                                struct osd_device *dev,
3959                                struct osd_idmap_cache *oic,
3960                                struct lu_fid *fid, __u32 ino)
3961 {
3962         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3963         struct inode            *inode;
3964         int                      rc;
3965
3966         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3967         inode = osd_iget(oti, dev, &oic->oic_lid);
3968         if (IS_ERR(inode)) {
3969                 fid_zero(&oic->oic_fid);
3970                 return PTR_ERR(inode);
3971         }
3972
3973         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3974         iput(inode);
3975         if (rc != 0)
3976                 fid_zero(&oic->oic_fid);
3977         else
3978                 *fid = oic->oic_fid = lma->lma_self_fid;
3979         return rc;
3980 }
3981
3982 void osd_add_oi_cache(struct osd_thread_info *info, struct osd_device *osd,
3983                       struct osd_inode_id *id, const struct lu_fid *fid)
3984 {
3985         CDEBUG(D_INODE, "add "DFID" %u:%u to info %p\n", PFID(fid),
3986                id->oii_ino, id->oii_gen, info);
3987         info->oti_cache.oic_lid = *id;
3988         info->oti_cache.oic_fid = *fid;
3989         info->oti_cache.oic_dev = osd;
3990 }
3991
3992 /**
3993  * Get parent FID from the linkEA.
3994  *
3995  * For a directory which parent resides on remote MDT, to satisfy the
3996  * local e2fsck, we insert it into the /REMOTE_PARENT_DIR locally. On
3997  * the other hand, to make the lookup(..) on the directory can return
3998  * the real parent FID, we append the real parent FID after its ".."
3999  * name entry in the /REMOTE_PARENT_DIR.
4000  *
4001  * Unfortunately, such PFID-in-dirent cannot be preserved via file-level
4002  * backup. So after the restore, we cannot get the right parent FID from
4003  * its ".." name entry in the /REMOTE_PARENT_DIR. Under such case, since
4004  * we have stored the real parent FID in the directory object's linkEA,
4005  * we can parse the linkEA for the real parent FID.
4006  *
4007  * \param[in] env       pointer to the thread context
4008  * \param[in] obj       pointer to the object to be handled
4009  * \param[out]fid       pointer to the buffer to hold the parent FID
4010  *
4011  * \retval              0 for getting the real parent FID successfully
4012  * \retval              negative error number on failure
4013  */
4014 static int osd_get_pfid_from_linkea(const struct lu_env *env,
4015                                     struct osd_object *obj,
4016                                     struct lu_fid *fid)
4017 {
4018         struct osd_thread_info  *oti    = osd_oti_get(env);
4019         struct lu_buf           *buf    = &oti->oti_big_buf;
4020         struct dentry           *dentry = &oti->oti_obj_dentry;
4021         struct inode            *inode  = obj->oo_inode;
4022         struct linkea_data       ldata  = { NULL };
4023         int                      rc;
4024         ENTRY;
4025
4026         fid_zero(fid);
4027         if (!S_ISDIR(inode->i_mode))
4028                 RETURN(-EIO);
4029
4030 again:
4031         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4032                              buf->lb_buf, buf->lb_len);
4033         if (rc == -ERANGE) {
4034                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LINK,
4035                                      NULL, 0);
4036                 if (rc > 0) {
4037                         lu_buf_realloc(buf, rc);
4038                         if (buf->lb_buf == NULL)
4039                                 RETURN(-ENOMEM);
4040
4041                         goto again;
4042                 }
4043         }
4044
4045         if (unlikely(rc == 0))
4046                 RETURN(-ENODATA);
4047
4048         if (rc < 0)
4049                 RETURN(rc);
4050
4051         if (unlikely(buf->lb_buf == NULL)) {
4052                 lu_buf_realloc(buf, rc);
4053                 if (buf->lb_buf == NULL)
4054                         RETURN(-ENOMEM);
4055
4056                 goto again;
4057         }
4058
4059         ldata.ld_buf = buf;
4060         rc = linkea_init(&ldata);
4061         if (rc == 0) {
4062                 linkea_first_entry(&ldata);
4063                 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, NULL, fid);
4064         }
4065
4066         RETURN(rc);
4067 }
4068
4069 /**
4070  * Calls ->lookup() to find dentry. From dentry get inode and
4071  * read inode's ea to get fid. This is required for  interoperability
4072  * mode (b11826)
4073  *
4074  * \retval   0, on success
4075  * \retval -ve, on error
4076  */
4077 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
4078                              struct dt_rec *rec, const struct dt_key *key)
4079 {
4080         struct inode                    *dir    = obj->oo_inode;
4081         struct dentry                   *dentry;
4082         struct ldiskfs_dir_entry_2      *de;
4083         struct buffer_head              *bh;
4084         struct lu_fid                   *fid = (struct lu_fid *) rec;
4085         struct htree_lock               *hlock = NULL;
4086         int                             ino;
4087         int                             rc;
4088         ENTRY;
4089
4090         LASSERT(dir->i_op != NULL);
4091         LASSERT(dir->i_op->lookup != NULL);
4092
4093         dentry = osd_child_dentry_get(env, obj,
4094                                       (char *)key, strlen((char *)key));
4095
4096         if (obj->oo_hl_head != NULL) {
4097                 hlock = osd_oti_get(env)->oti_hlock;
4098                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4099                                    dir, LDISKFS_HLOCK_LOOKUP);
4100         } else {
4101                 down_read(&obj->oo_ext_idx_sem);
4102         }
4103
4104         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
4105         if (bh) {
4106                 struct osd_thread_info *oti = osd_oti_get(env);
4107                 struct osd_inode_id *id = &oti->oti_id;
4108                 struct osd_idmap_cache *oic = &oti->oti_cache;
4109                 struct osd_device *dev = osd_obj2dev(obj);
4110
4111                 ino = le32_to_cpu(de->inode);
4112                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
4113                         brelse(bh);
4114                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
4115                         GOTO(out, rc);
4116                 }
4117
4118                 rc = osd_get_fid_from_dentry(de, rec);
4119
4120                 /* done with de, release bh */
4121                 brelse(bh);
4122                 if (rc != 0) {
4123                         if (unlikely(ino == osd_remote_parent_ino(dev))) {
4124                                 const char *name = (const char *)key;
4125
4126                                 /* If the parent is on remote MDT, and there
4127                                  * is no FID-in-dirent, then we have to get
4128                                  * the parent FID from the linkEA.  */
4129                                 if (likely(strlen(name) == 2 &&
4130                                            name[0] == '.' && name[1] == '.'))
4131                                         rc = osd_get_pfid_from_linkea(env, obj,
4132                                                                       fid);
4133                         } else {
4134                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4135                         }
4136                 } else {
4137                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4138                 }
4139
4140                 if (rc != 0) {
4141                         fid_zero(&oic->oic_fid);
4142                         GOTO(out, rc);
4143                 }
4144
4145                 if (osd_remote_fid(env, dev, fid))
4146                         GOTO(out, rc = 0);
4147
4148                 osd_add_oi_cache(osd_oti_get(env), osd_obj2dev(obj), id, fid);
4149                 osd_consistency_check(oti, dev, oic);
4150         } else {
4151                 rc = -ENOENT;
4152         }
4153
4154         GOTO(out, rc);
4155
4156 out:
4157         if (hlock != NULL)
4158                 ldiskfs_htree_unlock(hlock);
4159         else
4160                 up_read(&obj->oo_ext_idx_sem);
4161         return rc;
4162 }
4163
4164 /**
4165  * Find the osd object for given fid.
4166  *
4167  * \param fid need to find the osd object having this fid
4168  *
4169  * \retval osd_object on success
4170  * \retval        -ve on error
4171  */
4172 static struct osd_object *osd_object_find(const struct lu_env *env,
4173                                           struct dt_object *dt,
4174                                           const struct lu_fid *fid)
4175 {
4176         struct lu_device  *ludev = dt->do_lu.lo_dev;
4177         struct osd_object *child = NULL;
4178         struct lu_object  *luch;
4179         struct lu_object  *lo;
4180
4181         /*
4182          * at this point topdev might not exist yet
4183          * (i.e. MGS is preparing profiles). so we can
4184          * not rely on topdev and instead lookup with
4185          * our device passed as topdev. this can't work
4186          * if the object isn't cached yet (as osd doesn't
4187          * allocate lu_header). IOW, the object must be
4188          * in the cache, otherwise lu_object_alloc() crashes
4189          * -bzzz
4190          */
4191         luch = lu_object_find_at(env, ludev->ld_site->ls_top_dev == NULL ?
4192                                  ludev : ludev->ld_site->ls_top_dev,
4193                                  fid, NULL);
4194         if (!IS_ERR(luch)) {
4195                 if (lu_object_exists(luch)) {
4196                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
4197                         if (lo != NULL)
4198                                 child = osd_obj(lo);
4199                         else
4200                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
4201                                                 "lu_object can't be located"
4202                                                 DFID"\n", PFID(fid));
4203
4204                         if (child == NULL) {
4205                                 lu_object_put(env, luch);
4206                                 CERROR("Unable to get osd_object\n");
4207                                 child = ERR_PTR(-ENOENT);
4208                         }
4209                 } else {
4210                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
4211                                         "lu_object does not exists "DFID"\n",
4212                                         PFID(fid));
4213                         lu_object_put(env, luch);
4214                         child = ERR_PTR(-ENOENT);
4215                 }
4216         } else {
4217                 child = ERR_CAST(luch);
4218         }
4219
4220         return child;
4221 }
4222
4223 /**
4224  * Put the osd object once done with it.
