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));
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));
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 static int osd_declare_object_create(const struct lu_env *env,
2009                                      struct dt_object *dt,
2010                                      struct lu_attr *attr,
2011                                      struct dt_allocation_hint *hint,
2012                                      struct dt_object_format *dof,
2013                                      struct thandle *handle)
2014 {
2015         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
2016         struct osd_thandle      *oh;
2017         int                      rc;
2018         ENTRY;
2019
2020         LASSERT(handle != NULL);
2021
2022         oh = container_of0(handle, struct osd_thandle, ot_super);
2023         LASSERT(oh->ot_handle == NULL);
2024
2025         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2026                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2027         if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2028                            lu_object_fid(&dt->do_lu)))
2029                 /* Reuse idle OI block may cause additional one OI block
2030                  * to be changed. */
2031                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2032                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2033
2034         /* If this is directory, then we expect . and .. to be inserted as
2035          * well. The one directory block always needs to be created for the
2036          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2037          * block), there is no danger of needing a tree for the first block.
2038          */
2039         if (attr && S_ISDIR(attr->la_mode)) {
2040                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2041                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2042                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2043         }
2044
2045         if (!attr)
2046                 RETURN(0);
2047
2048         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2049                                    false, false, NULL, false);
2050         if (rc != 0)
2051                 RETURN(rc);
2052
2053         /* It does fld look up inside declare, and the result will be
2054          * added to fld cache, so the following fld lookup inside insert
2055          * does not need send RPC anymore, so avoid send rpc with holding
2056          * transaction */
2057         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2058                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2059                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2060                                lu_object_fid(&dt->do_lu), range);
2061
2062
2063         RETURN(rc);
2064 }
2065
2066 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2067                              struct lu_attr *attr,
2068                              struct dt_allocation_hint *hint,
2069                              struct dt_object_format *dof,
2070                              struct thandle *th)
2071 {
2072         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2073         struct osd_object      *obj    = osd_dt_obj(dt);
2074         struct osd_thread_info *info   = osd_oti_get(env);
2075         int result;
2076
2077         ENTRY;
2078
2079         LINVRNT(osd_invariant(obj));
2080         LASSERT(!dt_object_exists(dt));
2081         LASSERT(osd_write_locked(env, obj));
2082         LASSERT(th != NULL);
2083
2084         if (unlikely(fid_is_acct(fid)))
2085                 /* Quota files can't be created from the kernel any more,
2086                  * 'tune2fs -O quota' will take care of creating them */
2087                 RETURN(-EPERM);
2088
2089         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2090         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2091
2092         result = __osd_object_create(info, obj, attr, hint, dof, th);
2093         if (result == 0)
2094                 result = __osd_oi_insert(env, obj, fid, th);
2095
2096         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2097         LASSERT(osd_invariant(obj));
2098         RETURN(result);
2099 }
2100
2101 /**
2102  * Called to destroy on-disk representation of the object
2103  *
2104  * Concurrency: must be locked
2105  */
2106 static int osd_declare_object_destroy(const struct lu_env *env,
2107                                       struct dt_object *dt,
2108                                       struct thandle *th)
2109 {
2110         struct osd_object  *obj = osd_dt_obj(dt);
2111         struct inode       *inode = obj->oo_inode;
2112         struct osd_thandle *oh;
2113         int                 rc;
2114         ENTRY;
2115
2116         oh = container_of0(th, struct osd_thandle, ot_super);
2117         LASSERT(oh->ot_handle == NULL);
2118         LASSERT(inode);
2119
2120         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2121                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2122         /* Recycle idle OI leaf may cause additional three OI blocks
