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                               "%lu/%#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                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
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                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
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         __u32   umask;
1951
1952         /* we drop umask so that permissions we pass are not affected */
1953         umask = current->fs->umask;
1954         current->fs->umask = 0;
1955
1956         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1957                                                   th);
1958         if (result == 0) {
1959                 osd_attr_init(info, obj, attr, dof);
1960                 osd_object_init0(obj);
1961                 /* bz 24037 */
1962                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1963                         unlock_new_inode(obj->oo_inode);
1964         }
1965
1966         /* restore previous umask value */
1967         current->fs->umask = umask;
1968
1969         return result;
1970 }
1971
1972 /**
1973  * Helper function for osd_object_create()
1974  *
1975  * \retval 0, on success
1976  */
1977 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1978                            const struct lu_fid *fid, struct thandle *th)
1979 {
1980         struct osd_thread_info *info = osd_oti_get(env);
1981         struct osd_inode_id    *id   = &info->oti_id;
1982         struct osd_device      *osd  = osd_obj2dev(obj);
1983
1984         LASSERT(obj->oo_inode != NULL);
1985
1986         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1987         return osd_oi_insert(info, osd, fid, id, th);
1988 }
1989
1990 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1991                    const struct lu_fid *fid, struct lu_seq_range *range)
1992 {
1993         struct seq_server_site  *ss = osd_seq_site(osd);
1994         int                     rc;
1995
1996         if (fid_is_idif(fid)) {
1997                 range->lsr_flags = LU_SEQ_RANGE_OST;
1998                 range->lsr_index = fid_idif_ost_idx(fid);
1999                 return 0;
2000         }
2001
2002         if (!fid_seq_in_fldb(fid_seq(fid))) {
2003                 range->lsr_flags = LU_SEQ_RANGE_MDT;
2004                 if (ss != NULL)
2005                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2006                          * at all */
2007                         range->lsr_index = ss->ss_node_id;
2008                 return 0;
2009         }
2010
2011         LASSERT(ss != NULL);
2012         range->lsr_flags = -1;
2013         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2014         if (rc != 0) {
2015                 CERROR("%s can not find "DFID": rc = %d\n",
2016                        osd_name(osd), PFID(fid), rc);
2017         }
2018         return rc;
2019 }
2020
2021 /*
2022  * Concurrency: no external locking is necessary.
2023  */
2024 static int osd_declare_object_create(const struct lu_env *env,
2025                                      struct dt_object *dt,
2026                                      struct lu_attr *attr,
2027                                      struct dt_allocation_hint *hint,
2028                                      struct dt_object_format *dof,
2029                                      struct thandle *handle)
2030 {
2031         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
2032         struct osd_thandle      *oh;
2033         int                      rc;
2034         ENTRY;
2035
2036         LASSERT(handle != NULL);
2037
2038         oh = container_of0(handle, struct osd_thandle, ot_super);
2039         LASSERT(oh->ot_handle == NULL);
2040
2041         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2042                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2043         if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2044                            lu_object_fid(&dt->do_lu)))
2045                 /* Reuse idle OI block may cause additional one OI block
2046                  * to be changed. */
2047                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2048                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2049
2050         /* If this is directory, then we expect . and .. to be inserted as
2051          * well. The one directory block always needs to be created for the
2052          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2053          * block), there is no danger of needing a tree for the first block.
2054          */
2055         if (attr && S_ISDIR(attr->la_mode)) {
2056                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2057                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2058                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2059         }
2060
2061         if (!attr)
2062                 RETURN(0);
2063
2064         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2065                                    false, false, NULL, false);
2066         if (rc != 0)
2067                 RETURN(rc);
2068
2069         /* It does fld look up inside declare, and the result will be
2070          * added to fld cache, so the following fld lookup inside insert
2071          * does not need send RPC anymore, so avoid send rpc with holding
2072          * transaction */
2073         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2074                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2075                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2076                                lu_object_fid(&dt->do_lu), range);
2077
2078
2079         RETURN(rc);
2080 }
2081
2082 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2083                              struct lu_attr *attr,
2084                              struct dt_allocation_hint *hint,
2085                              struct dt_object_format *dof,
2086                              struct thandle *th)
2087 {
2088         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2089         struct osd_object      *obj    = osd_dt_obj(dt);
2090         struct osd_thread_info *info   = osd_oti_get(env);
2091         int result;
2092
2093         ENTRY;
2094
2095         LINVRNT(osd_invariant(obj));
2096         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2097         LASSERT(osd_write_locked(env, obj));
2098         LASSERT(th != NULL);
2099
2100         if (unlikely(fid_is_acct(fid)))
2101                 /* Quota files can't be created from the kernel any more,
2102                  * 'tune2fs -O quota' will take care of creating them */
2103                 RETURN(-EPERM);
2104
2105         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2106         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2107
2108         result = __osd_object_create(info, obj, attr, hint, dof, th);
2109         if (result == 0)
2110                 result = __osd_oi_insert(env, obj, fid, th);
2111
2112         LASSERT(ergo(result == 0,
2113                      dt_object_exists(dt) && !dt_object_remote(dt)));
2114
2115         LASSERT(osd_invariant(obj));
2116         RETURN(result);
2117 }
2118
2119 /**
2120  * Called to destroy on-disk representation of the object
2121  *
2122  * Concurrency: must be locked
2123  */
2124 static int osd_declare_object_destroy(const struct lu_env *env,
2125                                       struct dt_object *dt,
2126                                       struct thandle *th)
2127 {
2128         struct osd_object  *obj = osd_dt_obj(dt);
2129         struct inode       *inode = obj->oo_inode;
2130         struct osd_thandle *oh;
2131         int                 rc;
2132         ENTRY;
2133
2134         oh = container_of0(th, struct osd_thandle, ot_super);
2135         LASSERT(oh->ot_handle == NULL);
2136         LASSERT(inode);
2137
2138         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2139                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2140         /* Recycle idle OI leaf may cause additional three OI blocks
2141          * to be changed. */
2142         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2143                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2144
2145         /* one less inode */
2146         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2147                                    false, true, NULL, false);
2148         if (rc)
2149                 RETURN(rc);
2150         /* data to be truncated */
2151         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2152                                    true, true, NULL, false);
2153         RETURN(rc);
2154 }
2155
2156 static int osd_object_destroy(const struct lu_env *env,
2157                               struct dt_object *dt,
2158                               struct thandle *th)
2159 {
2160         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2161         struct osd_object      *obj = osd_dt_obj(dt);
2162         struct inode           *inode = obj->oo_inode;
2163         struct osd_device      *osd = osd_obj2dev(obj);
2164         struct osd_thandle     *oh;
2165         int                     result;
2166         ENTRY;
2167
2168         oh = container_of0(th, struct osd_thandle, ot_super);
2169         LASSERT(oh->ot_handle);
2170         LASSERT(inode);
2171         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2172
2173         if (unlikely(fid_is_acct(fid)))
2174                 RETURN(-EPERM);
2175
2176         /* Parallel control for OI scrub. For most of cases, there is no
2177          * lock contention. So it will not affect unlink performance. */
2178         mutex_lock(&inode->i_mutex);
2179         if (S_ISDIR(inode->i_mode)) {
2180                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2181                 /* it will check/delete the inode from remote parent,
2182                  * how to optimize it? unlink performance impaction XXX */
2183                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2184                 if (result != 0 && result != -ENOENT) {
2185                         CERROR("%s: delete inode "DFID": rc = %d\n",
2186                                osd_name(osd), PFID(fid), result);
2187                 }
2188                 spin_lock(&obj->oo_guard);
2189                 clear_nlink(inode);
2190                 spin_unlock(&obj->oo_guard);
2191                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2192         }
2193
2194         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2195
2196         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2197         mutex_unlock(&inode->i_mutex);
2198
2199         /* XXX: add to ext3 orphan list */
2200         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2201
2202         /* not needed in the cache anymore */
2203         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2204
2205         RETURN(0);
2206 }
2207
2208 /**
2209  * Put the fid into lustre_mdt_attrs, and then place the structure
2210  * inode's ea. This fid should not be altered during the life time
2211  * of the inode.
2212  *
2213  * \retval +ve, on success
2214  * \retval -ve, on error
2215  *
2216  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2217  */
2218 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2219                    const struct lu_fid *fid, __u64 flags)
2220 {
2221         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2222         int                      rc;
2223
2224         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2225                 return 0;
2226
2227         lustre_lma_init(lma, fid, flags);
2228         lustre_lma_swab(lma);
2229
2230         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2231                              XATTR_CREATE);
2232         /* Someone may created the EA by race. */
2233         if (unlikely(rc == -EEXIST))
2234                 rc = 0;
2235         return rc;
2236 }
2237
2238 /**
2239  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2240  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2241  * To have compatilibility with 1.8 ldiskfs driver we need to have
2242  * magic number at start of fid data.
2243  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2244  * its inmemory API.
2245  */
2246 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2247                                   const struct dt_rec *fid)
2248 {
2249         if (!fid_is_namespace_visible((const struct lu_fid *)fid) ||
2250             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2251                 param->edp_magic = 0;
2252                 return;
2253         }
2254
2255         param->edp_magic = LDISKFS_LUFID_MAGIC;
2256         param->edp_len =  sizeof(struct lu_fid) + 1;
2257         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2258 }
2259
2260 /**
2261  * Try to read the fid from inode ea into dt_rec.
2262  *
2263  * \param fid object fid.
2264  *
2265  * \retval 0 on success
2266  */
2267 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2268                           __u32 ino, struct lu_fid *fid,
2269                           struct osd_inode_id *id)
2270 {
2271         struct osd_thread_info *info  = osd_oti_get(env);
2272         struct inode           *inode;
2273         ENTRY;
2274
2275         osd_id_gen(id, ino, OSD_OII_NOGEN);
2276         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2277         if (IS_ERR(inode))
2278                 RETURN(PTR_ERR(inode));
2279
2280         iput(inode);
2281         RETURN(0);
2282 }
2283
2284 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2285                                         struct inode *dir,
2286                                         struct inode  *parent_dir,
2287                                         const struct dt_rec *dot_fid,
2288                                         const struct dt_rec *dot_dot_fid,
2289                                         struct osd_thandle *oth)
2290 {
2291         struct ldiskfs_dentry_param *dot_ldp;
2292         struct ldiskfs_dentry_param *dot_dot_ldp;
2293
2294         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2295         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2296
2297         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2298         dot_ldp->edp_magic = 0;
2299         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2300                                         dir, dot_ldp, dot_dot_ldp);
2301 }
2302
2303 /**
2304  * Create an local agent inode for remote entry
2305  */
2306 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2307                                                   struct osd_device *osd,
2308                                                   struct osd_object *pobj,
2309                                                   const struct lu_fid *fid,
2310                                                   struct thandle *th)
2311 {
2312         struct osd_thread_info  *info = osd_oti_get(env);
2313         struct inode            *local;
2314         struct osd_thandle      *oh;
2315         int                     rc;
2316         ENTRY;
2317
2318         LASSERT(th);
2319         oh = container_of(th, struct osd_thandle, ot_super);
2320         LASSERT(oh->ot_handle->h_transaction != NULL);
2321
2322         /* FIXME: Insert index api needs to know the mode of
2323          * the remote object. Just use S_IFDIR for now */
2324         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2325         if (IS_ERR(local)) {
2326                 CERROR("%s: create local error %d\n", osd_name(osd),
2327                        (int)PTR_ERR(local));
2328                 RETURN(local);
2329         }
2330
2331         /* Set special LMA flag for local agent inode */
2332         rc = osd_ea_fid_set(info, local, fid, LMAI_AGENT);
2333         if (rc != 0) {
2334                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2335                        osd_name(osd), PFID(fid), rc);
2336                 RETURN(ERR_PTR(rc));
2337         }
2338
2339         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2340                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2341                 (const struct dt_rec *)fid, oh);
2342         if (rc != 0) {
2343                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2344                         osd_name(osd), PFID(fid), rc);
2345                 RETURN(ERR_PTR(rc));
2346         }
2347
2348         RETURN(local);
2349 }
2350
2351 /**
2352  * Delete local agent inode for remote entry
2353  */
2354 static int osd_delete_local_agent_inode(const struct lu_env *env,
2355                                         struct osd_device *osd,
2356                                         const struct lu_fid *fid,
2357                                         __u32 ino, struct osd_thandle *oh)
2358 {
2359         struct osd_thread_info  *oti = osd_oti_get(env);
2360         struct osd_inode_id     *id = &oti->oti_id;
2361         struct inode            *inode;
2362         ENTRY;
2363
2364         id->oii_ino = le32_to_cpu(ino);
2365         id->oii_gen = OSD_OII_NOGEN;
2366         inode = osd_iget(oti, osd, id);
2367         if (IS_ERR(inode)) {
2368                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2369                        PFID(fid), id->oii_ino, id->oii_gen);
2370                 RETURN(PTR_ERR(inode));
2371         }
2372
2373         clear_nlink(inode);
2374         mark_inode_dirty(inode);
2375         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2376                 osd_name(osd), PFID(fid), inode->i_ino);
2377         iput(inode);
2378         RETURN(0);
2379 }
2380
2381 /**
2382  * OSD layer object create function for interoperability mode (b11826).
2383  * This is mostly similar to osd_object_create(). Only difference being, fid is
2384  * inserted into inode ea here.