4225  *
4226  * \param obj osd object that needs to be put
4227  */
4228 static inline void osd_object_put(const struct lu_env *env,
4229                                   struct osd_object *obj)
4230 {
4231         lu_object_put(env, &obj->oo_dt.do_lu);
4232 }
4233
4234 static int osd_index_declare_ea_insert(const struct lu_env *env,
4235                                        struct dt_object *dt,
4236                                        const struct dt_rec *rec,
4237                                        const struct dt_key *key,
4238                                        struct thandle *handle)
4239 {
4240         struct osd_thandle      *oh;
4241         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
4242         struct lu_fid           *fid = (struct lu_fid *)rec;
4243         int                      credits, rc = 0;
4244         ENTRY;
4245
4246         LASSERT(!dt_object_remote(dt));
4247         LASSERT(handle != NULL);
4248
4249         oh = container_of0(handle, struct osd_thandle, ot_super);
4250         LASSERT(oh->ot_handle == NULL);
4251
4252         credits = osd_dto_credits_noquota[DTO_INDEX_INSERT];
4253         if (fid != NULL) {
4254                 rc = osd_remote_fid(env, osd, fid);
4255                 if (unlikely(rc < 0))
4256                         RETURN(rc);
4257                 if (rc > 0) {
4258                         /* a reference to remote inode is represented by an
4259                          * agent inode which we have to create */
4260                         credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
4261                         credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
4262                 }
4263                 rc = 0;
4264         }
4265
4266         osd_trans_declare_op(env, oh, OSD_OT_INSERT, credits);
4267
4268         if (osd_dt_obj(dt)->oo_inode != NULL) {
4269                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
4270
4271                 /* We ignore block quota on meta pool (MDTs), so needn't
4272                  * calculate how many blocks will be consumed by this index
4273                  * insert */
4274                 rc = osd_declare_inode_qid(env, i_uid_read(inode),
4275                                            i_gid_read(inode), 0, oh,
4276                                            osd_dt_obj(dt), true, NULL, false);
4277         }
4278
4279         RETURN(rc);
4280 }
4281
4282 /**
4283  * Index add function for interoperability mode (b11826).
4284  * It will add the directory entry.This entry is needed to
4285  * maintain name->fid mapping.
4286  *
4287  * \param key it is key i.e. file entry to be inserted
4288  * \param rec it is value of given key i.e. fid
4289  *
4290  * \retval   0, on success
4291  * \retval -ve, on error
4292  */
4293 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
4294                                const struct dt_rec *rec,
4295                                const struct dt_key *key, struct thandle *th,
4296                                int ignore_quota)
4297 {
4298         struct osd_object       *obj = osd_dt_obj(dt);
4299         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
4300         struct dt_insert_rec    *rec1   = (struct dt_insert_rec *)rec;
4301         const struct lu_fid     *fid    = rec1->rec_fid;
4302         const char              *name = (const char *)key;
4303         struct osd_thread_info  *oti   = osd_oti_get(env);
4304         struct osd_inode_id     *id    = &oti->oti_id;
4305         struct inode            *child_inode = NULL;
4306         struct osd_object       *child = NULL;
4307         int                     rc;
4308         ENTRY;
4309
4310         if (!dt_object_exists(dt))
4311                 RETURN(-ENOENT);
4312
4313         LASSERT(osd_invariant(obj));
4314         LASSERT(!dt_object_remote(dt));
4315         LASSERT(th != NULL);
4316
4317         osd_trans_exec_op(env, th, OSD_OT_INSERT);
4318
4319         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!\n", PFID(fid));
4320
4321         rc = osd_remote_fid(env, osd, fid);
4322         if (rc < 0) {
4323                 CERROR("%s: Can not find object "DFID" rc %d\n",
4324                        osd_name(osd), PFID(fid), rc);
4325                 RETURN(rc);
4326         }
4327
4328         if (rc == 1) {
4329                 /* Insert remote entry */
4330                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
4331                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
4332                         struct osd_thandle      *oh;
4333
4334                         /* If parent on remote MDT, we need put this object
4335                          * under AGENT */
4336                         oh = container_of(th, typeof(*oh), ot_super);
4337                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
4338                         if (rc != 0) {
4339                                 CERROR("%s: add "DFID" error: rc = %d\n",
4340                                        osd_name(osd),
4341                                        PFID(lu_object_fid(&dt->do_lu)), rc);
4342                                 RETURN(rc);
4343                         }
4344
4345                         child_inode = igrab(omm->omm_remote_parent->d_inode);
4346                 } else {
4347                         child_inode = osd_create_local_agent_inode(env, osd,
4348                                         obj, fid, rec1->rec_type & S_IFMT, th);
4349                         if (IS_ERR(child_inode))
4350                                 RETURN(PTR_ERR(child_inode));
4351                 }
4352         } else {
4353                 /* Insert local entry */
4354                 child = osd_object_find(env, dt, fid);
4355                 if (IS_ERR(child)) {
4356                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
4357                                osd_name(osd), PFID(fid),
4358                                id->oii_ino, id->oii_gen,
4359                                (int)PTR_ERR(child));
4360                         RETURN(PTR_ERR(child));
4361                 }
4362                 child_inode = igrab(child->oo_inode);
4363         }
4364
4365         rc = osd_ea_add_rec(env, obj, child_inode, name, fid, th);
4366
4367         CDEBUG(D_INODE, "parent %lu insert %s:%lu rc = %d\n",
4368                obj->oo_inode->i_ino, name, child_inode->i_ino, rc);
4369
4370         iput(child_inode);
4371         if (child != NULL)
4372                 osd_object_put(env, child);
4373         LASSERT(osd_invariant(obj));
4374         osd_trans_exec_check(env, th, OSD_OT_INSERT);
4375         RETURN(rc);
4376 }
4377
4378 /**
4379  *  Initialize osd Iterator for given osd index object.
4380  *
4381  *  \param  dt      osd index object
4382  */
4383
4384 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
4385                                      struct dt_object *dt,
4386                                      __u32 unused)
4387 {
4388         struct osd_it_iam      *it;
4389         struct osd_object      *obj = osd_dt_obj(dt);
4390         struct lu_object       *lo  = &dt->do_lu;
4391         struct iam_path_descr  *ipd;
4392         struct iam_container   *bag = &obj->oo_dir->od_container;
4393
4394         if (!dt_object_exists(dt))
4395                 return ERR_PTR(-ENOENT);
4396
4397         OBD_ALLOC_PTR(it);
4398         if (it == NULL)
4399                 return ERR_PTR(-ENOMEM);
4400
4401         ipd = osd_it_ipd_get(env, bag);
4402         if (likely(ipd != NULL)) {
4403                 it->oi_obj = obj;
4404                 it->oi_ipd = ipd;
4405                 lu_object_get(lo);
4406                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4407                 return (struct dt_it *)it;
4408         } else {
4409                 OBD_FREE_PTR(it);
4410                 return ERR_PTR(-ENOMEM);
4411         }
4412 }
4413
4414 /**
4415  * free given Iterator.
4416  */
4417
4418 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4419 {
4420         struct osd_it_iam       *it  = (struct osd_it_iam *)di;
4421         struct osd_object       *obj = it->oi_obj;
4422
4423         iam_it_fini(&it->oi_it);
4424         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4425         lu_object_put(env, &obj->oo_dt.do_lu);
4426         OBD_FREE_PTR(it);
4427 }
4428
4429 /**
4430  *  Move Iterator to record specified by \a key
4431  *
4432  *  \param  di      osd iterator
4433  *  \param  key     key for index
4434  *
4435  *  \retval +ve  di points to record with least key not larger than key
4436  *  \retval  0   di points to exact matched key
4437  *  \retval -ve  failure
4438  */
4439
4440 static int osd_it_iam_get(const struct lu_env *env,
4441                           struct dt_it *di, const struct dt_key *key)
4442 {
4443         struct osd_thread_info  *oti = osd_oti_get(env);
4444         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4445
4446         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4447                 /* swab quota uid/gid */
4448                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4449                 key = (struct dt_key *)&oti->oti_quota_id;
4450         }
4451
4452         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4453 }
4454
4455 /**
4456  *  Release Iterator
4457  *
4458  *  \param  di      osd iterator
4459  */
4460 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4461 {
4462         struct osd_it_iam *it = (struct osd_it_iam *)di;
4463
4464         iam_it_put(&it->oi_it);
4465 }
4466
4467 /**
4468  *  Move iterator by one record
4469  *
4470  *  \param  di      osd iterator
4471  *
4472  *  \retval +1   end of container reached
4473  *  \retval  0   success
4474  *  \retval -ve  failure
4475  */
4476
4477 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4478 {
4479         struct osd_it_iam *it = (struct osd_it_iam *)di;
4480
4481         return iam_it_next(&it->oi_it);
4482 }
4483
4484 /**
4485  * Return pointer to the key under iterator.
4486  */
4487
4488 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4489                                  const struct dt_it *di)
4490 {
4491         struct osd_thread_info *oti = osd_oti_get(env);
4492         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4493         struct osd_object      *obj = it->oi_obj;
4494         struct dt_key          *key;
4495
4496         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4497
4498         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4499                 /* swab quota uid/gid */
4500                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4501                 key = (struct dt_key *)&oti->oti_quota_id;
4502         }
4503
4504         return key;
4505 }
4506
4507 /**
4508  * Return size of key under iterator (in bytes)
4509  */
4510
4511 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4512 {
4513         struct osd_it_iam *it = (struct osd_it_iam *)di;
4514
4515         return iam_it_key_size(&it->oi_it);
4516 }
4517
4518 static inline void
4519 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4520 {
4521         /* check if file type is required */
4522         if (ent->lde_attrs & LUDA_TYPE) {
4523                 struct luda_type *lt;
4524                 int align = sizeof(*lt) - 1;
4525
4526                 len = (len + align) & ~align;
4527                 lt = (struct luda_type *)(ent->lde_name + len);
4528                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4529         }
4530
4531         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4532 }
4533
4534 /**
4535  * build lu direct from backend fs dirent.