2123          * to be changed. */
2124         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2125                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2126
2127         /* one less inode */
2128         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2129                                    false, true, NULL, false);
2130         if (rc)
2131                 RETURN(rc);
2132         /* data to be truncated */
2133         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2134                                    true, true, NULL, false);
2135         RETURN(rc);
2136 }
2137
2138 static int osd_object_destroy(const struct lu_env *env,
2139                               struct dt_object *dt,
2140                               struct thandle *th)
2141 {
2142         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2143         struct osd_object      *obj = osd_dt_obj(dt);
2144         struct inode           *inode = obj->oo_inode;
2145         struct osd_device      *osd = osd_obj2dev(obj);
2146         struct osd_thandle     *oh;
2147         int                     result;
2148         ENTRY;
2149
2150         oh = container_of0(th, struct osd_thandle, ot_super);
2151         LASSERT(oh->ot_handle);
2152         LASSERT(inode);
2153         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2154
2155         if (unlikely(fid_is_acct(fid)))
2156                 RETURN(-EPERM);
2157
2158         /* Parallel control for OI scrub. For most of cases, there is no
2159          * lock contention. So it will not affect unlink performance. */
2160         mutex_lock(&inode->i_mutex);
2161         if (S_ISDIR(inode->i_mode)) {
2162                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2163                 /* it will check/delete the agent inode for every dir
2164                  * destory, how to optimize it? unlink performance
2165                  * impaction XXX */
2166                 result = osd_delete_agent_inode(env, osd, obj, oh);
2167                 if (result != 0 && result != -ENOENT) {
2168                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
2169                                osd_name(osd), PFID(fid), result);
2170                 }
2171                 spin_lock(&obj->oo_guard);
2172                 clear_nlink(inode);
2173                 spin_unlock(&obj->oo_guard);
2174                 inode->i_sb->s_op->dirty_inode(inode);
2175         }
2176
2177         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2178
2179         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2180         mutex_unlock(&inode->i_mutex);
2181
2182         /* XXX: add to ext3 orphan list */
2183         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2184
2185         /* not needed in the cache anymore */
2186         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2187
2188         RETURN(0);
2189 }
2190
2191 static inline int __osd_xattr_set(struct osd_thread_info *info,
2192                                   struct inode *inode, const char *name,
2193                                   const void *buf, int buflen, int fl)
2194 {
2195         struct dentry *dentry = &info->oti_child_dentry;
2196
2197         ll_vfs_dq_init(inode);
2198         dentry->d_inode = inode;
2199         return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
2200 }
2201
2202 /**
2203  * Put the fid into lustre_mdt_attrs, and then place the structure
2204  * inode's ea. This fid should not be altered during the life time
2205  * of the inode.
2206  *
2207  * \retval +ve, on success
2208  * \retval -ve, on error
2209  *
2210  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2211  */
2212 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2213                    const struct lu_fid *fid)
2214 {
2215         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2216         int                      rc;
2217
2218         lustre_lma_init(lma, fid);
2219         lustre_lma_swab(lma);
2220
2221         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2222                              XATTR_CREATE);
2223         /* Someone may created the EA by race. */
2224         if (unlikely(rc == -EEXIST))
2225                 rc = 0;
2226         return rc;
2227 }
2228
2229 /**
2230  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2231  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2232  * To have compatilibility with 1.8 ldiskfs driver we need to have
2233  * magic number at start of fid data.
2234  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2235  * its inmemory API.
2236  */
2237 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2238                                   const struct dt_rec *fid)
2239 {
2240         if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) {
2241                 param->edp_magic = 0;
2242                 return;
2243         }
2244
2245         param->edp_magic = LDISKFS_LUFID_MAGIC;
2246         param->edp_len =  sizeof(struct lu_fid) + 1;
2247         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2248 }
2249
2250 /**
2251  * Try to read the fid from inode ea into dt_rec.
2252  *
2253  * \param fid object fid.
