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, 2012, 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_MDS
  45
  46 #include <linux/module.h>
  47
  48 /* LUSTRE_VERSION_CODE */
  49 #include <lustre_ver.h>
  50 /* prerequisite for linux/xattr.h */
  51 #include <linux/types.h>
  52 /* prerequisite for linux/xattr.h */
  53 #include <linux/fs.h>
  54 /* XATTR_{REPLACE,CREATE} */
  55 #include <linux/xattr.h>
  56
  57 /*
  58  * struct OBD_{ALLOC,FREE}*()
  59  * OBD_FAIL_CHECK
  60  */
  61 #include <obd_support.h>
  62 /* struct ptlrpc_thread */
  63 #include <lustre_net.h>
  64 #include <lustre_fid.h>
  65
  66 #include "osd_internal.h"
  67
  68 /* llo_* api support */
  69 #include <md_object.h>
  70 #include <lustre_quota.h>
  71
  72 int ldiskfs_pdo = 1;
  73 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  74                 "ldiskfs with parallel directory operations");
  75
  76 static const char dot[] = ".";
  77 static const char dotdot[] = "..";
  78 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  79
  80 static const struct lu_object_operations      osd_lu_obj_ops;
  81 static const struct dt_object_operations      osd_obj_ops;
  82 static const struct dt_object_operations      osd_obj_ea_ops;
  83 static const struct dt_object_operations      osd_obj_otable_it_ops;
  84 static const struct dt_index_operations       osd_index_iam_ops;
  85 static const struct dt_index_operations       osd_index_ea_ops;
  86
  87 #ifdef OSD_TRACK_DECLARES
  88 int osd_trans_declare_op2rb[] = {
  89         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
  90         [OSD_OT_PUNCH]          = OSD_OT_MAX,
  91         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
  92         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
  93         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
  94         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
  95         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
  96         [OSD_OT_WRITE]          = OSD_OT_WRITE,
  97         [OSD_OT_INSERT]         = OSD_OT_DELETE,
  98         [OSD_OT_DELETE]         = OSD_OT_INSERT,
  99         [OSD_OT_QUOTA]          = OSD_OT_MAX,
 100 };
 101 #endif
 102
 103 static int osd_has_index(const struct osd_object *obj)
 104 {
 105         return obj->oo_dt.do_index_ops != NULL;
 106 }
 107
 108 static int osd_object_invariant(const struct lu_object *l)
 109 {
 110         return osd_invariant(osd_obj(l));
 111 }
 112
 113 /*
 114  * Concurrency: doesn't matter
 115  */
 116 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 117 {
 118         return osd_oti_get(env)->oti_r_locks > 0;
 119 }
 120
 121 /*
 122  * Concurrency: doesn't matter
 123  */
 124 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 125 {
 126         struct osd_thread_info *oti = osd_oti_get(env);
 127         return oti->oti_w_locks > 0 && o->oo_owner == env;
 128 }
 129
 130 /*
 131  * Concurrency: doesn't access mutable data
 132  */
 133 static int osd_root_get(const struct lu_env *env,
 134                         struct dt_device *dev, struct lu_fid *f)
 135 {
 136         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 137         return 0;
 138 }
 139
 140 /*
 141  * OSD object methods.
 142  */
 143
 144 /*
 145  * Concurrency: no concurrent access is possible that early in object
 146  * life-cycle.
 147  */
 148 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 149                                           const struct lu_object_header *hdr,
 150                                           struct lu_device *d)
 151 {
 152         struct osd_object *mo;
 153
 154         OBD_ALLOC_PTR(mo);
 155         if (mo != NULL) {
 156                 struct lu_object *l;
 157
 158                 l = &mo->oo_dt.do_lu;
 159                 dt_object_init(&mo->oo_dt, NULL, d);
 160                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
 161                 l->lo_ops = &osd_lu_obj_ops;
 162                 init_rwsem(&mo->oo_sem);
 163                 init_rwsem(&mo->oo_ext_idx_sem);
 164                 spin_lock_init(&mo->oo_guard);
 165                 return l;
 166         } else {
 167                 return NULL;
 168         }
 169 }
 170
 171 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
 172                                   const char *name, void *buf, int len)
 173 {
 174         dentry->d_inode = inode;
 175         return inode->i_op->getxattr(dentry, name, buf, len);
 176 }
 177
 178 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
 179                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
 180 {
 181         int rc;
 182
 183         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma,
 184                              sizeof(*lma));
 185         if (rc == -ERANGE) {
 186                 /* try with old lma size */
 187                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
 188                                            info->oti_mdt_attrs_old,
 189                                            LMA_OLD_SIZE);
 190                 if (rc > 0)
 191                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
 192         }
 193         if (rc > 0) {
 194                 /* Check LMA compatibility */
 195                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 196                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 197                               "%lx/%#x\n",
 198                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 199                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 200                                                         ~LMA_INCOMPAT_SUPP);
 201                         rc = -ENOSYS;
 202                 } else {
 203                         lustre_lma_swab(lma);
 204                         rc = 0;
 205                 }
 206         } else if (rc == 0) {
 207                 rc = -ENODATA;
 208         }
 209
 210         return rc;
 211 }
 212
 213 /*
 214  * retrieve object from backend ext fs.
 215  **/
 216 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 217                        struct osd_inode_id *id)
 218 {
 219         struct inode *inode = NULL;
 220
 221         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 222         if (IS_ERR(inode)) {
 223                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 224                        id->oii_ino, PTR_ERR(inode));
 225         } else if (id->oii_gen != OSD_OII_NOGEN &&
 226                    inode->i_generation != id->oii_gen) {
 227                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 228                        "gen1 = %u\n",
 229                        id->oii_ino, id->oii_gen, inode->i_generation);
 230                 iput(inode);
 231                 inode = ERR_PTR(-ESTALE);
 232         } else if (inode->i_nlink == 0) {
 233                 /* due to parallel readdir and unlink,
 234                 * we can have dead inode here. */
 235                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 236                 make_bad_inode(inode);
 237                 iput(inode);
 238                 inode = ERR_PTR(-ESTALE);
 239         } else if (is_bad_inode(inode)) {
 240                 CWARN("%.16s: bad inode: ino = %u\n",
 241                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 242                 iput(inode);
 243                 inode = ERR_PTR(-ENOENT);
 244         } else {
 245                 if (id->oii_gen == OSD_OII_NOGEN)
 246                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 247
 248                 /* Do not update file c/mtime in ldiskfs.
 249                  * NB: we don't have any lock to protect this because we don't
 250                  * have reference on osd_object now, but contention with
 251                  * another lookup + attr_set can't happen in the tiny window
 252                  * between if (...) and set S_NOCMTIME. */
 253                 if (!(inode->i_flags & S_NOCMTIME))
 254                         inode->i_flags |= S_NOCMTIME;
 255         }
 256         return inode;
 257 }
 258
 259 static struct inode *
 260 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 261              struct osd_inode_id *id, struct lu_fid *fid)
 262 {
 263         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 264         struct inode            *inode;
 265         int                      rc;
 266
 267         inode = osd_iget(info, dev, id);
 268         if (IS_ERR(inode))
 269                 return inode;
 270
 271         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 272         if (rc == 0) {
 273                 *fid = lma->lma_self_fid;
 274         } else if (rc == -ENODATA) {
 275                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
 276                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
 277                 else
 278                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
 279         } else {
 280                 iput(inode);
 281                 inode = ERR_PTR(rc);
 282         }
 283         return inode;
 284 }
 285
 286 static struct inode *
 287 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 288                 struct osd_inode_id *id, const struct lu_fid *fid)
 289 {
 290         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 291         struct inode            *inode;
 292         int                      rc;
 293
 294         inode = osd_iget(info, dev, id);
 295         if (IS_ERR(inode))
 296                 return inode;
 297
 298         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
 299         if (rc == -ENODATA)
 300                 return inode;
 301
 302         if (rc != 0) {
 303                 iput(inode);
 304                 return ERR_PTR(rc);
 305         }
 306
 307         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 308                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 309                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 310                 iput(inode);
 311                 return ERR_PTR(-EREMCHG);
 312         }
 313
 314         return inode;
 315 }
 316
 317 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 318                           const struct lu_fid *fid,
 319                           const struct lu_object_conf *conf)
 320 {
 321         struct osd_thread_info *info;
 322         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 323         struct osd_device      *dev;
 324         struct osd_idmap_cache *oic;
 325         struct osd_inode_id    *id;
 326         struct inode           *inode;
 327         struct osd_scrub       *scrub;
 328         struct scrub_file      *sf;
 329         int                     result;
 330         int                     verify = 0;
 331         ENTRY;
 332
 333         LINVRNT(osd_invariant(obj));
 334         LASSERT(obj->oo_inode == NULL);
 335         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 336
 337         dev = osd_dev(ldev);
 338         scrub = &dev->od_scrub;
 339         sf = &scrub->os_file;
 340         info = osd_oti_get(env);
 341         LASSERT(info);
 342         oic = &info->oti_cache;
 343
 344         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 345                 RETURN(-ENOENT);
 346
 347         /* Search order: 1. per-thread cache. */
 348         if (lu_fid_eq(fid, &oic->oic_fid)) {
 349                 id = &oic->oic_lid;
 350                 goto iget;
 351         }
 352
 353         id = &info->oti_id;
 354         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 355                 /* Search order: 2. OI scrub pending list. */
 356                 result = osd_oii_lookup(dev, fid, id);
 357                 if (result == 0)
 358                         goto iget;
 359         }
 360
 361         if (sf->sf_flags & SF_INCONSISTENT)
 362                 verify = 1;
 363
 364         /*
 365          * Objects are created as locking anchors or place holders for objects
 366          * yet to be created. No need to osd_oi_lookup() at here because FID
 367          * shouldn't never be re-used, if it's really a duplicate FID from
 368          * unexpected reason, we should be able to detect it later by calling
 369          * do_create->osd_oi_insert()
 370          */
 371         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 372                 GOTO(out, result = 0);
 373
 374         /* Search order: 3. OI files. */
 375         result = osd_oi_lookup(info, dev, fid, id, true);
 376         if (result == -ENOENT) {
 377                 if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
 378                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 379                                       sf->sf_oi_bitmap))
 380                         GOTO(out, result = 0);
 381
 382                 goto trigger;
 383         }
 384
 385         if (result != 0)
 386                 GOTO(out, result);
 387
 388 iget:
 389         if (verify == 0)
 390                 inode = osd_iget(info, dev, id);
 391         else
 392                 inode = osd_iget_verify(info, dev, id, fid);
 393         if (IS_ERR(inode)) {
 394                 result = PTR_ERR(inode);
 395                 if (result == -ENOENT || result == -ESTALE) {
 396                         fid_zero(&oic->oic_fid);
 397                         result = 0;
 398                 } else if (result == -EREMCHG) {
 399
 400 trigger:
 401                         if (thread_is_running(&scrub->os_thread)) {
 402                                 result = -EINPROGRESS;
 403                         } else if (!dev->od_noscrub) {
 404                                 result = osd_scrub_start(dev);
 405                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 406                                                "for "DFID", rc = %d [1]\n",
 407                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 408                                                s_volume_name,PFID(fid), result);
 409                                 if (result == 0 || result == -EALREADY)
 410                                         result = -EINPROGRESS;
 411                                 else
 412                                         result = -EREMCHG;
 413                         }
 414                 }
 415
 416                 GOTO(out, result);
 417         }
 418
 419         obj->oo_inode = inode;
 420         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 421
 422         obj->oo_compat_dot_created = 1;
 423         obj->oo_compat_dotdot_created = 1;
 424
 425         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 426                 GOTO(out, result = 0);
 427
 428         LASSERT(obj->oo_hl_head == NULL);
 429         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 430         if (obj->oo_hl_head == NULL) {
 431                 obj->oo_inode = NULL;
 432                 iput(inode);
 433                 GOTO(out, result = -ENOMEM);
 434         }
 435         GOTO(out, result = 0);
 436
 437 out:
 438         LINVRNT(osd_invariant(obj));
 439         return result;
 440 }
 441
 442 /*
 443  * Concurrency: shouldn't matter.
 444  */
 445 static void osd_object_init0(struct osd_object *obj)
 446 {
 447         LASSERT(obj->oo_inode != NULL);
 448         obj->oo_dt.do_body_ops = &osd_body_ops;
 449         obj->oo_dt.do_lu.lo_header->loh_attr |=
 450                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 451 }
 452
 453 /*
 454  * Concurrency: no concurrent access is possible that early in object
 455  * life-cycle.
 456  */
 457 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 458                            const struct lu_object_conf *conf)
 459 {
 460         struct osd_object *obj = osd_obj(l);
 461         int result;
 462
 463         LINVRNT(osd_invariant(obj));
 464
 465         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 466                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 467                 l->lo_header->loh_attr |= LOHA_EXISTS;
 468                 return 0;
 469         }
 470
 471         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 472         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 473         if (result == 0 && obj->oo_inode != NULL)
 474                 osd_object_init0(obj);
 475
 476         LINVRNT(osd_invariant(obj));
 477         return result;
 478 }
 479
 480 /*
 481  * Concurrency: no concurrent access is possible that late in object
 482  * life-cycle.
 483  */
 484 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 485 {
 486         struct osd_object *obj = osd_obj(l);
 487
 488         LINVRNT(osd_invariant(obj));
 489
 490         dt_object_fini(&obj->oo_dt);
 491         if (obj->oo_hl_head != NULL)
 492                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 493         OBD_FREE_PTR(obj);
 494 }
 495
 496 /*
 497  * Concurrency: no concurrent access is possible that late in object
 498  * life-cycle.
 499  */
 500 static void osd_index_fini(struct osd_object *o)
 501 {
 502         struct iam_container *bag;
 503
 504         if (o->oo_dir != NULL) {
 505                 bag = &o->oo_dir->od_container;
 506                 if (o->oo_inode != NULL) {
 507                         if (bag->ic_object == o->oo_inode)
 508                                 iam_container_fini(bag);
 509                 }
 510                 OBD_FREE_PTR(o->oo_dir);
 511                 o->oo_dir = NULL;
 512         }
 513 }
 514
 515 /*
 516  * Concurrency: no concurrent access is possible that late in object
 517  * life-cycle (for all existing callers, that is. New callers have to provide
 518  * their own locking.)
 519  */
 520 static int osd_inode_unlinked(const struct inode *inode)
 521 {
 522         return inode->i_nlink == 0;
 523 }
 524
 525 enum {
 526         OSD_TXN_OI_DELETE_CREDITS    = 20,
 527         OSD_TXN_INODE_DELETE_CREDITS = 20
 528 };
 529
 530 /*
 531  * Journal
 532  */
 533
 534 #if OSD_THANDLE_STATS
 535 /**
 536  * Set time when the handle is allocated
 537  */
 538 static void osd_th_alloced(struct osd_thandle *oth)
 539 {
 540         oth->oth_alloced = cfs_time_current();
 541 }
 542
 543 /**
 544  * Set time when the handle started
 545  */
 546 static void osd_th_started(struct osd_thandle *oth)
 547 {
 548         oth->oth_started = cfs_time_current();
 549 }
 550
 551 /**
 552  * Helper function to convert time interval to microseconds packed in
 553  * long int.
 554  */
 555 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 556 {
 557         struct timeval val;
 558
 559         cfs_duration_usec(cfs_time_sub(end, start), &val);
 560         return val.tv_sec * 1000000 + val.tv_usec;
 561 }
 562
 563 /**
 564  * Check whether the we deal with this handle for too long.
 565  */
 566 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 567                                 cfs_time_t alloced, cfs_time_t started,
 568                                 cfs_time_t closed)
 569 {
 570         cfs_time_t now = cfs_time_current();
 571
 572         LASSERT(dev != NULL);
 573
 574         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 575                             interval_to_usec(alloced, started));
 576         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 577                             interval_to_usec(started, closed));
 578         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 579                             interval_to_usec(closed, now));
 580
 581         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 582                 CWARN("transaction handle %p was open for too long: "
 583                       "now "CFS_TIME_T" ,"
 584                       "alloced "CFS_TIME_T" ,"
 585                       "started "CFS_TIME_T" ,"
 586                       "closed "CFS_TIME_T"\n",
 587                       oth, now, alloced, started, closed);
 588                 libcfs_debug_dumpstack(NULL);
 589         }
 590 }
 591
 592 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 593 {                                                                       \
 594         cfs_time_t __closed = cfs_time_current();                       \
 595         cfs_time_t __alloced = oth->oth_alloced;                        \
 596         cfs_time_t __started = oth->oth_started;                        \
 597                                                                         \
 598         expr;                                                           \
 599         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 600 }
 601
 602 #else /* OSD_THANDLE_STATS */
 603
 604 #define osd_th_alloced(h)                  do {} while(0)
 605 #define osd_th_started(h)                  do {} while(0)
 606 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 607
 608 #endif /* OSD_THANDLE_STATS */
 609
 610 /*
 611  * Concurrency: doesn't access mutable data.
 612  */
 613 static int osd_param_is_not_sane(const struct osd_device *dev,
 614                                  const struct thandle *th)
 615 {
 616         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
 617
 618         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
 619 }
 620
 621 /*
 622  * Concurrency: shouldn't matter.
 623  */
 624 static void osd_trans_commit_cb(struct super_block *sb,
 625                                 struct ldiskfs_journal_cb_entry *jcb, int error)
 626 {
 627         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 628         struct thandle     *th  = &oh->ot_super;
 629         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 630         struct dt_txn_commit_cb *dcb, *tmp;
 631
 632         LASSERT(oh->ot_handle == NULL);
 633
 634         if (error)
 635                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 636
 637         dt_txn_hook_commit(th);
 638
 639         /* call per-transaction callbacks if any */
 640         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 641                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 642                          "commit callback entry: magic=%x name='%s'\n",
 643                          dcb->dcb_magic, dcb->dcb_name);
 644                 cfs_list_del_init(&dcb->dcb_linkage);
 645                 dcb->dcb_func(NULL, th, dcb, error);
 646         }
 647
 648         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 649         lu_device_put(lud);
 650         th->th_dev = NULL;
 651
 652         lu_context_exit(&th->th_ctx);
 653         lu_context_fini(&th->th_ctx);
 654         OBD_FREE_PTR(oh);
 655 }
 656
 657 static struct thandle *osd_trans_create(const struct lu_env *env,
 658                                         struct dt_device *d)
 659 {
 660         struct osd_thread_info *oti = osd_oti_get(env);
 661         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 662         struct osd_thandle     *oh;
 663         struct thandle         *th;
 664         ENTRY;
 665
 666         /* on pending IO in this thread should left from prev. request */
 667         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 668
 669         th = ERR_PTR(-ENOMEM);
 670         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 671         if (oh != NULL) {
 672                 oh->ot_quota_trans = &oti->oti_quota_trans;
 673                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
 674                 th = &oh->ot_super;
 675                 th->th_dev = d;
 676                 th->th_result = 0;
 677                 th->th_tags = LCT_TX_HANDLE;
 678                 oh->ot_credits = 0;
 679                 oti->oti_dev = osd_dt_dev(d);
 680                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 681                 osd_th_alloced(oh);
 682
 683                 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
 684                 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
 685                 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
 686                 oti->oti_rollback = false;
 687         }
 688         RETURN(th);
 689 }
 690
 691 /*
 692  * Concurrency: shouldn't matter.
 693  */
 694 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 695                     struct thandle *th)
 696 {
 697         struct osd_thread_info *oti = osd_oti_get(env);
 698         struct osd_device  *dev = osd_dt_dev(d);
 699         handle_t           *jh;
 700         struct osd_thandle *oh;
 701         int rc;
 702
 703         ENTRY;
 704
 705         LASSERT(current->journal_info == NULL);
 706
 707         oh = container_of0(th, struct osd_thandle, ot_super);
 708         LASSERT(oh != NULL);
 709         LASSERT(oh->ot_handle == NULL);
 710
 711         rc = dt_txn_hook_start(env, d, th);
 712         if (rc != 0)
 713                 GOTO(out, rc);
 714
 715         if (unlikely(osd_param_is_not_sane(dev, th))) {
 716                 static unsigned long last_printed;
 717                 static int last_credits;
 718
 719                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 720                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 721                       oh->ot_credits,
 722                       osd_journal(dev)->j_max_transaction_buffers);
 723 #ifdef OSD_TRACK_DECLARES
 724                 CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
 725                       oti->oti_declare_ops[OSD_OT_CREATE],
 726                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
 727                       oti->oti_declare_ops[OSD_OT_DELETE],
 728                       oti->oti_declare_ops_cred[OSD_OT_DELETE],
 729                       oti->oti_declare_ops[OSD_OT_DESTROY],
 730                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 731                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
 732                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
 733                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
 734                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
 735                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
 736                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
 737                       oti->oti_declare_ops[OSD_OT_WRITE],
 738                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
 739                       oti->oti_declare_ops[OSD_OT_PUNCH],
 740                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
 741                       oti->oti_declare_ops[OSD_OT_QUOTA],
 742                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
 743                 CWARN("  insert: %u/%u, delete: %u/%u\n",
 744                       oti->oti_declare_ops[OSD_OT_INSERT],
 745                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
 746                       oti->oti_declare_ops[OSD_OT_DESTROY],
 747                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
 748                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
 749                       oti->oti_declare_ops[OSD_OT_REF_ADD],
 750                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
 751                       oti->oti_declare_ops[OSD_OT_REF_DEL],
 752                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
 753
 754                 if (last_credits != oh->ot_credits &&
 755                     time_after(jiffies, last_printed + 60 * HZ)) {
 756                         libcfs_debug_dumpstack(NULL);
 757                         last_credits = oh->ot_credits;
 758                         last_printed = jiffies;
 759                 }
 760 #endif
 761                 /* XXX Limit the credits to 'max_transaction_buffers', and
 762                  *     let the underlying filesystem to catch the error if
 763                  *     we really need so many credits.