4536  */
4537
4538 static inline void
4539 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4540                    char *name, __u16 namelen, __u16 type, __u32 attr)
4541 {
4542         ent->lde_attrs = attr | LUDA_FID;
4543         fid_cpu_to_le(&ent->lde_fid, fid);
4544
4545         ent->lde_hash = cpu_to_le64(offset);
4546         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4547
4548         strncpy(ent->lde_name, name, namelen);
4549         ent->lde_name[namelen] = '\0';
4550         ent->lde_namelen = cpu_to_le16(namelen);
4551
4552         /* append lustre attributes */
4553         osd_it_append_attrs(ent, namelen, type);
4554 }
4555
4556 /**
4557  * Return pointer to the record under iterator.
4558  */
4559 static int osd_it_iam_rec(const struct lu_env *env,
4560                           const struct dt_it *di,
4561                           struct dt_rec *dtrec, __u32 attr)
4562 {
4563         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4564         struct osd_thread_info *info = osd_oti_get(env);
4565         ENTRY;
4566
4567         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4568                 const struct osd_fid_pack *rec;
4569                 struct lu_fid             *fid = &info->oti_fid;
4570                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4571                 char                      *name;
4572                 int                        namelen;
4573                 __u64                      hash;
4574                 int                        rc;
4575
4576                 name = (char *)iam_it_key_get(&it->oi_it);
4577                 if (IS_ERR(name))
4578                         RETURN(PTR_ERR(name));
4579
4580                 namelen = iam_it_key_size(&it->oi_it);
4581
4582                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4583                 if (IS_ERR(rec))
4584                         RETURN(PTR_ERR(rec));
4585
4586                 rc = osd_fid_unpack(fid, rec);
4587                 if (rc)
4588                         RETURN(rc);
4589
4590                 hash = iam_it_store(&it->oi_it);
4591
4592                 /* IAM does not store object type in IAM index (dir) */
4593                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4594                                    0, LUDA_FID);
4595         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4596                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4597                            (struct iam_rec *)dtrec);
4598                 osd_quota_unpack(it->oi_obj, dtrec);
4599         } else {
4600                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4601                            (struct iam_rec *)dtrec);
4602         }
4603
4604         RETURN(0);
4605 }
4606
4607 /**
4608  * Returns cookie for current Iterator position.
4609  */
4610 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4611 {
4612         struct osd_it_iam *it = (struct osd_it_iam *)di;
4613
4614         return iam_it_store(&it->oi_it);
4615 }
4616
4617 /**
4618  * Restore iterator from cookie.
4619  *
4620  * \param  di      osd iterator
4621  * \param  hash    Iterator location cookie
4622  *
4623  * \retval +ve  di points to record with least key not larger than key.
4624  * \retval  0   di points to exact matched key
4625  * \retval -ve  failure
4626  */
4627
4628 static int osd_it_iam_load(const struct lu_env *env,
4629                            const struct dt_it *di, __u64 hash)
4630 {
4631         struct osd_it_iam *it = (struct osd_it_iam *)di;
4632
4633         return iam_it_load(&it->oi_it, hash);
4634 }
4635
4636 static const struct dt_index_operations osd_index_iam_ops = {
4637         .dio_lookup         = osd_index_iam_lookup,
4638         .dio_declare_insert = osd_index_declare_iam_insert,
4639         .dio_insert         = osd_index_iam_insert,
4640         .dio_declare_delete = osd_index_declare_iam_delete,
4641         .dio_delete         = osd_index_iam_delete,
4642         .dio_it     = {
4643                 .init     = osd_it_iam_init,
4644                 .fini     = osd_it_iam_fini,
4645                 .get      = osd_it_iam_get,
4646                 .put      = osd_it_iam_put,
4647                 .next     = osd_it_iam_next,
4648                 .key      = osd_it_iam_key,
4649                 .key_size = osd_it_iam_key_size,
4650                 .rec      = osd_it_iam_rec,
4651                 .store    = osd_it_iam_store,
4652                 .load     = osd_it_iam_load
4653         }
4654 };
4655
4656
4657 /**
4658  * Creates or initializes iterator context.
4659  *
4660  * \retval struct osd_it_ea, iterator structure on success
4661  *
4662  */
4663 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4664                                     struct dt_object *dt,
4665                                     __u32 attr)
4666 {
4667         struct osd_object       *obj  = osd_dt_obj(dt);
4668         struct osd_thread_info  *info = osd_oti_get(env);
4669         struct osd_it_ea        *oie;
4670         struct file             *file;
4671         struct lu_object        *lo   = &dt->do_lu;
4672         struct dentry           *obj_dentry;
4673         ENTRY;
4674
4675         if (!dt_object_exists(dt) || obj->oo_destroyed)
4676                 RETURN(ERR_PTR(-ENOENT));
4677
4678         OBD_SLAB_ALLOC_PTR_GFP(oie, osd_itea_cachep, GFP_NOFS);
4679         if (oie == NULL)
4680                 RETURN(ERR_PTR(-ENOMEM));
4681         obj_dentry = &oie->oie_dentry;
4682
4683         obj_dentry->d_inode = obj->oo_inode;
4684         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4685         obj_dentry->d_name.hash = 0;
4686
4687         oie->oie_rd_dirent       = 0;
4688         oie->oie_it_dirent       = 0;
4689         oie->oie_dirent          = NULL;
4690         if (unlikely(!info->oti_it_ea_buf_used)) {
4691                 oie->oie_buf = info->oti_it_ea_buf;
4692                 info->oti_it_ea_buf_used = 1;
4693         } else {
4694                 OBD_ALLOC(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4695                 if (oie->oie_buf == NULL)
4696                         RETURN(ERR_PTR(-ENOMEM));
4697         }
4698         oie->oie_obj             = obj;
4699
4700         file = &oie->oie_file;
4701
4702         /* Only FMODE_64BITHASH or FMODE_32BITHASH should be set, NOT both. */
4703         if (attr & LUDA_64BITHASH)
4704                 file->f_mode    = FMODE_64BITHASH;
4705         else
4706                 file->f_mode    = FMODE_32BITHASH;
4707         file->f_path.dentry     = obj_dentry;
4708         file->f_mapping         = obj->oo_inode->i_mapping;
4709         file->f_op              = obj->oo_inode->i_fop;
4710         set_file_inode(file, obj->oo_inode);
4711
4712         lu_object_get(lo);
4713         RETURN((struct dt_it *) oie);
4714 }
4715
4716 /**
4717  * Destroy or finishes iterator context.
4718  *
4719  * \param di iterator structure to be destroyed
4720  */
4721 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4722 {
4723         struct osd_thread_info  *info = osd_oti_get(env);
4724         struct osd_it_ea        *oie    = (struct osd_it_ea *)di;
4725         struct osd_object       *obj    = oie->oie_obj;
4726         struct inode            *inode  = obj->oo_inode;
4727
4728         ENTRY;
4729         oie->oie_file.f_op->release(inode, &oie->oie_file);
4730         lu_object_put(env, &obj->oo_dt.do_lu);
4731         if (unlikely(oie->oie_buf != info->oti_it_ea_buf))
4732                 OBD_FREE(oie->oie_buf, OSD_IT_EA_BUFSIZE);
4733         else
4734                 info->oti_it_ea_buf_used = 0;
4735         OBD_SLAB_FREE_PTR(oie, osd_itea_cachep);
4736         EXIT;
4737 }
4738
4739 /**
4740  * It position the iterator at given key, so that next lookup continues from
4741  * that key Or it is similar to dio_it->load() but based on a key,
4742  * rather than file position.
4743  *
4744  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4745  * to the beginning.
4746  *
4747  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4748  */
4749 static int osd_it_ea_get(const struct lu_env *env,
4750                          struct dt_it *di, const struct dt_key *key)
4751 {
4752         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4753
4754         ENTRY;
4755         LASSERT(((const char *)key)[0] == '\0');
4756         it->oie_file.f_pos      = 0;
4757         it->oie_rd_dirent       = 0;
4758         it->oie_it_dirent       = 0;
4759         it->oie_dirent          = NULL;
4760
4761         RETURN(+1);
4762 }
4763
4764 /**
4765  * Does nothing
4766  */
4767 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4768 {
4769 }
4770
4771 struct osd_filldir_cbs {
4772 #ifdef HAVE_DIR_CONTEXT
4773         struct dir_context ctx;
4774 #endif
4775         struct osd_it_ea  *it;
4776 };
4777 /**
4778  * It is called internally by ->readdir(). It fills the
4779  * iterator's in-memory data structure with required
4780  * information i.e. name, namelen, rec_size etc.
4781  *
4782  * \param buf in which information to be filled in.
4783  * \param name name of the file in given dir
4784  *
4785  * \retval 0 on success
4786  * \retval 1 on buffer full
4787  */
4788 static int osd_ldiskfs_filldir(void *buf, const char *name, int namelen,
4789                                loff_t offset, __u64 ino,
4790                                unsigned d_type)
4791 {
4792         struct osd_it_ea        *it   = ((struct osd_filldir_cbs *)buf)->it;
4793         struct osd_object       *obj  = it->oie_obj;
4794         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4795         struct lu_fid           *fid  = &ent->oied_fid;
4796         struct osd_fid_pack     *rec;
4797         ENTRY;
4798
4799         /* this should never happen */
4800         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4801                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4802                 RETURN(-EIO);
4803         }
4804
4805         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4806             OSD_IT_EA_BUFSIZE)
4807                 RETURN(1);
4808
4809         /* "." is just the object itself. */
4810         if (namelen == 1 && name[0] == '.') {
4811                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4812         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4813                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4814                 if (osd_fid_unpack(fid, rec) != 0)
4815                         fid_zero(fid);
4816         } else {
4817                 fid_zero(fid);
4818         }
4819         d_type &= ~LDISKFS_DIRENT_LUFID;
4820
4821         /* NOT export local root. */
4822         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4823                 ino = obj->oo_inode->i_ino;
4824                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4825         }
4826
4827         ent->oied_ino     = ino;
4828         ent->oied_off     = offset;
4829         ent->oied_namelen = namelen;
4830         ent->oied_type    = d_type;
4831
4832         memcpy(ent->oied_name, name, namelen);
4833
4834         it->oie_rd_dirent++;
4835         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4836         RETURN(0);
4837 }
4838
4839 /**
4840  * Calls ->readdir() to load a directory entry at a time
4841  * and stored it in iterator's in-memory data structure.