2254  *
2255  * \retval 0 on success
2256  */
2257 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2258                           __u32 ino, struct lu_fid *fid,
2259                           struct osd_inode_id *id)
2260 {
2261         struct osd_thread_info *info  = osd_oti_get(env);
2262         struct inode           *inode;
2263         ENTRY;
2264
2265         osd_id_gen(id, ino, OSD_OII_NOGEN);
2266         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2267         if (IS_ERR(inode))
2268                 RETURN(PTR_ERR(inode));
2269
2270         iput(inode);
2271         RETURN(0);
2272 }
2273
2274 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2275                                         struct inode *dir,
2276                                         struct inode  *parent_dir,
2277                                         const struct dt_rec *dot_fid,
2278                                         const struct dt_rec *dot_dot_fid,
2279                                         struct osd_thandle *oth)
2280 {
2281         struct ldiskfs_dentry_param *dot_ldp;
2282         struct ldiskfs_dentry_param *dot_dot_ldp;
2283
2284         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2285         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2286
2287         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2288         dot_ldp->edp_magic = 0;
2289         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2290                                         dir, dot_ldp, dot_dot_ldp);
2291 }
2292
2293 /**
2294  * Create an local inode for remote entry
2295  */
2296 static struct inode *osd_create_remote_inode(const struct lu_env *env,
2297                                              struct osd_device *osd,
2298                                              struct osd_object *pobj,
2299                                              const struct lu_fid *fid,
2300                                              struct thandle *th)
2301 {
2302         struct osd_thread_info  *info = osd_oti_get(env);
2303         struct inode            *local;
2304         struct osd_thandle      *oh;
2305         int                     rc;
2306         ENTRY;
2307
2308         LASSERT(th);
2309         oh = container_of(th, struct osd_thandle, ot_super);
2310         LASSERT(oh->ot_handle->h_transaction != NULL);
2311
2312         /* FIXME: Insert index api needs to know the mode of
2313          * the remote object. Just use S_IFDIR for now */
2314         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2315         if (IS_ERR(local)) {
2316                 CERROR("%s: create local error %d\n", osd_name(osd),
2317                        (int)PTR_ERR(local));
2318                 RETURN(local);
2319         }
2320
2321         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2322                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2323                 (const struct dt_rec *)fid, oh);
2324         if (rc != 0) {
2325                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2326                         osd_name(osd), PFID(fid), rc);
2327                 RETURN(ERR_PTR(rc));
2328         }
2329
2330         RETURN(local);
2331 }
2332
2333 /**
2334  * Delete local inode for remote entry
2335  */
2336 static int osd_delete_remote_inode(const struct lu_env *env,
2337                                    struct osd_device *osd,
2338                                    const struct lu_fid *fid,
2339                                     __u32 ino, struct osd_thandle *oh)
2340 {
2341         struct osd_thread_info  *oti = osd_oti_get(env);
2342         struct osd_inode_id     *id = &oti->oti_id;
2343         struct inode            *inode;
2344         ENTRY;
2345
2346         id->oii_ino = le32_to_cpu(ino);
2347         id->oii_gen = OSD_OII_NOGEN;
2348         inode = osd_iget(oti, osd, id);
2349         if (IS_ERR(inode)) {
2350                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2351                        PFID(fid), id->oii_ino, id->oii_gen);
2352                 RETURN(PTR_ERR(inode));
2353         }
2354
2355         clear_nlink(inode);
2356         mark_inode_dirty(inode);
2357         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2358                 osd_name(osd), PFID(fid), inode->i_ino);
2359         iput(inode);
2360         RETURN(0);
2361 }
2362
2363 /**
2364  * OSD layer object create function for interoperability mode (b11826).
2365  * This is mostly similar to osd_object_create(). Only difference being, fid is
2366  * inserted into inode ea here.
2367  *
2368  * \retval   0, on success
2369  * \retval -ve, on error
2370  */
2371 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2372                                 struct lu_attr *attr,
2373                                 struct dt_allocation_hint *hint,
2374                                 struct dt_object_format *dof,
2375                                 struct thandle *th)
2376 {
2377         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2378         struct osd_object      *obj    = osd_dt_obj(dt);
2379         struct osd_thread_info *info   = osd_oti_get(env);
2380         int                     result;
2381
2382         ENTRY;
2383
2384         LASSERT(osd_invariant(obj));
2385         LASSERT(!dt_object_exists(dt));
2386         LASSERT(osd_write_locked(env, obj));
2387         LASSERT(th != NULL);
2388
2389         if (unlikely(fid_is_acct(fid)))
2390                 /* Quota files can't be created from the kernel any more,
2391                  * 'tune2fs -O quota' will take care of creating them */
2392                 RETURN(-EPERM);
2393
2394         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2395         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2396
2397         result = __osd_object_create(info, obj, attr, hint, dof, th);
2398         if ((result == 0) &&
2399             (fid_is_last_id(fid) ||
2400              !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)))
2401                 result = osd_ea_fid_set(info, obj->oo_inode, fid);
2402
2403         if (result == 0)
2404                 result = __osd_oi_insert(env, obj, fid, th);
2405
2406         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2407         LINVRNT(osd_invariant(obj));
2408         RETURN(result);
2409 }
2410
2411 static int osd_declare_object_ref_add(const struct lu_env *env,
2412                                       struct dt_object *dt,
2413                                       struct thandle *handle)
2414 {
2415         struct osd_thandle       *oh;
2416
2417         /* it's possible that object doesn't exist yet */
2418         LASSERT(handle != NULL);
2419
2420         oh = container_of0(handle, struct osd_thandle, ot_super);
2421         LASSERT(oh->ot_handle == NULL);
2422
2423         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2424                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2425
2426         return 0;
2427 }
2428
2429 /*
2430  * Concurrency: @dt is write locked.