 764                  *
 765                  *     This should be removed when we can calculate the
 766                  *     credits precisely. */
 767                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 768         }
 769
 770         /*
 771          * XXX temporary stuff. Some abstraction layer should
 772          * be used.
 773          */
 774         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 775         osd_th_started(oh);
 776         if (!IS_ERR(jh)) {
 777                 oh->ot_handle = jh;
 778                 LASSERT(oti->oti_txns == 0);
 779                 lu_context_init(&th->th_ctx, th->th_tags);
 780                 lu_context_enter(&th->th_ctx);
 781
 782                 lu_device_get(&d->dd_lu_dev);
 783                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 784                                              "osd-tx", th);
 785                 oti->oti_txns++;
 786                 rc = 0;
 787         } else {
 788                 rc = PTR_ERR(jh);
 789         }
 790 out:
 791         RETURN(rc);
 792 }
 793
 794 /*
 795  * Concurrency: shouldn't matter.
 796  */
 797 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 798 {
 799         int                     rc = 0;
 800         struct osd_thandle     *oh;
 801         struct osd_thread_info *oti = osd_oti_get(env);
 802         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 803         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
 804         ENTRY;
 805
 806         oh = container_of0(th, struct osd_thandle, ot_super);
 807
 808         if (qsd != NULL)
 809                 /* inform the quota slave device that the transaction is
 810                  * stopping */
 811                 qsd_op_end(env, qsd, oh->ot_quota_trans);
 812         oh->ot_quota_trans = NULL;
 813
 814         if (oh->ot_handle != NULL) {
 815                 handle_t *hdl = oh->ot_handle;
 816
 817                 /*
 818                  * add commit callback
 819                  * notice we don't do this in osd_trans_start()
 820                  * as underlying transaction can change during truncate
 821                  */
 822                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
 823                                          &oh->ot_jcb);
 824
 825                 LASSERT(oti->oti_txns == 1);
 826                 oti->oti_txns--;
 827                 rc = dt_txn_hook_stop(env, th);
 828                 if (rc != 0)
 829                         CERROR("Failure in transaction hook: %d\n", rc);
 830
 831                 /* hook functions might modify th_sync */
 832                 hdl->h_sync = th->th_sync;
 833
 834                 oh->ot_handle = NULL;
 835                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 836                                   rc = ldiskfs_journal_stop(hdl));
 837                 if (rc != 0)
 838                         CERROR("Failure to stop transaction: %d\n", rc);
 839         } else {
 840                 OBD_FREE_PTR(oh);
 841         }
 842
 843         /* as we want IO to journal and data IO be concurrent, we don't block
 844          * awaiting data IO completion in osd_do_bio(), instead we wait here
 845          * once transaction is submitted to the journal. all reqular requests
 846          * don't do direct IO (except read/write), thus this wait_event becomes
 847          * no-op for them.
 848          *
 849          * IMPORTANT: we have to wait till any IO submited by the thread is
 850          * completed otherwise iobuf may be corrupted by different request
 851          */
 852         cfs_wait_event(iobuf->dr_wait,
 853                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 854         if (!rc)
 855                 rc = iobuf->dr_error;
 856
 857         RETURN(rc);
 858 }
 859
 860 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 861 {
 862         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 863                                                ot_super);
 864
 865         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 866         LASSERT(&dcb->dcb_func != NULL);
 867         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 868
 869         return 0;
 870 }
 871
 872 /*
 873  * Called just before object is freed. Releases all resources except for
 874  * object itself (that is released by osd_object_free()).
 875  *
 876  * Concurrency: no concurrent access is possible that late in object
 877  * life-cycle.
 878  */
 879 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 880 {
 881         struct osd_object *obj   = osd_obj(l);
 882         struct inode      *inode = obj->oo_inode;
 883
 884         LINVRNT(osd_invariant(obj));
 885
 886         /*
 887          * If object is unlinked remove fid->ino mapping from object index.
 888          */
 889
 890         osd_index_fini(obj);
 891         if (inode != NULL) {
 892                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
 893                 qid_t                    uid = inode->i_uid;
 894                 qid_t                    gid = inode->i_gid;
 895
 896                 iput(inode);
 897                 obj->oo_inode = NULL;
 898
 899                 if (qsd != NULL) {
 900                         struct osd_thread_info  *info = osd_oti_get(env);
 901                         struct lquota_id_info   *qi = &info->oti_qi;
 902
 903                         /* Release granted quota to master if necessary */
 904                         qi->lqi_id.qid_uid = uid;
 905                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
 906
 907                         qi->lqi_id.qid_uid = gid;
 908                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
 909                 }
 910         }
 911 }
 912
 913 /*
 914  * Concurrency: ->loo_object_release() is called under site spin-lock.
 915  */
 916 static void osd_object_release(const struct lu_env *env,
 917                                struct lu_object *l)
 918 {
 919 }
 920
 921 /*
 922  * Concurrency: shouldn't matter.
 923  */
 924 static int osd_object_print(const struct lu_env *env, void *cookie,
 925                             lu_printer_t p, const struct lu_object *l)
 926 {
 927         struct osd_object *o = osd_obj(l);
 928         struct iam_descr  *d;
 929
 930         if (o->oo_dir != NULL)
 931                 d = o->oo_dir->od_container.ic_descr;
 932         else
 933                 d = NULL;
 934         return (*p)(env, cookie,
 935                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 936                     o, o->oo_inode,
 937                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 938                     o->oo_inode ? o->oo_inode->i_generation : 0,
 939                     d ? d->id_ops->id_name : "plain");
 940 }
 941
 942 /*
 943  * Concurrency: shouldn't matter.
 944  */
 945 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 946                struct obd_statfs *sfs)
 947 {
 948         struct osd_device  *osd = osd_dt_dev(d);
 949         struct super_block *sb = osd_sb(osd);
 950         struct kstatfs     *ksfs;
 951         int result = 0;
 952
 953         if (unlikely(osd->od_mnt == NULL))
 954                 return -EINPROGRESS;
 955
 956         /* osd_lproc.c call this without env, allocate ksfs for that case */
 957         if (unlikely(env == NULL)) {
 958                 OBD_ALLOC_PTR(ksfs);
 959                 if (ksfs == NULL)
 960                         return -ENOMEM;
 961         } else {
 962                 ksfs = &osd_oti_get(env)->oti_ksfs;
 963         }
 964
 965         spin_lock(&osd->od_osfs_lock);
 966         /* cache 1 second */
 967         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 968                 result = sb->s_op->statfs(sb->s_root, ksfs);
 969                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 970                         osd->od_osfs_age = cfs_time_current_64();
 971                         statfs_pack(&osd->od_statfs, ksfs);
 972                         if (sb->s_flags & MS_RDONLY)
 973                                 sfs->os_state = OS_STATE_READONLY;
 974                 }
 975         }
 976
 977         if (likely(result == 0))
 978                 *sfs = osd->od_statfs;
 979         spin_unlock(&osd->od_osfs_lock);
 980
 981         if (unlikely(env == NULL))
 982                 OBD_FREE_PTR(ksfs);
 983
 984         return result;
 985 }
 986
 987 /**
 988  * Estimate space needed for file creations. We assume the largest filename
 989  * which is 2^64 - 1, hence a filename of 20 chars.
 990  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
 991  */
 992 #ifdef __LDISKFS_DIR_REC_LEN
 993 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
 994 #else
 995 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
 996 #endif
 997
 998 /*
 999  * Concurrency: doesn't access mutable data.
1000  */
1001 static void osd_conf_get(const struct lu_env *env,
1002                          const struct dt_device *dev,
1003                          struct dt_device_param *param)
1004 {
1005         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1006
1007         /*
1008          * XXX should be taken from not-yet-existing fs abstraction layer.
1009          */
1010         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1011         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1012         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1013         param->ddp_block_shift  = sb->s_blocksize_bits;
1014         param->ddp_mount_type     = LDD_MT_LDISKFS;
1015         param->ddp_maxbytes       = sb->s_maxbytes;
1016         /* Overhead estimate should be fairly accurate, so we really take a tiny
1017          * error margin which also avoids fragmenting the filesystem too much */
1018         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1019         /* inode are statically allocated, so per-inode space consumption
1020          * is the space consumed by the directory entry */
1021         param->ddp_inodespace     = PER_OBJ_USAGE;
1022         /* per-fragment overhead to be used by the client code */
1023         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1024         param->ddp_mntopts      = 0;
1025         if (test_opt(sb, XATTR_USER))
1026                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1027         if (test_opt(sb, POSIX_ACL))
1028                 param->ddp_mntopts |= MNTOPT_ACL;
1029
1030 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1031         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1032                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1033         else
1034 #endif
1035                 param->ddp_max_ea_size = sb->s_blocksize;
1036
1037 }
1038
1039 /*
1040  * Concurrency: shouldn't matter.
1041  */
1042 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1043 {
1044         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1045         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1046 }
1047
1048 /**
1049  * Start commit for OSD device.
1050  *
1051  * An implementation of dt_commit_async method for OSD device.
1052  * Asychronously starts underlayng fs sync and thereby a transaction
1053  * commit.
1054  *
1055  * \param env environment
1056  * \param d dt device
1057  *
1058  * \see dt_device_operations
1059  */
1060 static int osd_commit_async(const struct lu_env *env,
1061                             struct dt_device *d)
1062 {
1063         struct super_block *s = osd_sb(osd_dt_dev(d));
1064         ENTRY;
1065
1066         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1067         RETURN(s->s_op->sync_fs(s, 0));
1068 }
1069
1070 /*
1071  * Concurrency: shouldn't matter.
1072  */
1073
1074 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1075 {
1076         struct super_block *sb = osd_sb(osd_dt_dev(d));
1077         struct block_device *dev = sb->s_bdev;
1078 #ifdef HAVE_DEV_SET_RDONLY
1079         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1080         int rc = 0;
1081 #else
1082         int rc = -EOPNOTSUPP;
1083 #endif
1084         ENTRY;
1085
1086 #ifdef HAVE_DEV_SET_RDONLY
1087         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1088
1089         if (jdev && (jdev != dev)) {
1090                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1091                        (long)jdev);
1092                 dev_set_rdonly(jdev);
1093         }
1094         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1095         dev_set_rdonly(dev);
1096 #else
1097         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1098                osd_dt_dev(d)->od_svname, (long)dev, rc);
1099 #endif
1100         RETURN(rc);
1101 }
1102
1103 /*
1104  * Concurrency: serialization provided by callers.
1105  */
1106 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1107                               int mode, unsigned long timeout, __u32 alg,
1108                               struct lustre_capa_key *keys)
1109 {
1110         struct osd_device *dev = osd_dt_dev(d);
1111         ENTRY;
1112
1113         dev->od_fl_capa = mode;
1114         dev->od_capa_timeout = timeout;
1115         dev->od_capa_alg = alg;
1116         dev->od_capa_keys = keys;
1117         RETURN(0);
1118 }
1119
1120 /**
1121  * Note: we do not count into QUOTA here.
1122  * If we mount with --data_journal we may need more.
1123  */
1124 const int osd_dto_credits_noquota[DTO_NR] = {
1125         /**
1126          * Insert/Delete.
1127          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1128          * SINGLEDATA_TRANS_BLOCKS(8)
1129          * XXX Note: maybe iam need more, since iam have more level than
1130          *           EXT3 htree.
1131          */
1132         [DTO_INDEX_INSERT]  = 16,
1133         [DTO_INDEX_DELETE]  = 16,
1134         /**
1135          * Used for OI scrub
1136          */
1137         [DTO_INDEX_UPDATE]  = 16,
1138         /**
1139          * Create a object. The same as create object in EXT3.
1140          * DATA_TRANS_BLOCKS(14) +
1141          * INDEX_EXTRA_BLOCKS(8) +
1142          * 3(inode bits, groups, GDT)
1143          */
1144         [DTO_OBJECT_CREATE] = 25,
1145         /**
1146          * XXX: real credits to be fixed
1147          */
1148         [DTO_OBJECT_DELETE] = 25,
1149         /**
1150          * Attr set credits (inode)
1151          */
1152         [DTO_ATTR_SET_BASE] = 1,
1153         /**
1154          * Xattr set. The same as xattr of EXT3.
1155          * DATA_TRANS_BLOCKS(14)
1156          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1157          * are also counted in. Do not know why?
1158          */
1159         [DTO_XATTR_SET]     = 14,
1160         [DTO_LOG_REC]       = 14,
1161         /**
1162          * credits for inode change during write.
1163          */
1164         [DTO_WRITE_BASE]    = 3,
1165         /**
1166          * credits for single block write.
1167          */
1168         [DTO_WRITE_BLOCK]   = 14,
1169         /**
1170          * Attr set credits for chown.
1171          * This is extra credits for setattr, and it is null without quota
1172          */
1173         [DTO_ATTR_SET_CHOWN]= 0
1174 };
1175
1176 static const struct dt_device_operations osd_dt_ops = {
1177         .dt_root_get       = osd_root_get,
1178         .dt_statfs         = osd_statfs,
1179         .dt_trans_create   = osd_trans_create,
1180         .dt_trans_start    = osd_trans_start,
1181         .dt_trans_stop     = osd_trans_stop,
1182         .dt_trans_cb_add   = osd_trans_cb_add,
1183         .dt_conf_get       = osd_conf_get,
1184         .dt_sync           = osd_sync,
1185         .dt_ro             = osd_ro,
1186         .dt_commit_async   = osd_commit_async,
1187         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1188 };
1189
1190 static void osd_object_read_lock(const struct lu_env *env,
1191                                  struct dt_object *dt, unsigned role)
1192 {
1193         struct osd_object *obj = osd_dt_obj(dt);
1194         struct osd_thread_info *oti = osd_oti_get(env);
1195
1196         LINVRNT(osd_invariant(obj));
1197
1198         LASSERT(obj->oo_owner != env);
1199         down_read_nested(&obj->oo_sem, role);
1200
1201         LASSERT(obj->oo_owner == NULL);
1202         oti->oti_r_locks++;
1203 }
1204
1205 static void osd_object_write_lock(const struct lu_env *env,
1206                                   struct dt_object *dt, unsigned role)
1207 {
1208         struct osd_object *obj = osd_dt_obj(dt);
1209         struct osd_thread_info *oti = osd_oti_get(env);
1210
1211         LINVRNT(osd_invariant(obj));
1212
1213         LASSERT(obj->oo_owner != env);
1214         down_write_nested(&obj->oo_sem, role);
1215
1216         LASSERT(obj->oo_owner == NULL);
1217         obj->oo_owner = env;
1218         oti->oti_w_locks++;
1219 }
1220
1221 static void osd_object_read_unlock(const struct lu_env *env,
1222                                    struct dt_object *dt)
1223 {
1224         struct osd_object *obj = osd_dt_obj(dt);
1225         struct osd_thread_info *oti = osd_oti_get(env);
1226
1227         LINVRNT(osd_invariant(obj));
1228
1229         LASSERT(oti->oti_r_locks > 0);
1230         oti->oti_r_locks--;
1231         up_read(&obj->oo_sem);
1232 }
1233
1234 static void osd_object_write_unlock(const struct lu_env *env,
1235                                     struct dt_object *dt)
1236 {
1237         struct osd_object *obj = osd_dt_obj(dt);
1238         struct osd_thread_info *oti = osd_oti_get(env);
1239
1240         LINVRNT(osd_invariant(obj));
1241
1242         LASSERT(obj->oo_owner == env);
1243         LASSERT(oti->oti_w_locks > 0);
1244         oti->oti_w_locks--;
1245         obj->oo_owner = NULL;
1246         up_write(&obj->oo_sem);
1247 }
1248
1249 static int osd_object_write_locked(const struct lu_env *env,
1250                                    struct dt_object *dt)
1251 {
1252         struct osd_object *obj = osd_dt_obj(dt);
1253
1254         LINVRNT(osd_invariant(obj));
1255
1256         return obj->oo_owner == env;
1257 }
1258
1259 static int capa_is_sane(const struct lu_env *env,
1260                         struct osd_device *dev,
1261                         struct lustre_capa *capa,
1262                         struct lustre_capa_key *keys)
1263 {
1264         struct osd_thread_info *oti = osd_oti_get(env);
1265         struct lustre_capa *tcapa = &oti->oti_capa;
1266         struct obd_capa *oc;
1267         int i, rc = 0;
1268         ENTRY;
1269
1270         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1271         if (oc) {
1272                 if (capa_is_expired(oc)) {
1273                         DEBUG_CAPA(D_ERROR, capa, "expired");
1274                         rc = -ESTALE;
1275                 }
1276                 capa_put(oc);
1277                 RETURN(rc);
1278         }
1279
1280         if (capa_is_expired_sec(capa)) {
1281                 DEBUG_CAPA(D_ERROR, capa, "expired");
1282                 RETURN(-ESTALE);
1283         }
1284
1285         spin_lock(&capa_lock);
1286         for (i = 0; i < 2; i++) {
1287                 if (keys[i].lk_keyid == capa->lc_keyid) {
1288                         oti->oti_capa_key = keys[i];
1289                         break;
1290                 }
1291         }
1292         spin_unlock(&capa_lock);
1293
1294         if (i == 2) {
1295                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1296                 RETURN(-ESTALE);
1297         }
1298
1299         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1300         if (rc)
1301                 RETURN(rc);
1302
1303         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1304                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1305                 RETURN(-EACCES);
1306         }
1307
1308         oc = capa_add(dev->od_capa_hash, capa);
1309         capa_put(oc);
1310
1311         RETURN(0);
1312 }
1313
1314 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1315                     struct lustre_capa *capa, __u64 opc)
1316 {
1317         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1318         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1319         struct md_capainfo *ci;
1320         int rc;
1321
1322         if (!dev->od_fl_capa)
1323                 return 0;
1324
1325         if (capa == BYPASS_CAPA)
1326                 return 0;
1327
1328         ci = md_capainfo(env);
1329         if (unlikely(!ci))
1330                 return 0;
1331
1332         if (ci->mc_auth == LC_ID_NONE)
1333                 return 0;
1334
1335         if (!capa) {
1336                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1337                 return -EACCES;
1338         }
1339
1340         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1341                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1342                            PFID(fid));
1343                 return -EACCES;
1344         }
1345
1346         if (!capa_opc_supported(capa, opc)) {
1347                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1348                 return -EACCES;
1349         }
1350
1351         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1352                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1353                 return -EACCES;
1354         }
1355
1356         return 0;
1357 }
1358
1359 static struct timespec *osd_inode_time(const struct lu_env *env,
1360                                        struct inode *inode, __u64 seconds)
1361 {
1362         struct osd_thread_info  *oti = osd_oti_get(env);
1363         struct timespec         *t   = &oti->oti_time;
1364
1365         t->tv_sec = seconds;
1366         t->tv_nsec = 0;
1367         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1368         return t;
1369 }
1370
1371
1372 static void osd_inode_getattr(const struct lu_env *env,
1373                               struct inode *inode, struct lu_attr *attr)
1374 {
1375         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1376                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1377                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1378                                LA_TYPE;
1379
1380         attr->la_atime      = LTIME_S(inode->i_atime);
1381         attr->la_mtime      = LTIME_S(inode->i_mtime);
1382         attr->la_ctime      = LTIME_S(inode->i_ctime);
1383         attr->la_mode       = inode->i_mode;
1384         attr->la_size       = i_size_read(inode);
1385         attr->la_blocks     = inode->i_blocks;
1386         attr->la_uid        = inode->i_uid;
1387         attr->la_gid        = inode->i_gid;
1388         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1389         attr->la_nlink      = inode->i_nlink;
1390         attr->la_rdev       = inode->i_rdev;
1391         attr->la_blksize    = 1 << inode->i_blkbits;
1392         attr->la_blkbits    = inode->i_blkbits;
1393 }
1394
1395 static int osd_attr_get(const struct lu_env *env,
1396                         struct dt_object *dt,
1397                         struct lu_attr *attr,
1398                         struct lustre_capa *capa)
1399 {
1400         struct osd_object *obj = osd_dt_obj(dt);
1401
1402         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1403         LINVRNT(osd_invariant(obj));
1404
1405         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1406                 return -EACCES;
1407
1408         spin_lock(&obj->oo_guard);
1409         osd_inode_getattr(env, obj->oo_inode, attr);
1410         spin_unlock(&obj->oo_guard);
1411         return 0;
1412 }
1413
1414 static int osd_declare_attr_set(const struct lu_env *env,
1415                                 struct dt_object *dt,
1416                                 const struct lu_attr *attr,
1417                                 struct thandle *handle)
1418 {
1419         struct osd_thandle     *oh;
1420         struct osd_object      *obj;
1421         struct osd_thread_info *info = osd_oti_get(env);
1422         struct lquota_id_info  *qi = &info->oti_qi;
1423         long long               bspace;
1424         int                     rc = 0;
1425         bool                    allocated;
1426         ENTRY;
1427
1428         LASSERT(dt != NULL);
1429         LASSERT(handle != NULL);
1430
1431         obj = osd_dt_obj(dt);
1432         LASSERT(osd_invariant(obj));
1433
1434         oh = container_of0(handle, struct osd_thandle, ot_super);
1435         LASSERT(oh->ot_handle == NULL);
1436
1437         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1438                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1439
1440         if (attr == NULL || obj->oo_inode == NULL)
1441                 RETURN(rc);
1442
1443         bspace   = obj->oo_inode->i_blocks;
1444         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1445         bspace   = toqb(bspace);
1446
1447         /* Changing ownership is always preformed by super user, it should not
1448          * fail with EDQUOT.