4842  *
4843  * \param di iterator's in memory structure
4844  *
4845  * \retval   0 on success
4846  * \retval -ve on error
4847  * \retval +1 reach the end of entry
4848  */
4849 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4850                                const struct dt_it *di)
4851 {
4852         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4853         struct osd_object  *obj   = it->oie_obj;
4854         struct inode       *inode = obj->oo_inode;
4855         struct htree_lock  *hlock = NULL;
4856         struct file        *filp  = &it->oie_file;
4857         int                 rc = 0;
4858         struct osd_filldir_cbs buf = {
4859 #ifdef HAVE_DIR_CONTEXT
4860                 .ctx.actor = osd_ldiskfs_filldir,
4861 #endif
4862                 .it = it
4863         };
4864
4865         ENTRY;
4866         it->oie_dirent = it->oie_buf;
4867         it->oie_rd_dirent = 0;
4868
4869         if (obj->oo_hl_head != NULL) {
4870                 hlock = osd_oti_get(env)->oti_hlock;
4871                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4872                                    inode, LDISKFS_HLOCK_READDIR);
4873         } else {
4874                 down_read(&obj->oo_ext_idx_sem);
4875         }
4876
4877 #ifdef HAVE_DIR_CONTEXT
4878         buf.ctx.pos = filp->f_pos;
4879         rc = inode->i_fop->iterate(filp, &buf.ctx);
4880         filp->f_pos = buf.ctx.pos;
4881 #else
4882         rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
4883 #endif
4884
4885         if (hlock != NULL)
4886                 ldiskfs_htree_unlock(hlock);
4887         else
4888                 up_read(&obj->oo_ext_idx_sem);
4889
4890         if (it->oie_rd_dirent == 0) {
4891                 /*If it does not get any dirent, it means it has been reached
4892                  *to the end of the dir */
4893                 it->oie_file.f_pos = ldiskfs_get_htree_eof(&it->oie_file);
4894                 if (rc == 0)
4895                         rc = 1;
4896         } else {
4897                 it->oie_dirent = it->oie_buf;
4898                 it->oie_it_dirent = 1;
4899         }
4900
4901         RETURN(rc);
4902 }
4903
4904 /**
4905  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4906  * to load a directory entry at a time and stored it in
4907  * iterator's in-memory data structure.
4908  *
4909  * \param di iterator's in memory structure
4910  *
4911  * \retval +ve iterator reached to end
4912  * \retval   0 iterator not reached to end
4913  * \retval -ve on error
4914  */
4915 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4916 {
4917         struct osd_it_ea *it = (struct osd_it_ea *)di;
4918         int rc;
4919
4920         ENTRY;
4921
4922         if (it->oie_it_dirent < it->oie_rd_dirent) {
4923                 it->oie_dirent =
4924                         (void *) it->oie_dirent +
4925                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4926                                        it->oie_dirent->oied_namelen);
4927                 it->oie_it_dirent++;
4928                 RETURN(0);
4929         } else {
4930                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4931                         rc = +1;
4932                 else
4933                         rc = osd_ldiskfs_it_fill(env, di);
4934         }
4935
4936         RETURN(rc);
4937 }
4938
4939 /**
4940  * Returns the key at current position from iterator's in memory structure.
4941  *
4942  * \param di iterator's in memory structure
4943  *
4944  * \retval key i.e. struct dt_key on success
4945  */
4946 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4947                                     const struct dt_it *di)
4948 {
4949         struct osd_it_ea *it = (struct osd_it_ea *)di;
4950
4951         return (struct dt_key *)it->oie_dirent->oied_name;
4952 }
4953
4954 /**
4955  * Returns the key's size at current position from iterator's in memory structure.
4956  *
4957  * \param di iterator's in memory structure
4958  *
4959  * \retval key_size i.e. struct dt_key on success
4960  */
4961 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4962 {
4963         struct osd_it_ea *it = (struct osd_it_ea *)di;
4964
4965         return it->oie_dirent->oied_namelen;
4966 }
4967
4968 static int
4969 osd_dirent_update(handle_t *jh, struct super_block *sb,
4970                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4971                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4972 {
4973         struct osd_fid_pack *rec;
4974         int                  rc;
4975         ENTRY;
4976
4977         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4978         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4979
4980         rc = ldiskfs_journal_get_write_access(jh, bh);
4981         if (rc != 0)
4982                 RETURN(rc);
4983
4984         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4985         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4986         rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
4987
4988         RETURN(rc);
4989 }
4990
4991 static inline int
4992 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4993 {
4994         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4995             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4996                 return 1;
4997         else
4998                 return 0;
4999 }
5000
5001 static inline int
5002 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
5003 {
5004         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
5005                  "dot_dotdot = %d\n", dot_dotdot);
5006
5007         if (LDISKFS_DIR_REC_LEN(de) >=
5008             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
5009                 return 1;
5010         else
5011                 return 0;
5012 }
5013
5014 static int
5015 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
5016                     struct inode *dir, struct inode *inode,
5017                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
5018                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
5019                     struct htree_lock *hlock)
5020 {
5021         struct dentry               *dentry;
5022         struct osd_fid_pack         *rec;
5023         struct ldiskfs_dentry_param *ldp;
5024         int                          rc;
5025         ENTRY;
5026
5027         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
5028                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5029                 RETURN(0);
5030
5031         /* There is enough space to hold the FID-in-dirent. */
5032         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
5033                                  dir->i_sb->s_blocksize)) {
5034                 rc = ldiskfs_journal_get_write_access(jh, bh);
5035                 if (rc != 0)
5036                         RETURN(rc);
5037
5038                 de->name[de->name_len] = 0;
5039                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
5040                 rec->fp_len = sizeof(struct lu_fid) + 1;
5041                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
5042                 de->file_type |= LDISKFS_DIRENT_LUFID;
5043
5044                 rc = ldiskfs_handle_dirty_metadata(jh, NULL, bh);
5045
5046                 RETURN(rc);
5047         }
5048
5049         rc = ldiskfs_delete_entry(jh, dir, de, bh);
5050         if (rc != 0)
5051                 RETURN(rc);
5052
5053         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
5054                                            ent->oied_namelen);
5055         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
5056         osd_get_ldiskfs_dirent_param(ldp, fid);
5057         dentry->d_fsdata = (void *)ldp;
5058         ll_vfs_dq_init(dir);
5059         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
5060         /* It is too bad, we cannot reinsert the name entry back.
5061          * That means we lose it! */
5062         if (rc != 0)
5063                 CDEBUG(D_LFSCK, "%.16s: fail to reinsert the dirent, "
5064                        "dir = %lu/%u, name = %.*s, "DFID": rc = %d\n",
5065                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
5066                        dir->i_ino, dir->i_generation,
5067                        ent->oied_namelen, ent->oied_name, PFID(fid), rc);
5068
5069         RETURN(rc);
5070 }
5071
5072 static int
5073 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
5074                         struct osd_it_ea *it, struct lu_fid *fid,
5075                         struct osd_inode_id *id, __u32 *attr)
5076 {
5077         struct osd_thread_info     *info        = osd_oti_get(env);
5078         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
5079         struct osd_device          *dev         = osd_obj2dev(obj);
5080         struct super_block         *sb          = osd_sb(dev);
5081         const char                 *devname     =
5082                                         LDISKFS_SB(sb)->s_es->s_volume_name;
5083         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
5084         struct inode               *dir         = obj->oo_inode;
5085         struct htree_lock          *hlock       = NULL;
5086         struct buffer_head         *bh          = NULL;
5087         handle_t                   *jh          = NULL;
5088         struct ldiskfs_dir_entry_2 *de;
5089         struct dentry              *dentry;
5090         struct inode               *inode;
5091         int                         credits;
5092         int                         rc;
5093         int                         dot_dotdot  = 0;
5094         bool                        dirty       = false;
5095         ENTRY;
5096
5097         if (ent->oied_name[0] == '.') {
5098                 if (ent->oied_namelen == 1)
5099                         dot_dotdot = 1;
5100                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
5101                         dot_dotdot = 2;
5102         }
5103
5104         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
5105                                       ent->oied_namelen);
5106
5107         /* We need to ensure that the name entry is still valid.
5108          * Because it may be removed or renamed by other already.
5109          *
5110          * The unlink or rename operation will start journal before PDO lock,
5111          * so to avoid deadlock, here we need to start journal handle before
5112          * related PDO lock also. But because we do not know whether there
5113          * will be something to be repaired before PDO lock, we just start
5114          * journal without conditions.
5115          *
5116          * We may need to remove the name entry firstly, then insert back.
5117          * One credit is for user quota file update.
5118          * One credit is for group quota file update.
5119          * Two credits are for dirty inode. */
5120         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
5121                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
5122
5123 again:
5124         if (dev->od_dirent_journal != 0) {
5125                 jh = osd_journal_start_sb(sb, LDISKFS_HT_MISC, credits);
5126                 if (IS_ERR(jh)) {
5127                         rc = PTR_ERR(jh);
5128                         CDEBUG(D_LFSCK, "%.16s: fail to start trans for dirent "
5129                                "check_repair, dir = %lu/%u, credits = %d, "
5130                                "name = %.*s: rc = %d\n",
5131                                devname, dir->i_ino, dir->i_generation, credits,
5132                                ent->oied_namelen, ent->oied_name, rc);
5133                         RETURN(rc);
5134                 }
5135
5136                 if (obj->oo_hl_head != NULL) {
5137                         hlock = osd_oti_get(env)->oti_hlock;
5138                         /* "0" means exclusive lock for the whole directory.