2431  */
2432 static int osd_object_ref_add(const struct lu_env *env,
2433                               struct dt_object *dt, struct thandle *th)
2434 {
2435         struct osd_object *obj = osd_dt_obj(dt);
2436         struct inode      *inode = obj->oo_inode;
2437
2438         LINVRNT(osd_invariant(obj));
2439         LASSERT(dt_object_exists(dt));
2440         LASSERT(osd_write_locked(env, obj));
2441         LASSERT(th != NULL);
2442
2443         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2444
2445         /*
2446          * DIR_NLINK feature is set for compatibility reasons if:
2447          * 1) nlinks > LDISKFS_LINK_MAX, or
2448          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2449          *
2450          * It is easier to always set this flag (rather than check and set),
2451          * since it has less overhead, and the superblock will be dirtied
2452          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2453          * do not actually care whether this flag is set or not.
2454          */
2455         spin_lock(&obj->oo_guard);
2456         /* inc_nlink from 0 may cause WARN_ON */
2457         if(inode->i_nlink == 0)
2458                 set_nlink(inode, 1);
2459         else
2460                 inc_nlink(inode);
2461         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2462                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2463                     inode->i_nlink == 2)
2464                         set_nlink(inode, 1);
2465         }
2466         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2467         spin_unlock(&obj->oo_guard);
2468         inode->i_sb->s_op->dirty_inode(inode);
2469         LINVRNT(osd_invariant(obj));
2470
2471         return 0;
2472 }
2473
2474 static int osd_declare_object_ref_del(const struct lu_env *env,
2475                                       struct dt_object *dt,
2476                                       struct thandle *handle)
2477 {
2478         struct osd_thandle *oh;
2479
2480         LASSERT(dt_object_exists(dt));
2481         LASSERT(handle != NULL);
2482
2483         oh = container_of0(handle, struct osd_thandle, ot_super);
2484         LASSERT(oh->ot_handle == NULL);
2485
2486         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2487                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2488
2489         return 0;
2490 }
2491
2492 /*
2493  * Concurrency: @dt is write locked.
2494  */
2495 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2496                               struct thandle *th)
2497 {
2498         struct osd_object *obj = osd_dt_obj(dt);
2499         struct inode      *inode = obj->oo_inode;
2500
2501         LINVRNT(osd_invariant(obj));
2502         LASSERT(dt_object_exists(dt));
2503         LASSERT(osd_write_locked(env, obj));
2504         LASSERT(th != NULL);
2505
2506         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2507
2508         spin_lock(&obj->oo_guard);
2509         LASSERT(inode->i_nlink > 0);
2510         drop_nlink(inode);
2511         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2512          * then the nlink count is 1. Don't let it be set to 0 or the directory
2513          * inode will be deleted incorrectly. */
2514         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2515                 set_nlink(inode, 1);
2516         spin_unlock(&obj->oo_guard);
2517         inode->i_sb->s_op->dirty_inode(inode);
2518         LINVRNT(osd_invariant(obj));
2519
2520         return 0;
2521 }
2522
2523 /*
2524  * Get the 64-bit version for an inode.
2525  */
2526 static int osd_object_version_get(const struct lu_env *env,
2527                                   struct dt_object *dt, dt_obj_version_t *ver)
2528 {
2529         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2530
2531         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2532                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2533         *ver = LDISKFS_I(inode)->i_fs_version;
2534         return 0;
2535 }
2536
2537 /*
2538  * Concurrency: @dt is read locked.
2539  */
2540 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2541                          struct lu_buf *buf, const char *name,
2542                          struct lustre_capa *capa)
2543 {
2544         struct osd_object      *obj    = osd_dt_obj(dt);
2545         struct inode           *inode  = obj->oo_inode;
2546         struct osd_thread_info *info   = osd_oti_get(env);
2547         struct dentry          *dentry = &info->oti_obj_dentry;
2548
2549         /* version get is not real XATTR but uses xattr API */
2550         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2551                 /* for version we are just using xattr API but change inode
2552                  * field instead */
2553                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2554                 osd_object_version_get(env, dt, buf->lb_buf);
2555                 return sizeof(dt_obj_version_t);
2556         }
2557
2558         LASSERT(dt_object_exists(dt));
2559         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2560
2561         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2562                 return -EACCES;
2563
2564         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2565 }
2566
2567
2568 static int osd_declare_xattr_set(const struct lu_env *env,
2569                                  struct dt_object *dt,
2570                                  const struct lu_buf *buf, const char *name,
2571                                  int fl, struct thandle *handle)
2572 {
2573         struct osd_thandle *oh;
2574
2575         LASSERT(handle != NULL);
2576
2577         oh = container_of0(handle, struct osd_thandle, ot_super);
2578         LASSERT(oh->ot_handle == NULL);
2579
2580         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2581                              strcmp(name, XATTR_NAME_VERSION) == 0 ?