1449          *
1450          * We still need to call the osd_declare_qid() to calculate the journal
1451          * credits for updating quota accounting files and to trigger quota
1452          * space adjustment once the operation is completed.*/
1453         if ((attr->la_valid & LA_UID) != 0 &&
1454              attr->la_uid != obj->oo_inode->i_uid) {
1455                 qi->lqi_type = USRQUOTA;
1456
1457                 /* inode accounting */
1458                 qi->lqi_is_blk = false;
1459
1460                 /* one more inode for the new owner ... */
1461                 qi->lqi_id.qid_uid = attr->la_uid;
1462                 qi->lqi_space      = 1;
1463                 allocated = (attr->la_uid == 0) ? true : false;
1464                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1465                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1466                         rc = 0;
1467                 if (rc)
1468                         RETURN(rc);
1469
1470                 /* and one less inode for the current uid */
1471                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1472                 qi->lqi_space      = -1;
1473                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1474                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1475                         rc = 0;
1476                 if (rc)
1477                         RETURN(rc);
1478
1479                 /* block accounting */
1480                 qi->lqi_is_blk = true;
1481
1482                 /* more blocks for the new owner ... */
1483                 qi->lqi_id.qid_uid = attr->la_uid;
1484                 qi->lqi_space      = bspace;
1485                 allocated = (attr->la_uid == 0) ? true : false;
1486                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1487                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1488                         rc = 0;
1489                 if (rc)
1490                         RETURN(rc);
1491
1492                 /* and finally less blocks for the current owner */
1493                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1494                 qi->lqi_space      = -bspace;
1495                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1496                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1497                         rc = 0;
1498                 if (rc)
1499                         RETURN(rc);
1500         }
1501
1502         if (attr->la_valid & LA_GID &&
1503             attr->la_gid != obj->oo_inode->i_gid) {
1504                 qi->lqi_type = GRPQUOTA;
1505
1506                 /* inode accounting */
1507                 qi->lqi_is_blk = false;
1508
1509                 /* one more inode for the new group owner ... */
1510                 qi->lqi_id.qid_gid = attr->la_gid;
1511                 qi->lqi_space      = 1;
1512                 allocated = (attr->la_gid == 0) ? true : false;
1513                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1514                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1515                         rc = 0;
1516                 if (rc)
1517                         RETURN(rc);
1518
1519                 /* and one less inode for the current gid */
1520                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1521                 qi->lqi_space      = -1;
1522                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1523                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1524                         rc = 0;
1525                 if (rc)
1526                         RETURN(rc);
1527
1528                 /* block accounting */
1529                 qi->lqi_is_blk = true;
1530
1531                 /* more blocks for the new owner ... */
1532                 qi->lqi_id.qid_gid = attr->la_gid;
1533                 qi->lqi_space      = bspace;
1534                 allocated = (attr->la_gid == 0) ? true : false;
1535                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1536                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1537                         rc = 0;
1538                 if (rc)
1539                         RETURN(rc);
1540
1541                 /* and finally less blocks for the current owner */
1542                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1543                 qi->lqi_space      = -bspace;
1544                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1545                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1546                         rc = 0;
1547                 if (rc)
1548                         RETURN(rc);
1549         }
1550
1551         RETURN(rc);
1552 }
1553
1554 static int osd_inode_setattr(const struct lu_env *env,
1555                              struct inode *inode, const struct lu_attr *attr)
1556 {
1557         __u64 bits;
1558
1559         bits = attr->la_valid;
1560
1561         if (bits & LA_ATIME)
1562                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1563         if (bits & LA_CTIME)
1564                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1565         if (bits & LA_MTIME)
1566                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1567         if (bits & LA_SIZE) {
1568                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1569                 i_size_write(inode, attr->la_size);
1570         }
1571
1572 #if 0
1573         /* OSD should not change "i_blocks" which is used by quota.
1574          * "i_blocks" should be changed by ldiskfs only. */
1575         if (bits & LA_BLOCKS)
1576                 inode->i_blocks = attr->la_blocks;
1577 #endif
1578         if (bits & LA_MODE)
1579                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1580                         (attr->la_mode & ~S_IFMT);
1581         if (bits & LA_UID)
1582                 inode->i_uid    = attr->la_uid;
1583         if (bits & LA_GID)
1584                 inode->i_gid    = attr->la_gid;
1585         if (bits & LA_NLINK)
1586                 set_nlink(inode, attr->la_nlink);
1587         if (bits & LA_RDEV)
1588                 inode->i_rdev   = attr->la_rdev;
1589
1590         if (bits & LA_FLAGS) {
1591                 /* always keep S_NOCMTIME */
1592                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1593                                  S_NOCMTIME;
1594         }
1595         return 0;
1596 }
1597
1598 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1599 {
1600         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1601             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1602                 struct iattr    iattr;
1603                 int             rc;
1604
1605                 iattr.ia_valid = 0;
1606                 if (attr->la_valid & LA_UID)
1607                         iattr.ia_valid |= ATTR_UID;
1608                 if (attr->la_valid & LA_GID)
1609                         iattr.ia_valid |= ATTR_GID;
1610                 iattr.ia_uid = attr->la_uid;
1611                 iattr.ia_gid = attr->la_gid;
1612
1613                 rc = ll_vfs_dq_transfer(inode, &iattr);
1614                 if (rc) {
1615                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1616                                "enforcement enabled on the ldiskfs filesystem?",
1617                                inode->i_sb->s_id, rc);
1618                         return rc;
1619                 }
1620         }
1621         return 0;
1622 }
1623
1624 static int osd_attr_set(const struct lu_env *env,
1625                         struct dt_object *dt,
1626                         const struct lu_attr *attr,
1627                         struct thandle *handle,
1628                         struct lustre_capa *capa)
1629 {
1630         struct osd_object *obj = osd_dt_obj(dt);
1631         struct inode      *inode;
1632         int rc;
1633
1634         LASSERT(handle != NULL);
1635         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1636         LASSERT(osd_invariant(obj));
1637
1638         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1639                 return -EACCES;
1640
1641         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1642
1643         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1644                 struct osd_thread_info  *oti  = osd_oti_get(env);
1645                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1646                 struct lu_fid           *fid1 = &oti->oti_fid;
1647                 struct osd_inode_id     *id   = &oti->oti_id;
1648                 struct iam_path_descr   *ipd;
1649                 struct iam_container    *bag;
1650                 struct osd_thandle      *oh;
1651                 int                      rc;
1652
1653                 fid_cpu_to_be(fid1, fid0);
1654                 memset(id, 1, sizeof(*id));
1655                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1656                                   fid0)->oi_dir.od_container;
1657                 ipd = osd_idx_ipd_get(env, bag);
1658                 if (unlikely(ipd == NULL))
1659                         RETURN(-ENOMEM);
1660
1661                 oh = container_of0(handle, struct osd_thandle, ot_super);
1662                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1663                                 (const struct iam_rec *)id, ipd);
1664                 osd_ipd_put(env, bag, ipd);
1665                 return(rc > 0 ? 0 : rc);
1666         }
1667
1668         inode = obj->oo_inode;
1669         ll_vfs_dq_init(inode);
1670
1671         rc = osd_quota_transfer(inode, attr);
1672         if (rc)
1673                 return rc;
1674
1675         spin_lock(&obj->oo_guard);
1676         rc = osd_inode_setattr(env, inode, attr);
1677         spin_unlock(&obj->oo_guard);
1678
1679         if (!rc)
1680                 inode->i_sb->s_op->dirty_inode(inode);
1681         return rc;
1682 }
1683
1684 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1685                                     struct osd_object *obj,
1686                                     const char *name, const int namelen)
1687 {
1688         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1689 }
1690
1691 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1692                       cfs_umode_t mode,
1693                       struct dt_allocation_hint *hint,
1694                       struct thandle *th)
1695 {
1696         int result;
1697         struct osd_device  *osd = osd_obj2dev(obj);
1698         struct osd_thandle *oth;
1699         struct dt_object   *parent = NULL;
1700         struct inode       *inode;
1701
1702         LINVRNT(osd_invariant(obj));
1703         LASSERT(obj->oo_inode == NULL);
1704         LASSERT(obj->oo_hl_head == NULL);
1705
1706         if (S_ISDIR(mode) && ldiskfs_pdo) {
1707                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1708                 if (obj->oo_hl_head == NULL)
1709                         return -ENOMEM;
1710         }
1711
1712         oth = container_of(th, struct osd_thandle, ot_super);
1713         LASSERT(oth->ot_handle->h_transaction != NULL);
1714
1715         if (hint && hint->dah_parent)
1716                 parent = hint->dah_parent;
1717
1718         inode = ldiskfs_create_inode(oth->ot_handle,
1719                                      parent ? osd_dt_obj(parent)->oo_inode :
1720                                               osd_sb(osd)->s_root->d_inode,
1721                                      mode);
1722         if (!IS_ERR(inode)) {
1723                 /* Do not update file c/mtime in ldiskfs.
1724                  * NB: don't need any lock because no contention at this
1725                  * early stage */
1726                 inode->i_flags |= S_NOCMTIME;
1727
1728                 /* For new created object, it must be consistent,
1729                  * and it is unnecessary to scrub against it. */
1730                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1731                 obj->oo_inode = inode;
1732                 result = 0;
1733         } else {
1734                 if (obj->oo_hl_head != NULL) {
1735                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1736                         obj->oo_hl_head = NULL;
1737                 }
1738                 result = PTR_ERR(inode);
1739         }
1740         LINVRNT(osd_invariant(obj));
1741         return result;
1742 }
1743
1744 enum {
1745         OSD_NAME_LEN = 255
1746 };
1747
1748 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1749                      struct lu_attr *attr,
1750                      struct dt_allocation_hint *hint,
1751                      struct dt_object_format *dof,
1752                      struct thandle *th)
1753 {
1754         int result;
1755         struct osd_thandle *oth;
1756         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1757
1758         LASSERT(S_ISDIR(attr->la_mode));
1759
1760         oth = container_of(th, struct osd_thandle, ot_super);
1761         LASSERT(oth->ot_handle->h_transaction != NULL);
1762         result = osd_mkfile(info, obj, mode, hint, th);
1763
1764         return result;
1765 }
1766
1767 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1768                         struct lu_attr *attr,
1769                         struct dt_allocation_hint *hint,
1770                         struct dt_object_format *dof,
1771                         struct thandle *th)
1772 {
1773         int result;
1774         struct osd_thandle *oth;
1775         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1776
1777         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1778
1779         LASSERT(S_ISREG(attr->la_mode));
1780
1781         oth = container_of(th, struct osd_thandle, ot_super);
1782         LASSERT(oth->ot_handle->h_transaction != NULL);
1783
1784         result = osd_mkfile(info, obj, mode, hint, th);
1785         if (result == 0) {
1786                 LASSERT(obj->oo_inode != NULL);
1787                 if (feat->dif_flags & DT_IND_VARKEY)
1788                         result = iam_lvar_create(obj->oo_inode,
1789                                                  feat->dif_keysize_max,
1790                                                  feat->dif_ptrsize,
1791                                                  feat->dif_recsize_max,
1792                                                  oth->ot_handle);
1793                 else
1794                         result = iam_lfix_create(obj->oo_inode,
1795                                                  feat->dif_keysize_max,
1796                                                  feat->dif_ptrsize,
1797                                                  feat->dif_recsize_max,
1798                                                  oth->ot_handle);
1799
1800         }
1801         return result;
1802 }
1803
1804 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1805                      struct lu_attr *attr,
1806                      struct dt_allocation_hint *hint,
1807                      struct dt_object_format *dof,
1808                      struct thandle *th)
1809 {
1810         LASSERT(S_ISREG(attr->la_mode));
1811         return osd_mkfile(info, obj, (attr->la_mode &
1812                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1813 }
1814
1815 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1816                      struct lu_attr *attr,
1817                      struct dt_allocation_hint *hint,
1818                      struct dt_object_format *dof,
1819                      struct thandle *th)
1820 {
1821         LASSERT(S_ISLNK(attr->la_mode));
1822         return osd_mkfile(info, obj, (attr->la_mode &
1823                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1824 }
1825
1826 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1827                      struct lu_attr *attr,
1828                      struct dt_allocation_hint *hint,
1829                      struct dt_object_format *dof,
1830                      struct thandle *th)
1831 {
1832         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1833         int result;
1834
1835         LINVRNT(osd_invariant(obj));
1836         LASSERT(obj->oo_inode == NULL);
1837         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1838                 S_ISFIFO(mode) || S_ISSOCK(mode));
1839
1840         result = osd_mkfile(info, obj, mode, hint, th);
1841         if (result == 0) {
1842                 LASSERT(obj->oo_inode != NULL);
1843                 /*
1844                  * This inode should be marked dirty for i_rdev.  Currently
1845                  * that is done in the osd_attr_init().
1846                  */
1847                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1848                                    attr->la_rdev);
1849         }
1850         LINVRNT(osd_invariant(obj));
1851         return result;
1852 }
1853
1854 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1855                               struct lu_attr *,
1856                               struct dt_allocation_hint *hint,
1857                               struct dt_object_format *dof,
1858                               struct thandle *);
1859
1860 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1861 {
1862         osd_obj_type_f result;
1863
1864         switch (type) {
1865         case DFT_DIR:
1866                 result = osd_mkdir;
1867                 break;
1868         case DFT_REGULAR:
1869                 result = osd_mkreg;
1870                 break;
1871         case DFT_SYM:
1872                 result = osd_mksym;
1873                 break;
1874         case DFT_NODE:
1875                 result = osd_mknod;
1876                 break;
1877         case DFT_INDEX:
1878                 result = osd_mk_index;
1879                 break;
1880
1881         default:
1882                 LBUG();
1883                 break;
1884         }
1885         return result;
1886 }
1887
1888
1889 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1890                         struct dt_object *parent, struct dt_object *child,
1891                         cfs_umode_t child_mode)
1892 {
1893         LASSERT(ah);
1894
1895         memset(ah, 0, sizeof(*ah));
1896         ah->dah_parent = parent;
1897         ah->dah_mode = child_mode;
1898 }
1899
1900 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1901                           struct lu_attr *attr, struct dt_object_format *dof)
1902 {
1903         struct inode   *inode = obj->oo_inode;
1904         __u64           valid = attr->la_valid;
1905         int             result;
1906
1907         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1908
1909         if (dof->dof_type != DFT_NODE)
1910                 attr->la_valid &= ~LA_RDEV;
1911         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1912                 attr->la_valid &= ~LA_ATIME;
1913         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1914                 attr->la_valid &= ~LA_CTIME;
1915         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1916                 attr->la_valid &= ~LA_MTIME;
1917
1918         result = osd_quota_transfer(inode, attr);
1919         if (result)
1920                 return;
1921
1922         if (attr->la_valid != 0) {
1923                 result = osd_inode_setattr(info->oti_env, inode, attr);
1924                 /*
1925                  * The osd_inode_setattr() should always succeed here.  The
1926                  * only error that could be returned is EDQUOT when we are
1927                  * trying to change the UID or GID of the inode. However, this
1928                  * should not happen since quota enforcement is no longer
1929                  * enabled on ldiskfs (lquota takes care of it).
1930                  */
1931                 LASSERTF(result == 0, "%d", result);
1932                 inode->i_sb->s_op->dirty_inode(inode);
1933         }
1934
1935         attr->la_valid = valid;
1936 }
1937
1938 /**
1939  * Helper function for osd_object_create()
1940  *
1941  * \retval 0, on success
1942  */
1943 static int __osd_object_create(struct osd_thread_info *info,
1944                                struct osd_object *obj, struct lu_attr *attr,
1945                                struct dt_allocation_hint *hint,
1946                                struct dt_object_format *dof,
1947                                struct thandle *th)
1948 {
1949         int     result;
1950
1951         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1952                                                   th);
1953         if (result == 0) {
1954                 osd_attr_init(info, obj, attr, dof);
1955                 osd_object_init0(obj);
1956                 /* bz 24037 */
1957                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1958                         unlock_new_inode(obj->oo_inode);
1959         }
1960
1961         return result;
1962 }
1963
1964 /**
1965  * Helper function for osd_object_create()
1966  *
1967  * \retval 0, on success
1968  */
1969 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1970                            const struct lu_fid *fid, struct thandle *th)
1971 {
1972         struct osd_thread_info *info = osd_oti_get(env);
1973         struct osd_inode_id    *id   = &info->oti_id;
1974         struct osd_device      *osd  = osd_obj2dev(obj);
1975
1976         LASSERT(obj->oo_inode != NULL);
1977
1978         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1979         return osd_oi_insert(info, osd, fid, id, th);
1980 }
1981
1982 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1983                    const struct lu_fid *fid, struct lu_seq_range *range)
1984 {
1985         struct seq_server_site  *ss = osd_seq_site(osd);
1986         int                     rc;
1987
1988         if (fid_is_idif(fid)) {
1989                 range->lsr_flags = LU_SEQ_RANGE_OST;
1990                 range->lsr_index = fid_idif_ost_idx(fid);
1991                 return 0;
1992         }
1993
1994         if (!fid_seq_in_fldb(fid_seq(fid))) {
1995                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1996                 if (ss != NULL)
1997                         /* FIXME: If ss is NULL, it suppose not get lsr_index
1998                          * at all */
1999                         range->lsr_index = ss->ss_node_id;
2000                 return 0;
2001         }
2002
2003         LASSERT(ss != NULL);
2004         range->lsr_flags = -1;
2005         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2006         if (rc != 0) {
2007                 CERROR("%s can not find "DFID": rc = %d\n",
2008                        osd_name(osd), PFID(fid), rc);
2009         }
2010         return rc;
2011 }
2012
2013 /*
2014  * Concurrency: no external locking is necessary.
2015  */
2016 static int osd_declare_object_create(const struct lu_env *env,
2017                                      struct dt_object *dt,
2018                                      struct lu_attr *attr,
2019                                      struct dt_allocation_hint *hint,
2020                                      struct dt_object_format *dof,
2021                                      struct thandle *handle)
2022 {
2023         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
2024         struct osd_thandle      *oh;
2025         int                      rc;
2026         ENTRY;
2027
2028         LASSERT(handle != NULL);
2029
2030         oh = container_of0(handle, struct osd_thandle, ot_super);
2031         LASSERT(oh->ot_handle == NULL);
2032
2033         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2034                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2035         if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2036                            lu_object_fid(&dt->do_lu)))
2037                 /* Reuse idle OI block may cause additional one OI block
2038                  * to be changed. */
2039                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2040                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2041
2042         /* If this is directory, then we expect . and .. to be inserted as
2043          * well. The one directory block always needs to be created for the
2044          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2045          * block), there is no danger of needing a tree for the first block.
2046          */
2047         if (attr && S_ISDIR(attr->la_mode)) {
2048                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2049                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2050                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2051         }
2052
2053         if (!attr)
2054                 RETURN(0);
2055
2056         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2057                                    false, false, NULL, false);
2058         if (rc != 0)
2059                 RETURN(rc);
2060
2061         /* It does fld look up inside declare, and the result will be
2062          * added to fld cache, so the following fld lookup inside insert
2063          * does not need send RPC anymore, so avoid send rpc with holding
2064          * transaction */
2065         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2066                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2067                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2068                                lu_object_fid(&dt->do_lu), range);
2069
2070
2071         RETURN(rc);
2072 }
2073
2074 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2075                              struct lu_attr *attr,
2076                              struct dt_allocation_hint *hint,
2077                              struct dt_object_format *dof,
2078                              struct thandle *th)
2079 {
2080         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2081         struct osd_object      *obj    = osd_dt_obj(dt);
2082         struct osd_thread_info *info   = osd_oti_get(env);
2083         int result;
2084
2085         ENTRY;
2086
2087         LINVRNT(osd_invariant(obj));
2088         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2089         LASSERT(osd_write_locked(env, obj));
2090         LASSERT(th != NULL);
2091
2092         if (unlikely(fid_is_acct(fid)))
2093                 /* Quota files can't be created from the kernel any more,
2094                  * 'tune2fs -O quota' will take care of creating them */
2095                 RETURN(-EPERM);
2096
2097         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2098         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2099
2100         result = __osd_object_create(info, obj, attr, hint, dof, th);
2101         if (result == 0)
2102                 result = __osd_oi_insert(env, obj, fid, th);
2103
2104         LASSERT(ergo(result == 0,
2105                      dt_object_exists(dt) && !dt_object_remote(dt)));
2106
2107         LASSERT(osd_invariant(obj));
2108         RETURN(result);
2109 }
2110
2111 /**
2112  * Called to destroy on-disk representation of the object
2113  *
2114  * Concurrency: must be locked
2115  */
2116 static int osd_declare_object_destroy(const struct lu_env *env,
2117                                       struct dt_object *dt,
2118                                       struct thandle *th)
2119 {
2120         struct osd_object  *obj = osd_dt_obj(dt);
2121         struct inode       *inode = obj->oo_inode;
2122         struct osd_thandle *oh;
2123         int                 rc;
2124         ENTRY;
2125
2126         oh = container_of0(th, struct osd_thandle, ot_super);
2127         LASSERT(oh->ot_handle == NULL);
2128         LASSERT(inode);
2129
2130         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2131                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2132         /* Recycle idle OI leaf may cause additional three OI blocks
2133          * to be changed. */
2134         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2135                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2136
2137         /* one less inode */
2138         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2139                                    false, true, NULL, false);
2140         if (rc)
2141                 RETURN(rc);
2142         /* data to be truncated */
2143         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2144                                    true, true, NULL, false);
2145         RETURN(rc);
2146 }
2147
2148 static int osd_object_destroy(const struct lu_env *env,
2149                               struct dt_object *dt,
2150                               struct thandle *th)
2151 {
2152         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2153         struct osd_object      *obj = osd_dt_obj(dt);
2154         struct inode           *inode = obj->oo_inode;
2155         struct osd_device      *osd = osd_obj2dev(obj);
2156         struct osd_thandle     *oh;
2157         int                     result;
2158         ENTRY;
2159
2160         oh = container_of0(th, struct osd_thandle, ot_super);
2161         LASSERT(oh->ot_handle);
2162         LASSERT(inode);
2163         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2164
2165         if (unlikely(fid_is_acct(fid)))
2166                 RETURN(-EPERM);
2167
2168         /* Parallel control for OI scrub. For most of cases, there is no
2169          * lock contention. So it will not affect unlink performance. */
2170         mutex_lock(&inode->i_mutex);
2171         if (S_ISDIR(inode->i_mode)) {
2172                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2173                 /* it will check/delete the agent inode for every dir
2174                  * destory, how to optimize it? unlink performance
2175                  * impaction XXX */
2176                 result = osd_delete_from_agent(env, osd, obj, oh);
2177                 if (result != 0 && result != -ENOENT) {
2178                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
2179                                osd_name(osd), PFID(fid), result);
2180                 }
2181                 spin_lock(&obj->oo_guard);
2182                 clear_nlink(inode);
2183                 spin_unlock(&obj->oo_guard);
2184                 inode->i_sb->s_op->dirty_inode(inode);
2185         }
2186
2187         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2188
2189         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2190         mutex_unlock(&inode->i_mutex);
2191
2192         /* XXX: add to ext3 orphan list */
2193         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2194
2195         /* not needed in the cache anymore */
2196         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2197
2198         RETURN(0);
2199 }
2200
2201 static inline int __osd_xattr_set(struct osd_thread_info *info,
2202                                   struct inode *inode, const char *name,
2203                                   const void *buf, int buflen, int fl)
2204 {
2205         struct dentry *dentry = &info->oti_child_dentry;
2206
2207         ll_vfs_dq_init(inode);
2208         dentry->d_inode = inode;
2209         return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
2210 }
2211
2212 /**
2213  * Put the fid into lustre_mdt_attrs, and then place the structure
2214  * inode's ea. This fid should not be altered during the life time
2215  * of the inode.