2385  *
2386  * \retval   0, on success
2387  * \retval -ve, on error
2388  */
2389 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2390                                 struct lu_attr *attr,
2391                                 struct dt_allocation_hint *hint,
2392                                 struct dt_object_format *dof,
2393                                 struct thandle *th)
2394 {
2395         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2396         struct osd_object      *obj    = osd_dt_obj(dt);
2397         struct osd_thread_info *info   = osd_oti_get(env);
2398         int                     result;
2399
2400         ENTRY;
2401
2402         LASSERT(osd_invariant(obj));
2403         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2404         LASSERT(osd_write_locked(env, obj));
2405         LASSERT(th != NULL);
2406
2407         if (unlikely(fid_is_acct(fid)))
2408                 /* Quota files can't be created from the kernel any more,
2409                  * 'tune2fs -O quota' will take care of creating them */
2410                 RETURN(-EPERM);
2411
2412         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2413         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2414
2415         result = __osd_object_create(info, obj, attr, hint, dof, th);
2416         if (result == 0)
2417                 result = osd_ea_fid_set(info, obj->oo_inode, fid, 0);
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         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
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         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
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         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
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         ENTRY;
2635
2636         LASSERT(handle != NULL);
2637
2638         /* version set is not real XATTR */
2639         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2640                 /* for version we are just using xattr API but change inode
2641                  * field instead */
2642                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2643                 osd_object_version_set(env, dt, buf->lb_buf);
2644                 return sizeof(dt_obj_version_t);
2645         }
2646
2647         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2648                 return -EACCES;
2649
2650         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2651         if (fl & LU_XATTR_REPLACE)
2652                 fs_flags |= XATTR_REPLACE;
2653
2654         if (fl & LU_XATTR_CREATE)
2655                 fs_flags |= XATTR_CREATE;
2656
2657         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2658                                fs_flags);
2659 }
2660
2661 /*
2662  * Concurrency: @dt is read locked.
2663  */
2664 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2665                           struct lu_buf *buf, struct lustre_capa *capa)
2666 {
2667         struct osd_object      *obj    = osd_dt_obj(dt);
2668         struct inode           *inode  = obj->oo_inode;
2669         struct osd_thread_info *info   = osd_oti_get(env);
2670         struct dentry          *dentry = &info->oti_obj_dentry;
2671
2672         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2673         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2674         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2675
2676         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2677                 return -EACCES;
2678
2679         dentry->d_inode = inode;
2680         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2681 }
2682
2683 static int osd_declare_xattr_del(const struct lu_env *env,
2684                                  struct dt_object *dt, const char *name,
2685                                  struct thandle *handle)
2686 {
2687         struct osd_thandle *oh;
2688
2689         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2690         LASSERT(handle != NULL);
2691
2692         oh = container_of0(handle, struct osd_thandle, ot_super);
2693         LASSERT(oh->ot_handle == NULL);
2694
2695         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2696                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2697
2698         return 0;
2699 }
2700
2701 /*
2702  * Concurrency: @dt is write locked.
2703  */
2704 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2705                          const char *name, struct thandle *handle,
2706                          struct lustre_capa *capa)
2707 {
2708         struct osd_object      *obj    = osd_dt_obj(dt);
2709         struct inode           *inode  = obj->oo_inode;
2710         struct osd_thread_info *info   = osd_oti_get(env);
2711         struct dentry          *dentry = &info->oti_obj_dentry;
2712         int                     rc;
2713
2714         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2715         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2716         LASSERT(handle != NULL);
2717
2718         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2719                 return -EACCES;
2720
2721         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2722
2723         ll_vfs_dq_init(inode);
2724         dentry->d_inode = inode;
2725         rc = inode->i_op->removexattr(dentry, name);
2726         return rc;
2727 }
2728
2729 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2730                                      struct dt_object *dt,
2731                                      struct lustre_capa *old,
2732                                      __u64 opc)
2733 {
2734         struct osd_thread_info *info = osd_oti_get(env);
2735         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2736         struct osd_object *obj = osd_dt_obj(dt);
2737         struct osd_device *dev = osd_obj2dev(obj);
2738         struct lustre_capa_key *key = &info->oti_capa_key;
2739         struct lustre_capa *capa = &info->oti_capa;
2740         struct obd_capa *oc;
2741         struct md_capainfo *ci;
2742         int rc;
2743         ENTRY;
2744
2745         if (!dev->od_fl_capa)
2746                 RETURN(ERR_PTR(-ENOENT));
2747
2748         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2749         LINVRNT(osd_invariant(obj));
2750
2751         /* renewal sanity check */
2752         if (old && osd_object_auth(env, dt, old, opc))
2753                 RETURN(ERR_PTR(-EACCES));
2754
2755         ci = md_capainfo(env);
2756         if (unlikely(!ci))
2757                 RETURN(ERR_PTR(-ENOENT));
2758
2759         switch (ci->mc_auth) {
2760         case LC_ID_NONE:
2761                 RETURN(NULL);
2762         case LC_ID_PLAIN:
2763                 capa->lc_uid = obj->oo_inode->i_uid;
2764                 capa->lc_gid = obj->oo_inode->i_gid;
2765                 capa->lc_flags = LC_ID_PLAIN;
2766                 break;
2767         case LC_ID_CONVERT: {
2768                 __u32 d[4], s[4];
2769
2770                 s[0] = obj->oo_inode->i_uid;
2771                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2772                 s[2] = obj->oo_inode->i_gid;
2773                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2774                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2775                 if (unlikely(rc))
2776                         RETURN(ERR_PTR(rc));
2777
2778                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2779                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2780                 capa->lc_flags = LC_ID_CONVERT;
2781                 break;
2782         }
2783         default:
2784                 RETURN(ERR_PTR(-EINVAL));
2785         }
2786
2787         capa->lc_fid = *fid;
2788         capa->lc_opc = opc;
2789         capa->lc_flags |= dev->od_capa_alg << 24;
2790         capa->lc_timeout = dev->od_capa_timeout;
2791         capa->lc_expiry = 0;
2792
2793         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2794         if (oc) {
2795                 LASSERT(!capa_is_expired(oc));
2796                 RETURN(oc);
2797         }
2798
2799         spin_lock(&capa_lock);
2800         *key = dev->od_capa_keys[1];
2801         spin_unlock(&capa_lock);
2802
2803         capa->lc_keyid = key->lk_keyid;
2804         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2805
2806         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2807         if (rc) {
2808                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2809                 RETURN(ERR_PTR(rc));
2810         }
2811
2812         oc = capa_add(dev->od_capa_hash, capa);
2813         RETURN(oc);
2814 }
2815
2816 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2817 {
2818         struct osd_object       *obj    = osd_dt_obj(dt);
2819         struct inode            *inode  = obj->oo_inode;
2820         struct osd_thread_info  *info   = osd_oti_get(env);
2821         struct dentry           *dentry = &info->oti_obj_dentry;
2822         struct file             *file   = &info->oti_file;
2823         int                     rc;
2824
2825         ENTRY;
2826
2827         dentry->d_inode = inode;
2828         file->f_dentry = dentry;
2829         file->f_mapping = inode->i_mapping;
2830         file->f_op = inode->i_fop;
2831 #ifndef HAVE_FILE_FSYNC_4ARGS
2832         mutex_lock(&inode->i_mutex);
2833 #endif
2834         rc = do_fsync(file, 0);
2835 #ifndef HAVE_FILE_FSYNC_4ARGS
2836         mutex_unlock(&inode->i_mutex);
2837 #endif
2838         RETURN(rc);
2839 }
2840
2841 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2842                         void **data)
2843 {
2844         struct osd_object *obj = osd_dt_obj(dt);
2845         ENTRY;
2846
2847         *data = (void *)obj->oo_inode;
2848         RETURN(0);
2849 }
2850
2851 /*
2852  * Index operations.
2853  */
2854
2855 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2856                            const struct dt_index_features *feat)
2857 {
2858         struct iam_descr *descr;
2859
2860         if (osd_object_is_root(o))
2861                 return feat == &dt_directory_features;
2862
2863         LASSERT(o->oo_dir != NULL);
2864
2865         descr = o->oo_dir->od_container.ic_descr;
2866         if (feat == &dt_directory_features) {
2867                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2868                         return 1;
2869                 else
2870                         return 0;
2871         } else {
2872                 return
2873                         feat->dif_keysize_min <= descr->id_key_size &&
2874                         descr->id_key_size <= feat->dif_keysize_max &&
2875                         feat->dif_recsize_min <= descr->id_rec_size &&
2876                         descr->id_rec_size <= feat->dif_recsize_max &&
2877                         !(feat->dif_flags & (DT_IND_VARKEY |
2878                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2879                         ergo(feat->dif_flags & DT_IND_UPDATE,
2880                              1 /* XXX check that object (and file system) is
2881                                 * writable */);
2882         }
2883 }
2884
2885 static int osd_iam_container_init(const struct lu_env *env,
2886                                   struct osd_object *obj,
2887                                   struct osd_directory *dir)
2888 {
2889         struct iam_container *bag = &dir->od_container;
2890         int result;
2891
2892         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2893         if (result != 0)
2894                 return result;
2895
2896         result = iam_container_setup(bag);
2897         if (result == 0)
2898                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2899         else
2900                 iam_container_fini(bag);
2901
2902         return result;
2903 }
2904
2905
2906 /*
2907  * Concurrency: no external locking is necessary.
2908  */
2909 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2910                          const struct dt_index_features *feat)
2911 {
2912         int                      result;
2913         int                      skip_iam = 0;
2914         struct osd_object       *obj = osd_dt_obj(dt);
2915
2916         LINVRNT(osd_invariant(obj));
2917
2918         if (osd_object_is_root(obj)) {
2919                 dt->do_index_ops = &osd_index_ea_ops;
2920                 result = 0;
2921         } else if (feat == &dt_directory_features) {
2922                 dt->do_index_ops = &osd_index_ea_ops;
2923                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2924                         result = 0;
2925                 else
2926                         result = -ENOTDIR;
2927                 skip_iam = 1;
2928         } else if (unlikely(feat == &dt_otable_features)) {
2929                 dt->do_index_ops = &osd_otable_ops;
2930                 return 0;
2931         } else if (unlikely(feat == &dt_acct_features)) {
2932                 dt->do_index_ops = &osd_acct_index_ops;
2933                 result = 0;
2934                 skip_iam = 1;
2935         } else if (!osd_has_index(obj)) {
2936                 struct osd_directory *dir;
2937
2938                 OBD_ALLOC_PTR(dir);
2939                 if (dir != NULL) {
2940
2941                         spin_lock(&obj->oo_guard);
2942                         if (obj->oo_dir == NULL)
2943                                 obj->oo_dir = dir;
2944                         else
2945                                 /*
2946                                  * Concurrent thread allocated container data.
2947                                  */
2948                                 OBD_FREE_PTR(dir);
2949                         spin_unlock(&obj->oo_guard);
2950                         /*
2951                          * Now, that we have container data, serialize its
2952                          * initialization.
2953                          */
2954                         down_write(&obj->oo_ext_idx_sem);
2955                         /*
2956                          * recheck under lock.