2582                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2583                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2584
2585         return 0;
2586 }
2587
2588 /*
2589  * Set the 64-bit version for object
2590  */
2591 static void osd_object_version_set(const struct lu_env *env,
2592                                    struct dt_object *dt,
2593                                    dt_obj_version_t *new_version)
2594 {
2595         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2596
2597         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2598                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2599
2600         LDISKFS_I(inode)->i_fs_version = *new_version;
2601         /** Version is set after all inode operations are finished,
2602          *  so we should mark it dirty here */
2603         inode->i_sb->s_op->dirty_inode(inode);
2604 }
2605
2606 /*
2607  * Concurrency: @dt is write locked.
2608  */
2609 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2610                          const struct lu_buf *buf, const char *name, int fl,
2611                          struct thandle *handle, struct lustre_capa *capa)
2612 {
2613         struct osd_object      *obj      = osd_dt_obj(dt);
2614         struct inode           *inode    = obj->oo_inode;
2615         struct osd_thread_info *info     = osd_oti_get(env);
2616         int                     fs_flags = 0;
2617
2618         LASSERT(handle != NULL);
2619
2620         /* version set is not real XATTR */
2621         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2622                 /* for version we are just using xattr API but change inode
2623                  * field instead */
2624                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2625                 osd_object_version_set(env, dt, buf->lb_buf);
2626                 return sizeof(dt_obj_version_t);
2627         }
2628
2629         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2630                 return -EACCES;
2631
2632         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2633         if (fl & LU_XATTR_REPLACE)
2634                 fs_flags |= XATTR_REPLACE;
2635
2636         if (fl & LU_XATTR_CREATE)
2637                 fs_flags |= XATTR_CREATE;
2638
2639         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2640                                fs_flags);
2641 }
2642
2643 /*
2644  * Concurrency: @dt is read locked.
2645  */
2646 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2647                           struct lu_buf *buf, struct lustre_capa *capa)
2648 {
2649         struct osd_object      *obj    = osd_dt_obj(dt);
2650         struct inode           *inode  = obj->oo_inode;
2651         struct osd_thread_info *info   = osd_oti_get(env);
2652         struct dentry          *dentry = &info->oti_obj_dentry;
2653
2654         LASSERT(dt_object_exists(dt));
2655         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2656         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2657
2658         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2659                 return -EACCES;
2660
2661         dentry->d_inode = inode;
2662         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2663 }
2664
2665 static int osd_declare_xattr_del(const struct lu_env *env,
2666                                  struct dt_object *dt, const char *name,
2667                                  struct thandle *handle)
2668 {
2669         struct osd_thandle *oh;
2670
2671         LASSERT(dt_object_exists(dt));
2672         LASSERT(handle != NULL);
2673
2674         oh = container_of0(handle, struct osd_thandle, ot_super);
2675         LASSERT(oh->ot_handle == NULL);
2676
2677         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2678                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2679
2680         return 0;
2681 }
2682
2683 /*
2684  * Concurrency: @dt is write locked.