2216  *
2217  * \retval +ve, on success
2218  * \retval -ve, on error
2219  *
2220  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2221  */
2222 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2223                    const struct lu_fid *fid)
2224 {
2225         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2226         int                      rc;
2227
2228         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2229                 return 0;
2230
2231         if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid))
2232                 return 0;
2233
2234         lustre_lma_init(lma, fid);
2235         lustre_lma_swab(lma);
2236
2237         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2238                              XATTR_CREATE);
2239         /* Someone may created the EA by race. */
2240         if (unlikely(rc == -EEXIST))
2241                 rc = 0;
2242         return rc;
2243 }
2244
2245 /**
2246  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2247  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2248  * To have compatilibility with 1.8 ldiskfs driver we need to have
2249  * magic number at start of fid data.
2250  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2251  * its inmemory API.
2252  */
2253 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2254                                   const struct dt_rec *fid)
2255 {
2256         if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) {
2257                 param->edp_magic = 0;
2258                 return;
2259         }
2260
2261         param->edp_magic = LDISKFS_LUFID_MAGIC;
2262         param->edp_len =  sizeof(struct lu_fid) + 1;
2263         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2264 }
2265
2266 /**
2267  * Try to read the fid from inode ea into dt_rec.
2268  *
2269  * \param fid object fid.
2270  *
2271  * \retval 0 on success
2272  */
2273 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2274                           __u32 ino, struct lu_fid *fid,
2275                           struct osd_inode_id *id)
2276 {
2277         struct osd_thread_info *info  = osd_oti_get(env);
2278         struct inode           *inode;
2279         ENTRY;
2280
2281         osd_id_gen(id, ino, OSD_OII_NOGEN);
2282         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2283         if (IS_ERR(inode))
2284                 RETURN(PTR_ERR(inode));
2285
2286         iput(inode);
2287         RETURN(0);
2288 }
2289
2290 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2291                                         struct inode *dir,
2292                                         struct inode  *parent_dir,
2293                                         const struct dt_rec *dot_fid,
2294                                         const struct dt_rec *dot_dot_fid,
2295                                         struct osd_thandle *oth)
2296 {
2297         struct ldiskfs_dentry_param *dot_ldp;
2298         struct ldiskfs_dentry_param *dot_dot_ldp;
2299
2300         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2301         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2302
2303         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2304         dot_ldp->edp_magic = 0;
2305         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2306                                         dir, dot_ldp, dot_dot_ldp);
2307 }
2308
2309 /**
2310  * Create an local inode for remote entry
2311  */
2312 static struct inode *osd_create_remote_inode(const struct lu_env *env,
2313                                              struct osd_device *osd,
2314                                              struct osd_object *pobj,
2315                                              const struct lu_fid *fid,
2316                                              struct thandle *th)
2317 {
2318         struct osd_thread_info  *info = osd_oti_get(env);
2319         struct inode            *local;
2320         struct osd_thandle      *oh;
2321         int                     rc;
2322         ENTRY;
2323
2324         LASSERT(th);
2325         oh = container_of(th, struct osd_thandle, ot_super);
2326         LASSERT(oh->ot_handle->h_transaction != NULL);
2327
2328         /* FIXME: Insert index api needs to know the mode of
2329          * the remote object. Just use S_IFDIR for now */
2330         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2331         if (IS_ERR(local)) {
2332                 CERROR("%s: create local error %d\n", osd_name(osd),
2333                        (int)PTR_ERR(local));
2334                 RETURN(local);
2335         }
2336
2337         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2338                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2339                 (const struct dt_rec *)fid, oh);
2340         if (rc != 0) {
2341                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2342                         osd_name(osd), PFID(fid), rc);
2343                 RETURN(ERR_PTR(rc));
2344         }
2345
2346         RETURN(local);
2347 }
2348
2349 /**
2350  * Delete local inode for remote entry
2351  */
2352 static int osd_delete_remote_inode(const struct lu_env *env,
2353                                    struct osd_device *osd,
2354                                    const struct lu_fid *fid,
2355                                     __u32 ino, struct osd_thandle *oh)
2356 {
2357         struct osd_thread_info  *oti = osd_oti_get(env);
2358         struct osd_inode_id     *id = &oti->oti_id;
2359         struct inode            *inode;
2360         ENTRY;
2361
2362         id->oii_ino = le32_to_cpu(ino);
2363         id->oii_gen = OSD_OII_NOGEN;
2364         inode = osd_iget(oti, osd, id);
2365         if (IS_ERR(inode)) {
2366                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2367                        PFID(fid), id->oii_ino, id->oii_gen);
2368                 RETURN(PTR_ERR(inode));
2369         }
2370
2371         clear_nlink(inode);
2372         mark_inode_dirty(inode);
2373         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2374                 osd_name(osd), PFID(fid), inode->i_ino);
2375         iput(inode);
2376         RETURN(0);
2377 }
2378
2379 /**
2380  * OSD layer object create function for interoperability mode (b11826).
2381  * This is mostly similar to osd_object_create(). Only difference being, fid is
2382  * inserted into inode ea here.
2383  *
2384  * \retval   0, on success
2385  * \retval -ve, on error
2386  */
2387 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2388                                 struct lu_attr *attr,
2389                                 struct dt_allocation_hint *hint,
2390                                 struct dt_object_format *dof,
2391                                 struct thandle *th)
2392 {
2393         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2394         struct osd_object      *obj    = osd_dt_obj(dt);
2395         struct osd_thread_info *info   = osd_oti_get(env);
2396         int                     result;
2397
2398         ENTRY;
2399
2400         LASSERT(osd_invariant(obj));
2401         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2402         LASSERT(osd_write_locked(env, obj));
2403         LASSERT(th != NULL);
2404
2405         if (unlikely(fid_is_acct(fid)))
2406                 /* Quota files can't be created from the kernel any more,
2407                  * 'tune2fs -O quota' will take care of creating them */
2408                 RETURN(-EPERM);
2409
2410         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2411         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2412
2413         result = __osd_object_create(info, obj, attr, hint, dof, th);
2414         if ((result == 0) &&
2415             (fid_is_last_id(fid) ||
2416              !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)))
2417                 result = osd_ea_fid_set(info, obj->oo_inode, fid);
2418
2419         if (result == 0)
2420                 result = __osd_oi_insert(env, obj, fid, th);
2421
2422         LASSERT(ergo(result == 0,
2423                      dt_object_exists(dt) && !dt_object_remote(dt)));
2424         LINVRNT(osd_invariant(obj));
2425         RETURN(result);
2426 }
2427
2428 static int osd_declare_object_ref_add(const struct lu_env *env,
2429                                       struct dt_object *dt,
2430                                       struct thandle *handle)
2431 {
2432         struct osd_thandle       *oh;
2433
2434         /* it's possible that object doesn't exist yet */
2435         LASSERT(handle != NULL);
2436
2437         oh = container_of0(handle, struct osd_thandle, ot_super);
2438         LASSERT(oh->ot_handle == NULL);
2439
2440         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2441                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2442
2443         return 0;
2444 }
2445
2446 /*
2447  * Concurrency: @dt is write locked.
2448  */
2449 static int osd_object_ref_add(const struct lu_env *env,
2450                               struct dt_object *dt, struct thandle *th)
2451 {
2452         struct osd_object *obj = osd_dt_obj(dt);
2453         struct inode      *inode = obj->oo_inode;
2454
2455         LINVRNT(osd_invariant(obj));
2456         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2457         LASSERT(osd_write_locked(env, obj));
2458         LASSERT(th != NULL);
2459
2460         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2461
2462         /*
2463          * DIR_NLINK feature is set for compatibility reasons if:
2464          * 1) nlinks > LDISKFS_LINK_MAX, or
2465          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2466          *
2467          * It is easier to always set this flag (rather than check and set),
2468          * since it has less overhead, and the superblock will be dirtied
2469          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2470          * do not actually care whether this flag is set or not.
2471          */
2472         spin_lock(&obj->oo_guard);
2473         /* inc_nlink from 0 may cause WARN_ON */
2474         if(inode->i_nlink == 0)
2475                 set_nlink(inode, 1);
2476         else
2477                 inc_nlink(inode);
2478         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2479                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2480                     inode->i_nlink == 2)
2481                         set_nlink(inode, 1);
2482         }
2483         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2484         spin_unlock(&obj->oo_guard);
2485         inode->i_sb->s_op->dirty_inode(inode);
2486         LINVRNT(osd_invariant(obj));
2487
2488         return 0;
2489 }
2490
2491 static int osd_declare_object_ref_del(const struct lu_env *env,
2492                                       struct dt_object *dt,
2493                                       struct thandle *handle)
2494 {
2495         struct osd_thandle *oh;
2496
2497         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2498         LASSERT(handle != NULL);
2499
2500         oh = container_of0(handle, struct osd_thandle, ot_super);
2501         LASSERT(oh->ot_handle == NULL);
2502
2503         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2504                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2505
2506         return 0;
2507 }
2508
2509 /*
2510  * Concurrency: @dt is write locked.
2511  */
2512 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2513                               struct thandle *th)
2514 {
2515         struct osd_object *obj = osd_dt_obj(dt);
2516         struct inode      *inode = obj->oo_inode;
2517
2518         LINVRNT(osd_invariant(obj));
2519         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2520         LASSERT(osd_write_locked(env, obj));
2521         LASSERT(th != NULL);
2522
2523         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2524
2525         spin_lock(&obj->oo_guard);
2526         LASSERT(inode->i_nlink > 0);
2527         drop_nlink(inode);
2528         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2529          * then the nlink count is 1. Don't let it be set to 0 or the directory
2530          * inode will be deleted incorrectly. */
2531         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2532                 set_nlink(inode, 1);
2533         spin_unlock(&obj->oo_guard);
2534         inode->i_sb->s_op->dirty_inode(inode);
2535         LINVRNT(osd_invariant(obj));
2536
2537         return 0;
2538 }
2539
2540 /*
2541  * Get the 64-bit version for an inode.
2542  */
2543 static int osd_object_version_get(const struct lu_env *env,
2544                                   struct dt_object *dt, dt_obj_version_t *ver)
2545 {
2546         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2547
2548         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2549                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2550         *ver = LDISKFS_I(inode)->i_fs_version;
2551         return 0;
2552 }
2553
2554 /*
2555  * Concurrency: @dt is read locked.
2556  */
2557 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2558                          struct lu_buf *buf, const char *name,
2559                          struct lustre_capa *capa)
2560 {
2561         struct osd_object      *obj    = osd_dt_obj(dt);
2562         struct inode           *inode  = obj->oo_inode;
2563         struct osd_thread_info *info   = osd_oti_get(env);
2564         struct dentry          *dentry = &info->oti_obj_dentry;
2565
2566         /* version get is not real XATTR but uses xattr API */
2567         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2568                 /* for version we are just using xattr API but change inode
2569                  * field instead */
2570                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2571                 osd_object_version_get(env, dt, buf->lb_buf);
2572                 return sizeof(dt_obj_version_t);
2573         }
2574
2575         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2576         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2577
2578         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2579                 return -EACCES;
2580
2581         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2582 }
2583
2584
2585 static int osd_declare_xattr_set(const struct lu_env *env,
2586                                  struct dt_object *dt,
2587                                  const struct lu_buf *buf, const char *name,
2588                                  int fl, struct thandle *handle)
2589 {
2590         struct osd_thandle *oh;
2591
2592         LASSERT(handle != NULL);
2593
2594         oh = container_of0(handle, struct osd_thandle, ot_super);
2595         LASSERT(oh->ot_handle == NULL);
2596
2597         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2598                              strcmp(name, XATTR_NAME_VERSION) == 0 ?
2599                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2600                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2601
2602         return 0;
2603 }
2604
2605 /*
2606  * Set the 64-bit version for object
2607  */
2608 static void osd_object_version_set(const struct lu_env *env,
2609                                    struct dt_object *dt,
2610                                    dt_obj_version_t *new_version)
2611 {
2612         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2613
2614         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2615                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2616
2617         LDISKFS_I(inode)->i_fs_version = *new_version;
2618         /** Version is set after all inode operations are finished,
2619          *  so we should mark it dirty here */
2620         inode->i_sb->s_op->dirty_inode(inode);
2621 }
2622
2623 /*
2624  * Concurrency: @dt is write locked.
2625  */
2626 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2627                          const struct lu_buf *buf, const char *name, int fl,
2628                          struct thandle *handle, struct lustre_capa *capa)
2629 {
2630         struct osd_object      *obj      = osd_dt_obj(dt);
2631         struct inode           *inode    = obj->oo_inode;
2632         struct osd_thread_info *info     = osd_oti_get(env);
2633         int                     fs_flags = 0;
2634
2635         LASSERT(handle != NULL);
2636
2637         /* version set is not real XATTR */
2638         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2639                 /* for version we are just using xattr API but change inode
2640                  * field instead */
2641                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2642                 osd_object_version_set(env, dt, buf->lb_buf);
2643                 return sizeof(dt_obj_version_t);
2644         }
2645
2646         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2647                 return -EACCES;
2648
2649         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2650         if (fl & LU_XATTR_REPLACE)
2651                 fs_flags |= XATTR_REPLACE;
2652
2653         if (fl & LU_XATTR_CREATE)
2654                 fs_flags |= XATTR_CREATE;
2655
2656         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2657                                fs_flags);
2658 }
2659
2660 /*
2661  * Concurrency: @dt is read locked.
2662  */
2663 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2664                           struct lu_buf *buf, struct lustre_capa *capa)
2665 {
2666         struct osd_object      *obj    = osd_dt_obj(dt);
2667         struct inode           *inode  = obj->oo_inode;
2668         struct osd_thread_info *info   = osd_oti_get(env);
2669         struct dentry          *dentry = &info->oti_obj_dentry;
2670
2671         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2672         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2673         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2674
2675         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2676                 return -EACCES;
2677
2678         dentry->d_inode = inode;
2679         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2680 }
2681
2682 static int osd_declare_xattr_del(const struct lu_env *env,
2683                                  struct dt_object *dt, const char *name,
2684                                  struct thandle *handle)
2685 {
2686         struct osd_thandle *oh;
2687
2688         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2689         LASSERT(handle != NULL);
2690
2691         oh = container_of0(handle, struct osd_thandle, ot_super);
2692         LASSERT(oh->ot_handle == NULL);
2693
2694         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2695                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2696
2697         return 0;
2698 }
2699
2700 /*
2701  * Concurrency: @dt is write locked.
2702  */
2703 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2704                          const char *name, struct thandle *handle,
2705                          struct lustre_capa *capa)
2706 {
2707         struct osd_object      *obj    = osd_dt_obj(dt);
2708         struct inode           *inode  = obj->oo_inode;
2709         struct osd_thread_info *info   = osd_oti_get(env);
2710         struct dentry          *dentry = &info->oti_obj_dentry;
2711         int                     rc;
2712
2713         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2714         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2715         LASSERT(handle != NULL);
2716
2717         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2718                 return -EACCES;
2719
2720         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2721
2722         ll_vfs_dq_init(inode);
2723         dentry->d_inode = inode;
2724         rc = inode->i_op->removexattr(dentry, name);
2725         return rc;
2726 }
2727
2728 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2729                                      struct dt_object *dt,
2730                                      struct lustre_capa *old,
2731                                      __u64 opc)
2732 {
2733         struct osd_thread_info *info = osd_oti_get(env);
2734         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2735         struct osd_object *obj = osd_dt_obj(dt);
2736         struct osd_device *dev = osd_obj2dev(obj);
2737         struct lustre_capa_key *key = &info->oti_capa_key;
2738         struct lustre_capa *capa = &info->oti_capa;
2739         struct obd_capa *oc;
2740         struct md_capainfo *ci;
2741         int rc;
2742         ENTRY;
2743
2744         if (!dev->od_fl_capa)
2745                 RETURN(ERR_PTR(-ENOENT));
2746
2747         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2748         LINVRNT(osd_invariant(obj));
2749
2750         /* renewal sanity check */
2751         if (old && osd_object_auth(env, dt, old, opc))
2752                 RETURN(ERR_PTR(-EACCES));
2753
2754         ci = md_capainfo(env);
2755         if (unlikely(!ci))
2756                 RETURN(ERR_PTR(-ENOENT));
2757
2758         switch (ci->mc_auth) {
2759         case LC_ID_NONE:
2760                 RETURN(NULL);
2761         case LC_ID_PLAIN:
2762                 capa->lc_uid = obj->oo_inode->i_uid;
2763                 capa->lc_gid = obj->oo_inode->i_gid;
2764                 capa->lc_flags = LC_ID_PLAIN;
2765                 break;
2766         case LC_ID_CONVERT: {
2767                 __u32 d[4], s[4];
2768
2769                 s[0] = obj->oo_inode->i_uid;
2770                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2771                 s[2] = obj->oo_inode->i_gid;
2772                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2773                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2774                 if (unlikely(rc))
2775                         RETURN(ERR_PTR(rc));
2776
2777                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2778                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2779                 capa->lc_flags = LC_ID_CONVERT;
2780                 break;
2781         }
2782         default:
2783                 RETURN(ERR_PTR(-EINVAL));
2784         }
2785
2786         capa->lc_fid = *fid;
2787         capa->lc_opc = opc;
2788         capa->lc_flags |= dev->od_capa_alg << 24;
2789         capa->lc_timeout = dev->od_capa_timeout;
2790         capa->lc_expiry = 0;
2791
2792         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2793         if (oc) {
2794                 LASSERT(!capa_is_expired(oc));
2795                 RETURN(oc);
2796         }
2797
2798         spin_lock(&capa_lock);
2799         *key = dev->od_capa_keys[1];
2800         spin_unlock(&capa_lock);
2801
2802         capa->lc_keyid = key->lk_keyid;
2803         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2804
2805         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2806         if (rc) {
2807                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2808                 RETURN(ERR_PTR(rc));
2809         }
2810
2811         oc = capa_add(dev->od_capa_hash, capa);
2812         RETURN(oc);
2813 }
2814
2815 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2816 {
2817         struct osd_object       *obj    = osd_dt_obj(dt);
2818         struct inode            *inode  = obj->oo_inode;
2819         struct osd_thread_info  *info   = osd_oti_get(env);
2820         struct dentry           *dentry = &info->oti_obj_dentry;
2821         struct file             *file   = &info->oti_file;
2822         int                     rc;
2823
2824         ENTRY;
2825
2826         dentry->d_inode = inode;
2827         file->f_dentry = dentry;
2828         file->f_mapping = inode->i_mapping;
2829         file->f_op = inode->i_fop;
2830 #ifndef HAVE_FILE_FSYNC_4ARGS
2831         mutex_lock(&inode->i_mutex);
2832 #endif
2833         rc = cfs_do_fsync(file, 0);
2834 #ifndef HAVE_FILE_FSYNC_4ARGS
2835         mutex_unlock(&inode->i_mutex);
2836 #endif
2837         RETURN(rc);
2838 }
2839
2840 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2841                         void **data)
2842 {
2843         struct osd_object *obj = osd_dt_obj(dt);
2844         ENTRY;
2845
2846         *data = (void *)obj->oo_inode;
2847         RETURN(0);
2848 }
2849
2850 /*
2851  * Index operations.
