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