2957                          */
2958                         if (!osd_has_index(obj))
2959                                 result = osd_iam_container_init(env, obj, dir);
2960                         else
2961                                 result = 0;
2962                         up_write(&obj->oo_ext_idx_sem);
2963                 } else {
2964                         result = -ENOMEM;
2965                 }
2966         } else {
2967                 result = 0;
2968         }
2969
2970         if (result == 0 && skip_iam == 0) {
2971                 if (!osd_iam_index_probe(env, obj, feat))
2972                         result = -ENOTDIR;
2973         }
2974         LINVRNT(osd_invariant(obj));
2975
2976         if (result == 0 && is_quota_glb_feat(feat) &&
2977             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
2978                 result = osd_quota_migration(env, dt, feat);
2979
2980         return result;
2981 }
2982
2983 static int osd_otable_it_attr_get(const struct lu_env *env,
2984                                  struct dt_object *dt,
2985                                  struct lu_attr *attr,
2986                                  struct lustre_capa *capa)
2987 {
2988         attr->la_valid = 0;
2989         return 0;
2990 }
2991
2992 static const struct dt_object_operations osd_obj_ops = {
2993         .do_read_lock         = osd_object_read_lock,
2994         .do_write_lock        = osd_object_write_lock,
2995         .do_read_unlock       = osd_object_read_unlock,
2996         .do_write_unlock      = osd_object_write_unlock,
2997         .do_write_locked      = osd_object_write_locked,
2998         .do_attr_get          = osd_attr_get,
2999         .do_declare_attr_set  = osd_declare_attr_set,
3000         .do_attr_set          = osd_attr_set,
3001         .do_ah_init           = osd_ah_init,
3002         .do_declare_create    = osd_declare_object_create,
3003         .do_create            = osd_object_create,
3004         .do_declare_destroy   = osd_declare_object_destroy,
3005         .do_destroy           = osd_object_destroy,
3006         .do_index_try         = osd_index_try,
3007         .do_declare_ref_add   = osd_declare_object_ref_add,
3008         .do_ref_add           = osd_object_ref_add,
3009         .do_declare_ref_del   = osd_declare_object_ref_del,
3010         .do_ref_del           = osd_object_ref_del,
3011         .do_xattr_get         = osd_xattr_get,
3012         .do_declare_xattr_set = osd_declare_xattr_set,
3013         .do_xattr_set         = osd_xattr_set,
3014         .do_declare_xattr_del = osd_declare_xattr_del,
3015         .do_xattr_del         = osd_xattr_del,
3016         .do_xattr_list        = osd_xattr_list,
3017         .do_capa_get          = osd_capa_get,
3018         .do_object_sync       = osd_object_sync,
3019         .do_data_get          = osd_data_get,
3020 };
3021
3022 /**
3023  * dt_object_operations for interoperability mode
3024  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3025  */
3026 static const struct dt_object_operations osd_obj_ea_ops = {
3027         .do_read_lock         = osd_object_read_lock,
3028         .do_write_lock        = osd_object_write_lock,
3029         .do_read_unlock       = osd_object_read_unlock,
3030         .do_write_unlock      = osd_object_write_unlock,
3031         .do_write_locked      = osd_object_write_locked,
3032         .do_attr_get          = osd_attr_get,
3033         .do_declare_attr_set  = osd_declare_attr_set,
3034         .do_attr_set          = osd_attr_set,
3035         .do_ah_init           = osd_ah_init,
3036         .do_declare_create    = osd_declare_object_create,
3037         .do_create            = osd_object_ea_create,
3038         .do_declare_destroy   = osd_declare_object_destroy,
3039         .do_destroy           = osd_object_destroy,
3040         .do_index_try         = osd_index_try,
3041         .do_declare_ref_add   = osd_declare_object_ref_add,
3042         .do_ref_add           = osd_object_ref_add,
3043         .do_declare_ref_del   = osd_declare_object_ref_del,
3044         .do_ref_del           = osd_object_ref_del,
3045         .do_xattr_get         = osd_xattr_get,
3046         .do_declare_xattr_set = osd_declare_xattr_set,
3047         .do_xattr_set         = osd_xattr_set,
3048         .do_declare_xattr_del = osd_declare_xattr_del,
3049         .do_xattr_del         = osd_xattr_del,
3050         .do_xattr_list        = osd_xattr_list,
3051         .do_capa_get          = osd_capa_get,
3052         .do_object_sync       = osd_object_sync,
3053         .do_data_get          = osd_data_get,
3054 };
3055
3056 static const struct dt_object_operations osd_obj_otable_it_ops = {
3057         .do_attr_get    = osd_otable_it_attr_get,
3058         .do_index_try   = osd_index_try,
3059 };
3060
3061 static int osd_index_declare_iam_delete(const struct lu_env *env,
3062                                         struct dt_object *dt,
3063                                         const struct dt_key *key,
3064                                         struct thandle *handle)
3065 {
3066         struct osd_thandle    *oh;
3067
3068         oh = container_of0(handle, struct osd_thandle, ot_super);
3069         LASSERT(oh->ot_handle == NULL);
3070
3071         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3072                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3073
3074         return 0;
3075 }
3076
3077 /**
3078  *      delete a (key, value) pair from index \a dt specified by \a key
3079  *
3080  *      \param  dt      osd index object
3081  *      \param  key     key for index
3082  *      \param  rec     record reference
3083  *      \param  handle  transaction handler
3084  *
3085  *      \retval  0  success
3086  *      \retval -ve   failure
3087  */
3088
3089 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3090                                 const struct dt_key *key,
3091                                 struct thandle *handle,
3092                                 struct lustre_capa *capa)
3093 {
3094         struct osd_thread_info *oti = osd_oti_get(env);
3095         struct osd_object      *obj = osd_dt_obj(dt);
3096         struct osd_thandle     *oh;
3097         struct iam_path_descr  *ipd;
3098         struct iam_container   *bag = &obj->oo_dir->od_container;
3099         int                     rc;
3100
3101         ENTRY;
3102
3103         LINVRNT(osd_invariant(obj));
3104         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3105         LASSERT(bag->ic_object == obj->oo_inode);
3106         LASSERT(handle != NULL);
3107
3108         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3109                 RETURN(-EACCES);
3110
3111         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3112
3113         ipd = osd_idx_ipd_get(env, bag);
3114         if (unlikely(ipd == NULL))
3115                 RETURN(-ENOMEM);
3116
3117         oh = container_of0(handle, struct osd_thandle, ot_super);
3118         LASSERT(oh->ot_handle != NULL);
3119         LASSERT(oh->ot_handle->h_transaction != NULL);
3120
3121         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3122                 /* swab quota uid/gid provided by caller */
3123                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3124                 key = (const struct dt_key *)&oti->oti_quota_id;
3125         }
3126
3127         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3128         osd_ipd_put(env, bag, ipd);
3129         LINVRNT(osd_invariant(obj));
3130         RETURN(rc);
3131 }
3132
3133 static int osd_index_declare_ea_delete(const struct lu_env *env,
3134                                        struct dt_object *dt,
3135                                        const struct dt_key *key,
3136                                        struct thandle *handle)
3137 {
3138         struct osd_thandle *oh;
3139         struct inode       *inode;
3140         int                 rc;
3141         ENTRY;
3142
3143         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3144         LASSERT(handle != NULL);
3145
3146         oh = container_of0(handle, struct osd_thandle, ot_super);
3147         LASSERT(oh->ot_handle == NULL);
3148
3149         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3150                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3151
3152         inode = osd_dt_obj(dt)->oo_inode;
3153         LASSERT(inode);
3154
3155         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3156                                    true, true, NULL, false);
3157         RETURN(rc);
3158 }
3159
3160 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3161                                           struct dt_rec *fid)
3162 {
3163         struct osd_fid_pack *rec;
3164         int                  rc = -ENODATA;
3165
3166         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3167                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3168                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3169         }
3170         return rc;
3171 }
3172
3173 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3174                           struct lu_fid *fid)
3175 {
3176         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
3177         struct seq_server_site  *ss = osd_seq_site(osd);
3178         int                     rc;
3179         ENTRY;
3180
3181         /* Those FID seqs, which are not in FLDB, must be local seq */
3182         if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
3183                 RETURN(0);
3184
3185         rc = osd_fld_lookup(env, osd, fid, range);
3186         if (rc != 0) {
3187                 CERROR("%s: Can not lookup fld for "DFID"\n",
3188                        osd_name(osd), PFID(fid));
3189                 RETURN(rc);
3190         }
3191
3192         RETURN(ss->ss_node_id != range->lsr_index);
3193 }
3194
3195 /**
3196  * Index delete function for interoperability mode (b11826).
3197  * It will remove the directory entry added by osd_index_ea_insert().
3198  * This entry is needed to maintain name->fid mapping.
3199  *
3200  * \param key,  key i.e. file entry to be deleted
3201  *
3202  * \retval   0, on success
3203  * \retval -ve, on error
3204  */
3205 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3206                                const struct dt_key *key,
3207                                struct thandle *handle,
3208                                struct lustre_capa *capa)
3209 {
3210         struct osd_object          *obj    = osd_dt_obj(dt);
3211         struct inode               *dir    = obj->oo_inode;
3212         struct dentry              *dentry;
3213         struct osd_thandle         *oh;
3214         struct ldiskfs_dir_entry_2 *de = NULL;
3215         struct buffer_head         *bh;
3216         struct htree_lock          *hlock = NULL;
3217         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3218         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3219         int                        rc;
3220         ENTRY;
3221
3222         LINVRNT(osd_invariant(obj));
3223         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3224         LASSERT(handle != NULL);
3225
3226         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3227
3228         oh = container_of(handle, struct osd_thandle, ot_super);
3229         LASSERT(oh->ot_handle != NULL);
3230         LASSERT(oh->ot_handle->h_transaction != NULL);
3231
3232         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3233                 RETURN(-EACCES);
3234
3235         ll_vfs_dq_init(dir);
3236         dentry = osd_child_dentry_get(env, obj,
3237                                       (char *)key, strlen((char *)key));
3238
3239         if (obj->oo_hl_head != NULL) {
3240                 hlock = osd_oti_get(env)->oti_hlock;
3241                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3242                                    dir, LDISKFS_HLOCK_DEL);
3243         } else {
3244                 down_write(&obj->oo_ext_idx_sem);
3245         }
3246
3247         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3248         if (bh) {
3249                 __u32 ino = 0;
3250
3251                 /* Tried to delete local agent inode for the entries,
3252                  * If it tries to delete .., which only happens during
3253                  * rename or move the dir to /ORPHAN, even .. might
3254                  * point to the remote parent, but it do not have
3255                  * local agent inode, so we do not need check .. at all,
3256                  * Note: we need delete entry first, then inode to keep
3257                  * lfsck safe. */
3258                 if (strcmp((char *)key, dotdot) != 0) {
3259                         LASSERT(de != NULL);
3260                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3261                         if (rc == 0 &&
3262                             unlikely(osd_remote_fid(env, osd, fid)))
3263                                 /* Need to delete agent inode */
3264                                 ino = le32_to_cpu(de->inode);
3265                 }
3266                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3267                 brelse(bh);
3268                 if (rc == 0 && unlikely(ino != 0)) {
3269                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3270                                                           oh);
3271                         if (rc != 0)
3272                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3273                                        osd_name(osd), PFID(fid), rc);
3274                 }
3275         } else {
3276                 rc = -ENOENT;
3277         }
3278         if (hlock != NULL)
3279                 ldiskfs_htree_unlock(hlock);
3280         else
3281                 up_write(&obj->oo_ext_idx_sem);
3282
3283         if (rc != 0)
3284                 GOTO(out, rc);
3285
3286         /* For inode on the remote MDT, .. will point to
3287          * /Agent directory, Check whether it needs to delete
3288          * from agent directory */
3289         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3290                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3291                                                    oh);
3292                 if (rc != 0 && rc != -ENOENT) {
3293                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3294                                osd_name(osd), PFID(fid), rc);
3295                 }
3296
3297                 if (rc == -ENOENT)
3298                         rc = 0;
3299
3300                 GOTO(out, rc);
3301         }
3302 out:
3303
3304         LASSERT(osd_invariant(obj));
3305         RETURN(rc);
3306 }
3307
3308 /**
3309  *      Lookup index for \a key and copy record to \a rec.
3310  *
3311  *      \param  dt      osd index object
3312  *      \param  key     key for index
3313  *      \param  rec     record reference
3314  *
3315  *      \retval  +ve  success : exact mach
3316  *      \retval  0    return record with key not greater than \a key
3317  *      \retval -ve   failure
3318  */
3319 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3320                                 struct dt_rec *rec, const struct dt_key *key,
3321                                 struct lustre_capa *capa)
3322 {
3323         struct osd_object      *obj = osd_dt_obj(dt);
3324         struct iam_path_descr  *ipd;
3325         struct iam_container   *bag = &obj->oo_dir->od_container;
3326         struct osd_thread_info *oti = osd_oti_get(env);
3327         struct iam_iterator    *it = &oti->oti_idx_it;
3328         struct iam_rec         *iam_rec;
3329         int                     rc;
3330
3331         ENTRY;
3332
3333         LASSERT(osd_invariant(obj));
3334         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3335         LASSERT(bag->ic_object == obj->oo_inode);
3336
3337         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3338                 RETURN(-EACCES);
3339
3340         ipd = osd_idx_ipd_get(env, bag);
3341         if (IS_ERR(ipd))
3342                 RETURN(-ENOMEM);
3343
3344         /* got ipd now we can start iterator. */
3345         iam_it_init(it, bag, 0, ipd);
3346
3347         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3348                 /* swab quota uid/gid provided by caller */
3349                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3350                 key = (const struct dt_key *)&oti->oti_quota_id;
3351         }
3352
3353         rc = iam_it_get(it, (struct iam_key *)key);
3354         if (rc >= 0) {
3355                 if (S_ISDIR(obj->oo_inode->i_mode))
3356                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3357                 else
3358                         iam_rec = (struct iam_rec *) rec;
3359
3360                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3361
3362                 if (S_ISDIR(obj->oo_inode->i_mode))
3363                         osd_fid_unpack((struct lu_fid *) rec,
3364                                        (struct osd_fid_pack *)iam_rec);
3365                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3366                         osd_quota_unpack(obj, rec);
3367         }
3368
3369         iam_it_put(it);
3370         iam_it_fini(it);
3371         osd_ipd_put(env, bag, ipd);
3372
3373         LINVRNT(osd_invariant(obj));
3374
3375         RETURN(rc);
3376 }
3377
3378 static int osd_index_declare_iam_insert(const struct lu_env *env,
3379                                         struct dt_object *dt,
3380                                         const struct dt_rec *rec,
3381                                         const struct dt_key *key,
3382                                         struct thandle *handle)
3383 {
3384         struct osd_thandle *oh;
3385
3386         LASSERT(handle != NULL);
3387
3388         oh = container_of0(handle, struct osd_thandle, ot_super);
3389         LASSERT(oh->ot_handle == NULL);
3390
3391         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3392                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3393
3394         return 0;
3395 }
3396
3397 /**
3398  *      Inserts (key, value) pair in \a dt index object.
3399  *
3400  *      \param  dt      osd index object
3401  *      \param  key     key for index
3402  *      \param  rec     record reference
3403  *      \param  th      transaction handler
3404  *
3405  *      \retval  0  success
3406  *      \retval -ve failure
3407  */
3408 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3409                                 const struct dt_rec *rec,
3410                                 const struct dt_key *key, struct thandle *th,
3411                                 struct lustre_capa *capa, int ignore_quota)
3412 {
3413         struct osd_object     *obj = osd_dt_obj(dt);
3414         struct iam_path_descr *ipd;
3415         struct osd_thandle    *oh;
3416         struct iam_container  *bag = &obj->oo_dir->od_container;
3417         struct osd_thread_info *oti = osd_oti_get(env);
3418         struct iam_rec         *iam_rec;
3419         int                     rc;
3420
3421         ENTRY;
3422
3423         LINVRNT(osd_invariant(obj));
3424         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3425         LASSERT(bag->ic_object == obj->oo_inode);
3426         LASSERT(th != NULL);
3427
3428         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3429                 RETURN(-EACCES);
3430
3431         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3432
3433         ipd = osd_idx_ipd_get(env, bag);
3434         if (unlikely(ipd == NULL))
3435                 RETURN(-ENOMEM);
3436
3437         oh = container_of0(th, struct osd_thandle, ot_super);
3438         LASSERT(oh->ot_handle != NULL);
3439         LASSERT(oh->ot_handle->h_transaction != NULL);
3440         if (S_ISDIR(obj->oo_inode->i_mode)) {
3441                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3442                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3443         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3444                 /* pack quota uid/gid */
3445                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3446                 key = (const struct dt_key *)&oti->oti_quota_id;
3447                 /* pack quota record */
3448                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3449                 iam_rec = (struct iam_rec *)rec;
3450         } else {
3451                 iam_rec = (struct iam_rec *)rec;
3452         }
3453
3454         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3455                         iam_rec, ipd);
3456         osd_ipd_put(env, bag, ipd);
3457         LINVRNT(osd_invariant(obj));
3458         RETURN(rc);
3459 }
3460
3461 /**
3462  * Calls ldiskfs_add_entry() to add directory entry
3463  * into the directory. This is required for
3464  * interoperability mode (b11826)
3465  *
3466  * \retval   0, on success
3467  * \retval -ve, on error
3468  */
3469 static int __osd_ea_add_rec(struct osd_thread_info *info,
3470                             struct osd_object *pobj, struct inode  *cinode,
3471                             const char *name, const struct dt_rec *fid,
3472                             struct htree_lock *hlock, struct thandle *th)
3473 {
3474         struct ldiskfs_dentry_param *ldp;
3475         struct dentry               *child;
3476         struct osd_thandle          *oth;
3477         int                          rc;
3478
3479         oth = container_of(th, struct osd_thandle, ot_super);
3480         LASSERT(oth->ot_handle != NULL);
3481         LASSERT(oth->ot_handle->h_transaction != NULL);
3482         LASSERT(pobj->oo_inode);
3483
3484         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3485         if (unlikely(pobj->oo_inode ==
3486                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3487                 ldp->edp_magic = 0;
3488         else
3489                 osd_get_ldiskfs_dirent_param(ldp, fid);
3490         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3491         child->d_fsdata = (void *)ldp;
3492         ll_vfs_dq_init(pobj->oo_inode);
3493         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3494
3495         RETURN(rc);
3496 }
3497
3498 /**
3499  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3500  * into the directory.Also sets flags into osd object to
3501  * indicate dot and dotdot are created. This is required for
3502  * interoperability mode (b11826)
3503  *
3504  * \param dir   directory for dot and dotdot fixup.