2852  */
2853
2854 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2855                            const struct dt_index_features *feat)
2856 {
2857         struct iam_descr *descr;
2858
2859         if (osd_object_is_root(o))
2860                 return feat == &dt_directory_features;
2861
2862         LASSERT(o->oo_dir != NULL);
2863
2864         descr = o->oo_dir->od_container.ic_descr;
2865         if (feat == &dt_directory_features) {
2866                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2867                         return 1;
2868                 else
2869                         return 0;
2870         } else {
2871                 return
2872                         feat->dif_keysize_min <= descr->id_key_size &&
2873                         descr->id_key_size <= feat->dif_keysize_max &&
2874                         feat->dif_recsize_min <= descr->id_rec_size &&
2875                         descr->id_rec_size <= feat->dif_recsize_max &&
2876                         !(feat->dif_flags & (DT_IND_VARKEY |
2877                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2878                         ergo(feat->dif_flags & DT_IND_UPDATE,
2879                              1 /* XXX check that object (and file system) is
2880                                 * writable */);
2881         }
2882 }
2883
2884 static int osd_iam_container_init(const struct lu_env *env,
2885                                   struct osd_object *obj,
2886                                   struct osd_directory *dir)
2887 {
2888         struct iam_container *bag = &dir->od_container;
2889         int result;
2890
2891         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2892         if (result != 0)
2893                 return result;
2894
2895         result = iam_container_setup(bag);
2896         if (result == 0)
2897                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2898         else
2899                 iam_container_fini(bag);
2900
2901         return result;
2902 }
2903
2904
2905 /*
2906  * Concurrency: no external locking is necessary.
2907  */
2908 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2909                          const struct dt_index_features *feat)
2910 {
2911         int                      result;
2912         int                      skip_iam = 0;
2913         struct osd_object       *obj = osd_dt_obj(dt);
2914
2915         LINVRNT(osd_invariant(obj));
2916
2917         if (osd_object_is_root(obj)) {
2918                 dt->do_index_ops = &osd_index_ea_ops;
2919                 result = 0;
2920         } else if (feat == &dt_directory_features) {
2921                 dt->do_index_ops = &osd_index_ea_ops;
2922                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2923                         result = 0;
2924                 else
2925                         result = -ENOTDIR;
2926                 skip_iam = 1;
2927         } else if (unlikely(feat == &dt_otable_features)) {
2928                 dt->do_index_ops = &osd_otable_ops;
2929                 return 0;
2930         } else if (unlikely(feat == &dt_acct_features)) {
2931                 dt->do_index_ops = &osd_acct_index_ops;
2932                 result = 0;
2933                 skip_iam = 1;
2934         } else if (!osd_has_index(obj)) {
2935                 struct osd_directory *dir;
2936
2937                 OBD_ALLOC_PTR(dir);
2938                 if (dir != NULL) {
2939
2940                         spin_lock(&obj->oo_guard);
2941                         if (obj->oo_dir == NULL)
2942                                 obj->oo_dir = dir;
2943                         else
2944                                 /*
2945                                  * Concurrent thread allocated container data.
2946                                  */
2947                                 OBD_FREE_PTR(dir);
2948                         spin_unlock(&obj->oo_guard);
2949                         /*
2950                          * Now, that we have container data, serialize its
2951                          * initialization.
2952                          */
2953                         down_write(&obj->oo_ext_idx_sem);
2954                         /*
2955                          * recheck under lock.
2956                          */
2957                         if (!osd_has_index(obj))
2958                                 result = osd_iam_container_init(env, obj, dir);
2959                         else
2960                                 result = 0;
2961                         up_write(&obj->oo_ext_idx_sem);
2962                 } else {
2963                         result = -ENOMEM;
2964                 }
2965         } else {
2966                 result = 0;
2967         }
2968
2969         if (result == 0 && skip_iam == 0) {
2970                 if (!osd_iam_index_probe(env, obj, feat))
2971                         result = -ENOTDIR;
2972         }
2973         LINVRNT(osd_invariant(obj));
2974
2975         if (result == 0 && is_quota_glb_feat(feat) &&
2976             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
2977                 result = osd_quota_migration(env, dt, feat);
2978
2979         return result;
2980 }
2981
2982 static int osd_otable_it_attr_get(const struct lu_env *env,
2983                                  struct dt_object *dt,
2984                                  struct lu_attr *attr,
2985                                  struct lustre_capa *capa)
2986 {
2987         attr->la_valid = 0;
2988         return 0;
2989 }
2990
2991 static const struct dt_object_operations osd_obj_ops = {
2992         .do_read_lock         = osd_object_read_lock,
2993         .do_write_lock        = osd_object_write_lock,
2994         .do_read_unlock       = osd_object_read_unlock,
2995         .do_write_unlock      = osd_object_write_unlock,
2996         .do_write_locked      = osd_object_write_locked,
2997         .do_attr_get          = osd_attr_get,
2998         .do_declare_attr_set  = osd_declare_attr_set,
2999         .do_attr_set          = osd_attr_set,
3000         .do_ah_init           = osd_ah_init,
3001         .do_declare_create    = osd_declare_object_create,
3002         .do_create            = osd_object_create,
3003         .do_declare_destroy   = osd_declare_object_destroy,
3004         .do_destroy           = osd_object_destroy,
3005         .do_index_try         = osd_index_try,
3006         .do_declare_ref_add   = osd_declare_object_ref_add,
3007         .do_ref_add           = osd_object_ref_add,
3008         .do_declare_ref_del   = osd_declare_object_ref_del,
3009         .do_ref_del           = osd_object_ref_del,
3010         .do_xattr_get         = osd_xattr_get,
3011         .do_declare_xattr_set = osd_declare_xattr_set,
3012         .do_xattr_set         = osd_xattr_set,
3013         .do_declare_xattr_del = osd_declare_xattr_del,
3014         .do_xattr_del         = osd_xattr_del,
3015         .do_xattr_list        = osd_xattr_list,
3016         .do_capa_get          = osd_capa_get,
3017         .do_object_sync       = osd_object_sync,
3018         .do_data_get          = osd_data_get,
3019 };
3020
3021 /**
3022  * dt_object_operations for interoperability mode
3023  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3024  */
3025 static const struct dt_object_operations osd_obj_ea_ops = {
3026         .do_read_lock         = osd_object_read_lock,
3027         .do_write_lock        = osd_object_write_lock,
3028         .do_read_unlock       = osd_object_read_unlock,
3029         .do_write_unlock      = osd_object_write_unlock,
3030         .do_write_locked      = osd_object_write_locked,
3031         .do_attr_get          = osd_attr_get,
3032         .do_declare_attr_set  = osd_declare_attr_set,
3033         .do_attr_set          = osd_attr_set,
3034         .do_ah_init           = osd_ah_init,
3035         .do_declare_create    = osd_declare_object_create,
3036         .do_create            = osd_object_ea_create,
3037         .do_declare_destroy   = osd_declare_object_destroy,
3038         .do_destroy           = osd_object_destroy,
3039         .do_index_try         = osd_index_try,
3040         .do_declare_ref_add   = osd_declare_object_ref_add,
3041         .do_ref_add           = osd_object_ref_add,
3042         .do_declare_ref_del   = osd_declare_object_ref_del,
3043         .do_ref_del           = osd_object_ref_del,
3044         .do_xattr_get         = osd_xattr_get,
3045         .do_declare_xattr_set = osd_declare_xattr_set,
3046         .do_xattr_set         = osd_xattr_set,
3047         .do_declare_xattr_del = osd_declare_xattr_del,
3048         .do_xattr_del         = osd_xattr_del,
3049         .do_xattr_list        = osd_xattr_list,
3050         .do_capa_get          = osd_capa_get,
3051         .do_object_sync       = osd_object_sync,
3052         .do_data_get          = osd_data_get,
3053 };
3054
3055 static const struct dt_object_operations osd_obj_otable_it_ops = {
3056         .do_attr_get    = osd_otable_it_attr_get,
3057         .do_index_try   = osd_index_try,
3058 };
3059
3060 static int osd_index_declare_iam_delete(const struct lu_env *env,
3061                                         struct dt_object *dt,
3062                                         const struct dt_key *key,
3063                                         struct thandle *handle)
3064 {
3065         struct osd_thandle    *oh;
3066
3067         oh = container_of0(handle, struct osd_thandle, ot_super);
3068         LASSERT(oh->ot_handle == NULL);
3069
3070         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3071                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3072
3073         return 0;
3074 }
3075
3076 /**
3077  *      delete a (key, value) pair from index \a dt specified by \a key
3078  *
3079  *      \param  dt      osd index object
3080  *      \param  key     key for index
3081  *      \param  rec     record reference
3082  *      \param  handle  transaction handler
3083  *
3084  *      \retval  0  success
3085  *      \retval -ve   failure
3086  */
3087
3088 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3089                                 const struct dt_key *key,
3090                                 struct thandle *handle,
3091                                 struct lustre_capa *capa)
3092 {
3093         struct osd_thread_info *oti = osd_oti_get(env);
3094         struct osd_object      *obj = osd_dt_obj(dt);
3095         struct osd_thandle     *oh;
3096         struct iam_path_descr  *ipd;
3097         struct iam_container   *bag = &obj->oo_dir->od_container;
3098         int                     rc;
3099
3100         ENTRY;
3101
3102         LINVRNT(osd_invariant(obj));
3103         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3104         LASSERT(bag->ic_object == obj->oo_inode);
3105         LASSERT(handle != NULL);
3106
3107         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3108                 RETURN(-EACCES);
3109
3110         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3111
3112         ipd = osd_idx_ipd_get(env, bag);
3113         if (unlikely(ipd == NULL))
3114                 RETURN(-ENOMEM);
3115
3116         oh = container_of0(handle, struct osd_thandle, ot_super);
3117         LASSERT(oh->ot_handle != NULL);
3118         LASSERT(oh->ot_handle->h_transaction != NULL);
3119
3120         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3121                 /* swab quota uid/gid provided by caller */
3122                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3123                 key = (const struct dt_key *)&oti->oti_quota_id;
3124         }
3125
3126         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3127         osd_ipd_put(env, bag, ipd);
3128         LINVRNT(osd_invariant(obj));
3129         RETURN(rc);
3130 }
3131
3132 static int osd_index_declare_ea_delete(const struct lu_env *env,
3133                                        struct dt_object *dt,
3134                                        const struct dt_key *key,
3135                                        struct thandle *handle)
3136 {
3137         struct osd_thandle *oh;
3138         struct inode       *inode;
3139         int                 rc;
3140         ENTRY;
3141
3142         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3143         LASSERT(handle != NULL);
3144
3145         oh = container_of0(handle, struct osd_thandle, ot_super);
3146         LASSERT(oh->ot_handle == NULL);
3147
3148         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3149                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3150
3151         inode = osd_dt_obj(dt)->oo_inode;
3152         LASSERT(inode);
3153
3154         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3155                                    true, true, NULL, false);
3156         RETURN(rc);
3157 }
3158
3159 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3160                                           struct dt_rec *fid)
3161 {
3162         struct osd_fid_pack *rec;
3163         int                  rc = -ENODATA;
3164
3165         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3166                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3167                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3168         }
3169         return rc;
3170 }
3171
3172 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3173                           struct lu_fid *fid)
3174 {
3175         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
3176         struct seq_server_site  *ss = osd_seq_site(osd);
3177         int                     rc;
3178         ENTRY;
3179
3180         /* Those FID seqs, which are not in FLDB, must be local seq */
3181         if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
3182                 RETURN(0);
3183
3184         rc = osd_fld_lookup(env, osd, fid, range);
3185         if (rc != 0) {
3186                 CERROR("%s: Can not lookup fld for "DFID"\n",
3187                        osd_name(osd), PFID(fid));
3188                 RETURN(rc);
3189         }
3190
3191         RETURN(ss->ss_node_id != range->lsr_index);
3192 }
3193
3194 /**
3195  * Index delete function for interoperability mode (b11826).
3196  * It will remove the directory entry added by osd_index_ea_insert().
3197  * This entry is needed to maintain name->fid mapping.
3198  *
3199  * \param key,  key i.e. file entry to be deleted
3200  *
3201  * \retval   0, on success
3202  * \retval -ve, on error
3203  */
3204 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3205                                const struct dt_key *key,
3206                                struct thandle *handle,
3207                                struct lustre_capa *capa)
3208 {
3209         struct osd_object          *obj    = osd_dt_obj(dt);
3210         struct inode               *dir    = obj->oo_inode;
3211         struct dentry              *dentry;
3212         struct osd_thandle         *oh;
3213         struct ldiskfs_dir_entry_2 *de = NULL;
3214         struct buffer_head         *bh;
3215         struct htree_lock          *hlock = NULL;
3216         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3217         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3218         int                        rc;
3219         ENTRY;
3220
3221         LINVRNT(osd_invariant(obj));
3222         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3223         LASSERT(handle != NULL);
3224
3225         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3226
3227         oh = container_of(handle, struct osd_thandle, ot_super);
3228         LASSERT(oh->ot_handle != NULL);
3229         LASSERT(oh->ot_handle->h_transaction != NULL);
3230
3231         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3232                 RETURN(-EACCES);
3233
3234         ll_vfs_dq_init(dir);
3235         dentry = osd_child_dentry_get(env, obj,
3236                                       (char *)key, strlen((char *)key));
3237
3238         if (obj->oo_hl_head != NULL) {
3239                 hlock = osd_oti_get(env)->oti_hlock;
3240                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3241                                    dir, LDISKFS_HLOCK_DEL);
3242         } else {
3243                 down_write(&obj->oo_ext_idx_sem);
3244         }
3245
3246         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3247         if (bh) {
3248                 rc = ldiskfs_delete_entry(oh->ot_handle,
3249                                           dir, de, bh);
3250                 brelse(bh);
3251         } else {
3252                 rc = -ENOENT;
3253         }
3254         if (hlock != NULL)
3255                 ldiskfs_htree_unlock(hlock);
3256         else
3257                 up_write(&obj->oo_ext_idx_sem);
3258
3259         if (rc != 0)
3260                 GOTO(out, rc);
3261
3262         /* For inode on the remote MDT, .. will point to
3263          * /Agent directory. So do not try to lookup/delete
3264          * remote inode for .. */
3265         if (strcmp((char *)key, dotdot) == 0)
3266                 GOTO(out, rc = 0);
3267
3268         LASSERT(de != NULL);
3269         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3270         if (rc == 0 && osd_remote_fid(env, osd, fid)) {
3271                 __u32 ino = le32_to_cpu(de->inode);
3272
3273                 rc = osd_delete_remote_inode(env, osd, fid, ino, oh);
3274                 if (rc != 0)
3275                         CERROR("%s: del local inode "DFID": rc = %d\n",
3276                                 osd_name(osd), PFID(fid), rc);
3277         } else {
3278                 if (rc == -ENODATA)
3279                         rc = 0;
3280         }
3281 out:
3282
3283         LASSERT(osd_invariant(obj));
3284         RETURN(rc);
3285 }
3286
3287 /**
3288  *      Lookup index for \a key and copy record to \a rec.
3289  *
3290  *      \param  dt      osd index object
3291  *      \param  key     key for index
3292  *      \param  rec     record reference
3293  *
3294  *      \retval  +ve  success : exact mach
3295  *      \retval  0    return record with key not greater than \a key
3296  *      \retval -ve   failure
3297  */
3298 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3299                                 struct dt_rec *rec, const struct dt_key *key,
3300                                 struct lustre_capa *capa)
3301 {
3302         struct osd_object      *obj = osd_dt_obj(dt);
3303         struct iam_path_descr  *ipd;
3304         struct iam_container   *bag = &obj->oo_dir->od_container;
3305         struct osd_thread_info *oti = osd_oti_get(env);
3306         struct iam_iterator    *it = &oti->oti_idx_it;
3307         struct iam_rec         *iam_rec;
3308         int                     rc;
3309
3310         ENTRY;
3311
3312         LASSERT(osd_invariant(obj));
3313         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3314         LASSERT(bag->ic_object == obj->oo_inode);
3315
3316         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3317                 RETURN(-EACCES);
3318
3319         ipd = osd_idx_ipd_get(env, bag);
3320         if (IS_ERR(ipd))
3321                 RETURN(-ENOMEM);
3322
3323         /* got ipd now we can start iterator. */
3324         iam_it_init(it, bag, 0, ipd);
3325
3326         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3327                 /* swab quota uid/gid provided by caller */
3328                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3329                 key = (const struct dt_key *)&oti->oti_quota_id;
3330         }
3331
3332         rc = iam_it_get(it, (struct iam_key *)key);
3333         if (rc >= 0) {
3334                 if (S_ISDIR(obj->oo_inode->i_mode))
3335                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3336                 else
3337                         iam_rec = (struct iam_rec *) rec;
3338
3339                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3340
3341                 if (S_ISDIR(obj->oo_inode->i_mode))
3342                         osd_fid_unpack((struct lu_fid *) rec,
3343                                        (struct osd_fid_pack *)iam_rec);
3344                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3345                         osd_quota_unpack(obj, rec);
3346         }
3347
3348         iam_it_put(it);
3349         iam_it_fini(it);
3350         osd_ipd_put(env, bag, ipd);
3351
3352         LINVRNT(osd_invariant(obj));
3353
3354         RETURN(rc);
3355 }
3356
3357 static int osd_index_declare_iam_insert(const struct lu_env *env,
3358                                         struct dt_object *dt,
3359                                         const struct dt_rec *rec,
3360                                         const struct dt_key *key,
3361                                         struct thandle *handle)
3362 {
3363         struct osd_thandle *oh;
3364
3365         LASSERT(handle != NULL);
3366
3367         oh = container_of0(handle, struct osd_thandle, ot_super);
3368         LASSERT(oh->ot_handle == NULL);
3369
3370         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3371                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3372
3373         return 0;
3374 }
3375
3376 /**
3377  *      Inserts (key, value) pair in \a dt index object.
3378  *
3379  *      \param  dt      osd index object
3380  *      \param  key     key for index
3381  *      \param  rec     record reference
3382  *      \param  th      transaction handler
3383  *
3384  *      \retval  0  success
3385  *      \retval -ve failure
3386  */
3387 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3388                                 const struct dt_rec *rec,
3389                                 const struct dt_key *key, struct thandle *th,
3390                                 struct lustre_capa *capa, int ignore_quota)
3391 {
3392         struct osd_object     *obj = osd_dt_obj(dt);
3393         struct iam_path_descr *ipd;
3394         struct osd_thandle    *oh;
3395         struct iam_container  *bag = &obj->oo_dir->od_container;
3396         struct osd_thread_info *oti = osd_oti_get(env);
3397         struct iam_rec         *iam_rec;
3398         int                     rc;
3399
3400         ENTRY;
3401
3402         LINVRNT(osd_invariant(obj));
3403         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3404         LASSERT(bag->ic_object == obj->oo_inode);
3405         LASSERT(th != NULL);
3406
3407         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3408                 RETURN(-EACCES);
3409
3410         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3411
3412         ipd = osd_idx_ipd_get(env, bag);
3413         if (unlikely(ipd == NULL))
3414                 RETURN(-ENOMEM);
3415
3416         oh = container_of0(th, struct osd_thandle, ot_super);
3417         LASSERT(oh->ot_handle != NULL);
3418         LASSERT(oh->ot_handle->h_transaction != NULL);
3419         if (S_ISDIR(obj->oo_inode->i_mode)) {
3420                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3421                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3422         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3423                 /* pack quota uid/gid */
3424                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3425                 key = (const struct dt_key *)&oti->oti_quota_id;
3426                 /* pack quota record */
3427                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3428                 iam_rec = (struct iam_rec *)rec;
3429         } else {
3430                 iam_rec = (struct iam_rec *)rec;
3431         }
3432
3433         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3434                         iam_rec, ipd);
3435         osd_ipd_put(env, bag, ipd);
3436         LINVRNT(osd_invariant(obj));
3437         RETURN(rc);
3438 }
3439
3440 /**
3441  * Calls ldiskfs_add_entry() to add directory entry
3442  * into the directory. This is required for
3443  * interoperability mode (b11826)
3444  *
3445  * \retval   0, on success
3446  * \retval -ve, on error
3447  */
3448 static int __osd_ea_add_rec(struct osd_thread_info *info,
3449                             struct osd_object *pobj, struct inode  *cinode,
3450                             const char *name, const struct dt_rec *fid,
3451                             struct htree_lock *hlock, struct thandle *th)
3452 {
3453         struct ldiskfs_dentry_param *ldp;
3454         struct dentry               *child;
3455         struct osd_thandle          *oth;
3456         int                          rc;
3457
3458         oth = container_of(th, struct osd_thandle, ot_super);
3459         LASSERT(oth->ot_handle != NULL);
3460         LASSERT(oth->ot_handle->h_transaction != NULL);
3461         LASSERT(pobj->oo_inode);
3462
3463         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3464         if (unlikely(pobj->oo_inode ==
3465                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3466                 ldp->edp_magic = 0;
3467         else
3468                 osd_get_ldiskfs_dirent_param(ldp, fid);
3469         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3470         child->d_fsdata = (void *)ldp;
3471         ll_vfs_dq_init(pobj->oo_inode);
3472         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3473
3474         RETURN(rc);
3475 }
3476
3477 /**
3478  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3479  * into the directory.Also sets flags into osd object to
3480  * indicate dot and dotdot are created. This is required for
3481  * interoperability mode (b11826)
3482  *
3483  * \param dir   directory for dot and dotdot fixup.