5139                          * We need to prevent others access such name entry
5140                          * during the delete + insert. Neither HLOCK_ADD nor
5141                          * HLOCK_DEL cannot guarantee the atomicity. */
5142                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
5143                 } else {
5144                         down_write(&obj->oo_ext_idx_sem);
5145                 }
5146         } else {
5147                 if (obj->oo_hl_head != NULL) {
5148                         hlock = osd_oti_get(env)->oti_hlock;
5149                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
5150                                            LDISKFS_HLOCK_LOOKUP);
5151                 } else {
5152                         down_read(&obj->oo_ext_idx_sem);
5153                 }
5154         }
5155
5156         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
5157         /* For dot/dotdot entry, if there is not enough space to hold the
5158          * FID-in-dirent, just keep them there. It only happens when the
5159          * device upgraded from 1.8 or restored from MDT file-level backup.
5160          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
5161          * and needs to get FID from LMA when readdir, it will not affect the
5162          * performance much. */
5163         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
5164             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
5165                 *attr |= LUDA_IGNORE;
5166                 GOTO(out_journal, rc = 0);
5167         }
5168
5169         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
5170         inode = osd_iget(info, dev, id);
5171         if (IS_ERR(inode)) {
5172                 rc = PTR_ERR(inode);
5173                 if (rc == -ENOENT || rc == -ESTALE)
5174                         rc = 1;
5175                 else
5176                         CDEBUG(D_LFSCK, "%.16s: fail to iget for dirent "
5177                                "check_repair, dir = %lu/%u, name = %.*s: "
5178                                "rc = %d\n",
5179                                devname, dir->i_ino, dir->i_generation,
5180                                ent->oied_namelen, ent->oied_name, rc);
5181
5182                 GOTO(out_journal, rc);
5183         }
5184
5185         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
5186         if (rc == 0) {
5187                 LASSERT(!(lma->lma_compat & LMAC_NOT_IN_OI));
5188
5189                 if (fid_is_sane(fid)) {
5190                         /* FID-in-dirent is valid. */
5191                         if (lu_fid_eq(fid, &lma->lma_self_fid))
5192                                 GOTO(out_inode, rc = 0);
5193
5194                         /* Do not repair under dryrun mode. */
5195                         if (*attr & LUDA_VERIFY_DRYRUN) {
5196                                 *attr |= LUDA_REPAIR;
5197                                 GOTO(out_inode, rc = 0);
5198                         }
5199
5200                         if (dev->od_dirent_journal == 0) {
5201                                 iput(inode);
5202                                 brelse(bh);
5203                                 if (hlock != NULL)
5204                                         ldiskfs_htree_unlock(hlock);
5205                                 else
5206                                         up_read(&obj->oo_ext_idx_sem);
5207                                 dev->od_dirent_journal = 1;
5208                                 goto again;
5209                         }
5210
5211                         *fid = lma->lma_self_fid;
5212                         dirty = true;
5213                         /* Update the FID-in-dirent. */
5214                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
5215                         if (rc == 0)
5216                                 *attr |= LUDA_REPAIR;
5217                         else
5218                                 CDEBUG(D_LFSCK, "%.16s: fail to update FID "
5219                                        "in the dirent, dir = %lu/%u, "
5220                                        "name = %.*s, "DFID": rc = %d\n",
5221                                        devname, dir->i_ino, dir->i_generation,
5222                                        ent->oied_namelen, ent->oied_name,
5223                                        PFID(fid), rc);
5224                 } else {
5225                         /* Do not repair under dryrun mode. */
5226                         if (*attr & LUDA_VERIFY_DRYRUN) {
5227                                 *fid = lma->lma_self_fid;
5228                                 *attr |= LUDA_REPAIR;
5229                                 GOTO(out_inode, rc = 0);
5230                         }
5231
5232                         if (dev->od_dirent_journal == 0) {
5233                                 iput(inode);
5234                                 brelse(bh);
5235                                 if (hlock != NULL)
5236                                         ldiskfs_htree_unlock(hlock);
5237                                 else
5238                                         up_read(&obj->oo_ext_idx_sem);
5239                                 dev->od_dirent_journal = 1;
5240                                 goto again;
5241                         }
5242
5243                         *fid = lma->lma_self_fid;
5244                         dirty = true;
5245                         /* Append the FID-in-dirent. */
5246                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5247                                                  fid, bh, de, hlock);
5248                         if (rc == 0)
5249                                 *attr |= LUDA_REPAIR;
5250                         else
5251                                 CDEBUG(D_LFSCK, "%.16s: fail to append FID "
5252                                        "after the dirent, dir = %lu/%u, "
5253                                        "name = %.*s, "DFID": rc = %d\n",
5254                                        devname, dir->i_ino, dir->i_generation,
5255                                        ent->oied_namelen, ent->oied_name,
5256                                        PFID(fid), rc);
5257                 }
5258         } else if (rc == -ENODATA) {
5259                 /* Do not repair under dryrun mode. */
5260                 if (*attr & LUDA_VERIFY_DRYRUN) {
5261                         if (fid_is_sane(fid)) {
5262                                 *attr |= LUDA_REPAIR;
5263                         } else {
5264                                 lu_igif_build(fid, inode->i_ino,
5265                                               inode->i_generation);
5266                                 *attr |= LUDA_UPGRADE;
5267                         }
5268                         GOTO(out_inode, rc = 0);
5269                 }
5270
5271                 if (dev->od_dirent_journal == 0) {
5272                         iput(inode);
5273                         brelse(bh);
5274                         if (hlock != NULL)
5275                                 ldiskfs_htree_unlock(hlock);
5276                         else
5277                                 up_read(&obj->oo_ext_idx_sem);
5278                         dev->od_dirent_journal = 1;
5279                         goto again;
5280                 }
5281
5282                 dirty = true;
5283                 if (unlikely(fid_is_sane(fid))) {
5284                         /* FID-in-dirent exists, but FID-in-LMA is lost.
5285                          * Trust the FID-in-dirent, and add FID-in-LMA. */
5286                         rc = osd_ea_fid_set(info, inode, fid, 0, 0);
5287                         if (rc == 0)
5288                                 *attr |= LUDA_REPAIR;
5289                         else
5290                                 CDEBUG(D_LFSCK, "%.16s: fail to set LMA for "
5291                                        "update dirent, dir = %lu/%u, "
5292                                        "name = %.*s, "DFID": rc = %d\n",
5293                                        devname, dir->i_ino, dir->i_generation,
5294                                        ent->oied_namelen, ent->oied_name,
5295                                        PFID(fid), rc);
5296                 } else {
5297                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
5298                         /* It is probably IGIF object. Only aappend the
5299                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
5300                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
5301                                                  fid, bh, de, hlock);
5302                         if (rc == 0)
5303                                 *attr |= LUDA_UPGRADE;
5304                         else
5305                                 CDEBUG(D_LFSCK, "%.16s: fail to append IGIF "
5306                                        "after the dirent, dir = %lu/%u, "
5307                                        "name = %.*s, "DFID": rc = %d\n",
5308                                        devname, dir->i_ino, dir->i_generation,
5309                                        ent->oied_namelen, ent->oied_name,
5310                                        PFID(fid), rc);
5311                 }
5312         }
5313
5314         GOTO(out_inode, rc);
5315
5316 out_inode:
5317         iput(inode);
5318
5319 out_journal:
5320         brelse(bh);
5321         if (hlock != NULL) {
5322                 ldiskfs_htree_unlock(hlock);
5323         } else {
5324                 if (dev->od_dirent_journal != 0)
5325                         up_write(&obj->oo_ext_idx_sem);
5326                 else
5327                         up_read(&obj->oo_ext_idx_sem);
5328         }
5329         if (jh != NULL)
5330                 ldiskfs_journal_stop(jh);
5331         if (rc >= 0 && !dirty)
5332                 dev->od_dirent_journal = 0;
5333         return rc;
5334 }
5335
5336 /**
5337  * Returns the value at current position from iterator's in memory structure.
5338  *
5339  * \param di struct osd_it_ea, iterator's in memory structure
5340  * \param attr attr requested for dirent.
5341  * \param lde lustre dirent
5342  *
5343  * \retval   0 no error and \param lde has correct lustre dirent.
5344  * \retval -ve on error
5345  */
5346 static inline int osd_it_ea_rec(const struct lu_env *env,
5347                                 const struct dt_it *di,
5348                                 struct dt_rec *dtrec, __u32 attr)
5349 {
5350         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
5351         struct osd_object      *obj   = it->oie_obj;
5352         struct osd_device      *dev   = osd_obj2dev(obj);
5353         struct osd_thread_info *oti   = osd_oti_get(env);
5354         struct osd_inode_id    *id    = &oti->oti_id;
5355         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
5356         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
5357         __u32                   ino   = it->oie_dirent->oied_ino;
5358         int                     rc    = 0;
5359         ENTRY;
5360
5361         LASSERT(obj->oo_inode != dev->od_mdt_map->omm_remote_parent->d_inode);
5362
5363         if (attr & LUDA_VERIFY) {
5364                 if (unlikely(ino == osd_remote_parent_ino(dev))) {
5365                         attr |= LUDA_IGNORE;
5366                         /* If the parent is on remote MDT, and there
5367                          * is no FID-in-dirent, then we have to get
5368                          * the parent FID from the linkEA.  */
5369                         if (!fid_is_sane(fid) &&
5370                             it->oie_dirent->oied_namelen == 2 &&
5371                             it->oie_dirent->oied_name[0] == '.' &&
5372                             it->oie_dirent->oied_name[1] == '.')
5373                                 osd_get_pfid_from_linkea(env, obj, fid);
5374                 } else {
5375                         rc = osd_dirent_check_repair(env, obj, it, fid, id,
5376                                                      &attr);
5377                 }
5378         } else {
5379                 attr &= ~LU_DIRENT_ATTRS_MASK;
5380                 if (!fid_is_sane(fid)) {
5381                         bool is_dotdot = false;
5382                         if (it->oie_dirent->oied_namelen == 2 &&
5383                             it->oie_dirent->oied_name[0] == '.' &&
5384                             it->oie_dirent->oied_name[1] == '.')