3505  * \param obj   child object for linking
3506  *
3507  * \retval   0, on success
3508  * \retval -ve, on error
3509  */
3510 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3511                               struct osd_object *dir,
3512                               struct inode  *parent_dir, const char *name,
3513                               const struct dt_rec *dot_fid,
3514                               const struct dt_rec *dot_dot_fid,
3515                               struct thandle *th)
3516 {
3517         struct inode                *inode = dir->oo_inode;
3518         struct osd_thandle          *oth;
3519         int result = 0;
3520
3521         oth = container_of(th, struct osd_thandle, ot_super);
3522         LASSERT(oth->ot_handle->h_transaction != NULL);
3523         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3524
3525         if (strcmp(name, dot) == 0) {
3526                 if (dir->oo_compat_dot_created) {
3527                         result = -EEXIST;
3528                 } else {
3529                         LASSERT(inode == parent_dir);
3530                         dir->oo_compat_dot_created = 1;
3531                         result = 0;
3532                 }
3533         } else if (strcmp(name, dotdot) == 0) {
3534                 if (!dir->oo_compat_dot_created)
3535                         return -EINVAL;
3536                 /* in case of rename, dotdot is already created */
3537                 if (dir->oo_compat_dotdot_created) {
3538                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3539                                                 dot_dot_fid, NULL, th);
3540                 }
3541
3542                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3543                                                 parent_dir, dot_fid,
3544                                                 dot_dot_fid, oth);
3545                 if (result == 0)
3546                         dir->oo_compat_dotdot_created = 1;
3547         }
3548
3549         return result;
3550 }
3551
3552
3553 /**
3554  * It will call the appropriate osd_add* function and return the
3555  * value, return by respective functions.
3556  */
3557 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3558                           struct inode *cinode, const char *name,
3559                           const struct dt_rec *fid, struct thandle *th)
3560 {
3561         struct osd_thread_info *info   = osd_oti_get(env);
3562         struct htree_lock      *hlock;
3563         int                     rc;
3564
3565         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3566
3567         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3568                                                    name[2] =='\0'))) {
3569                 if (hlock != NULL) {
3570                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3571                                            pobj->oo_inode, 0);
3572                 } else {
3573                         down_write(&pobj->oo_ext_idx_sem);
3574                 }
3575                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3576                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3577                                         fid, th);
3578         } else {
3579                 if (hlock != NULL) {
3580                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3581                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3582                 } else {
3583                         down_write(&pobj->oo_ext_idx_sem);
3584                 }
3585
3586                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_INDIR)) {
3587                         struct lu_fid *tfid = &info->oti_fid;
3588
3589                         *tfid = *(const struct lu_fid *)fid;
3590                         tfid->f_ver = ~0;
3591                         rc = __osd_ea_add_rec(info, pobj, cinode, name,
3592                                               (const struct dt_rec *)tfid,
3593                                               hlock, th);
3594                 } else {
3595                         rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3596                                               hlock, th);
3597                 }
3598         }
3599         if (hlock != NULL)
3600                 ldiskfs_htree_unlock(hlock);
3601         else
3602                 up_write(&pobj->oo_ext_idx_sem);
3603
3604         return rc;
3605 }
3606
3607 static void
3608 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3609                       struct osd_idmap_cache *oic)
3610 {
3611         struct osd_scrub    *scrub = &dev->od_scrub;
3612         struct lu_fid       *fid   = &oic->oic_fid;
3613         struct osd_inode_id *id    = &oti->oti_id;
3614         int                  once  = 0;
3615         int                  rc;
3616         ENTRY;
3617
3618         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3619                 RETURN_EXIT;
3620
3621 again:
3622         rc = osd_oi_lookup(oti, dev, fid, id, true);
3623         if (rc != 0 && rc != -ENOENT)
3624                 RETURN_EXIT;
3625
3626         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3627                 RETURN_EXIT;
3628
3629         if (thread_is_running(&scrub->os_thread)) {
3630                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3631                 /* There is race condition between osd_oi_lookup and OI scrub.
3632                  * The OI scrub finished just after osd_oi_lookup() failure.
3633                  * Under such case, it is unnecessary to trigger OI scrub again,
3634                  * but try to call osd_oi_lookup() again. */
3635                 if (unlikely(rc == -EAGAIN))
3636                         goto again;
3637
3638                 RETURN_EXIT;
3639         }
3640
3641         if (!dev->od_noscrub && ++once == 1) {
3642                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3643                        PFID(fid));
3644                 rc = osd_scrub_start(dev);
3645                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3646                                ", rc = %d [2]\n",
3647                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3648                                PFID(fid), rc);
3649                 if (rc == 0)
3650                         goto again;
3651         }
3652
3653         EXIT;
3654 }
3655
3656 static int osd_fail_fid_lookup(struct osd_thread_info *oti,
3657                                struct osd_device *dev,
3658                                struct osd_idmap_cache *oic,
3659                                struct lu_fid *fid, __u32 ino)
3660 {
3661         struct lustre_mdt_attrs *lma   = &oti->oti_mdt_attrs;
3662         struct inode            *inode;
3663         int                      rc;
3664
3665         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3666         inode = osd_iget(oti, dev, &oic->oic_lid);
3667         if (IS_ERR(inode)) {
3668                 fid_zero(&oic->oic_fid);
3669                 return PTR_ERR(inode);
3670         }
3671
3672         rc = osd_get_lma(oti, inode, &oti->oti_obj_dentry, lma);
3673         iput(inode);
3674         if (rc != 0)
3675                 fid_zero(&oic->oic_fid);
3676         else
3677                 *fid = oic->oic_fid = lma->lma_self_fid;
3678         return rc;
3679 }
3680
3681 /**
3682  * Calls ->lookup() to find dentry. From dentry get inode and
3683  * read inode's ea to get fid. This is required for  interoperability
3684  * mode (b11826)
3685  *
3686  * \retval   0, on success
3687  * \retval -ve, on error
3688  */
3689 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3690                              struct dt_rec *rec, const struct dt_key *key)
3691 {
3692         struct inode               *dir    = obj->oo_inode;
3693         struct dentry              *dentry;
3694         struct ldiskfs_dir_entry_2 *de;
3695         struct buffer_head         *bh;
3696         struct lu_fid              *fid = (struct lu_fid *) rec;
3697         struct htree_lock          *hlock = NULL;
3698         int                         ino;
3699         int                         rc;
3700         ENTRY;
3701
3702         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3703
3704         dentry = osd_child_dentry_get(env, obj,
3705                                       (char *)key, strlen((char *)key));
3706
3707         if (obj->oo_hl_head != NULL) {
3708                 hlock = osd_oti_get(env)->oti_hlock;
3709                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3710                                    dir, LDISKFS_HLOCK_LOOKUP);
3711         } else {
3712                 down_read(&obj->oo_ext_idx_sem);
3713         }
3714
3715         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3716         if (bh) {
3717                 struct osd_thread_info *oti = osd_oti_get(env);
3718                 struct osd_inode_id *id = &oti->oti_id;
3719                 struct osd_idmap_cache *oic = &oti->oti_cache;
3720                 struct osd_device *dev = osd_obj2dev(obj);
3721                 struct osd_scrub *scrub = &dev->od_scrub;
3722                 struct scrub_file *sf = &scrub->os_file;
3723
3724                 ino = le32_to_cpu(de->inode);
3725                 if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP)) {
3726                         brelse(bh);
3727                         rc = osd_fail_fid_lookup(oti, dev, oic, fid, ino);
3728                         GOTO(out, rc);
3729                 }
3730
3731                 rc = osd_get_fid_from_dentry(de, rec);
3732
3733                 /* done with de, release bh */
3734                 brelse(bh);
3735                 if (rc != 0)
3736                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
3737                 else
3738                         osd_id_gen(id, ino, OSD_OII_NOGEN);
3739                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
3740                         fid_zero(&oic->oic_fid);
3741                         GOTO(out, rc);
3742                 }
3743
3744                 oic->oic_lid = *id;
3745                 oic->oic_fid = *fid;
3746                 if ((scrub->os_pos_current <= ino) &&
3747                     ((sf->sf_flags & SF_INCONSISTENT) ||
3748                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
3749                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3750                                       sf->sf_oi_bitmap)))
3751                         osd_consistency_check(oti, dev, oic);
3752         } else {
3753                 rc = -ENOENT;
3754         }
3755
3756         GOTO(out, rc);
3757
3758 out:
3759         if (hlock != NULL)
3760                 ldiskfs_htree_unlock(hlock);
3761         else
3762                 up_read(&obj->oo_ext_idx_sem);
3763         return rc;
3764 }
3765
3766 /**
3767  * Find the osd object for given fid.
3768  *
3769  * \param fid need to find the osd object having this fid
3770  *
3771  * \retval osd_object on success
3772  * \retval        -ve on error
3773  */
3774 struct osd_object *osd_object_find(const struct lu_env *env,
3775                                    struct dt_object *dt,
3776                                    const struct lu_fid *fid)
3777 {
3778         struct lu_device  *ludev = dt->do_lu.lo_dev;
3779         struct osd_object *child = NULL;
3780         struct lu_object  *luch;
3781         struct lu_object  *lo;
3782
3783         /*
3784          * at this point topdev might not exist yet
3785          * (i.e. MGS is preparing profiles). so we can
3786          * not rely on topdev and instead lookup with
3787          * our device passed as topdev. this can't work
3788          * if the object isn't cached yet (as osd doesn't
3789          * allocate lu_header). IOW, the object must be
3790          * in the cache, otherwise lu_object_alloc() crashes
3791          * -bzzz
3792          */
3793         luch = lu_object_find_at(env, ludev, fid, NULL);
3794         if (!IS_ERR(luch)) {
3795                 if (lu_object_exists(luch)) {
3796                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3797                         if (lo != NULL)
3798                                 child = osd_obj(lo);
3799                         else
3800                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3801                                                 "lu_object can't be located"
3802                                                 DFID"\n", PFID(fid));
3803
3804                         if (child == NULL) {
3805                                 lu_object_put(env, luch);
3806                                 CERROR("Unable to get osd_object\n");
3807                                 child = ERR_PTR(-ENOENT);
3808                         }
3809                 } else {
3810                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3811                                         "lu_object does not exists "DFID"\n",
3812                                         PFID(fid));
3813                         lu_object_put(env, luch);
3814                         child = ERR_PTR(-ENOENT);
3815                 }
3816         } else
3817                 child = (void *)luch;
3818
3819         return child;
3820 }
3821
3822 /**
3823  * Put the osd object once done with it.
3824  *
3825  * \param obj osd object that needs to be put
3826  */
3827 static inline void osd_object_put(const struct lu_env *env,
3828                                   struct osd_object *obj)
3829 {
3830         lu_object_put(env, &obj->oo_dt.do_lu);
3831 }
3832
3833 static int osd_index_declare_ea_insert(const struct lu_env *env,
3834                                        struct dt_object *dt,
3835                                        const struct dt_rec *rec,
3836                                        const struct dt_key *key,
3837                                        struct thandle *handle)
3838 {
3839         struct osd_thandle      *oh;
3840         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
3841         struct lu_fid           *fid = (struct lu_fid *)rec;
3842         int                     rc;
3843         ENTRY;
3844
3845         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3846         LASSERT(handle != NULL);
3847
3848         oh = container_of0(handle, struct osd_thandle, ot_super);
3849         LASSERT(oh->ot_handle == NULL);
3850
3851         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3852                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3853
3854         if (osd_dt_obj(dt)->oo_inode == NULL) {
3855                 const char *name  = (const char *)key;
3856                 /* Object is not being created yet. Only happens when
3857                  *     1. declare directory create
3858                  *     2. declare insert .
3859                  *     3. declare insert ..
3860                  */
3861                 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
3862         } else {
3863                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3864
3865                 /* We ignore block quota on meta pool (MDTs), so needn't
3866                  * calculate how many blocks will be consumed by this index
3867                  * insert */
3868                 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
3869                                            oh, true, true, NULL, false);
3870         }
3871
3872         if (fid == NULL)
3873                 RETURN(0);
3874
3875         rc = osd_remote_fid(env, osd, fid);
3876         if (rc <= 0)
3877                 RETURN(rc);
3878
3879         rc = 0;
3880
3881         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3882                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
3883         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3884                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3885         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3886                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3887
3888         RETURN(rc);
3889 }
3890
3891 /**
3892  * Index add function for interoperability mode (b11826).
3893  * It will add the directory entry.This entry is needed to
3894  * maintain name->fid mapping.