3484  * \param obj   child object for linking
3485  *
3486  * \retval   0, on success
3487  * \retval -ve, on error
3488  */
3489 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3490                               struct osd_object *dir,
3491                               struct inode  *parent_dir, const char *name,
3492                               const struct dt_rec *dot_fid,
3493                               const struct dt_rec *dot_dot_fid,
3494                               struct thandle *th)
3495 {
3496         struct inode                *inode = dir->oo_inode;
3497         struct osd_thandle          *oth;
3498         int result = 0;
3499
3500         oth = container_of(th, struct osd_thandle, ot_super);
3501         LASSERT(oth->ot_handle->h_transaction != NULL);
3502         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3503
3504         if (strcmp(name, dot) == 0) {
3505                 if (dir->oo_compat_dot_created) {
3506                         result = -EEXIST;
3507                 } else {
3508                         LASSERT(inode == parent_dir);
3509                         dir->oo_compat_dot_created = 1;
3510                         result = 0;
3511                 }
3512         } else if (strcmp(name, dotdot) == 0) {
3513                 if (!dir->oo_compat_dot_created)
3514                         return -EINVAL;
3515                 /* in case of rename, dotdot is already created */
3516                 if (dir->oo_compat_dotdot_created) {
3517                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3518                                                 dot_dot_fid, NULL, th);
3519                 }
3520
3521                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3522                                                 parent_dir, dot_fid,
3523                                                 dot_dot_fid, oth);
3524                 if (result == 0)
3525                         dir->oo_compat_dotdot_created = 1;
3526         }
3527
3528         return result;
3529 }
3530
3531
3532 /**
3533  * It will call the appropriate osd_add* function and return the
3534  * value, return by respective functions.
3535  */
3536 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3537                           struct inode *cinode, const char *name,
3538                           const struct dt_rec *fid, struct thandle *th)
3539 {
3540         struct osd_thread_info *info   = osd_oti_get(env);
3541         struct htree_lock      *hlock;
3542         int                     rc;
3543
3544         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3545
3546         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3547                                                    name[2] =='\0'))) {
3548                 if (hlock != NULL) {
3549                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3550                                            pobj->oo_inode, 0);
3551                 } else {
3552                         down_write(&pobj->oo_ext_idx_sem);
3553                 }
3554                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3555                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3556                                         fid, th);
3557         } else {
3558                 if (hlock != NULL) {
3559                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3560                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3561                 } else {
3562                         down_write(&pobj->oo_ext_idx_sem);
3563                 }
3564
3565                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3566                         struct lu_fid *tfid = &info->oti_fid;
3567
3568                         *tfid = *(const struct lu_fid *)fid;
3569                         tfid->f_ver = ~0;
3570                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3571                                               (const struct dt_rec *)tfid,
3572                                               hlock, th);
3573                 } else {
3574                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3575                                               hlock, th);
3576                 }
3577         }
3578         if (hlock != NULL)
3579                 ldiskfs_htree_unlock(hlock);
3580         else
3581                 up_write(&pobj->oo_ext_idx_sem);
3582
3583         return rc;
3584 }
3585
3586 static void
3587 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3588                       struct osd_idmap_cache *oic)
3589 {
3590         struct osd_scrub    *scrub = &dev->od_scrub;
3591         struct lu_fid       *fid   = &oic->oic_fid;
3592         struct osd_inode_id *id    = &oti->oti_id;
3593         int                  once  = 0;
3594         int                  rc;
3595         ENTRY;
3596
3597         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3598                 RETURN_EXIT;
3599
3600 again:
3601         rc = osd_oi_lookup(oti, dev, fid, id, true);
3602         if (rc != 0 && rc != -ENOENT)
3603                 RETURN_EXIT;
3604
3605         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3606                 RETURN_EXIT;
3607
3608         if (thread_is_running(&scrub->os_thread)) {
3609                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3610                 /* There is race condition between osd_oi_lookup and OI scrub.
3611                  * The OI scrub finished just after osd_oi_lookup() failure.
3612                  * Under such case, it is unnecessary to trigger OI scrub again,
3613                  * but try to call osd_oi_lookup() again. */
3614                 if (unlikely(rc == -EAGAIN))
3615                         goto again;
3616
3617                 RETURN_EXIT;
3618         }
3619
3620         if (!dev->od_noscrub && ++once == 1) {
3621                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3622                        PFID(fid));
3623                 rc = osd_scrub_start(dev);
3624                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3625                                ", rc = %d [2]\n",
3626                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3627                                PFID(fid), rc);
3628                 if (rc == 0)
3629                         goto again;
3630         }
3631
3632         EXIT;
3633 }
3634
3635 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3636                                struct osd_device *dev,
3637                                struct osd_idmap_cache *oic,
3638                                struct lu_fid *fid, __u32 ino)
3639 {
3640         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3641         struct inode            *inode;
3642         int                      rc;
3643
3644         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3645         inode = osd_iget(oti, dev, &oic->oic_lid);
3646         if (IS_ERR(inode)) {
3647                 fid_zero(&oic->oic_fid);
3648                 return PTR_ERR(inode);
3649         }
3650
3651         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3652         iput(inode);
3653         if (rc != 0)
3654                 fid_zero(&oic->oic_fid);
3655         else
3656                 *fid = oic->oic_fid = lma->lma_self_fid;
3657         return rc;
3658 }
3659
3660 /**
3661  * Calls ->lookup() to find dentry. From dentry get inode and
3662  * read inode's ea to get fid. This is required for  interoperability
3663  * mode (b11826)
3664  *
3665  * \retval   0, on success
3666  * \retval -ve, on error
3667  */
3668 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3669                              struct dt_rec *rec, const struct dt_key *key)
3670 {
3671         struct inode               *dir    = obj->oo_inode;
3672         struct dentry              *dentry;
3673         struct ldiskfs_dir_entry_2 *de;
3674         struct buffer_head         *bh;
3675         struct lu_fid              *fid = (struct lu_fid *) rec;
3676         struct htree_lock          *hlock = NULL;
3677         int                         ino;
3678         int                         rc;
3679         ENTRY;
3680
3681         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3682
3683         dentry = osd_child_dentry_get(env, obj,
3684                                       (char *)key, strlen((char *)key));
3685
3686         if (obj->oo_hl_head != NULL) {
3687                 hlock = osd_oti_get(env)->oti_hlock;
3688                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3689                                    dir, LDISKFS_HLOCK_LOOKUP);
3690         } else {
3691                 down_read(&obj->oo_ext_idx_sem);
3692         }
3693
3694         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3695         if (bh) {
3696                 struct osd_thread_info *oti = osd_oti_get(env);
3697                 struct osd_inode_id *id = &oti->oti_id;
3698                 struct osd_idmap_cache *oic = &oti->oti_cache;
3699                 struct osd_device *dev = osd_obj2dev(obj);
3700                 struct osd_scrub *scrub = &dev->od_scrub;
3701                 struct scrub_file *sf = &scrub->os_file;
3702
3703                 ino = le32_to_cpu(de->inode);
3704                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
3705                         brelse(bh);
3706                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
3707                         GOTO(out, rc);
3708                 }
3709
3710                 rc = osd_get_fid_from_dentry(de, rec);
3711
3712                 /* done with de, release bh */
3713                 brelse(bh);
3714                 if (rc != 0)
3715                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
3716                 else
3717                         osd_id_gen(id, ino, OSD_OII_NOGEN);
3718                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
3719                         fid_zero(&oic->oic_fid);
3720                         GOTO(out, rc);
3721                 }
3722
3723                 oic->oic_lid = *id;
3724                 oic->oic_fid = *fid;
3725                 if ((scrub->os_pos_current <= ino) &&
3726                     ((sf->sf_flags & SF_INCONSISTENT) ||
3727                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
3728                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3729                                       sf->sf_oi_bitmap)))
3730                         osd_consistency_check(oti, dev, oic);
3731         } else {
3732                 rc = -ENOENT;
3733         }
3734
3735         GOTO(out, rc);
3736
3737 out:
3738         if (hlock != NULL)
3739                 ldiskfs_htree_unlock(hlock);
3740         else
3741                 up_read(&obj->oo_ext_idx_sem);
3742         return rc;
3743 }
3744
3745 /**
3746  * Find the osd object for given fid.
3747  *
3748  * \param fid need to find the osd object having this fid
3749  *
3750  * \retval osd_object on success
3751  * \retval        -ve on error
3752  */
3753 struct osd_object *osd_object_find(const struct lu_env *env,
3754                                    struct dt_object *dt,
3755                                    const struct lu_fid *fid)
3756 {
3757         struct lu_device  *ludev = dt->do_lu.lo_dev;
3758         struct osd_object *child = NULL;
3759         struct lu_object  *luch;
3760         struct lu_object  *lo;
3761
3762         /*
3763          * at this point topdev might not exist yet
3764          * (i.e. MGS is preparing profiles). so we can
3765          * not rely on topdev and instead lookup with
3766          * our device passed as topdev. this can't work
3767          * if the object isn't cached yet (as osd doesn't
3768          * allocate lu_header). IOW, the object must be
3769          * in the cache, otherwise lu_object_alloc() crashes
3770          * -bzzz
3771          */
3772         luch = lu_object_find_at(env, ludev, fid, NULL);
3773         if (!IS_ERR(luch)) {
3774                 if (lu_object_exists(luch)) {
3775                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3776                         if (lo != NULL)
3777                                 child = osd_obj(lo);
3778                         else
3779                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3780                                                 "lu_object can't be located"
3781                                                 DFID"\n", PFID(fid));
3782
3783                         if (child == NULL) {
3784                                 lu_object_put(env, luch);
3785                                 CERROR("Unable to get osd_object\n");
3786                                 child = ERR_PTR(-ENOENT);
3787                         }
3788                 } else {
3789                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3790                                         "lu_object does not exists "DFID"\n",
3791                                         PFID(fid));
3792                         lu_object_put(env, luch);
3793                         child = ERR_PTR(-ENOENT);
3794                 }
3795         } else
3796                 child = (void *)luch;
3797
3798         return child;
3799 }
3800
3801 /**
3802  * Put the osd object once done with it.
3803  *
3804  * \param obj osd object that needs to be put
3805  */
3806 static inline void osd_object_put(const struct lu_env *env,
3807                                   struct osd_object *obj)
3808 {
3809         lu_object_put(env, &obj->oo_dt.do_lu);
3810 }
3811
3812 static int osd_index_declare_ea_insert(const struct lu_env *env,
3813                                        struct dt_object *dt,
3814                                        const struct dt_rec *rec,
3815                                        const struct dt_key *key,
3816                                        struct thandle *handle)
3817 {
3818         struct osd_thandle      *oh;
3819         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
3820         struct lu_fid           *fid = (struct lu_fid *)rec;
3821         int                     rc;
3822         ENTRY;
3823
3824         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3825         LASSERT(handle != NULL);
3826
3827         oh = container_of0(handle, struct osd_thandle, ot_super);
3828         LASSERT(oh->ot_handle == NULL);
3829
3830         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3831                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3832
3833         if (osd_dt_obj(dt)->oo_inode == NULL) {
3834                 const char *name  = (const char *)key;
3835                 /* Object is not being created yet. Only happens when
3836                  *     1. declare directory create
3837                  *     2. declare insert .
3838                  *     3. declare insert ..
3839                  */
3840                 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
3841         } else {
3842                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3843
3844                 /* We ignore block quota on meta pool (MDTs), so needn't
3845                  * calculate how many blocks will be consumed by this index
3846                  * insert */
3847                 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
3848                                            oh, true, true, NULL, false);
3849         }
3850
3851         if (fid == NULL)
3852                 RETURN(0);
3853
3854         rc = osd_remote_fid(env, osd, fid);
3855         if (rc <= 0)
3856                 RETURN(rc);
3857
3858         rc = 0;
3859
3860         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3861                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
3862         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3863                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3864         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3865                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3866
3867         RETURN(rc);
3868 }
3869
3870 /**
3871  * Index add function for interoperability mode (b11826).
3872  * It will add the directory entry.This entry is needed to
3873  * maintain name->fid mapping.
3874  *
3875  * \param key it is key i.e. file entry to be inserted
3876  * \param rec it is value of given key i.e. fid
3877  *
3878  * \retval   0, on success
3879  * \retval -ve, on error
3880  */
3881 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3882                                const struct dt_rec *rec,
3883                                const struct dt_key *key, struct thandle *th,
3884                                struct lustre_capa *capa, int ignore_quota)
3885 {
3886         struct osd_object       *obj = osd_dt_obj(dt);
3887         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3888         struct lu_fid           *fid = (struct lu_fid *) rec;
3889         const char              *name = (const char *)key;
3890         struct osd_thread_info  *oti   = osd_oti_get(env);
3891         struct osd_inode_id     *id    = &oti->oti_id;
3892         struct inode            *child_inode = NULL;
3893         struct osd_object       *child = NULL;
3894         int                     rc;
3895         ENTRY;
3896
3897         LASSERT(osd_invariant(obj));
3898         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3899         LASSERT(th != NULL);
3900
3901         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3902
3903         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3904                 RETURN(-EACCES);
3905
3906         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
3907
3908         rc = osd_remote_fid(env, osd, fid);
3909         if (rc < 0) {
3910                 CERROR("%s: Can not find object "DFID" rc %d\n",
3911                        osd_name(osd), PFID(fid), rc);
3912                 RETURN(rc);
3913         }
3914
3915         if (rc == 1) {
3916                 /* Insert remote entry */
3917                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
3918                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
3919                         struct osd_thandle      *oh;
3920
3921                         /* If parent on remote MDT, we need put this object
3922                          * under AGENT */
3923                         oh = container_of(th, typeof(*oh), ot_super);
3924                         rc = osd_add_to_agent(env, osd, obj, oh);
3925                         if (rc != 0) {
3926                                 CERROR("%s: add agent "DFID" error: rc = %d\n",
3927                                        osd_name(osd),
3928                                        PFID(lu_object_fid(&dt->do_lu)), rc);
3929                                 RETURN(rc);
3930                         }
3931
3932                         child_inode = igrab(omm->omm_agent_dentry->d_inode);
3933                 } else {
3934                         child_inode = osd_create_remote_inode(env, osd, obj,
3935                                                               fid, th);
3936                         if (IS_ERR(child_inode))
3937                                 RETURN(PTR_ERR(child_inode));
3938                 }
3939         } else {
3940                 /* Insert local entry */
3941                 child = osd_object_find(env, dt, fid);
3942                 if (IS_ERR(child)) {
3943                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
3944                                osd_name(osd), PFID(fid),
3945                                id->oii_ino, id->oii_gen,
3946                                (int)PTR_ERR(child_inode));
3947                         RETURN(PTR_ERR(child_inode));
3948                 }
3949                 child_inode = igrab(child->oo_inode);
3950         }
3951
3952         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
3953
3954         iput(child_inode);
3955         if (child != NULL)
3956                 osd_object_put(env, child);
3957         LASSERT(osd_invariant(obj));
3958         RETURN(rc);
3959 }
3960
3961 /**
3962  *  Initialize osd Iterator for given osd index object.
3963  *
3964  *  \param  dt      osd index object
3965  */
3966
3967 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3968                                      struct dt_object *dt,
3969                                      __u32 unused,
3970                                      struct lustre_capa *capa)
3971 {
3972         struct osd_it_iam      *it;
3973         struct osd_thread_info *oti = osd_oti_get(env);
3974         struct osd_object      *obj = osd_dt_obj(dt);
3975         struct lu_object       *lo  = &dt->do_lu;
3976         struct iam_path_descr  *ipd;
3977         struct iam_container   *bag = &obj->oo_dir->od_container;
3978
3979         LASSERT(lu_object_exists(lo));
3980
3981         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3982                 return ERR_PTR(-EACCES);
3983
3984         it = &oti->oti_it;
3985         ipd = osd_it_ipd_get(env, bag);
3986         if (likely(ipd != NULL)) {
3987                 it->oi_obj = obj;
3988                 it->oi_ipd = ipd;
3989                 lu_object_get(lo);
3990                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3991                 return (struct dt_it *)it;
3992         }
3993         return ERR_PTR(-ENOMEM);
3994 }
3995
3996 /**
3997  * free given Iterator.
3998  */
3999
4000 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4001 {
4002         struct osd_it_iam *it = (struct osd_it_iam *)di;
4003         struct osd_object *obj = it->oi_obj;
4004
4005         iam_it_fini(&it->oi_it);
4006         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4007         lu_object_put(env, &obj->oo_dt.do_lu);
4008 }
4009
4010 /**
4011  *  Move Iterator to record specified by \a key
4012  *
4013  *  \param  di      osd iterator
4014  *  \param  key     key for index
4015  *
4016  *  \retval +ve  di points to record with least key not larger than key
4017  *  \retval  0   di points to exact matched key
4018  *  \retval -ve  failure
4019  */
4020
4021 static int osd_it_iam_get(const struct lu_env *env,
4022                           struct dt_it *di, const struct dt_key *key)
4023 {
4024         struct osd_thread_info  *oti = osd_oti_get(env);
4025         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4026
4027         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4028                 /* swab quota uid/gid */
4029                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4030                 key = (struct dt_key *)&oti->oti_quota_id;
4031         }
4032
4033         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4034 }
4035
4036 /**
4037  *  Release Iterator
4038  *
4039  *  \param  di      osd iterator
4040  */
4041 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4042 {
4043         struct osd_it_iam *it = (struct osd_it_iam *)di;
4044
4045         iam_it_put(&it->oi_it);
4046 }
4047
4048 /**
4049  *  Move iterator by one record
4050  *
4051  *  \param  di      osd iterator
4052  *
4053  *  \retval +1   end of container reached
4054  *  \retval  0   success
4055  *  \retval -ve  failure
4056  */
4057
4058 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4059 {
4060         struct osd_it_iam *it = (struct osd_it_iam *)di;
4061
4062         return iam_it_next(&it->oi_it);
4063 }
4064
4065 /**
4066  * Return pointer to the key under iterator.
4067  */
4068
4069 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4070                                  const struct dt_it *di)
4071 {
4072         struct osd_thread_info *oti = osd_oti_get(env);
4073         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4074         struct osd_object      *obj = it->oi_obj;
4075         struct dt_key          *key;
4076
4077         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4078
4079         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4080                 /* swab quota uid/gid */
4081                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4082                 key = (struct dt_key *)&oti->oti_quota_id;
4083         }
4084
4085         return key;
4086 }
4087
4088 /**
4089  * Return size of key under iterator (in bytes)
4090  */
4091
4092 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4093 {
4094         struct osd_it_iam *it = (struct osd_it_iam *)di;
4095
4096         return iam_it_key_size(&it->oi_it);
4097 }
4098
4099 static inline void
4100 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4101 {
4102         /* check if file type is required */
4103         if (ent->lde_attrs & LUDA_TYPE) {
4104                 int align = sizeof(struct luda_type) - 1;
4105                 struct luda_type *lt;
4106
4107                 len = (len + align) & ~align;
4108                 lt = (struct luda_type *)(ent->lde_name + len);
4109                 lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
4110         }
4111
4112         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4113 }
4114
4115 /**
4116  * build lu direct from backend fs dirent.
4117  */
4118
4119 static inline void
4120 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4121                    char *name, __u16 namelen, __u16 type, __u32 attr)
4122 {
4123         ent->lde_attrs = attr | LUDA_FID;
4124         fid_cpu_to_le(&ent->lde_fid, fid);
4125
4126         ent->lde_hash = cpu_to_le64(offset);
4127         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4128
4129         strncpy(ent->lde_name, name, namelen);
4130         ent->lde_name[namelen] = '\0';
4131         ent->lde_namelen = cpu_to_le16(namelen);
4132
4133         /* append lustre attributes */
4134         osd_it_append_attrs(ent, namelen, type);
4135 }
4136
4137 /**
4138  * Return pointer to the record under iterator.
4139  */
4140 static int osd_it_iam_rec(const struct lu_env *env,
4141                           const struct dt_it *di,
4142                           struct dt_rec *dtrec, __u32 attr)
4143 {
4144         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4145         struct osd_thread_info *info = osd_oti_get(env);
4146         ENTRY;
4147
4148         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4149                 const struct osd_fid_pack *rec;
4150                 struct lu_fid             *fid = &info->oti_fid;
4151                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4152                 char                      *name;
4153                 int                        namelen;
4154                 __u64                      hash;
4155                 int                        rc;
4156
4157                 name = (char *)iam_it_key_get(&it->oi_it);
4158                 if (IS_ERR(name))
4159                         RETURN(PTR_ERR(name));
4160
4161                 namelen = iam_it_key_size(&it->oi_it);
4162
4163                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4164                 if (IS_ERR(rec))
4165                         RETURN(PTR_ERR(rec));
4166
4167                 rc = osd_fid_unpack(fid, rec);
4168                 if (rc)
4169                         RETURN(rc);
4170
4171                 hash = iam_it_store(&it->oi_it);
4172
4173                 /* IAM does not store object type in IAM index (dir) */
4174                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4175                                    0, LUDA_FID);
4176         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4177                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4178                            (struct iam_rec *)dtrec);
4179                 osd_quota_unpack(it->oi_obj, dtrec);
4180         } else {
4181                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4182                            (struct iam_rec *)dtrec);
4183         }
4184
4185         RETURN(0);
4186 }
4187
4188 /**
4189  * Returns cookie for current Iterator position.
4190  */
4191 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4192 {
4193         struct osd_it_iam *it = (struct osd_it_iam *)di;
4194
4195         return iam_it_store(&it->oi_it);
4196 }
4197
4198 /**
4199  * Restore iterator from cookie.
4200  *
4201  * \param  di      osd iterator
4202  * \param  hash    Iterator location cookie
4203  *
4204  * \retval +ve  di points to record with least key not larger than key.