5385                                 is_dotdot = true;
5386                         /* If the parent is on remote MDT, and there
5387                          * is no FID-in-dirent, then we have to get
5388                          * the parent FID from the linkEA.  */
5389                         if (ino == osd_remote_parent_ino(dev) && is_dotdot) {
5390                                 rc = osd_get_pfid_from_linkea(env, obj, fid);
5391                         } else {
5392                                 if (is_dotdot == false &&
5393                                     OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
5394                                         RETURN(-ENOENT);
5395
5396                                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
5397                         }
5398                 } else {
5399                         osd_id_gen(id, ino, OSD_OII_NOGEN);
5400                 }
5401         }
5402
5403         /* Pack the entry anyway, at least the offset is right. */
5404         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
5405                            it->oie_dirent->oied_name,
5406                            it->oie_dirent->oied_namelen,
5407                            it->oie_dirent->oied_type, attr);
5408
5409         if (rc < 0)
5410                 RETURN(rc);
5411
5412         if (osd_remote_fid(env, dev, fid))
5413                 RETURN(0);
5414
5415         if (likely(!(attr & LUDA_IGNORE) && rc == 0))
5416                 osd_add_oi_cache(oti, dev, id, fid);
5417
5418         RETURN(rc > 0 ? 0 : rc);
5419 }
5420
5421 /**
5422  * Returns the record size size at current position.
5423  *
5424  * This function will return record(lu_dirent) size in bytes.
5425  *
5426  * \param[in] env       execution environment
5427  * \param[in] di        iterator's in memory structure
5428  * \param[in] attr      attribute of the entry, only requires LUDA_TYPE to
5429  *                      calculate the lu_dirent size.
5430  *
5431  * \retval      record size(in bytes & in memory) of the current lu_dirent
5432  *              entry.
5433  */
5434 static int osd_it_ea_rec_size(const struct lu_env *env, const struct dt_it *di,
5435                               __u32 attr)
5436 {
5437         struct osd_it_ea *it = (struct osd_it_ea *)di;
5438
5439         return lu_dirent_calc_size(it->oie_dirent->oied_namelen, attr);
5440 }
5441
5442 /**
5443  * Returns a cookie for current position of the iterator head, so that
5444  * user can use this cookie to load/start the iterator next time.
5445  *
5446  * \param di iterator's in memory structure
5447  *
5448  * \retval cookie for current position, on success
5449  */
5450 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
5451 {
5452         struct osd_it_ea *it = (struct osd_it_ea *)di;
5453
5454         return it->oie_dirent->oied_off;
5455 }
5456
5457 /**
5458  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
5459  * to load a directory entry at a time and stored it i inn,
5460  * in iterator's in-memory data structure.
5461  *
5462  * \param di struct osd_it_ea, iterator's in memory structure
5463  *
5464  * \retval +ve on success
5465  * \retval -ve on error
5466  */
5467 static int osd_it_ea_load(const struct lu_env *env,
5468                           const struct dt_it *di, __u64 hash)
5469 {
5470         struct osd_it_ea *it = (struct osd_it_ea *)di;
5471         int rc;
5472
5473         ENTRY;
5474         it->oie_file.f_pos = hash;
5475
5476         rc =  osd_ldiskfs_it_fill(env, di);
5477         if (rc > 0)
5478                 rc = -ENODATA;
5479
5480         if (rc == 0)
5481                 rc = +1;
5482
5483         RETURN(rc);
5484 }
5485
5486 /**
5487  * Index lookup function for interoperability mode (b11826).
5488  *
5489  * \param key,  key i.e. file name to be searched
5490  *
5491  * \retval +ve, on success
5492  * \retval -ve, on error
5493  */
5494 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5495                                struct dt_rec *rec, const struct dt_key *key)
5496 {
5497         struct osd_object *obj = osd_dt_obj(dt);
5498         int rc = 0;
5499
5500         ENTRY;
5501
5502         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5503         LINVRNT(osd_invariant(obj));
5504
5505         rc = osd_ea_lookup_rec(env, obj, rec, key);
5506         if (rc == 0)
5507                 rc = +1;
5508         RETURN(rc);
5509 }
5510
5511 /**
5512  * Index and Iterator operations for interoperability
5513  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5514  */
5515 static const struct dt_index_operations osd_index_ea_ops = {
5516         .dio_lookup         = osd_index_ea_lookup,
5517         .dio_declare_insert = osd_index_declare_ea_insert,
5518         .dio_insert         = osd_index_ea_insert,
5519         .dio_declare_delete = osd_index_declare_ea_delete,
5520         .dio_delete         = osd_index_ea_delete,
5521         .dio_it     = {
5522                 .init     = osd_it_ea_init,
5523                 .fini     = osd_it_ea_fini,
5524                 .get      = osd_it_ea_get,
5525                 .put      = osd_it_ea_put,
5526                 .next     = osd_it_ea_next,
5527                 .key      = osd_it_ea_key,
5528                 .key_size = osd_it_ea_key_size,
5529                 .rec      = osd_it_ea_rec,
5530                 .rec_size = osd_it_ea_rec_size,
5531                 .store    = osd_it_ea_store,
5532                 .load     = osd_it_ea_load
5533         }
5534 };
5535
5536 static void *osd_key_init(const struct lu_context *ctx,
5537                           struct lu_context_key *key)
5538 {
5539         struct osd_thread_info *info;
5540
5541         OBD_ALLOC_PTR(info);
5542         if (info == NULL)
5543                 return ERR_PTR(-ENOMEM);
5544
5545         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5546         if (info->oti_it_ea_buf == NULL)
5547                 goto out_free_info;
5548
5549         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5550
5551         info->oti_hlock = ldiskfs_htree_lock_alloc();
5552         if (info->oti_hlock == NULL)
5553                 goto out_free_ea;
5554
5555         return info;
5556
5557  out_free_ea:
5558         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5559  out_free_info:
5560         OBD_FREE_PTR(info);
5561         return ERR_PTR(-ENOMEM);
5562 }
5563
5564 static void osd_key_fini(const struct lu_context *ctx,
5565                          struct lu_context_key *key, void* data)
5566 {
5567         struct osd_thread_info *info = data;
5568
5569         if (info->oti_inode != NULL)
5570                 OBD_FREE_PTR(info->oti_inode);
5571         if (info->oti_hlock != NULL)
5572                 ldiskfs_htree_lock_free(info->oti_hlock);
5573         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5574         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5575         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5576         lu_buf_free(&info->oti_big_buf);
5577         OBD_FREE_PTR(info);
5578 }
5579
5580 static void osd_key_exit(const struct lu_context *ctx,
5581                          struct lu_context_key *key, void *data)
5582 {
5583         struct osd_thread_info *info = data;
5584
5585         LASSERT(info->oti_r_locks == 0);
5586         LASSERT(info->oti_w_locks == 0);
5587         LASSERT(info->oti_txns    == 0);
5588 }
5589
5590 /* type constructor/destructor: osd_type_init, osd_type_fini */
5591 LU_TYPE_INIT_FINI(osd, &osd_key);
5592
5593 struct lu_context_key osd_key = {
5594         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5595         .lct_init = osd_key_init,
5596         .lct_fini = osd_key_fini,
5597         .lct_exit = osd_key_exit
5598 };
5599
5600
5601 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5602                            const char *name, struct lu_device *next)
5603 {
5604         struct osd_device *osd = osd_dev(d);
5605
5606         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5607             >= sizeof(osd->od_svname))
5608                 return -E2BIG;
5609         return osd_procfs_init(osd, name);
5610 }
5611
5612 static int osd_fid_init(const struct lu_env *env, struct osd_device *osd)
5613 {
5614         struct seq_server_site  *ss = osd_seq_site(osd);
5615         int                     rc;
5616         ENTRY;
5617
5618         if (osd->od_is_ost || osd->od_cl_seq != NULL)
5619                 RETURN(0);
5620
5621         if (unlikely(ss == NULL))
5622                 RETURN(-ENODEV);
5623
5624         OBD_ALLOC_PTR(osd->od_cl_seq);
5625         if (osd->od_cl_seq == NULL)
5626                 RETURN(-ENOMEM);
5627
5628         rc = seq_client_init(osd->od_cl_seq, NULL, LUSTRE_SEQ_METADATA,
5629                              osd->od_svname, ss->ss_server_seq);
5630         if (rc != 0) {
5631                 OBD_FREE_PTR(osd->od_cl_seq);
5632                 osd->od_cl_seq = NULL;
5633                 RETURN(rc);
5634         }
5635
5636         if (ss->ss_node_id == 0) {
5637                 /* If the OSD on the sequence controller(MDT0), then allocate
5638                  * sequence here, otherwise allocate sequence after connected
5639                  * to MDT0 (see mdt_register_lwp_callback()). */
5640                 rc = seq_server_alloc_meta(osd->od_cl_seq->lcs_srv,
5641                                    &osd->od_cl_seq->lcs_space, env);
5642         }
5643
5644         RETURN(rc);
5645 }
5646
5647 static void osd_fid_fini(const struct lu_env *env, struct osd_device *osd)
5648 {
5649         if (osd->od_cl_seq == NULL)
5650                 return;
5651
5652         seq_client_fini(osd->od_cl_seq);
5653         OBD_FREE_PTR(osd->od_cl_seq);
5654         osd->od_cl_seq = NULL;
5655 }
5656
5657 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5658 {
5659         ENTRY;
5660
5661         /* shutdown quota slave instance associated with the device */
5662         if (o->od_quota_slave != NULL) {
5663                 qsd_fini(env, o->od_quota_slave);
5664                 o->od_quota_slave = NULL;
5665         }
5666
5667         osd_fid_fini(env, o);
5668
5669         RETURN(0);
5670 }
5671
5672 static void osd_umount(const struct lu_env *env, struct osd_device *o)
5673 {
5674         ENTRY;
5675
5676         if (o->od_mnt != NULL) {
5677                 shrink_dcache_sb(osd_sb(o));
5678                 osd_sync(env, &o->od_dt_dev);
5679
5680                 mntput(o->od_mnt);
5681                 o->od_mnt = NULL;