3895  *
3896  * \param key it is key i.e. file entry to be inserted
3897  * \param rec it is value of given key i.e. fid
3898  *
3899  * \retval   0, on success
3900  * \retval -ve, on error
3901  */
3902 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3903                                const struct dt_rec *rec,
3904                                const struct dt_key *key, struct thandle *th,
3905                                struct lustre_capa *capa, int ignore_quota)
3906 {
3907         struct osd_object       *obj = osd_dt_obj(dt);
3908         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3909         struct lu_fid           *fid = (struct lu_fid *) rec;
3910         const char              *name = (const char *)key;
3911         struct osd_thread_info  *oti   = osd_oti_get(env);
3912         struct osd_inode_id     *id    = &oti->oti_id;
3913         struct inode            *child_inode = NULL;
3914         struct osd_object       *child = NULL;
3915         int                     rc;
3916         ENTRY;
3917
3918         LASSERT(osd_invariant(obj));
3919         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3920         LASSERT(th != NULL);
3921
3922         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3923
3924         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3925                 RETURN(-EACCES);
3926
3927         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
3928
3929         rc = osd_remote_fid(env, osd, fid);
3930         if (rc < 0) {
3931                 CERROR("%s: Can not find object "DFID" rc %d\n",
3932                        osd_name(osd), PFID(fid), rc);
3933                 RETURN(rc);
3934         }
3935
3936         if (rc == 1) {
3937                 /* Insert remote entry */
3938                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
3939                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
3940                         struct osd_thandle      *oh;
3941
3942                         /* If parent on remote MDT, we need put this object
3943                          * under AGENT */
3944                         oh = container_of(th, typeof(*oh), ot_super);
3945                         rc = osd_add_to_remote_parent(env, osd, obj, oh);
3946                         if (rc != 0) {
3947                                 CERROR("%s: add "DFID" error: rc = %d\n",
3948                                        osd_name(osd),
3949                                        PFID(lu_object_fid(&dt->do_lu)), rc);
3950                                 RETURN(rc);
3951                         }
3952
3953                         child_inode = igrab(omm->omm_remote_parent->d_inode);
3954                 } else {
3955                         child_inode = osd_create_local_agent_inode(env, osd,
3956                                                                    obj, fid,
3957                                                                    th);
3958                         if (IS_ERR(child_inode))
3959                                 RETURN(PTR_ERR(child_inode));
3960                 }
3961         } else {
3962                 /* Insert local entry */
3963                 child = osd_object_find(env, dt, fid);
3964                 if (IS_ERR(child)) {
3965                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
3966                                osd_name(osd), PFID(fid),
3967                                id->oii_ino, id->oii_gen,
3968                                (int)PTR_ERR(child_inode));
3969                         RETURN(PTR_ERR(child_inode));
3970                 }
3971                 child_inode = igrab(child->oo_inode);
3972         }
3973
3974         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
3975
3976         iput(child_inode);
3977         if (child != NULL)
3978                 osd_object_put(env, child);
3979         LASSERT(osd_invariant(obj));
3980         RETURN(rc);
3981 }
3982
3983 /**
3984  *  Initialize osd Iterator for given osd index object.
3985  *
3986  *  \param  dt      osd index object
3987  */
3988
3989 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3990                                      struct dt_object *dt,
3991                                      __u32 unused,
3992                                      struct lustre_capa *capa)
3993 {
3994         struct osd_it_iam      *it;
3995         struct osd_thread_info *oti = osd_oti_get(env);
3996         struct osd_object      *obj = osd_dt_obj(dt);
3997         struct lu_object       *lo  = &dt->do_lu;
3998         struct iam_path_descr  *ipd;
3999         struct iam_container   *bag = &obj->oo_dir->od_container;
4000
4001         LASSERT(lu_object_exists(lo));
4002
4003         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
4004                 return ERR_PTR(-EACCES);
4005
4006         it = &oti->oti_it;
4007         ipd = osd_it_ipd_get(env, bag);
4008         if (likely(ipd != NULL)) {
4009                 it->oi_obj = obj;
4010                 it->oi_ipd = ipd;
4011                 lu_object_get(lo);
4012                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
4013                 return (struct dt_it *)it;
4014         }
4015         return ERR_PTR(-ENOMEM);
4016 }
4017
4018 /**
4019  * free given Iterator.
4020  */
4021
4022 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
4023 {
4024         struct osd_it_iam *it = (struct osd_it_iam *)di;
4025         struct osd_object *obj = it->oi_obj;
4026
4027         iam_it_fini(&it->oi_it);
4028         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
4029         lu_object_put(env, &obj->oo_dt.do_lu);
4030 }
4031
4032 /**
4033  *  Move Iterator to record specified by \a key
4034  *
4035  *  \param  di      osd iterator
4036  *  \param  key     key for index
4037  *
4038  *  \retval +ve  di points to record with least key not larger than key
4039  *  \retval  0   di points to exact matched key
4040  *  \retval -ve  failure
4041  */
4042
4043 static int osd_it_iam_get(const struct lu_env *env,
4044                           struct dt_it *di, const struct dt_key *key)
4045 {
4046         struct osd_thread_info  *oti = osd_oti_get(env);
4047         struct osd_it_iam       *it = (struct osd_it_iam *)di;
4048
4049         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4050                 /* swab quota uid/gid */
4051                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
4052                 key = (struct dt_key *)&oti->oti_quota_id;
4053         }
4054
4055         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
4056 }
4057
4058 /**
4059  *  Release Iterator
4060  *
4061  *  \param  di      osd iterator
4062  */
4063 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
4064 {
4065         struct osd_it_iam *it = (struct osd_it_iam *)di;
4066
4067         iam_it_put(&it->oi_it);
4068 }
4069
4070 /**
4071  *  Move iterator by one record
4072  *
4073  *  \param  di      osd iterator
4074  *
4075  *  \retval +1   end of container reached
4076  *  \retval  0   success
4077  *  \retval -ve  failure
4078  */
4079
4080 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
4081 {
4082         struct osd_it_iam *it = (struct osd_it_iam *)di;
4083
4084         return iam_it_next(&it->oi_it);
4085 }
4086
4087 /**
4088  * Return pointer to the key under iterator.
4089  */
4090
4091 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4092                                  const struct dt_it *di)
4093 {
4094         struct osd_thread_info *oti = osd_oti_get(env);
4095         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4096         struct osd_object      *obj = it->oi_obj;
4097         struct dt_key          *key;
4098
4099         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4100
4101         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4102                 /* swab quota uid/gid */
4103                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4104                 key = (struct dt_key *)&oti->oti_quota_id;
4105         }
4106
4107         return key;
4108 }
4109
4110 /**
4111  * Return size of key under iterator (in bytes)
4112  */
4113
4114 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4115 {
4116         struct osd_it_iam *it = (struct osd_it_iam *)di;
4117
4118         return iam_it_key_size(&it->oi_it);
4119 }
4120
4121 static inline void
4122 osd_it_append_attrs(struct lu_dirent *ent, int len, __u16 type)
4123 {
4124         /* check if file type is required */
4125         if (ent->lde_attrs & LUDA_TYPE) {
4126                 struct luda_type *lt;
4127                 int align = sizeof(*lt) - 1;
4128
4129                 len = (len + align) & ~align;
4130                 lt = (struct luda_type *)(ent->lde_name + len);
4131                 lt->lt_type = cpu_to_le16(DTTOIF(type));
4132         }
4133
4134         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4135 }
4136
4137 /**
4138  * build lu direct from backend fs dirent.
4139  */
4140
4141 static inline void
4142 osd_it_pack_dirent(struct lu_dirent *ent, struct lu_fid *fid, __u64 offset,
4143                    char *name, __u16 namelen, __u16 type, __u32 attr)
4144 {
4145         ent->lde_attrs = attr | LUDA_FID;
4146         fid_cpu_to_le(&ent->lde_fid, fid);
4147
4148         ent->lde_hash = cpu_to_le64(offset);
4149         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4150
4151         strncpy(ent->lde_name, name, namelen);
4152         ent->lde_name[namelen] = '\0';
4153         ent->lde_namelen = cpu_to_le16(namelen);
4154
4155         /* append lustre attributes */
4156         osd_it_append_attrs(ent, namelen, type);
4157 }
4158
4159 /**
4160  * Return pointer to the record under iterator.
4161  */
4162 static int osd_it_iam_rec(const struct lu_env *env,
4163                           const struct dt_it *di,
4164                           struct dt_rec *dtrec, __u32 attr)
4165 {
4166         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4167         struct osd_thread_info *info = osd_oti_get(env);
4168         ENTRY;
4169
4170         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4171                 const struct osd_fid_pack *rec;
4172                 struct lu_fid             *fid = &info->oti_fid;
4173                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4174                 char                      *name;
4175                 int                        namelen;
4176                 __u64                      hash;
4177                 int                        rc;
4178
4179                 name = (char *)iam_it_key_get(&it->oi_it);
4180                 if (IS_ERR(name))
4181                         RETURN(PTR_ERR(name));
4182
4183                 namelen = iam_it_key_size(&it->oi_it);
4184
4185                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4186                 if (IS_ERR(rec))
4187                         RETURN(PTR_ERR(rec));
4188
4189                 rc = osd_fid_unpack(fid, rec);
4190                 if (rc)
4191                         RETURN(rc);
4192
4193                 hash = iam_it_store(&it->oi_it);
4194
4195                 /* IAM does not store object type in IAM index (dir) */
4196                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4197                                    0, LUDA_FID);
4198         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4199                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4200                            (struct iam_rec *)dtrec);
4201                 osd_quota_unpack(it->oi_obj, dtrec);
4202         } else {
4203                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4204                            (struct iam_rec *)dtrec);
4205         }
4206
4207         RETURN(0);
4208 }
4209
4210 /**
4211  * Returns cookie for current Iterator position.
4212  */
4213 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4214 {
4215         struct osd_it_iam *it = (struct osd_it_iam *)di;
4216
4217         return iam_it_store(&it->oi_it);
4218 }
4219
4220 /**
4221  * Restore iterator from cookie.
4222  *
4223  * \param  di      osd iterator
4224  * \param  hash    Iterator location cookie
4225  *
4226  * \retval +ve  di points to record with least key not larger than key.
4227  * \retval  0   di points to exact matched key
4228  * \retval -ve  failure
4229  */
4230
4231 static int osd_it_iam_load(const struct lu_env *env,
4232                            const struct dt_it *di, __u64 hash)
4233 {
4234         struct osd_it_iam *it = (struct osd_it_iam *)di;
4235
4236         return iam_it_load(&it->oi_it, hash);
4237 }
4238
4239 static const struct dt_index_operations osd_index_iam_ops = {
4240         .dio_lookup         = osd_index_iam_lookup,
4241         .dio_declare_insert = osd_index_declare_iam_insert,
4242         .dio_insert         = osd_index_iam_insert,
4243         .dio_declare_delete = osd_index_declare_iam_delete,
4244         .dio_delete         = osd_index_iam_delete,
4245         .dio_it     = {
4246                 .init     = osd_it_iam_init,
4247                 .fini     = osd_it_iam_fini,
4248                 .get      = osd_it_iam_get,
4249                 .put      = osd_it_iam_put,
4250                 .next     = osd_it_iam_next,
4251                 .key      = osd_it_iam_key,
4252                 .key_size = osd_it_iam_key_size,
4253                 .rec      = osd_it_iam_rec,
4254                 .store    = osd_it_iam_store,
4255                 .load     = osd_it_iam_load
4256         }
4257 };
4258
4259
4260 /**
4261  * Creates or initializes iterator context.
4262  *
4263  * \retval struct osd_it_ea, iterator structure on success
4264  *
4265  */
4266 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4267                                     struct dt_object *dt,
4268                                     __u32 attr,
4269                                     struct lustre_capa *capa)
4270 {
4271         struct osd_object       *obj  = osd_dt_obj(dt);
4272         struct osd_thread_info  *info = osd_oti_get(env);
4273         struct osd_it_ea        *it   = &info->oti_it_ea;
4274         struct lu_object        *lo   = &dt->do_lu;
4275         struct dentry           *obj_dentry = &info->oti_it_dentry;
4276         ENTRY;
4277         LASSERT(lu_object_exists(lo));
4278
4279         obj_dentry->d_inode = obj->oo_inode;
4280         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4281         obj_dentry->d_name.hash = 0;
4282
4283         it->oie_rd_dirent       = 0;
4284         it->oie_it_dirent       = 0;
4285         it->oie_dirent          = NULL;
4286         it->oie_buf             = info->oti_it_ea_buf;
4287         it->oie_obj             = obj;
4288         it->oie_file.f_pos      = 0;
4289         it->oie_file.f_dentry   = obj_dentry;
4290         if (attr & LUDA_64BITHASH)
4291                 it->oie_file.f_mode |= FMODE_64BITHASH;
4292         else
4293                 it->oie_file.f_mode |= FMODE_32BITHASH;
4294         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4295         it->oie_file.f_op         = obj->oo_inode->i_fop;
4296         it->oie_file.private_data = NULL;
4297         lu_object_get(lo);
4298         RETURN((struct dt_it *) it);
4299 }
4300
4301 /**
4302  * Destroy or finishes iterator context.
4303  *
4304  * \param di iterator structure to be destroyed
4305  */
4306 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4307 {
4308         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4309         struct osd_object    *obj  = it->oie_obj;
4310         struct inode       *inode  = obj->oo_inode;
4311
4312         ENTRY;
4313         it->oie_file.f_op->release(inode, &it->oie_file);
4314         lu_object_put(env, &obj->oo_dt.do_lu);
4315         EXIT;
4316 }
4317
4318 /**
4319  * It position the iterator at given key, so that next lookup continues from
4320  * that key Or it is similar to dio_it->load() but based on a key,
4321  * rather than file position.
4322  *
4323  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4324  * to the beginning.
4325  *
4326  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4327  */
4328 static int osd_it_ea_get(const struct lu_env *env,
4329                          struct dt_it *di, const struct dt_key *key)
4330 {
4331         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4332
4333         ENTRY;
4334         LASSERT(((const char *)key)[0] == '\0');
4335         it->oie_file.f_pos      = 0;
4336         it->oie_rd_dirent       = 0;
4337         it->oie_it_dirent       = 0;
4338         it->oie_dirent          = NULL;
4339
4340         RETURN(+1);
4341 }
4342
4343 /**
4344  * Does nothing
4345  */
4346 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4347 {
4348 }
4349
4350 /**
4351  * It is called internally by ->readdir(). It fills the
4352  * iterator's in-memory data structure with required
4353  * information i.e. name, namelen, rec_size etc.
4354  *
4355  * \param buf in which information to be filled in.
4356  * \param name name of the file in given dir
4357  *
4358  * \retval 0 on success
4359  * \retval 1 on buffer full
4360  */
4361 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4362                                loff_t offset, __u64 ino,
4363                                unsigned d_type)
4364 {
4365         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4366         struct osd_object       *obj  = it->oie_obj;
4367         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4368         struct lu_fid           *fid  = &ent->oied_fid;
4369         struct osd_fid_pack     *rec;
4370         ENTRY;
4371
4372         /* this should never happen */
4373         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4374                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4375                 RETURN(-EIO);
4376         }
4377
4378         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4379             OSD_IT_EA_BUFSIZE)
4380                 RETURN(1);
4381
4382         /* "." is just the object itself. */
4383         if (namelen == 1 && name[0] == '.') {
4384                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4385         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4386                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4387                 if (osd_fid_unpack(fid, rec) != 0)
4388                         fid_zero(fid);
4389         } else {
4390                 fid_zero(fid);
4391         }
4392         d_type &= ~LDISKFS_DIRENT_LUFID;
4393
4394         /* NOT export local root. */
4395         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4396                 ino = obj->oo_inode->i_ino;
4397                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4398         }
4399
4400         ent->oied_ino     = ino;
4401         ent->oied_off     = offset;
4402         ent->oied_namelen = namelen;
4403         ent->oied_type    = d_type;
4404
4405         memcpy(ent->oied_name, name, namelen);
4406
4407         it->oie_rd_dirent++;
4408         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4409         RETURN(0);
4410 }
4411
4412 /**
4413  * Calls ->readdir() to load a directory entry at a time
4414  * and stored it in iterator's in-memory data structure.