4205  * \retval  0   di points to exact matched key
4206  * \retval -ve  failure
4207  */
4208
4209 static int osd_it_iam_load(const struct lu_env *env,
4210                            const struct dt_it *di, __u64 hash)
4211 {
4212         struct osd_it_iam *it = (struct osd_it_iam *)di;
4213
4214         return iam_it_load(&it->oi_it, hash);
4215 }
4216
4217 static const struct dt_index_operations osd_index_iam_ops = {
4218         .dio_lookup         = osd_index_iam_lookup,
4219         .dio_declare_insert = osd_index_declare_iam_insert,
4220         .dio_insert         = osd_index_iam_insert,
4221         .dio_declare_delete = osd_index_declare_iam_delete,
4222         .dio_delete         = osd_index_iam_delete,
4223         .dio_it     = {
4224                 .init     = osd_it_iam_init,
4225                 .fini     = osd_it_iam_fini,
4226                 .get      = osd_it_iam_get,
4227                 .put      = osd_it_iam_put,
4228                 .next     = osd_it_iam_next,
4229                 .key      = osd_it_iam_key,
4230                 .key_size = osd_it_iam_key_size,
4231                 .rec      = osd_it_iam_rec,
4232                 .store    = osd_it_iam_store,
4233                 .load     = osd_it_iam_load
4234         }
4235 };
4236
4237
4238 /**
4239  * Creates or initializes iterator context.
4240  *
4241  * \retval struct osd_it_ea, iterator structure on success
4242  *
4243  */
4244 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4245                                     struct dt_object *dt,
4246                                     __u32 attr,
4247                                     struct lustre_capa *capa)
4248 {
4249         struct osd_object       *obj  = osd_dt_obj(dt);
4250         struct osd_thread_info  *info = osd_oti_get(env);
4251         struct osd_it_ea        *it   = &info->oti_it_ea;
4252         struct lu_object        *lo   = &dt->do_lu;
4253         struct dentry           *obj_dentry = &info->oti_it_dentry;
4254         ENTRY;
4255         LASSERT(lu_object_exists(lo));
4256
4257         obj_dentry->d_inode = obj->oo_inode;
4258         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4259         obj_dentry->d_name.hash = 0;
4260
4261         it->oie_rd_dirent       = 0;
4262         it->oie_it_dirent       = 0;
4263         it->oie_dirent          = NULL;
4264         it->oie_buf             = info->oti_it_ea_buf;
4265         it->oie_obj             = obj;
4266         it->oie_file.f_pos      = 0;
4267         it->oie_file.f_dentry   = obj_dentry;
4268         if (attr & LUDA_64BITHASH)
4269                 it->oie_file.f_mode |= FMODE_64BITHASH;
4270         else
4271                 it->oie_file.f_mode |= FMODE_32BITHASH;
4272         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4273         it->oie_file.f_op         = obj->oo_inode->i_fop;
4274         it->oie_file.private_data = NULL;
4275         lu_object_get(lo);
4276         RETURN((struct dt_it *) it);
4277 }
4278
4279 /**
4280  * Destroy or finishes iterator context.
4281  *
4282  * \param di iterator structure to be destroyed
4283  */
4284 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4285 {
4286         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4287         struct osd_object    *obj  = it->oie_obj;
4288         struct inode       *inode  = obj->oo_inode;
4289
4290         ENTRY;
4291         it->oie_file.f_op->release(inode, &it->oie_file);
4292         lu_object_put(env, &obj->oo_dt.do_lu);
4293         EXIT;
4294 }
4295
4296 /**
4297  * It position the iterator at given key, so that next lookup continues from
4298  * that key Or it is similar to dio_it->load() but based on a key,
4299  * rather than file position.
4300  *
4301  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4302  * to the beginning.
4303  *
4304  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4305  */
4306 static int osd_it_ea_get(const struct lu_env *env,
4307                          struct dt_it *di, const struct dt_key *key)
4308 {
4309         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4310
4311         ENTRY;
4312         LASSERT(((const char *)key)[0] == '\0');
4313         it->oie_file.f_pos      = 0;
4314         it->oie_rd_dirent       = 0;
4315         it->oie_it_dirent       = 0;
4316         it->oie_dirent          = NULL;
4317
4318         RETURN(+1);
4319 }
4320
4321 /**
4322  * Does nothing
4323  */
4324 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4325 {
4326 }
4327
4328 /**
4329  * It is called internally by ->readdir(). It fills the
4330  * iterator's in-memory data structure with required
4331  * information i.e. name, namelen, rec_size etc.
4332  *
4333  * \param buf in which information to be filled in.
4334  * \param name name of the file in given dir
4335  *
4336  * \retval 0 on success
4337  * \retval 1 on buffer full
4338  */
4339 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4340                                loff_t offset, __u64 ino,
4341                                unsigned d_type)
4342 {
4343         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4344         struct osd_object       *obj  = it->oie_obj;
4345         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4346         struct lu_fid           *fid  = &ent->oied_fid;
4347         struct osd_fid_pack     *rec;
4348         ENTRY;
4349
4350         /* this should never happen */
4351         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4352                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4353                 RETURN(-EIO);
4354         }
4355
4356         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4357             OSD_IT_EA_BUFSIZE)
4358                 RETURN(1);
4359
4360         /* "." is just the object itself. */
4361         if (namelen == 1 && name[0] == '.') {
4362                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4363         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4364                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4365                 if (osd_fid_unpack(fid, rec) != 0)
4366                         fid_zero(fid);
4367         } else {
4368                 fid_zero(fid);
4369         }
4370         d_type &= ~LDISKFS_DIRENT_LUFID;
4371
4372         /* NOT export local root. */
4373         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4374                 ino = obj->oo_inode->i_ino;
4375                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4376         }
4377
4378         ent->oied_ino     = ino;
4379         ent->oied_off     = offset;
4380         ent->oied_namelen = namelen;
4381         ent->oied_type    = d_type;
4382
4383         memcpy(ent->oied_name, name, namelen);
4384
4385         it->oie_rd_dirent++;
4386         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4387         RETURN(0);
4388 }
4389
4390 /**
4391  * Calls ->readdir() to load a directory entry at a time
4392  * and stored it in iterator's in-memory data structure.
4393  *
4394  * \param di iterator's in memory structure
4395  *
4396  * \retval   0 on success
4397  * \retval -ve on error
4398  */
4399 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4400                                const struct dt_it *di)
4401 {
4402         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4403         struct osd_object  *obj   = it->oie_obj;
4404         struct inode       *inode = obj->oo_inode;
4405         struct htree_lock  *hlock = NULL;
4406         int                 result = 0;
4407
4408         ENTRY;
4409         it->oie_dirent = it->oie_buf;
4410         it->oie_rd_dirent = 0;
4411
4412         if (obj->oo_hl_head != NULL) {
4413                 hlock = osd_oti_get(env)->oti_hlock;
4414                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4415                                    inode, LDISKFS_HLOCK_READDIR);
4416         } else {
4417                 down_read(&obj->oo_ext_idx_sem);
4418         }
4419
4420         result = inode->i_fop->readdir(&it->oie_file, it,
4421                                        (filldir_t) osd_ldiskfs_filldir);
4422
4423         if (hlock != NULL)
4424                 ldiskfs_htree_unlock(hlock);
4425         else
4426                 up_read(&obj->oo_ext_idx_sem);
4427
4428         if (it->oie_rd_dirent == 0) {
4429                 result = -EIO;
4430         } else {
4431                 it->oie_dirent = it->oie_buf;
4432                 it->oie_it_dirent = 1;
4433         }
4434
4435         RETURN(result);
4436 }
4437
4438 /**
4439  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4440  * to load a directory entry at a time and stored it in
4441  * iterator's in-memory data structure.
4442  *
4443  * \param di iterator's in memory structure
4444  *
4445  * \retval +ve iterator reached to end
4446  * \retval   0 iterator not reached to end
4447  * \retval -ve on error
4448  */
4449 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4450 {
4451         struct osd_it_ea *it = (struct osd_it_ea *)di;
4452         int rc;
4453
4454         ENTRY;
4455
4456         if (it->oie_it_dirent < it->oie_rd_dirent) {
4457                 it->oie_dirent =
4458                         (void *) it->oie_dirent +
4459                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4460                                        it->oie_dirent->oied_namelen);
4461                 it->oie_it_dirent++;
4462                 RETURN(0);
4463         } else {
4464                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4465                         rc = +1;
4466                 else
4467                         rc = osd_ldiskfs_it_fill(env, di);
4468         }
4469
4470         RETURN(rc);
4471 }
4472
4473 /**
4474  * Returns the key at current position from iterator's in memory structure.
4475  *
4476  * \param di iterator's in memory structure
4477  *
4478  * \retval key i.e. struct dt_key on success
4479  */
4480 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4481                                     const struct dt_it *di)
4482 {
4483         struct osd_it_ea *it = (struct osd_it_ea *)di;
4484
4485         return (struct dt_key *)it->oie_dirent->oied_name;
4486 }
4487
4488 /**
4489  * Returns the key's size at current position from iterator's in memory structure.
4490  *
4491  * \param di iterator's in memory structure
4492  *
4493  * \retval key_size i.e. struct dt_key on success
4494  */
4495 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4496 {
4497         struct osd_it_ea *it = (struct osd_it_ea *)di;
4498
4499         return it->oie_dirent->oied_namelen;
4500 }
4501
4502 static int
4503 osd_dirent_update(handle_t *jh, struct super_block *sb,
4504                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4505                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4506 {
4507         struct osd_fid_pack *rec;
4508         int                  rc;
4509         ENTRY;
4510
4511         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4512         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4513
4514         rc = ldiskfs_journal_get_write_access(jh, bh);
4515         if (rc != 0) {
4516                 CERROR("%.16s: fail to write access for update dirent: "
4517                        "name = %.*s, rc = %d\n",
4518                        LDISKFS_SB(sb)->s_es->s_volume_name,
4519                        ent->oied_namelen, ent->oied_name, rc);
4520                 RETURN(rc);
4521         }
4522
4523         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4524         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4525         rc = ldiskfs_journal_dirty_metadata(jh, bh);
4526         if (rc != 0)
4527                 CERROR("%.16s: fail to dirty metadata for update dirent: "
4528                        "name = %.*s, rc = %d\n",
4529                        LDISKFS_SB(sb)->s_es->s_volume_name,
4530                        ent->oied_namelen, ent->oied_name, rc);
4531
4532         RETURN(rc);
4533 }
4534
4535 static inline int
4536 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4537 {
4538         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4539             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4540                 return 1;
4541         else
4542                 return 0;
4543 }
4544
4545 static int
4546 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
4547                     struct inode *dir, struct inode *inode,
4548                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4549                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
4550                     struct htree_lock *hlock)
4551 {
4552         struct dentry               *dentry;
4553         struct osd_fid_pack         *rec;
4554         struct ldiskfs_dentry_param *ldp;
4555         int                          rc;
4556         ENTRY;
4557
4558         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
4559                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
4560                 RETURN(0);
4561
4562         /* There is enough space to hold the FID-in-dirent. */
4563         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
4564                                  dir->i_sb->s_blocksize)) {
4565                 rc = ldiskfs_journal_get_write_access(jh, bh);
4566                 if (rc != 0) {
4567                         CERROR("%.16s: fail to write access for reinsert "
4568                                "dirent: name = %.*s, rc = %d\n",
4569                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4570                                ent->oied_namelen, ent->oied_name, rc);
4571                         RETURN(rc);
4572                 }
4573
4574                 de->name[de->name_len] = 0;
4575                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4576                 rec->fp_len = sizeof(struct lu_fid) + 1;
4577                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4578                 de->file_type |= LDISKFS_DIRENT_LUFID;
4579
4580                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4581                 if (rc != 0)
4582                         CERROR("%.16s: fail to dirty metadata for reinsert "
4583                                "dirent: name = %.*s, rc = %d\n",
4584                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4585                                ent->oied_namelen, ent->oied_name, rc);
4586
4587                 RETURN(rc);
4588         }
4589
4590         rc = ldiskfs_delete_entry(jh, dir, de, bh);
4591         if (rc != 0) {
4592                 CERROR("%.16s: fail to delete entry for reinsert dirent: "
4593                        "name = %.*s, rc = %d\n",
4594                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4595                        ent->oied_namelen, ent->oied_name, rc);
4596                 RETURN(rc);
4597         }
4598
4599         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
4600                                            ent->oied_namelen);
4601         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
4602         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
4603         dentry->d_fsdata = (void *)ldp;
4604         ll_vfs_dq_init(dir);
4605         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
4606         /* It is too bad, we cannot reinsert the name entry back.
4607          * That means we lose it! */
4608         if (rc != 0)
4609                 CERROR("%.16s: fail to insert entry for reinsert dirent: "
4610                        "name = %.*s, rc = %d\n",
4611                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4612                        ent->oied_namelen, ent->oied_name, rc);
4613
4614         RETURN(rc);
4615 }
4616
4617 static int
4618 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
4619                         struct osd_it_ea *it, struct lu_fid *fid,
4620                         struct osd_inode_id *id, __u32 *attr)
4621 {
4622         struct osd_thread_info     *info        = osd_oti_get(env);
4623         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
4624         struct osd_device          *dev         = osd_obj2dev(obj);
4625         struct super_block         *sb          = osd_sb(dev);
4626         const char                 *devname     =
4627                                         LDISKFS_SB(sb)->s_es->s_volume_name;
4628         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
4629         struct inode               *dir         = obj->oo_inode;
4630         struct htree_lock          *hlock       = NULL;
4631         struct buffer_head         *bh          = NULL;
4632         handle_t                   *jh          = NULL;
4633         struct ldiskfs_dir_entry_2 *de;
4634         struct dentry              *dentry;
4635         struct inode               *inode;
4636         int                         credits;
4637         int                         rc;
4638         bool                        dirty       = false;
4639         bool                        is_dotdot   = false;
4640         ENTRY;
4641
4642         if (ent->oied_name[0] == '.') {
4643                 /* Skip dot entry, even if it has stale FID-in-dirent, because
4644                  * we do not use such FID-in-dirent anymore, it is harmless. */
4645                 if (ent->oied_namelen == 1)
4646                         RETURN(0);
4647
4648                 if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
4649                         is_dotdot = true;
4650         }
4651
4652         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
4653                                       ent->oied_namelen);
4654
4655         /* We need to ensure that the name entry is still valid.
4656          * Because it may be removed or renamed by other already.
4657          *
4658          * The unlink or rename operation will start journal before PDO lock,
4659          * so to avoid deadlock, here we need to start journal handle before
4660          * related PDO lock also. But because we do not know whether there
4661          * will be something to be repaired before PDO lock, we just start
4662          * journal without conditions.
4663          *
4664          * We may need to remove the name entry firstly, then insert back.
4665          * One credit is for user quota file update.
4666          * One credit is for group quota file update.
4667          * Two credits are for dirty inode. */
4668         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4669                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
4670
4671 again:
4672         if (dev->od_dirent_journal) {
4673                 jh = ldiskfs_journal_start_sb(sb, credits);
4674                 if (IS_ERR(jh)) {
4675                         rc = PTR_ERR(jh);
4676                         CERROR("%.16s: fail to start trans for dirent "
4677                                "check_repair: credits %d, name %.*s, rc %d\n",
4678                                devname, credits, ent->oied_namelen,
4679                                ent->oied_name, rc);
4680                         RETURN(rc);
4681                 }
4682         }
4683
4684         if (obj->oo_hl_head != NULL) {
4685                 hlock = osd_oti_get(env)->oti_hlock;
4686                 ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
4687                                    LDISKFS_HLOCK_DEL);
4688         } else {
4689                 down_write(&obj->oo_ext_idx_sem);
4690         }
4691
4692         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
4693         /* For dotdot entry, if there is not enough space to hold FID-in-dirent,
4694          * just keep it there. It only happens when the device upgraded from 1.8
4695          * or restored from MDT file-level backup. For the whole directory, only
4696          * dotdot entry has no FID-in-dirent and needs to get FID from LMA when
4697          * readdir, it will not affect the performance much. */
4698         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
4699             (is_dotdot && !osd_dirent_has_space(de->rec_len,
4700                                                 ent->oied_namelen,
4701                                                 sb->s_blocksize))) {
4702                 *attr |= LUDA_IGNORE;
4703                 GOTO(out_journal, rc = 0);
4704         }
4705
4706         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
4707         inode = osd_iget(info, dev, id);
4708         if (IS_ERR(inode)) {
4709                 rc = PTR_ERR(inode);
4710                 if (rc == -ENOENT || rc == -ESTALE) {
4711                         *attr |= LUDA_IGNORE;
4712                         rc = 0;
4713                 }
4714
4715                 GOTO(out_journal, rc);
4716         }
4717
4718         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
4719         if (rc == 0) {
4720                 if (fid_is_sane(fid)) {
4721                         /* FID-in-dirent is valid. */
4722                         if (lu_fid_eq(fid, &lma->lma_self_fid))
4723                                 GOTO(out_inode, rc = 0);
4724
4725                         /* Do not repair under dryrun mode. */
4726                         if (*attr & LUDA_VERIFY_DRYRUN) {
4727                                 *attr |= LUDA_REPAIR;
4728                                 GOTO(out_inode, rc = 0);
4729                         }
4730
4731                         if (!dev->od_dirent_journal) {
4732                                 iput(inode);
4733                                 brelse(bh);
4734                                 if (hlock != NULL)
4735                                         ldiskfs_htree_unlock(hlock);
4736                                 else
4737                                         up_write(&obj->oo_ext_idx_sem);
4738                                 dev->od_dirent_journal = 1;
4739                                 goto again;
4740                         }
4741
4742                         *fid = lma->lma_self_fid;
4743                         dirty = true;
4744                         /* Update the FID-in-dirent. */
4745                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
4746                         if (rc == 0)
4747                                 *attr |= LUDA_REPAIR;
4748                 } else {
4749                         /* Do not repair under dryrun mode. */
4750                         if (*attr & LUDA_VERIFY_DRYRUN) {
4751                                 *attr |= LUDA_REPAIR;
4752                                 GOTO(out_inode, rc = 0);
4753                         }
4754
4755                         if (!dev->od_dirent_journal) {
4756                                 iput(inode);
4757                                 brelse(bh);
4758                                 if (hlock != NULL)
4759                                         ldiskfs_htree_unlock(hlock);
4760                                 else
4761                                         up_write(&obj->oo_ext_idx_sem);
4762                                 dev->od_dirent_journal = 1;
4763                                 goto again;
4764                         }
4765
4766                         *fid = lma->lma_self_fid;
4767                         dirty = true;
4768                         /* Append the FID-in-dirent. */
4769                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4770                                                  fid, bh, de, hlock);
4771                         if (rc == 0)
4772                                 *attr |= LUDA_REPAIR;
4773                 }
4774         } else if (rc == -ENODATA) {
4775                 /* Do not repair under dryrun mode. */
4776                 if (*attr & LUDA_VERIFY_DRYRUN) {
4777                         if (fid_is_sane(fid))
4778                                 *attr |= LUDA_REPAIR;
4779                         else
4780                                 *attr |= LUDA_UPGRADE;
4781                         GOTO(out_inode, rc = 0);
4782                 }
4783
4784                 if (!dev->od_dirent_journal) {
4785                         iput(inode);
4786                         brelse(bh);
4787                         if (hlock != NULL)
4788                                 ldiskfs_htree_unlock(hlock);
4789                         else
4790                                 up_write(&obj->oo_ext_idx_sem);
4791                         dev->od_dirent_journal = 1;
4792                         goto again;
4793                 }
4794
4795                 dirty = true;
4796                 if (unlikely(fid_is_sane(fid))) {
4797                         /* FID-in-dirent exists, but FID-in-LMA is lost.
4798                          * Trust the FID-in-dirent, and add FID-in-LMA. */
4799                         rc = osd_ea_fid_set(info, inode, fid);
4800                         if (rc == 0)
4801                                 *attr |= LUDA_REPAIR;
4802                 } else {
4803                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
4804                         /* It is probably IGIF object. Only aappend the
4805                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
4806                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4807                                                  fid, bh, de, hlock);
4808                         if (rc == 0)
4809                                 *attr |= LUDA_UPGRADE;
4810                 }
4811         }
4812
4813         GOTO(out_inode, rc);
4814
4815 out_inode:
4816         iput(inode);
4817
4818 out_journal:
4819         brelse(bh);
4820         if (hlock != NULL)
4821                 ldiskfs_htree_unlock(hlock);
4822         else
4823                 up_write(&obj->oo_ext_idx_sem);
4824         if (jh != NULL)
4825                 ldiskfs_journal_stop(jh);
4826         if (rc >= 0 && !dirty)
4827                 dev->od_dirent_journal = 0;
4828         return rc;
4829 }
4830
4831 /**
4832  * Returns the value at current position from iterator's in memory structure.
4833  *
4834  * \param di struct osd_it_ea, iterator's in memory structure
4835  * \param attr attr requested for dirent.
4836  * \param lde lustre dirent
4837  *
4838  * \retval   0 no error and \param lde has correct lustre dirent.