5682         }
5683
5684         EXIT;
5685 }
5686
5687 static int osd_mount(const struct lu_env *env,
5688                      struct osd_device *o, struct lustre_cfg *cfg)
5689 {
5690         const char              *name  = lustre_cfg_string(cfg, 0);
5691         const char              *dev  = lustre_cfg_string(cfg, 1);
5692         const char              *opts;
5693         unsigned long            page, s_flags, lmd_flags = 0;
5694         struct page             *__page;
5695         struct file_system_type *type;
5696         char                    *options = NULL;
5697         char                    *str;
5698         struct osd_thread_info  *info = osd_oti_get(env);
5699         struct lu_fid           *fid = &info->oti_fid;
5700         struct inode            *inode;
5701         int                      rc = 0, force_over_128tb = 0;
5702         ENTRY;
5703
5704         if (o->od_mnt != NULL)
5705                 RETURN(0);
5706
5707         if (strlen(dev) >= sizeof(o->od_mntdev))
5708                 RETURN(-E2BIG);
5709         strcpy(o->od_mntdev, dev);
5710
5711         str = lustre_cfg_string(cfg, 2);
5712         s_flags = simple_strtoul(str, NULL, 0);
5713         str = strstr(str, ":");
5714         if (str)
5715                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5716         opts = lustre_cfg_string(cfg, 3);
5717 #ifdef __BIG_ENDIAN
5718         if (opts == NULL || strstr(opts, "bigendian_extents") == NULL) {
5719                 CERROR("%s: device %s extents feature is not guaranteed to "
5720                        "work on big-endian systems. Use \"bigendian_extents\" "
5721                        "mount option to override.\n", name, dev);
5722                 RETURN(-EINVAL);
5723         }
5724 #endif
5725         if (opts != NULL && strstr(opts, "force_over_128tb") != NULL)
5726                 force_over_128tb = 1;
5727
5728         __page = alloc_page(GFP_IOFS);
5729         if (__page == NULL)
5730                 GOTO(out, rc = -ENOMEM);
5731         page = (unsigned long)page_address(__page);
5732         options = (char *)page;
5733         *options = '\0';
5734         if (opts != NULL) {
5735                 /* strip out the options for back compatiblity */
5736                 static char *sout[] = {
5737                         "mballoc",
5738                         "iopen",
5739                         "noiopen",
5740                         "iopen_nopriv",
5741                         "extents",
5742                         "noextents",
5743                         /* strip out option we processed in osd */
5744                         "bigendian_extents",
5745                         "force_over_128tb",
5746                         NULL
5747                 };
5748                 strcat(options, opts);
5749                 for (rc = 0, str = options; sout[rc]; ) {
5750                         char *op = strstr(str, sout[rc]);
5751                         if (op == NULL) {
5752                                 rc++;
5753                                 str = options;
5754                                 continue;
5755                         }
5756                         if (op == options || *(op - 1) == ',') {
5757                                 str = op + strlen(sout[rc]);
5758                                 if (*str == ',' || *str == '\0') {
5759                                         *str == ',' ? str++ : str;
5760                                         memmove(op, str, strlen(str) + 1);
5761                                 }
5762                         }
5763                         for (str = op; *str != ',' && *str != '\0'; str++)
5764                                 ;
5765                 }
5766         } else {
5767                 strncat(options, "user_xattr,acl", 14);
5768         }
5769
5770         /* Glom up mount options */
5771         if (*options != '\0')
5772                 strcat(options, ",");
5773         strlcat(options, "no_mbcache", PAGE_CACHE_SIZE);
5774
5775         type = get_fs_type("ldiskfs");
5776         if (!type) {
5777                 CERROR("%s: cannot find ldiskfs module\n", name);
5778                 GOTO(out, rc = -ENODEV);
5779         }
5780
5781         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5782         module_put(type->owner);
5783
5784         if (IS_ERR(o->od_mnt)) {
5785                 rc = PTR_ERR(o->od_mnt);
5786                 o->od_mnt = NULL;
5787                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5788                 GOTO(out, rc);
5789         }
5790
5791         if (ldiskfs_blocks_count(LDISKFS_SB(osd_sb(o))->s_es) > (8ULL << 32) &&
5792             force_over_128tb == 0) {
5793                 CERROR("%s: device %s LDISKFS does not support filesystems "
5794                        "greater than 128TB and can cause data corruption. "
5795                        "Use \"force_over_128tb\" mount option to override.\n",
5796                        name, dev);
5797                 GOTO(out, rc = -EINVAL);
5798         }
5799
5800 #ifdef HAVE_DEV_SET_RDONLY
5801         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5802                 CERROR("%s: underlying device %s is marked as read-only. "
5803                        "Setup failed\n", name, dev);
5804                 GOTO(out_mnt, rc = -EROFS);
5805         }
5806 #endif
5807
5808         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5809                                         LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5810                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5811                 GOTO(out_mnt, rc = -EINVAL);
5812         }
5813
5814 #ifdef LDISKFS_MOUNT_DIRDATA
5815         if (LDISKFS_HAS_INCOMPAT_FEATURE(o->od_mnt->mnt_sb,
5816                                          LDISKFS_FEATURE_INCOMPAT_DIRDATA))
5817                 LDISKFS_SB(osd_sb(o))->s_mount_opt |= LDISKFS_MOUNT_DIRDATA;
5818         else if (!o->od_is_ost)
5819                 CWARN("%s: device %s was upgraded from Lustre-1.x without "
5820                       "enabling the dirdata feature. If you do not want to "
5821                       "downgrade to Lustre-1.x again, you can enable it via "
5822                       "'tune2fs -O dirdata device'\n", name, dev);
5823 #endif
5824         inode = osd_sb(o)->s_root->d_inode;
5825         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
5826         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
5827         if (rc != 0) {
5828                 CERROR("%s: failed to set lma on %s root inode\n", name, dev);
5829                 GOTO(out_mnt, rc);
5830         }
5831
5832         if (lmd_flags & LMD_FLG_NOSCRUB)
5833                 o->od_noscrub = 1;
5834
5835         GOTO(out, rc = 0);
5836
5837 out_mnt:
5838         mntput(o->od_mnt);
5839         o->od_mnt = NULL;
5840
5841 out:
5842         if (__page)
5843                 __free_page(__page);
5844
5845         return rc;
5846 }
5847
5848 static struct lu_device *osd_device_fini(const struct lu_env *env,
5849                                          struct lu_device *d)
5850 {
5851         struct osd_device *o = osd_dev(d);
5852         ENTRY;
5853
5854         osd_shutdown(env, o);
5855         osd_procfs_fini(o);
5856         osd_scrub_cleanup(env, o);
5857         osd_obj_map_fini(o);
5858         osd_umount(env, o);
5859
5860         RETURN(NULL);
5861 }
5862
5863 static int osd_device_init0(const struct lu_env *env,
5864                             struct osd_device *o,
5865                             struct lustre_cfg *cfg)
5866 {
5867         struct lu_device        *l = osd2lu_dev(o);
5868         struct osd_thread_info *info;
5869         int                     rc;
5870         int                     cplen = 0;
5871
5872         /* if the module was re-loaded, env can loose its keys */
5873         rc = lu_env_refill((struct lu_env *) env);
5874         if (rc)
5875                 GOTO(out, rc);
5876         info = osd_oti_get(env);
5877         LASSERT(info);
5878
5879         l->ld_ops = &osd_lu_ops;
5880         o->od_dt_dev.dd_ops = &osd_dt_ops;
5881
5882         spin_lock_init(&o->od_osfs_lock);
5883         mutex_init(&o->od_otable_mutex);
5884
5885         o->od_read_cache = 1;
5886         o->od_writethrough_cache = 1;
5887         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5888
5889         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5890                         sizeof(o->od_svname));
5891         if (cplen >= sizeof(o->od_svname)) {
5892                 rc = -E2BIG;
5893                 GOTO(out, rc);
5894         }
5895
5896         o->od_index = -1; /* -1 means index is invalid */
5897         rc = server_name2index(o->od_svname, &o->od_index, NULL);
5898         if (rc == LDD_F_SV_TYPE_OST)
5899                 o->od_is_ost = 1;
5900
5901         o->od_full_scrub_ratio = OFSR_DEFAULT;
5902         o->od_full_scrub_threshold_rate = FULL_SCRUB_THRESHOLD_RATE_DEFAULT;
5903         rc = osd_mount(env, o, cfg);
5904         if (rc != 0)
5905                 GOTO(out, rc);
5906
5907         rc = osd_obj_map_init(env, o);
5908         if (rc != 0)
5909                 GOTO(out_mnt, rc);
5910
5911         rc = lu_site_init(&o->od_site, l);
5912         if (rc != 0)
5913                 GOTO(out_compat, rc);
5914         o->od_site.ls_bottom_dev = l;
5915
5916         rc = lu_site_init_finish(&o->od_site);
5917         if (rc != 0)
5918                 GOTO(out_site, rc);
5919
5920         INIT_LIST_HEAD(&o->od_ios_list);
5921         /* setup scrub, including OI files initialization */
5922         rc = osd_scrub_setup(env, o);
5923         if (rc < 0)
5924                 GOTO(out_site, rc);
5925
5926         rc = osd_procfs_init(o, o->od_svname);
5927         if (rc != 0) {
5928                 CERROR("%s: can't initialize procfs: rc = %d\n",
5929                        o->od_svname, rc);
5930                 GOTO(out_scrub, rc);
5931         }
5932
5933         LASSERT(l->ld_site->ls_linkage.next != NULL);
5934         LASSERT(l->ld_site->ls_linkage.prev != NULL);
5935
5936         /* initialize quota slave instance */
5937         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5938                                      o->od_proc_entry);
5939         if (IS_ERR(o->od_quota_slave)) {
5940                 rc = PTR_ERR(o->od_quota_slave);
5941                 o->od_quota_slave = NULL;
5942                 GOTO(out_procfs, rc);
5943         }
5944
5945         RETURN(0);
5946
5947 out_procfs:
5948         osd_procfs_fini(o);
5949 out_scrub:
5950         osd_scrub_cleanup(env, o);
5951 out_site:
5952         lu_site_fini(&o->od_site);
5953 out_compat:
5954         osd_obj_map_fini(o);
5955 out_mnt:
5956         osd_umount(env, o);
5957 out:
5958         return rc;
5959 }
5960
5961 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5962                                           struct lu_device_type *t,
5963                                           struct lustre_cfg *cfg)
5964 {
5965         struct osd_device *o;
5966         int                rc;
5967
5968         OBD_ALLOC_PTR(o);
5969         if (o == NULL)
5970                 return ERR_PTR(-ENOMEM);
5971
5972         rc = dt_device_init(&o->od_dt_dev, t);
5973         if (rc == 0) {
5974                 /* Because the ctx might be revived in dt_device_init,
5975                  * refill the env here */
5976                 lu_env_refill((struct lu_env *)env);
5977                 rc = osd_device_init0(env, o, cfg);
5978                 if (rc)
5979                         dt_device_fini(&o->od_dt_dev);
5980         }
5981
5982         if (unlikely(rc != 0))
5983                 OBD_FREE_PTR(o);
5984
5985         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5986 }
5987
5988 static struct lu_device *osd_device_free(const struct lu_env *env,
5989                                          struct lu_device *d)
5990 {
5991         struct osd_device *o = osd_dev(d);
5992         ENTRY;
5993
5994         /* XXX: make osd top device in order to release reference */
5995         d->ld_site->ls_top_dev = d;
5996         lu_site_purge(env, d->ld_site, -1);
5997         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5998                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5999                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
6000         }
6001         lu_site_fini(&o->od_site);
6002         dt_device_fini(&o->od_dt_dev);
6003         OBD_FREE_PTR(o);
6004         RETURN(NULL);
6005 }
6006
6007 static int osd_process_config(const struct lu_env *env,
6008                               struct lu_device *d, struct lustre_cfg *cfg)
6009 {
6010         struct osd_device               *o = osd_dev(d);
6011         int                             rc;
6012         ENTRY;
6013
6014         switch (cfg->lcfg_command) {
6015         case LCFG_SETUP:
6016                 rc = osd_mount(env, o, cfg);
6017                 break;
6018         case LCFG_CLEANUP:
6019                 lu_dev_del_linkage(d->ld_site, d);
6020                 rc = osd_shutdown(env, o);
6021                 break;
6022         case LCFG_PARAM:
6023                 LASSERT(&o->od_dt_dev);
6024                 rc = class_process_proc_param(PARAM_OSD, lprocfs_osd_obd_vars,
6025                                               cfg, &o->od_dt_dev);
6026                 if (rc > 0 || rc == -ENOSYS)
6027                         rc = class_process_proc_param(PARAM_OST,
6028                                                       lprocfs_osd_obd_vars,
6029                                                       cfg, &o->od_dt_dev);
6030                 break;
6031         default:
6032                 rc = -ENOSYS;
6033         }
6034
6035         RETURN(rc);
6036 }
6037
6038 static int osd_recovery_complete(const struct lu_env *env,
6039                                  struct lu_device *d)
6040 {
6041         struct osd_device       *osd = osd_dev(d);
6042         int                      rc = 0;
6043         ENTRY;
6044
6045         if (osd->od_quota_slave == NULL)
6046                 RETURN(0);
6047
6048         /* start qsd instance on recovery completion, this notifies the quota
6049          * slave code that we are about to process new requests now */
6050         rc = qsd_start(env, osd->od_quota_slave);
6051         RETURN(rc);
6052 }
6053
6054 /*
6055  * we use exports to track all osd users
6056  */
6057 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
6058                            struct obd_device *obd, struct obd_uuid *cluuid,
6059                            struct obd_connect_data *data, void *localdata)
6060 {
6061         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
6062         struct lustre_handle  conn;
6063         int                   rc;
6064         ENTRY;
6065
6066         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
6067
6068         rc = class_connect(&conn, obd, cluuid);
6069         if (rc)
6070                 RETURN(rc);
6071
6072         *exp = class_conn2export(&conn);
6073
6074         spin_lock(&osd->od_osfs_lock);
6075         osd->od_connects++;
6076         spin_unlock(&osd->od_osfs_lock);
6077
6078         RETURN(0);
6079 }
6080
6081 /*
6082  * once last export (we don't count self-export) disappeared
6083  * osd can be released
6084  */
6085 static int osd_obd_disconnect(struct obd_export *exp)
6086 {
6087         struct obd_device *obd = exp->exp_obd;
6088         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6089         int                rc, release = 0;
6090         ENTRY;
6091
6092         /* Only disconnect the underlying layers on the final disconnect. */
6093         spin_lock(&osd->od_osfs_lock);
6094         osd->od_connects--;
6095         if (osd->od_connects == 0)
6096                 release = 1;
6097         spin_unlock(&osd->od_osfs_lock);
6098
6099         rc = class_disconnect(exp); /* bz 9811 */
6100
6101         if (rc == 0 && release)
6102                 class_manual_cleanup(obd);
6103         RETURN(rc);
6104 }
6105
6106 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
6107                        struct lu_device *dev)
6108 {
6109         struct osd_device       *osd    = osd_dev(dev);
6110         struct lr_server_data   *lsd    =
6111                         &osd->od_dt_dev.dd_lu_dev.ld_site->ls_tgt->lut_lsd;
6112         int                      result = 0;
6113         ENTRY;
6114
6115         if (osd->od_quota_slave != NULL) {
6116                 /* set up quota slave objects */
6117                 result = qsd_prepare(env, osd->od_quota_slave);
6118                 if (result != 0)
6119                         RETURN(result);
6120         }
6121
6122         if (lsd->lsd_feature_incompat & OBD_COMPAT_OST) {
6123 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
6124                 if (lsd->lsd_feature_rocompat & OBD_ROCOMPAT_IDX_IN_IDIF) {
6125                         osd->od_index_in_idif = 1;
6126                 } else {
6127                         osd->od_index_in_idif = 0;
6128                         result = osd_register_proc_index_in_idif(osd);
6129                         if (result != 0)
6130                                 RETURN(result);
6131                 }
6132 #else
6133                 osd->od_index_in_idif = 1;
6134 #endif
6135         }
6136
6137         result = osd_fid_init(env, osd);
6138
6139         RETURN(result);
6140 }
6141
6142 static int osd_fid_alloc(const struct lu_env *env, struct obd_export *exp,
6143                          struct lu_fid *fid, struct md_op_data *op_data)
6144 {
6145         struct osd_device *osd = osd_dev(exp->exp_obd->obd_lu_dev);
6146
6147         return seq_client_alloc_fid(env, osd->od_cl_seq, fid);
6148 }
6149
6150 static const struct lu_object_operations osd_lu_obj_ops = {
6151         .loo_object_init      = osd_object_init,
6152         .loo_object_delete    = osd_object_delete,
6153         .loo_object_release   = osd_object_release,
6154         .loo_object_free      = osd_object_free,
6155         .loo_object_print     = osd_object_print,
6156         .loo_object_invariant = osd_object_invariant
6157 };
6158
6159 const struct lu_device_operations osd_lu_ops = {
6160         .ldo_object_alloc      = osd_object_alloc,
6161         .ldo_process_config    = osd_process_config,
6162         .ldo_recovery_complete = osd_recovery_complete,
6163         .ldo_prepare           = osd_prepare,
6164 };
6165
6166 static const struct lu_device_type_operations osd_device_type_ops = {
6167         .ldto_init = osd_type_init,
6168         .ldto_fini = osd_type_fini,
6169
6170         .ldto_start = osd_type_start,
6171         .ldto_stop  = osd_type_stop,
6172
6173         .ldto_device_alloc = osd_device_alloc,
6174         .ldto_device_free  = osd_device_free,
6175
6176         .ldto_device_init    = osd_device_init,
6177         .ldto_device_fini    = osd_device_fini
6178 };
6179
6180 static struct lu_device_type osd_device_type = {
6181         .ldt_tags     = LU_DEVICE_DT,
6182         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
6183         .ldt_ops      = &osd_device_type_ops,
6184         .ldt_ctx_tags = LCT_LOCAL,
6185 };
6186
6187 static int osd_health_check(const struct lu_env *env, struct obd_device *obd)
6188 {
6189         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
6190         struct super_block *sb = osd_sb(osd);
6191
6192         return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
6193 }
6194
6195 /*
6196  * lprocfs legacy support.
6197  */
6198 static struct obd_ops osd_obd_device_ops = {
6199         .o_owner = THIS_MODULE,
6200         .o_connect      = osd_obd_connect,
6201         .o_disconnect   = osd_obd_disconnect,
6202         .o_fid_alloc    = osd_fid_alloc,
6203         .o_health_check = osd_health_check,
6204 };
6205
6206 static int __init osd_mod_init(void)
6207 {
6208         int rc;
6209
6210 #if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
6211         /* please, try to keep osd_thread_info smaller than a page */
6212         CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
6213 #endif
6214
6215         osd_oi_mod_init();
6216
6217         rc = lu_kmem_init(ldiskfs_caches);
6218         if (rc)
6219                 return rc;
6220
6221         rc = class_register_type(&osd_obd_device_ops, NULL, true,
6222                                  lprocfs_osd_module_vars,
6223                                  LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
6224         if (rc)
6225                 lu_kmem_fini(ldiskfs_caches);
6226         return rc;
6227 }
6228
6229 static void __exit osd_mod_exit(void)
6230 {
6231         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
6232         lu_kmem_fini(ldiskfs_caches);
6233 }
6234
6235 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
6236 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
6237 MODULE_VERSION(LUSTRE_VERSION_STRING);
6238 MODULE_LICENSE("GPL");
6239
6240 module_init(osd_mod_init);
6241 module_exit(osd_mod_exit);