4415  *
4416  * \param di iterator's in memory structure
4417  *
4418  * \retval   0 on success
4419  * \retval -ve on error
4420  */
4421 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4422                                const struct dt_it *di)
4423 {
4424         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4425         struct osd_object  *obj   = it->oie_obj;
4426         struct inode       *inode = obj->oo_inode;
4427         struct htree_lock  *hlock = NULL;
4428         int                 result = 0;
4429
4430         ENTRY;
4431         it->oie_dirent = it->oie_buf;
4432         it->oie_rd_dirent = 0;
4433
4434         if (obj->oo_hl_head != NULL) {
4435                 hlock = osd_oti_get(env)->oti_hlock;
4436                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4437                                    inode, LDISKFS_HLOCK_READDIR);
4438         } else {
4439                 down_read(&obj->oo_ext_idx_sem);
4440         }
4441
4442         result = inode->i_fop->readdir(&it->oie_file, it,
4443                                        (filldir_t) osd_ldiskfs_filldir);
4444
4445         if (hlock != NULL)
4446                 ldiskfs_htree_unlock(hlock);
4447         else
4448                 up_read(&obj->oo_ext_idx_sem);
4449
4450         if (it->oie_rd_dirent == 0) {
4451                 result = -EIO;
4452         } else {
4453                 it->oie_dirent = it->oie_buf;
4454                 it->oie_it_dirent = 1;
4455         }
4456
4457         RETURN(result);
4458 }
4459
4460 /**
4461  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4462  * to load a directory entry at a time and stored it in
4463  * iterator's in-memory data structure.
4464  *
4465  * \param di iterator's in memory structure
4466  *
4467  * \retval +ve iterator reached to end
4468  * \retval   0 iterator not reached to end
4469  * \retval -ve on error
4470  */
4471 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4472 {
4473         struct osd_it_ea *it = (struct osd_it_ea *)di;
4474         int rc;
4475
4476         ENTRY;
4477
4478         if (it->oie_it_dirent < it->oie_rd_dirent) {
4479                 it->oie_dirent =
4480                         (void *) it->oie_dirent +
4481                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4482                                        it->oie_dirent->oied_namelen);
4483                 it->oie_it_dirent++;
4484                 RETURN(0);
4485         } else {
4486                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4487                         rc = +1;
4488                 else
4489                         rc = osd_ldiskfs_it_fill(env, di);
4490         }
4491
4492         RETURN(rc);
4493 }
4494
4495 /**
4496  * Returns the key at current position from iterator's in memory structure.
4497  *
4498  * \param di iterator's in memory structure
4499  *
4500  * \retval key i.e. struct dt_key on success
4501  */
4502 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4503                                     const struct dt_it *di)
4504 {
4505         struct osd_it_ea *it = (struct osd_it_ea *)di;
4506
4507         return (struct dt_key *)it->oie_dirent->oied_name;
4508 }
4509
4510 /**
4511  * Returns the key's size at current position from iterator's in memory structure.
4512  *
4513  * \param di iterator's in memory structure
4514  *
4515  * \retval key_size i.e. struct dt_key on success
4516  */
4517 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4518 {
4519         struct osd_it_ea *it = (struct osd_it_ea *)di;
4520
4521         return it->oie_dirent->oied_namelen;
4522 }
4523
4524 static int
4525 osd_dirent_update(handle_t *jh, struct super_block *sb,
4526                   struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4527                   struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de)
4528 {
4529         struct osd_fid_pack *rec;
4530         int                  rc;
4531         ENTRY;
4532
4533         LASSERT(de->file_type & LDISKFS_DIRENT_LUFID);
4534         LASSERT(de->rec_len >= de->name_len + sizeof(struct osd_fid_pack));
4535
4536         rc = ldiskfs_journal_get_write_access(jh, bh);
4537         if (rc != 0) {
4538                 CERROR("%.16s: fail to write access for update dirent: "
4539                        "name = %.*s, rc = %d\n",
4540                        LDISKFS_SB(sb)->s_es->s_volume_name,
4541                        ent->oied_namelen, ent->oied_name, rc);
4542                 RETURN(rc);
4543         }
4544
4545         rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4546         fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4547         rc = ldiskfs_journal_dirty_metadata(jh, bh);
4548         if (rc != 0)
4549                 CERROR("%.16s: fail to dirty metadata for update dirent: "
4550                        "name = %.*s, rc = %d\n",
4551                        LDISKFS_SB(sb)->s_es->s_volume_name,
4552                        ent->oied_namelen, ent->oied_name, rc);
4553
4554         RETURN(rc);
4555 }
4556
4557 static inline int
4558 osd_dirent_has_space(__u16 reclen, __u16 namelen, unsigned blocksize)
4559 {
4560         if (ldiskfs_rec_len_from_disk(reclen, blocksize) >=
4561             __LDISKFS_DIR_REC_LEN(namelen + 1 + sizeof(struct osd_fid_pack)))
4562                 return 1;
4563         else
4564                 return 0;
4565 }
4566
4567 static inline int
4568 osd_dot_dotdot_has_space(struct ldiskfs_dir_entry_2 *de, int dot_dotdot)
4569 {
4570         LASSERTF(dot_dotdot == 1 || dot_dotdot == 2,
4571                  "dot_dotdot = %d\n", dot_dotdot);
4572
4573         if (LDISKFS_DIR_REC_LEN(de) >=
4574             __LDISKFS_DIR_REC_LEN(dot_dotdot + 1 + sizeof(struct osd_fid_pack)))
4575                 return 1;
4576         else
4577                 return 0;
4578 }
4579
4580 static int
4581 osd_dirent_reinsert(const struct lu_env *env, handle_t *jh,
4582                     struct inode *dir, struct inode *inode,
4583                     struct osd_it_ea_dirent *ent, struct lu_fid *fid,
4584                     struct buffer_head *bh, struct ldiskfs_dir_entry_2 *de,
4585                     struct htree_lock *hlock)
4586 {
4587         struct dentry               *dentry;
4588         struct osd_fid_pack         *rec;
4589         struct ldiskfs_dentry_param *ldp;
4590         int                          rc;
4591         ENTRY;
4592
4593         if (!LDISKFS_HAS_INCOMPAT_FEATURE(inode->i_sb,
4594                                           LDISKFS_FEATURE_INCOMPAT_DIRDATA))
4595                 RETURN(0);
4596
4597         /* There is enough space to hold the FID-in-dirent. */
4598         if (osd_dirent_has_space(de->rec_len, ent->oied_namelen,
4599                                  dir->i_sb->s_blocksize)) {
4600                 rc = ldiskfs_journal_get_write_access(jh, bh);
4601                 if (rc != 0) {
4602                         CERROR("%.16s: fail to write access for reinsert "
4603                                "dirent: name = %.*s, rc = %d\n",
4604                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4605                                ent->oied_namelen, ent->oied_name, rc);
4606                         RETURN(rc);
4607                 }
4608
4609                 de->name[de->name_len] = 0;
4610                 rec = (struct osd_fid_pack *)(de->name + de->name_len + 1);
4611                 rec->fp_len = sizeof(struct lu_fid) + 1;
4612                 fid_cpu_to_be((struct lu_fid *)rec->fp_area, fid);
4613                 de->file_type |= LDISKFS_DIRENT_LUFID;
4614
4615                 rc = ldiskfs_journal_dirty_metadata(jh, bh);
4616                 if (rc != 0)
4617                         CERROR("%.16s: fail to dirty metadata for reinsert "
4618                                "dirent: name = %.*s, rc = %d\n",
4619                                LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4620                                ent->oied_namelen, ent->oied_name, rc);
4621
4622                 RETURN(rc);
4623         }
4624
4625         rc = ldiskfs_delete_entry(jh, dir, de, bh);
4626         if (rc != 0) {
4627                 CERROR("%.16s: fail to delete entry for reinsert dirent: "
4628                        "name = %.*s, rc = %d\n",
4629                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4630                        ent->oied_namelen, ent->oied_name, rc);
4631                 RETURN(rc);
4632         }
4633
4634         dentry = osd_child_dentry_by_inode(env, dir, ent->oied_name,
4635                                            ent->oied_namelen);
4636         ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
4637         osd_get_ldiskfs_dirent_param(ldp, (const struct dt_rec *)fid);
4638         dentry->d_fsdata = (void *)ldp;
4639         ll_vfs_dq_init(dir);
4640         rc = osd_ldiskfs_add_entry(jh, dentry, inode, hlock);
4641         /* It is too bad, we cannot reinsert the name entry back.
4642          * That means we lose it! */
4643         if (rc != 0)
4644                 CERROR("%.16s: fail to insert entry for reinsert dirent: "
4645                        "name = %.*s, rc = %d\n",
4646                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
4647                        ent->oied_namelen, ent->oied_name, rc);
4648
4649         RETURN(rc);
4650 }
4651
4652 static int
4653 osd_dirent_check_repair(const struct lu_env *env, struct osd_object *obj,
4654                         struct osd_it_ea *it, struct lu_fid *fid,
4655                         struct osd_inode_id *id, __u32 *attr)
4656 {
4657         struct osd_thread_info     *info        = osd_oti_get(env);
4658         struct lustre_mdt_attrs    *lma         = &info->oti_mdt_attrs;
4659         struct osd_device          *dev         = osd_obj2dev(obj);
4660         struct super_block         *sb          = osd_sb(dev);
4661         const char                 *devname     =
4662                                         LDISKFS_SB(sb)->s_es->s_volume_name;
4663         struct osd_it_ea_dirent    *ent         = it->oie_dirent;
4664         struct inode               *dir         = obj->oo_inode;
4665         struct htree_lock          *hlock       = NULL;
4666         struct buffer_head         *bh          = NULL;
4667         handle_t                   *jh          = NULL;
4668         struct ldiskfs_dir_entry_2 *de;
4669         struct dentry              *dentry;
4670         struct inode               *inode;
4671         int                         credits;
4672         int                         rc;
4673         int                         dot_dotdot  = 0;
4674         bool                        dirty       = false;
4675         ENTRY;
4676
4677         if (ent->oied_name[0] == '.') {
4678                 if (ent->oied_namelen == 1)
4679                         dot_dotdot = 1;
4680                 else if (ent->oied_namelen == 2 && ent->oied_name[1] == '.')
4681                         dot_dotdot = 2;
4682         }
4683
4684         dentry = osd_child_dentry_get(env, obj, ent->oied_name,
4685                                       ent->oied_namelen);
4686
4687         /* We need to ensure that the name entry is still valid.
4688          * Because it may be removed or renamed by other already.
4689          *
4690          * The unlink or rename operation will start journal before PDO lock,
4691          * so to avoid deadlock, here we need to start journal handle before
4692          * related PDO lock also. But because we do not know whether there
4693          * will be something to be repaired before PDO lock, we just start
4694          * journal without conditions.
4695          *
4696          * We may need to remove the name entry firstly, then insert back.
4697          * One credit is for user quota file update.
4698          * One credit is for group quota file update.
4699          * Two credits are for dirty inode. */
4700         credits = osd_dto_credits_noquota[DTO_INDEX_DELETE] +
4701                   osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1 + 1 + 2;
4702
4703 again:
4704         if (dev->od_dirent_journal) {
4705                 jh = ldiskfs_journal_start_sb(sb, credits);
4706                 if (IS_ERR(jh)) {
4707                         rc = PTR_ERR(jh);
4708                         CERROR("%.16s: fail to start trans for dirent "
4709                                "check_repair: credits %d, name %.*s, rc %d\n",
4710                                devname, credits, ent->oied_namelen,
4711                                ent->oied_name, rc);
4712                         RETURN(rc);
4713                 }
4714
4715                 if (obj->oo_hl_head != NULL) {
4716                         hlock = osd_oti_get(env)->oti_hlock;
4717                         /* "0" means exclusive lock for the whole directory.
4718                          * We need to prevent others access such name entry
4719                          * during the delete + insert. Neither HLOCK_ADD nor
4720                          * HLOCK_DEL cannot guarantee the atomicity. */
4721                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir, 0);
4722                 } else {
4723                         down_write(&obj->oo_ext_idx_sem);
4724                 }
4725         } else {
4726                 if (obj->oo_hl_head != NULL) {
4727                         hlock = osd_oti_get(env)->oti_hlock;
4728                         ldiskfs_htree_lock(hlock, obj->oo_hl_head, dir,
4729                                            LDISKFS_HLOCK_LOOKUP);
4730                 } else {
4731                         down_read(&obj->oo_ext_idx_sem);
4732                 }
4733         }
4734
4735         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
4736         /* For dot/dotdot entry, if there is not enough space to hold the
4737          * FID-in-dirent, just keep them there. It only happens when the
4738          * device upgraded from 1.8 or restored from MDT file-level backup.