4839  * \retval -ve on error
4840  */
4841 static inline int osd_it_ea_rec(const struct lu_env *env,
4842                                 const struct dt_it *di,
4843                                 struct dt_rec *dtrec, __u32 attr)
4844 {
4845         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4846         struct osd_object      *obj   = it->oie_obj;
4847         struct osd_device      *dev   = osd_obj2dev(obj);
4848         struct osd_scrub       *scrub = &dev->od_scrub;
4849         struct scrub_file      *sf    = &scrub->os_file;
4850         struct osd_thread_info *oti   = osd_oti_get(env);
4851         struct osd_inode_id    *id    = &oti->oti_id;
4852         struct osd_idmap_cache *oic   = &oti->oti_cache;
4853         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4854         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4855         __u32                   ino   = it->oie_dirent->oied_ino;
4856         int                     rc    = 0;
4857         ENTRY;
4858
4859         if (attr & LUDA_VERIFY) {
4860                 attr |= LUDA_TYPE;
4861                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
4862                         attr |= LUDA_IGNORE;
4863                         rc = 0;
4864                         goto pack;
4865                 }
4866
4867                 rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr);
4868         } else {
4869                 attr &= ~LU_DIRENT_ATTRS_MASK;
4870                 if (!fid_is_sane(fid)) {
4871                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
4872                                 RETURN(-ENOENT);
4873
4874                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
4875                 } else {
4876                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4877                 }
4878         }
4879
4880         if (rc < 0)
4881                 RETURN(rc);
4882
4883 pack:
4884         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4885                            it->oie_dirent->oied_name,
4886                            it->oie_dirent->oied_namelen,
4887                            it->oie_dirent->oied_type, attr);
4888
4889         if (osd_remote_fid(env, dev, fid))
4890                 RETURN(0);
4891
4892         if (likely(!(attr & LUDA_IGNORE))) {
4893                 oic->oic_lid = *id;
4894                 oic->oic_fid = *fid;
4895         }
4896
4897         if (!(attr & LUDA_VERIFY) &&
4898             (scrub->os_pos_current <= ino) &&
4899             ((sf->sf_flags & SF_INCONSISTENT) ||
4900              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4901              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4902                 osd_consistency_check(oti, dev, oic);
4903
4904         RETURN(rc);
4905 }
4906
4907 /**
4908  * Returns a cookie for current position of the iterator head, so that
4909  * user can use this cookie to load/start the iterator next time.
4910  *
4911  * \param di iterator's in memory structure
4912  *
4913  * \retval cookie for current position, on success
4914  */
4915 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4916 {
4917         struct osd_it_ea *it = (struct osd_it_ea *)di;
4918
4919         return it->oie_dirent->oied_off;
4920 }
4921
4922 /**
4923  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4924  * to load a directory entry at a time and stored it i inn,
4925  * in iterator's in-memory data structure.
4926  *
4927  * \param di struct osd_it_ea, iterator's in memory structure
4928  *
4929  * \retval +ve on success
4930  * \retval -ve on error
4931  */
4932 static int osd_it_ea_load(const struct lu_env *env,
4933                           const struct dt_it *di, __u64 hash)
4934 {
4935         struct osd_it_ea *it = (struct osd_it_ea *)di;
4936         int rc;
4937
4938         ENTRY;
4939         it->oie_file.f_pos = hash;
4940
4941         rc =  osd_ldiskfs_it_fill(env, di);
4942         if (rc == 0)
4943                 rc = +1;
4944
4945         RETURN(rc);
4946 }
4947
4948 /**
4949  * Index lookup function for interoperability mode (b11826).
4950  *
4951  * \param key,  key i.e. file name to be searched
4952  *
4953  * \retval +ve, on success
4954  * \retval -ve, on error
4955  */
4956 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4957                                struct dt_rec *rec, const struct dt_key *key,
4958                                struct lustre_capa *capa)
4959 {
4960         struct osd_object *obj = osd_dt_obj(dt);
4961         int rc = 0;
4962
4963         ENTRY;
4964
4965         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4966         LINVRNT(osd_invariant(obj));
4967
4968         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4969                 return -EACCES;
4970
4971         rc = osd_ea_lookup_rec(env, obj, rec, key);
4972         if (rc == 0)
4973                 rc = +1;
4974         RETURN(rc);
4975 }
4976
4977 /**
4978  * Index and Iterator operations for interoperability
4979  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4980  */
4981 static const struct dt_index_operations osd_index_ea_ops = {
4982         .dio_lookup         = osd_index_ea_lookup,
4983         .dio_declare_insert = osd_index_declare_ea_insert,
4984         .dio_insert         = osd_index_ea_insert,
4985         .dio_declare_delete = osd_index_declare_ea_delete,
4986         .dio_delete         = osd_index_ea_delete,
4987         .dio_it     = {
4988                 .init     = osd_it_ea_init,
4989                 .fini     = osd_it_ea_fini,
4990                 .get      = osd_it_ea_get,
4991                 .put      = osd_it_ea_put,
4992                 .next     = osd_it_ea_next,
4993                 .key      = osd_it_ea_key,
4994                 .key_size = osd_it_ea_key_size,
4995                 .rec      = osd_it_ea_rec,
4996                 .store    = osd_it_ea_store,
4997                 .load     = osd_it_ea_load
4998         }
4999 };
5000
5001 static void *osd_key_init(const struct lu_context *ctx,
5002                           struct lu_context_key *key)
5003 {
5004         struct osd_thread_info *info;
5005
5006         OBD_ALLOC_PTR(info);
5007         if (info == NULL)
5008                 return ERR_PTR(-ENOMEM);
5009
5010         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5011         if (info->oti_it_ea_buf == NULL)
5012                 goto out_free_info;
5013
5014         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5015
5016         info->oti_hlock = ldiskfs_htree_lock_alloc();
5017         if (info->oti_hlock == NULL)
5018                 goto out_free_ea;
5019
5020         return info;
5021
5022  out_free_ea:
5023         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5024  out_free_info:
5025         OBD_FREE_PTR(info);
5026         return ERR_PTR(-ENOMEM);
5027 }
5028
5029 static void osd_key_fini(const struct lu_context *ctx,
5030                          struct lu_context_key *key, void* data)
5031 {
5032         struct osd_thread_info *info = data;
5033
5034         if (info->oti_hlock != NULL)
5035                 ldiskfs_htree_lock_free(info->oti_hlock);
5036         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5037         OBD_FREE_PTR(info);
5038 }
5039
5040 static void osd_key_exit(const struct lu_context *ctx,
5041                          struct lu_context_key *key, void *data)
5042 {
5043         struct osd_thread_info *info = data;
5044
5045         LASSERT(info->oti_r_locks == 0);
5046         LASSERT(info->oti_w_locks == 0);
5047         LASSERT(info->oti_txns    == 0);
5048 }
5049
5050 /* type constructor/destructor: osd_type_init, osd_type_fini */
5051 LU_TYPE_INIT_FINI(osd, &osd_key);
5052
5053 struct lu_context_key osd_key = {
5054         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5055         .lct_init = osd_key_init,
5056         .lct_fini = osd_key_fini,
5057         .lct_exit = osd_key_exit
5058 };
5059
5060
5061 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5062                            const char *name, struct lu_device *next)
5063 {
5064         struct osd_device *osd = osd_dev(d);
5065
5066         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5067             >= sizeof(osd->od_svname))
5068                 return -E2BIG;
5069         return osd_procfs_init(osd, name);
5070 }
5071
5072 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5073 {
5074         ENTRY;
5075
5076         osd_scrub_cleanup(env, o);
5077
5078         if (o->od_fsops) {
5079                 fsfilt_put_ops(o->od_fsops);
5080                 o->od_fsops = NULL;
5081         }
5082
5083         /* shutdown quota slave instance associated with the device */
5084         if (o->od_quota_slave != NULL) {
5085                 qsd_fini(env, o->od_quota_slave);
5086                 o->od_quota_slave = NULL;
5087         }
5088
5089         RETURN(0);
5090 }
5091
5092 static int osd_mount(const struct lu_env *env,
5093                      struct osd_device *o, struct lustre_cfg *cfg)
5094 {
5095         const char              *name  = lustre_cfg_string(cfg, 0);
5096         const char              *dev  = lustre_cfg_string(cfg, 1);
5097         const char              *opts;
5098         unsigned long            page, s_flags, lmd_flags = 0;
5099         struct page             *__page;
5100         struct file_system_type *type;
5101         char                    *options = NULL;
5102         char                    *str;
5103         int                       rc = 0;
5104         ENTRY;
5105
5106         if (o->od_mnt != NULL)
5107                 RETURN(0);
5108
5109         if (strlen(dev) >= sizeof(o->od_mntdev))
5110                 RETURN(-E2BIG);
5111         strcpy(o->od_mntdev, dev);
5112
5113         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
5114         if (o->od_fsops == NULL) {
5115                 CERROR("Can't find fsfilt_ldiskfs\n");
5116                 RETURN(-ENOTSUPP);
5117         }
5118
5119         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
5120         if (__page == NULL)
5121                 GOTO(out, rc = -ENOMEM);
5122
5123         str = lustre_cfg_string(cfg, 2);
5124         s_flags = simple_strtoul(str, NULL, 0);
5125         str = strstr(str, ":");
5126         if (str)
5127                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5128         opts = lustre_cfg_string(cfg, 3);
5129         page = (unsigned long)cfs_page_address(__page);
5130         options = (char *)page;
5131         *options = '\0';
5132         if (opts == NULL)
5133                 strcat(options, "user_xattr,acl");
5134         else
5135                 strcat(options, opts);
5136
5137         /* Glom up mount options */
5138         if (*options != '\0')
5139                 strcat(options, ",");
5140         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
5141
5142         type = get_fs_type("ldiskfs");
5143         if (!type) {
5144                 CERROR("%s: cannot find ldiskfs module\n", name);
5145                 GOTO(out, rc = -ENODEV);
5146         }
5147
5148         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5149         cfs_module_put(type->owner);
5150
5151         if (IS_ERR(o->od_mnt)) {
5152                 rc = PTR_ERR(o->od_mnt);
5153                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5154                 o->od_mnt = NULL;
5155                 GOTO(out, rc);
5156         }
5157
5158 #ifdef HAVE_DEV_SET_RDONLY
5159         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5160                 CERROR("%s: underlying device %s is marked as read-only. "
5161                        "Setup failed\n", name, dev);
5162                 mntput(o->od_mnt);
5163                 o->od_mnt = NULL;
5164                 GOTO(out, rc = -EROFS);
5165         }
5166 #endif
5167
5168         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5169             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5170                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5171                 mntput(o->od_mnt);
5172                 o->od_mnt = NULL;
5173                 GOTO(out, rc = -EINVAL);
5174         }
5175
5176         ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
5177                                 LDISKFS_STATE_LUSTRE_NO_OI);
5178         if (lmd_flags & LMD_FLG_NOSCRUB)
5179                 o->od_noscrub = 1;
5180
5181 out:
5182         if (__page)
5183                 OBD_PAGE_FREE(__page);
5184         if (rc)
5185                 fsfilt_put_ops(o->od_fsops);
5186
5187         RETURN(rc);
5188 }
5189
5190 static struct lu_device *osd_device_fini(const struct lu_env *env,
5191                                          struct lu_device *d)
5192 {
5193         int rc;
5194         ENTRY;
5195
5196         rc = osd_shutdown(env, osd_dev(d));
5197
5198         osd_obj_map_fini(osd_dev(d));
5199
5200         shrink_dcache_sb(osd_sb(osd_dev(d)));
5201         osd_sync(env, lu2dt_dev(d));
5202
5203         rc = osd_procfs_fini(osd_dev(d));
5204         if (rc) {
5205                 CERROR("proc fini error %d \n", rc);
5206                 RETURN (ERR_PTR(rc));
5207         }
5208
5209         if (osd_dev(d)->od_mnt) {
5210                 mntput(osd_dev(d)->od_mnt);
5211                 osd_dev(d)->od_mnt = NULL;
5212         }
5213
5214         RETURN(NULL);
5215 }
5216
5217 static int osd_device_init0(const struct lu_env *env,
5218                             struct osd_device *o,
5219                             struct lustre_cfg *cfg)
5220 {
5221         struct lu_device        *l = osd2lu_dev(o);
5222         struct osd_thread_info *info;
5223         int                     rc;
5224         int                     cplen = 0;
5225
5226         /* if the module was re-loaded, env can loose its keys */
5227         rc = lu_env_refill((struct lu_env *) env);
5228         if (rc)
5229                 GOTO(out, rc);
5230         info = osd_oti_get(env);
5231         LASSERT(info);
5232
5233         l->ld_ops = &osd_lu_ops;
5234         o->od_dt_dev.dd_ops = &osd_dt_ops;
5235
5236         spin_lock_init(&o->od_osfs_lock);
5237         mutex_init(&o->od_otable_mutex);
5238         o->od_osfs_age = cfs_time_shift_64(-1000);
5239
5240         o->od_capa_hash = init_capa_hash();
5241         if (o->od_capa_hash == NULL)
5242                 GOTO(out, rc = -ENOMEM);
5243
5244         o->od_read_cache = 1;
5245         o->od_writethrough_cache = 1;
5246         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5247
5248         rc = osd_mount(env, o, cfg);
5249         if (rc)
5250                 GOTO(out_capa, rc);
5251
5252         CFS_INIT_LIST_HEAD(&o->od_ios_list);
5253         /* setup scrub, including OI files initialization */
5254         rc = osd_scrub_setup(env, o);
5255         if (rc < 0)
5256                 GOTO(out_mnt, rc);
5257
5258         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5259                         sizeof(o->od_svname));
5260         if (cplen >= sizeof(o->od_svname)) {
5261                 rc = -E2BIG;
5262                 GOTO(out_mnt, rc);
5263         }
5264
5265         rc = osd_obj_map_init(o);
5266         if (rc != 0)
5267                 GOTO(out_scrub, rc);
5268
5269         rc = lu_site_init(&o->od_site, l);
5270         if (rc)
5271                 GOTO(out_compat, rc);
5272         o->od_site.ls_bottom_dev = l;
5273
5274         rc = lu_site_init_finish(&o->od_site);
5275         if (rc)
5276                 GOTO(out_site, rc);
5277
5278         rc = osd_procfs_init(o, o->od_svname);
5279         if (rc != 0) {
5280                 CERROR("%s: can't initialize procfs: rc = %d\n",
5281                        o->od_svname, rc);
5282                 GOTO(out_site, rc);
5283         }
5284
5285         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5286
5287         /* initialize quota slave instance */
5288         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5289                                      o->od_proc_entry);
5290         if (IS_ERR(o->od_quota_slave)) {
5291                 rc = PTR_ERR(o->od_quota_slave);
5292                 o->od_quota_slave = NULL;
5293                 GOTO(out_procfs, rc);
5294         }
5295
5296         RETURN(0);
5297 out_procfs:
5298         osd_procfs_fini(o);
5299 out_site:
5300         lu_site_fini(&o->od_site);
5301 out_compat:
5302         osd_obj_map_fini(o);
5303 out_scrub:
5304         osd_scrub_cleanup(env, o);
5305 out_mnt:
5306         osd_oi_fini(info, o);
5307         osd_shutdown(env, o);
5308         mntput(o->od_mnt);
5309         o->od_mnt = NULL;
5310 out_capa:
5311         cleanup_capa_hash(o->od_capa_hash);
5312 out:
5313         RETURN(rc);
5314 }
5315
5316 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5317                                           struct lu_device_type *t,
5318                                           struct lustre_cfg *cfg)
5319 {
5320         struct osd_device *o;
5321         int                rc;
5322
5323         OBD_ALLOC_PTR(o);
5324         if (o == NULL)
5325                 return ERR_PTR(-ENOMEM);
5326
5327         rc = dt_device_init(&o->od_dt_dev, t);
5328         if (rc == 0) {
5329                 /* Because the ctx might be revived in dt_device_init,
5330                  * refill the env here */
5331                 lu_env_refill((struct lu_env *)env);
5332                 rc = osd_device_init0(env, o, cfg);
5333                 if (rc)
5334                         dt_device_fini(&o->od_dt_dev);
5335         }
5336
5337         if (unlikely(rc != 0))
5338                 OBD_FREE_PTR(o);
5339
5340         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5341 }
5342
5343 static struct lu_device *osd_device_free(const struct lu_env *env,
5344                                          struct lu_device *d)
5345 {
5346         struct osd_device *o = osd_dev(d);
5347         ENTRY;
5348
5349         cleanup_capa_hash(o->od_capa_hash);
5350         /* XXX: make osd top device in order to release reference */
5351         d->ld_site->ls_top_dev = d;
5352         lu_site_purge(env, d->ld_site, -1);
5353         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5354                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5355                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5356         }
5357         lu_site_fini(&o->od_site);
5358         dt_device_fini(&o->od_dt_dev);
5359         OBD_FREE_PTR(o);
5360         RETURN(NULL);
5361 }
5362
5363 static int osd_process_config(const struct lu_env *env,
5364                               struct lu_device *d, struct lustre_cfg *cfg)
5365 {
5366         struct osd_device *o = osd_dev(d);
5367         int err;
5368         ENTRY;
5369
5370         switch(cfg->lcfg_command) {
5371         case LCFG_SETUP:
5372                 err = osd_mount(env, o, cfg);
5373                 break;
5374         case LCFG_CLEANUP:
5375                 lu_dev_del_linkage(d->ld_site, d);
5376                 err = osd_shutdown(env, o);
5377                 break;
5378         default:
5379                 err = -ENOSYS;
5380         }
5381
5382         RETURN(err);
5383 }
5384
5385 static int osd_recovery_complete(const struct lu_env *env,
5386                                  struct lu_device *d)
5387 {
5388         struct osd_device       *osd = osd_dev(d);
5389         int                      rc = 0;
5390         ENTRY;
5391
5392         if (osd->od_quota_slave == NULL)
5393                 RETURN(0);
5394
5395         /* start qsd instance on recovery completion, this notifies the quota
5396          * slave code that we are about to process new requests now */
5397         rc = qsd_start(env, osd->od_quota_slave);
5398         RETURN(rc);
5399 }
5400
5401 /*
5402  * we use exports to track all osd users
5403  */
5404 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5405                            struct obd_device *obd, struct obd_uuid *cluuid,
5406                            struct obd_connect_data *data, void *localdata)
5407 {
5408         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
5409         struct lustre_handle  conn;
5410         int                   rc;
5411         ENTRY;
5412
5413         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5414
5415         rc = class_connect(&conn, obd, cluuid);
5416         if (rc)
5417                 RETURN(rc);
5418
5419         *exp = class_conn2export(&conn);
5420
5421         spin_lock(&osd->od_osfs_lock);
5422         osd->od_connects++;
5423         spin_unlock(&osd->od_osfs_lock);
5424
5425         RETURN(0);
5426 }
5427
5428 /*
5429  * once last export (we don't count self-export) disappeared
5430  * osd can be released
5431  */
5432 static int osd_obd_disconnect(struct obd_export *exp)
5433 {
5434         struct obd_device *obd = exp->exp_obd;
5435         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5436         int                rc, release = 0;
5437         ENTRY;
5438
5439         /* Only disconnect the underlying layers on the final disconnect. */
5440         spin_lock(&osd->od_osfs_lock);
5441         osd->od_connects--;
5442         if (osd->od_connects == 0)
5443                 release = 1;
5444         spin_unlock(&osd->od_osfs_lock);
5445
5446         rc = class_disconnect(exp); /* bz 9811 */
5447
5448         if (rc == 0 && release)
5449                 class_manual_cleanup(obd);
5450         RETURN(rc);
5451 }
5452
5453 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5454                        struct lu_device *dev)
5455 {
5456         struct osd_device *osd = osd_dev(dev);
5457         int                result = 0;
5458         ENTRY;
5459
5460         if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
5461                 /* MDT/MDD still use old infrastructure to create
5462                  * special files */
5463                 result = llo_local_objects_setup(env, lu2md_dev(pdev),
5464                                                  lu2dt_dev(dev));
5465                 if (result)
5466                         RETURN(result);
5467         }
5468
5469         if (osd->od_quota_slave != NULL)
5470                 /* set up quota slave objects */
5471                 result = qsd_prepare(env, osd->od_quota_slave);
5472
5473         RETURN(result);
5474 }
5475
5476 static const struct lu_object_operations osd_lu_obj_ops = {
5477         .loo_object_init      = osd_object_init,
5478         .loo_object_delete    = osd_object_delete,
5479         .loo_object_release   = osd_object_release,
5480         .loo_object_free      = osd_object_free,
5481         .loo_object_print     = osd_object_print,
5482         .loo_object_invariant = osd_object_invariant
5483 };
5484
5485 const struct lu_device_operations osd_lu_ops = {
5486         .ldo_object_alloc      = osd_object_alloc,
5487         .ldo_process_config    = osd_process_config,
5488         .ldo_recovery_complete = osd_recovery_complete,
5489         .ldo_prepare           = osd_prepare,
5490 };
5491
5492 static const struct lu_device_type_operations osd_device_type_ops = {
5493         .ldto_init = osd_type_init,
5494         .ldto_fini = osd_type_fini,
5495
5496         .ldto_start = osd_type_start,
5497         .ldto_stop  = osd_type_stop,
5498
5499         .ldto_device_alloc = osd_device_alloc,
5500         .ldto_device_free  = osd_device_free,
5501
5502         .ldto_device_init    = osd_device_init,
5503         .ldto_device_fini    = osd_device_fini
5504 };
5505
5506 struct lu_device_type osd_device_type = {
5507         .ldt_tags     = LU_DEVICE_DT,
5508         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
5509         .ldt_ops      = &osd_device_type_ops,
5510         .ldt_ctx_tags = LCT_LOCAL,
5511 };
5512
5513 /*
5514  * lprocfs legacy support.
5515  */
5516 static struct obd_ops osd_obd_device_ops = {
5517         .o_owner = THIS_MODULE,
5518         .o_connect      = osd_obd_connect,
5519         .o_disconnect   = osd_obd_disconnect
5520 };
5521
5522 static int __init osd_mod_init(void)
5523 {
5524         struct lprocfs_static_vars lvars;
5525
5526         osd_oi_mod_init();
5527         lprocfs_osd_init_vars(&lvars);
5528         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
5529                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5530 }
5531
5532 static void __exit osd_mod_exit(void)
5533 {
5534         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
5535 }
5536
5537 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5538 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
5539 MODULE_LICENSE("GPL");
5540
5541 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);