2685  */
2686 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2687                          const char *name, struct thandle *handle,
2688                          struct lustre_capa *capa)
2689 {
2690         struct osd_object      *obj    = osd_dt_obj(dt);
2691         struct inode           *inode  = obj->oo_inode;
2692         struct osd_thread_info *info   = osd_oti_get(env);
2693         struct dentry          *dentry = &info->oti_obj_dentry;
2694         int                     rc;
2695
2696         LASSERT(dt_object_exists(dt));
2697         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2698         LASSERT(handle != NULL);
2699
2700         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2701                 return -EACCES;
2702
2703         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2704
2705         ll_vfs_dq_init(inode);
2706         dentry->d_inode = inode;
2707         rc = inode->i_op->removexattr(dentry, name);
2708         return rc;
2709 }
2710
2711 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2712                                      struct dt_object *dt,
2713                                      struct lustre_capa *old,
2714                                      __u64 opc)
2715 {
2716         struct osd_thread_info *info = osd_oti_get(env);
2717         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2718         struct osd_object *obj = osd_dt_obj(dt);
2719         struct osd_device *dev = osd_obj2dev(obj);
2720         struct lustre_capa_key *key = &info->oti_capa_key;
2721         struct lustre_capa *capa = &info->oti_capa;
2722         struct obd_capa *oc;
2723         struct md_capainfo *ci;
2724         int rc;
2725         ENTRY;
2726
2727         if (!dev->od_fl_capa)
2728                 RETURN(ERR_PTR(-ENOENT));
2729
2730         LASSERT(dt_object_exists(dt));
2731         LINVRNT(osd_invariant(obj));
2732
2733         /* renewal sanity check */
2734         if (old && osd_object_auth(env, dt, old, opc))
2735                 RETURN(ERR_PTR(-EACCES));
2736
2737         ci = md_capainfo(env);
2738         if (unlikely(!ci))
2739                 RETURN(ERR_PTR(-ENOENT));
2740
2741         switch (ci->mc_auth) {
2742         case LC_ID_NONE:
2743                 RETURN(NULL);
2744         case LC_ID_PLAIN:
2745                 capa->lc_uid = obj->oo_inode->i_uid;
2746                 capa->lc_gid = obj->oo_inode->i_gid;
2747                 capa->lc_flags = LC_ID_PLAIN;
2748                 break;
2749         case LC_ID_CONVERT: {
2750                 __u32 d[4], s[4];
2751
2752                 s[0] = obj->oo_inode->i_uid;
2753                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2754                 s[2] = obj->oo_inode->i_gid;
2755                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2756                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2757                 if (unlikely(rc))
2758                         RETURN(ERR_PTR(rc));
2759
2760                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2761                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2762                 capa->lc_flags = LC_ID_CONVERT;
2763                 break;
2764         }
2765         default:
2766                 RETURN(ERR_PTR(-EINVAL));
2767         }
2768
2769         capa->lc_fid = *fid;
2770         capa->lc_opc = opc;
2771         capa->lc_flags |= dev->od_capa_alg << 24;
2772         capa->lc_timeout = dev->od_capa_timeout;
2773         capa->lc_expiry = 0;
2774
2775         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2776         if (oc) {
2777                 LASSERT(!capa_is_expired(oc));
2778                 RETURN(oc);
2779         }
2780
2781         spin_lock(&capa_lock);
2782         *key = dev->od_capa_keys[1];
2783         spin_unlock(&capa_lock);
2784
2785         capa->lc_keyid = key->lk_keyid;
2786         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2787
2788         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2789         if (rc) {
2790                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2791                 RETURN(ERR_PTR(rc));
2792         }
2793
2794         oc = capa_add(dev->od_capa_hash, capa);
2795         RETURN(oc);
2796 }
2797
2798 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2799 {
2800         struct osd_object       *obj    = osd_dt_obj(dt);
2801         struct inode            *inode  = obj->oo_inode;
2802         struct osd_thread_info  *info   = osd_oti_get(env);
2803         struct dentry           *dentry = &info->oti_obj_dentry;
2804         struct file             *file   = &info->oti_file;
2805         int                     rc;
2806
2807         ENTRY;
2808
2809         dentry->d_inode = inode;
2810         file->f_dentry = dentry;
2811         file->f_mapping = inode->i_mapping;
2812         file->f_op = inode->i_fop;
2813         mutex_lock(&inode->i_mutex);
2814         rc = file->f_op->fsync(file, dentry, 0);
2815         mutex_unlock(&inode->i_mutex);
2816         RETURN(rc);
2817 }
2818
2819 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2820                         void **data)
2821 {
2822         struct osd_object *obj = osd_dt_obj(dt);
2823         ENTRY;
2824
2825         *data = (void *)obj->oo_inode;
2826         RETURN(0);
2827 }
2828
2829 /*
2830  * Index operations.