4739          * For the whole directory, only dot/dotdot entry have no FID-in-dirent
4740          * and needs to get FID from LMA when readdir, it will not affect the
4741          * performance much. */
4742         if ((bh == NULL) || (le32_to_cpu(de->inode) != ent->oied_ino) ||
4743             (dot_dotdot != 0 && !osd_dot_dotdot_has_space(de, dot_dotdot))) {
4744                 *attr |= LUDA_IGNORE;
4745                 GOTO(out_journal, rc = 0);
4746         }
4747
4748         osd_id_gen(id, ent->oied_ino, OSD_OII_NOGEN);
4749         inode = osd_iget(info, dev, id);
4750         if (IS_ERR(inode)) {
4751                 rc = PTR_ERR(inode);
4752                 if (rc == -ENOENT || rc == -ESTALE) {
4753                         *attr |= LUDA_IGNORE;
4754                         rc = 0;
4755                 }
4756
4757                 GOTO(out_journal, rc);
4758         }
4759
4760         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
4761         if (rc == 0) {
4762                 if (fid_is_sane(fid)) {
4763                         /* FID-in-dirent is valid. */
4764                         if (lu_fid_eq(fid, &lma->lma_self_fid))
4765                                 GOTO(out_inode, rc = 0);
4766
4767                         /* Do not repair under dryrun mode. */
4768                         if (*attr & LUDA_VERIFY_DRYRUN) {
4769                                 *attr |= LUDA_REPAIR;
4770                                 GOTO(out_inode, rc = 0);
4771                         }
4772
4773                         if (!dev->od_dirent_journal) {
4774                                 iput(inode);
4775                                 brelse(bh);
4776                                 if (hlock != NULL)
4777                                         ldiskfs_htree_unlock(hlock);
4778                                 else
4779                                         up_read(&obj->oo_ext_idx_sem);
4780                                 dev->od_dirent_journal = 1;
4781                                 goto again;
4782                         }
4783
4784                         *fid = lma->lma_self_fid;
4785                         dirty = true;
4786                         /* Update the FID-in-dirent. */
4787                         rc = osd_dirent_update(jh, sb, ent, fid, bh, de);
4788                         if (rc == 0)
4789                                 *attr |= LUDA_REPAIR;
4790                 } else {
4791                         /* Do not repair under dryrun mode. */
4792                         if (*attr & LUDA_VERIFY_DRYRUN) {
4793                                 *fid = lma->lma_self_fid;
4794                                 *attr |= LUDA_REPAIR;
4795                                 GOTO(out_inode, rc = 0);
4796                         }
4797
4798                         if (!dev->od_dirent_journal) {
4799                                 iput(inode);
4800                                 brelse(bh);
4801                                 if (hlock != NULL)
4802                                         ldiskfs_htree_unlock(hlock);
4803                                 else
4804                                         up_read(&obj->oo_ext_idx_sem);
4805                                 dev->od_dirent_journal = 1;
4806                                 goto again;
4807                         }
4808
4809                         *fid = lma->lma_self_fid;
4810                         dirty = true;
4811                         /* Append the FID-in-dirent. */
4812                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4813                                                  fid, bh, de, hlock);
4814                         if (rc == 0)
4815                                 *attr |= LUDA_REPAIR;
4816                 }
4817         } else if (rc == -ENODATA) {
4818                 /* Do not repair under dryrun mode. */
4819                 if (*attr & LUDA_VERIFY_DRYRUN) {
4820                         if (fid_is_sane(fid)) {
4821                                 *attr |= LUDA_REPAIR;
4822                         } else {
4823                                 lu_igif_build(fid, inode->i_ino,
4824                                               inode->i_generation);
4825                                 *attr |= LUDA_UPGRADE;
4826                         }
4827                         GOTO(out_inode, rc = 0);
4828                 }
4829
4830                 if (!dev->od_dirent_journal) {
4831                         iput(inode);
4832                         brelse(bh);
4833                         if (hlock != NULL)
4834                                 ldiskfs_htree_unlock(hlock);
4835                         else
4836                                 up_read(&obj->oo_ext_idx_sem);
4837                         dev->od_dirent_journal = 1;
4838                         goto again;
4839                 }
4840
4841                 dirty = true;
4842                 if (unlikely(fid_is_sane(fid))) {
4843                         /* FID-in-dirent exists, but FID-in-LMA is lost.
4844                          * Trust the FID-in-dirent, and add FID-in-LMA. */
4845                         rc = osd_ea_fid_set(info, inode, fid, 0);
4846                         if (rc == 0)
4847                                 *attr |= LUDA_REPAIR;
4848                 } else {
4849                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
4850                         /* It is probably IGIF object. Only aappend the
4851                          * FID-in-dirent. OI scrub will process FID-in-LMA. */
4852                         rc = osd_dirent_reinsert(env, jh, dir, inode, ent,
4853                                                  fid, bh, de, hlock);
4854                         if (rc == 0)
4855                                 *attr |= LUDA_UPGRADE;
4856                 }
4857         }
4858
4859         GOTO(out_inode, rc);
4860
4861 out_inode:
4862         iput(inode);
4863
4864 out_journal:
4865         brelse(bh);
4866         if (hlock != NULL) {
4867                 ldiskfs_htree_unlock(hlock);
4868         } else {
4869                 if (dev->od_dirent_journal)
4870                         up_write(&obj->oo_ext_idx_sem);
4871                 else
4872                         up_read(&obj->oo_ext_idx_sem);
4873         }
4874         if (jh != NULL)
4875                 ldiskfs_journal_stop(jh);
4876         if (rc >= 0 && !dirty)
4877                 dev->od_dirent_journal = 0;
4878         return rc;
4879 }
4880
4881 /**
4882  * Returns the value at current position from iterator's in memory structure.
4883  *
4884  * \param di struct osd_it_ea, iterator's in memory structure
4885  * \param attr attr requested for dirent.
4886  * \param lde lustre dirent
4887  *
4888  * \retval   0 no error and \param lde has correct lustre dirent.
4889  * \retval -ve on error
4890  */
4891 static inline int osd_it_ea_rec(const struct lu_env *env,
4892                                 const struct dt_it *di,
4893                                 struct dt_rec *dtrec, __u32 attr)
4894 {
4895         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4896         struct osd_object      *obj   = it->oie_obj;
4897         struct osd_device      *dev   = osd_obj2dev(obj);
4898         struct osd_scrub       *scrub = &dev->od_scrub;
4899         struct scrub_file      *sf    = &scrub->os_file;
4900         struct osd_thread_info *oti   = osd_oti_get(env);
4901         struct osd_inode_id    *id    = &oti->oti_id;
4902         struct osd_idmap_cache *oic   = &oti->oti_cache;
4903         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4904         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4905         __u32                   ino   = it->oie_dirent->oied_ino;
4906         int                     rc    = 0;
4907         ENTRY;
4908
4909         if (attr & LUDA_VERIFY) {
4910                 attr |= LUDA_TYPE;
4911                 if (unlikely(ino == osd_sb(dev)->s_root->d_inode->i_ino)) {
4912                         attr |= LUDA_IGNORE;
4913                         rc = 0;
4914                         goto pack;
4915                 }
4916
4917                 rc = osd_dirent_check_repair(env, obj, it, fid, id, &attr);
4918         } else {
4919                 attr &= ~LU_DIRENT_ATTRS_MASK;
4920                 if (!fid_is_sane(fid)) {
4921                         if (OBD_FAIL_CHECK(OBD_FAIL_FID_LOOKUP))
4922                                 RETURN(-ENOENT);
4923
4924                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
4925                 } else {
4926                         osd_id_gen(id, ino, OSD_OII_NOGEN);
4927                 }
4928         }
4929
4930         if (rc < 0)
4931                 RETURN(rc);
4932
4933 pack:
4934         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4935                            it->oie_dirent->oied_name,
4936                            it->oie_dirent->oied_namelen,
4937                            it->oie_dirent->oied_type, attr);
4938
4939         if (osd_remote_fid(env, dev, fid))
4940                 RETURN(0);
4941
4942         if (likely(!(attr & LUDA_IGNORE))) {
4943                 oic->oic_lid = *id;
4944                 oic->oic_fid = *fid;
4945         }
4946
4947         if (!(attr & LUDA_VERIFY) &&
4948             (scrub->os_pos_current <= ino) &&
4949             ((sf->sf_flags & SF_INCONSISTENT) ||
4950              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4951              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4952                 osd_consistency_check(oti, dev, oic);
4953
4954         RETURN(rc);
4955 }
4956
4957 /**
4958  * Returns a cookie for current position of the iterator head, so that
4959  * user can use this cookie to load/start the iterator next time.
4960  *
4961  * \param di iterator's in memory structure
4962  *
4963  * \retval cookie for current position, on success
4964  */
4965 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4966 {
4967         struct osd_it_ea *it = (struct osd_it_ea *)di;
4968
4969         return it->oie_dirent->oied_off;
4970 }
4971
4972 /**
4973  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4974  * to load a directory entry at a time and stored it i inn,
4975  * in iterator's in-memory data structure.
4976  *
4977  * \param di struct osd_it_ea, iterator's in memory structure
4978  *
4979  * \retval +ve on success
4980  * \retval -ve on error
4981  */
4982 static int osd_it_ea_load(const struct lu_env *env,
4983                           const struct dt_it *di, __u64 hash)
4984 {
4985         struct osd_it_ea *it = (struct osd_it_ea *)di;
4986         int rc;
4987
4988         ENTRY;
4989         it->oie_file.f_pos = hash;
4990
4991         rc =  osd_ldiskfs_it_fill(env, di);
4992         if (rc == 0)
4993                 rc = +1;
4994
4995         RETURN(rc);
4996 }
4997
4998 /**
4999  * Index lookup function for interoperability mode (b11826).
5000  *
5001  * \param key,  key i.e. file name to be searched
5002  *
5003  * \retval +ve, on success
5004  * \retval -ve, on error
5005  */
5006 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
5007                                struct dt_rec *rec, const struct dt_key *key,
5008                                struct lustre_capa *capa)
5009 {
5010         struct osd_object *obj = osd_dt_obj(dt);
5011         int rc = 0;
5012
5013         ENTRY;
5014
5015         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
5016         LINVRNT(osd_invariant(obj));
5017
5018         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
5019                 return -EACCES;
5020
5021         rc = osd_ea_lookup_rec(env, obj, rec, key);
5022         if (rc == 0)
5023                 rc = +1;
5024         RETURN(rc);
5025 }
5026
5027 /**
5028  * Index and Iterator operations for interoperability
5029  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
5030  */
5031 static const struct dt_index_operations osd_index_ea_ops = {
5032         .dio_lookup         = osd_index_ea_lookup,
5033         .dio_declare_insert = osd_index_declare_ea_insert,
5034         .dio_insert         = osd_index_ea_insert,
5035         .dio_declare_delete = osd_index_declare_ea_delete,
5036         .dio_delete         = osd_index_ea_delete,
5037         .dio_it     = {
5038                 .init     = osd_it_ea_init,
5039                 .fini     = osd_it_ea_fini,
5040                 .get      = osd_it_ea_get,
5041                 .put      = osd_it_ea_put,
5042                 .next     = osd_it_ea_next,
5043                 .key      = osd_it_ea_key,
5044                 .key_size = osd_it_ea_key_size,
5045                 .rec      = osd_it_ea_rec,
5046                 .store    = osd_it_ea_store,
5047                 .load     = osd_it_ea_load
5048         }
5049 };
5050
5051 static void *osd_key_init(const struct lu_context *ctx,
5052                           struct lu_context_key *key)
5053 {
5054         struct osd_thread_info *info;
5055
5056         OBD_ALLOC_PTR(info);
5057         if (info == NULL)
5058                 return ERR_PTR(-ENOMEM);
5059
5060         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5061         if (info->oti_it_ea_buf == NULL)
5062                 goto out_free_info;
5063
5064         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
5065
5066         info->oti_hlock = ldiskfs_htree_lock_alloc();
5067         if (info->oti_hlock == NULL)
5068                 goto out_free_ea;
5069
5070         return info;
5071
5072  out_free_ea:
5073         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5074  out_free_info:
5075         OBD_FREE_PTR(info);
5076         return ERR_PTR(-ENOMEM);
5077 }
5078
5079 static void osd_key_fini(const struct lu_context *ctx,
5080                          struct lu_context_key *key, void* data)
5081 {
5082         struct osd_thread_info *info = data;
5083
5084         if (info->oti_hlock != NULL)
5085                 ldiskfs_htree_lock_free(info->oti_hlock);
5086         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
5087         lu_buf_free(&info->oti_iobuf.dr_pg_buf);
5088         lu_buf_free(&info->oti_iobuf.dr_bl_buf);
5089         OBD_FREE_PTR(info);
5090 }
5091
5092 static void osd_key_exit(const struct lu_context *ctx,
5093                          struct lu_context_key *key, void *data)
5094 {
5095         struct osd_thread_info *info = data;
5096
5097         LASSERT(info->oti_r_locks == 0);
5098         LASSERT(info->oti_w_locks == 0);
5099         LASSERT(info->oti_txns    == 0);
5100 }
5101
5102 /* type constructor/destructor: osd_type_init, osd_type_fini */
5103 LU_TYPE_INIT_FINI(osd, &osd_key);
5104
5105 struct lu_context_key osd_key = {
5106         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
5107         .lct_init = osd_key_init,
5108         .lct_fini = osd_key_fini,
5109         .lct_exit = osd_key_exit
5110 };
5111
5112
5113 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
5114                            const char *name, struct lu_device *next)
5115 {
5116         struct osd_device *osd = osd_dev(d);
5117
5118         if (strlcpy(osd->od_svname, name, sizeof(osd->od_svname))
5119             >= sizeof(osd->od_svname))
5120                 return -E2BIG;
5121         return osd_procfs_init(osd, name);
5122 }
5123
5124 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
5125 {
5126         ENTRY;
5127
5128         osd_scrub_cleanup(env, o);
5129
5130         if (o->od_fsops) {
5131                 fsfilt_put_ops(o->od_fsops);
5132                 o->od_fsops = NULL;
5133         }
5134
5135         /* shutdown quota slave instance associated with the device */
5136         if (o->od_quota_slave != NULL) {
5137                 qsd_fini(env, o->od_quota_slave);
5138                 o->od_quota_slave = NULL;
5139         }
5140
5141         RETURN(0);
5142 }
5143
5144 static int osd_mount(const struct lu_env *env,
5145                      struct osd_device *o, struct lustre_cfg *cfg)
5146 {
5147         const char              *name  = lustre_cfg_string(cfg, 0);
5148         const char              *dev  = lustre_cfg_string(cfg, 1);
5149         const char              *opts;
5150         unsigned long            page, s_flags, lmd_flags = 0;
5151         struct page             *__page;
5152         struct file_system_type *type;
5153         char                    *options = NULL;
5154         char                    *str;
5155         int                       rc = 0;
5156         ENTRY;
5157
5158         if (o->od_mnt != NULL)
5159                 RETURN(0);
5160
5161         if (strlen(dev) >= sizeof(o->od_mntdev))
5162                 RETURN(-E2BIG);
5163         strcpy(o->od_mntdev, dev);
5164
5165         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
5166         if (o->od_fsops == NULL) {
5167                 CERROR("Can't find fsfilt_ldiskfs\n");
5168                 RETURN(-ENOTSUPP);
5169         }
5170
5171         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
5172         if (__page == NULL)
5173                 GOTO(out, rc = -ENOMEM);
5174
5175         str = lustre_cfg_string(cfg, 2);
5176         s_flags = simple_strtoul(str, NULL, 0);
5177         str = strstr(str, ":");
5178         if (str)
5179                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
5180         opts = lustre_cfg_string(cfg, 3);
5181         page = (unsigned long)cfs_page_address(__page);
5182         options = (char *)page;
5183         *options = '\0';
5184         if (opts == NULL)
5185                 strcat(options, "user_xattr,acl");
5186         else
5187                 strcat(options, opts);
5188
5189         /* Glom up mount options */
5190         if (*options != '\0')
5191                 strcat(options, ",");
5192         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
5193
5194         type = get_fs_type("ldiskfs");
5195         if (!type) {
5196                 CERROR("%s: cannot find ldiskfs module\n", name);
5197                 GOTO(out, rc = -ENODEV);
5198         }
5199
5200         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
5201         cfs_module_put(type->owner);
5202
5203         if (IS_ERR(o->od_mnt)) {
5204                 rc = PTR_ERR(o->od_mnt);
5205                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
5206                 o->od_mnt = NULL;
5207                 GOTO(out, rc);
5208         }
5209
5210 #ifdef HAVE_DEV_SET_RDONLY
5211         if (dev_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
5212                 CERROR("%s: underlying device %s is marked as read-only. "
5213                        "Setup failed\n", name, dev);
5214                 mntput(o->od_mnt);
5215                 o->od_mnt = NULL;
5216                 GOTO(out, rc = -EROFS);
5217         }
5218 #endif
5219
5220         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
5221             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
5222                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
5223                 mntput(o->od_mnt);
5224                 o->od_mnt = NULL;
5225                 GOTO(out, rc = -EINVAL);
5226         }
5227
5228         ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
5229                                 LDISKFS_STATE_LUSTRE_NO_OI);
5230         if (lmd_flags & LMD_FLG_NOSCRUB)
5231                 o->od_noscrub = 1;
5232
5233 out:
5234         if (__page)
5235                 OBD_PAGE_FREE(__page);
5236         if (rc)
5237                 fsfilt_put_ops(o->od_fsops);
5238
5239         RETURN(rc);
5240 }
5241
5242 static struct lu_device *osd_device_fini(const struct lu_env *env,
5243                                          struct lu_device *d)
5244 {
5245         int rc;
5246         ENTRY;
5247
5248         rc = osd_shutdown(env, osd_dev(d));
5249
5250         osd_obj_map_fini(osd_dev(d));
5251
5252         shrink_dcache_sb(osd_sb(osd_dev(d)));
5253         osd_sync(env, lu2dt_dev(d));
5254
5255         rc = osd_procfs_fini(osd_dev(d));
5256         if (rc) {
5257                 CERROR("proc fini error %d \n", rc);
5258                 RETURN (ERR_PTR(rc));
5259         }
5260
5261         if (osd_dev(d)->od_mnt) {
5262                 mntput(osd_dev(d)->od_mnt);
5263                 osd_dev(d)->od_mnt = NULL;
5264         }
5265
5266         RETURN(NULL);
5267 }
5268
5269 static int osd_device_init0(const struct lu_env *env,
5270                             struct osd_device *o,
5271                             struct lustre_cfg *cfg)
5272 {
5273         struct lu_device        *l = osd2lu_dev(o);
5274         struct osd_thread_info *info;
5275         int                     rc;
5276         int                     cplen = 0;
5277
5278         /* if the module was re-loaded, env can loose its keys */
5279         rc = lu_env_refill((struct lu_env *) env);
5280         if (rc)
5281                 GOTO(out, rc);
5282         info = osd_oti_get(env);
5283         LASSERT(info);
5284
5285         l->ld_ops = &osd_lu_ops;
5286         o->od_dt_dev.dd_ops = &osd_dt_ops;
5287
5288         spin_lock_init(&o->od_osfs_lock);
5289         mutex_init(&o->od_otable_mutex);
5290         o->od_osfs_age = cfs_time_shift_64(-1000);
5291
5292         o->od_capa_hash = init_capa_hash();
5293         if (o->od_capa_hash == NULL)
5294                 GOTO(out, rc = -ENOMEM);
5295
5296         o->od_read_cache = 1;
5297         o->od_writethrough_cache = 1;
5298         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
5299
5300         rc = osd_mount(env, o, cfg);
5301         if (rc)
5302                 GOTO(out_capa, rc);
5303
5304         CFS_INIT_LIST_HEAD(&o->od_ios_list);
5305         /* setup scrub, including OI files initialization */
5306         rc = osd_scrub_setup(env, o);
5307         if (rc < 0)
5308                 GOTO(out_mnt, rc);
5309
5310         cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
5311                         sizeof(o->od_svname));
5312         if (cplen >= sizeof(o->od_svname)) {
5313                 rc = -E2BIG;
5314                 GOTO(out_mnt, rc);
5315         }
5316
5317         rc = osd_obj_map_init(env, o);
5318         if (rc != 0)
5319                 GOTO(out_scrub, rc);
5320
5321         rc = lu_site_init(&o->od_site, l);
5322         if (rc)
5323                 GOTO(out_compat, rc);
5324         o->od_site.ls_bottom_dev = l;
5325
5326         rc = lu_site_init_finish(&o->od_site);
5327         if (rc)
5328                 GOTO(out_site, rc);
5329
5330         rc = osd_procfs_init(o, o->od_svname);
5331         if (rc != 0) {
5332                 CERROR("%s: can't initialize procfs: rc = %d\n",
5333                        o->od_svname, rc);
5334                 GOTO(out_site, rc);
5335         }
5336
5337         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
5338
5339         /* initialize quota slave instance */
5340         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
5341                                      o->od_proc_entry);
5342         if (IS_ERR(o->od_quota_slave)) {
5343                 rc = PTR_ERR(o->od_quota_slave);
5344                 o->od_quota_slave = NULL;
5345                 GOTO(out_procfs, rc);
5346         }
5347
5348         RETURN(0);
5349 out_procfs:
5350         osd_procfs_fini(o);
5351 out_site:
5352         lu_site_fini(&o->od_site);
5353 out_compat:
5354         osd_obj_map_fini(o);
5355 out_scrub:
5356         osd_scrub_cleanup(env, o);
5357 out_mnt:
5358         osd_oi_fini(info, o);
5359         osd_shutdown(env, o);
5360         mntput(o->od_mnt);
5361         o->od_mnt = NULL;
5362 out_capa:
5363         cleanup_capa_hash(o->od_capa_hash);
5364 out:
5365         RETURN(rc);
5366 }
5367
5368 static struct lu_device *osd_device_alloc(const struct lu_env *env,
5369                                           struct lu_device_type *t,
5370                                           struct lustre_cfg *cfg)
5371 {
5372         struct osd_device *o;
5373         int                rc;
5374
5375         OBD_ALLOC_PTR(o);
5376         if (o == NULL)
5377                 return ERR_PTR(-ENOMEM);
5378
5379         rc = dt_device_init(&o->od_dt_dev, t);
5380         if (rc == 0) {
5381                 /* Because the ctx might be revived in dt_device_init,
5382                  * refill the env here */
5383                 lu_env_refill((struct lu_env *)env);
5384                 rc = osd_device_init0(env, o, cfg);
5385                 if (rc)
5386                         dt_device_fini(&o->od_dt_dev);
5387         }
5388
5389         if (unlikely(rc != 0))
5390                 OBD_FREE_PTR(o);
5391
5392         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
5393 }
5394
5395 static struct lu_device *osd_device_free(const struct lu_env *env,
5396                                          struct lu_device *d)
5397 {
5398         struct osd_device *o = osd_dev(d);
5399         ENTRY;
5400
5401         cleanup_capa_hash(o->od_capa_hash);
5402         /* XXX: make osd top device in order to release reference */
5403         d->ld_site->ls_top_dev = d;
5404         lu_site_purge(env, d->ld_site, -1);
5405         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
5406                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
5407                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
5408         }
5409         lu_site_fini(&o->od_site);
5410         dt_device_fini(&o->od_dt_dev);
5411         OBD_FREE_PTR(o);
5412         RETURN(NULL);
5413 }
5414
5415 static int osd_process_config(const struct lu_env *env,
5416                               struct lu_device *d, struct lustre_cfg *cfg)
5417 {
5418         struct osd_device *o = osd_dev(d);
5419         int err;
5420         ENTRY;
5421
5422         switch(cfg->lcfg_command) {
5423         case LCFG_SETUP:
5424                 err = osd_mount(env, o, cfg);
5425                 break;
5426         case LCFG_CLEANUP:
5427                 lu_dev_del_linkage(d->ld_site, d);
5428                 err = osd_shutdown(env, o);
5429                 break;
5430         default:
5431                 err = -ENOSYS;
5432         }
5433
5434         RETURN(err);
5435 }
5436
5437 static int osd_recovery_complete(const struct lu_env *env,
5438                                  struct lu_device *d)
5439 {
5440         struct osd_device       *osd = osd_dev(d);
5441         int                      rc = 0;
5442         ENTRY;
5443
5444         if (osd->od_quota_slave == NULL)
5445                 RETURN(0);
5446
5447         /* start qsd instance on recovery completion, this notifies the quota
5448          * slave code that we are about to process new requests now */
5449         rc = qsd_start(env, osd->od_quota_slave);
5450         RETURN(rc);
5451 }
5452
5453 /*
5454  * we use exports to track all osd users
5455  */
5456 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
5457                            struct obd_device *obd, struct obd_uuid *cluuid,
5458                            struct obd_connect_data *data, void *localdata)
5459 {
5460         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
5461         struct lustre_handle  conn;
5462         int                   rc;
5463         ENTRY;
5464
5465         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
5466
5467         rc = class_connect(&conn, obd, cluuid);
5468         if (rc)
5469                 RETURN(rc);
5470
5471         *exp = class_conn2export(&conn);
5472
5473         spin_lock(&osd->od_osfs_lock);
5474         osd->od_connects++;
5475         spin_unlock(&osd->od_osfs_lock);
5476
5477         RETURN(0);
5478 }
5479
5480 /*
5481  * once last export (we don't count self-export) disappeared
5482  * osd can be released
5483  */
5484 static int osd_obd_disconnect(struct obd_export *exp)
5485 {
5486         struct obd_device *obd = exp->exp_obd;
5487         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5488         int                rc, release = 0;
5489         ENTRY;
5490
5491         /* Only disconnect the underlying layers on the final disconnect. */
5492         spin_lock(&osd->od_osfs_lock);
5493         osd->od_connects--;
5494         if (osd->od_connects == 0)
5495                 release = 1;
5496         spin_unlock(&osd->od_osfs_lock);
5497
5498         rc = class_disconnect(exp); /* bz 9811 */
5499
5500         if (rc == 0 && release)
5501                 class_manual_cleanup(obd);
5502         RETURN(rc);
5503 }
5504
5505 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5506                        struct lu_device *dev)
5507 {
5508         struct osd_device *osd = osd_dev(dev);
5509         int                result = 0;
5510         ENTRY;
5511
5512         if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
5513                 /* MDT/MDD still use old infrastructure to create
5514                  * special files */
5515                 result = llo_local_objects_setup(env, lu2md_dev(pdev),
5516                                                  lu2dt_dev(dev));
5517                 if (result)
5518                         RETURN(result);
5519         }
5520
5521         if (osd->od_quota_slave != NULL)
5522                 /* set up quota slave objects */
5523                 result = qsd_prepare(env, osd->od_quota_slave);
5524
5525         RETURN(result);
5526 }
5527
5528 static const struct lu_object_operations osd_lu_obj_ops = {
5529         .loo_object_init      = osd_object_init,
5530         .loo_object_delete    = osd_object_delete,
5531         .loo_object_release   = osd_object_release,
5532         .loo_object_free      = osd_object_free,
5533         .loo_object_print     = osd_object_print,
5534         .loo_object_invariant = osd_object_invariant
5535 };
5536
5537 const struct lu_device_operations osd_lu_ops = {
5538         .ldo_object_alloc      = osd_object_alloc,
5539         .ldo_process_config    = osd_process_config,
5540         .ldo_recovery_complete = osd_recovery_complete,
5541         .ldo_prepare           = osd_prepare,
5542 };
5543
5544 static const struct lu_device_type_operations osd_device_type_ops = {
5545         .ldto_init = osd_type_init,
5546         .ldto_fini = osd_type_fini,
5547
5548         .ldto_start = osd_type_start,
5549         .ldto_stop  = osd_type_stop,
5550
5551         .ldto_device_alloc = osd_device_alloc,
5552         .ldto_device_free  = osd_device_free,
5553
5554         .ldto_device_init    = osd_device_init,
5555         .ldto_device_fini    = osd_device_fini
5556 };
5557
5558 struct lu_device_type osd_device_type = {
5559         .ldt_tags     = LU_DEVICE_DT,
5560         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
5561         .ldt_ops      = &osd_device_type_ops,
5562         .ldt_ctx_tags = LCT_LOCAL,
5563 };
5564
5565 /*
5566  * lprocfs legacy support.
5567  */
5568 static struct obd_ops osd_obd_device_ops = {
5569         .o_owner = THIS_MODULE,
5570         .o_connect      = osd_obd_connect,
5571         .o_disconnect   = osd_obd_disconnect
5572 };
5573
5574 static int __init osd_mod_init(void)
5575 {
5576         struct lprocfs_static_vars lvars;
5577
5578         osd_oi_mod_init();
5579         lprocfs_osd_init_vars(&lvars);
5580         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
5581                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5582 }
5583
5584 static void __exit osd_mod_exit(void)
5585 {
5586         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
5587 }
5588
5589 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5590 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
5591 MODULE_LICENSE("GPL");
5592
5593 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);