2831  */
2832
2833 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2834                            const struct dt_index_features *feat)
2835 {
2836         struct iam_descr *descr;
2837
2838         if (osd_object_is_root(o))
2839                 return feat == &dt_directory_features;
2840
2841         LASSERT(o->oo_dir != NULL);
2842
2843         descr = o->oo_dir->od_container.ic_descr;
2844         if (feat == &dt_directory_features) {
2845                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2846                         return 1;
2847                 else
2848                         return 0;
2849         } else {
2850                 return
2851                         feat->dif_keysize_min <= descr->id_key_size &&
2852                         descr->id_key_size <= feat->dif_keysize_max &&
2853                         feat->dif_recsize_min <= descr->id_rec_size &&
2854                         descr->id_rec_size <= feat->dif_recsize_max &&
2855                         !(feat->dif_flags & (DT_IND_VARKEY |
2856                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2857                         ergo(feat->dif_flags & DT_IND_UPDATE,
2858                              1 /* XXX check that object (and file system) is
2859                                 * writable */);
2860         }
2861 }
2862
2863 static int osd_iam_container_init(const struct lu_env *env,
2864                                   struct osd_object *obj,
2865                                   struct osd_directory *dir)
2866 {
2867         struct iam_container *bag = &dir->od_container;
2868         int result;
2869
2870         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2871         if (result != 0)
2872                 return result;
2873
2874         result = iam_container_setup(bag);
2875         if (result == 0)
2876                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2877         else
2878                 iam_container_fini(bag);
2879
2880         return result;
2881 }
2882
2883
2884 /*
2885  * Concurrency: no external locking is necessary.
2886  */
2887 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2888                          const struct dt_index_features *feat)
2889 {
2890         int                      result;
2891         int                      skip_iam = 0;
2892         struct osd_object       *obj = osd_dt_obj(dt);
2893
2894         LINVRNT(osd_invariant(obj));
2895         LASSERT(dt_object_exists(dt));
2896
2897         if (osd_object_is_root(obj)) {
2898                 dt->do_index_ops = &osd_index_ea_ops;
2899                 result = 0;
2900         } else if (feat == &dt_directory_features) {
2901                 dt->do_index_ops = &osd_index_ea_ops;
2902                 if (S_ISDIR(obj->oo_inode->i_mode))
2903                         result = 0;
2904                 else
2905                         result = -ENOTDIR;
2906                 skip_iam = 1;
2907         } else if (unlikely(feat == &dt_otable_features)) {
2908                 dt->do_index_ops = &osd_otable_ops;
2909                 return 0;
2910         } else if (unlikely(feat == &dt_acct_features)) {
2911                 dt->do_index_ops = &osd_acct_index_ops;
2912                 result = 0;
2913                 skip_iam = 1;
2914         } else if (!osd_has_index(obj)) {
2915                 struct osd_directory *dir;
2916
2917                 OBD_ALLOC_PTR(dir);
2918                 if (dir != NULL) {
2919
2920                         spin_lock(&obj->oo_guard);
2921                         if (obj->oo_dir == NULL)
2922                                 obj->oo_dir = dir;
2923                         else
2924                                 /*
2925                                  * Concurrent thread allocated container data.
2926                                  */
2927                                 OBD_FREE_PTR(dir);
2928                         spin_unlock(&obj->oo_guard);
2929                         /*
2930                          * Now, that we have container data, serialize its
2931                          * initialization.
2932                          */
2933                         down_write(&obj->oo_ext_idx_sem);
2934                         /*
2935                          * recheck under lock.
2936                          */
2937                         if (!osd_has_index(obj))
2938                                 result = osd_iam_container_init(env, obj, dir);
2939                         else
2940                                 result = 0;
2941                         up_write(&obj->oo_ext_idx_sem);
2942                 } else {
2943                         result = -ENOMEM;
2944                 }
2945         } else {
2946                 result = 0;
2947         }
2948
2949         if (result == 0 && skip_iam == 0) {
2950                 if (!osd_iam_index_probe(env, obj, feat))
2951                         result = -ENOTDIR;
2952         }
2953         LINVRNT(osd_invariant(obj));
2954
2955         if (is_quota_glb_feat(feat))
2956                 result = osd_quota_migration(env, dt, feat);
2957
2958         return result;
2959 }
2960
2961 static int osd_otable_it_attr_get(const struct lu_env *env,
2962                                  struct dt_object *dt,
2963                                  struct lu_attr *attr,
2964                                  struct lustre_capa *capa)
2965 {
2966         attr->la_valid = 0;
2967         return 0;
2968 }
2969
2970 static const struct dt_object_operations osd_obj_ops = {
2971         .do_read_lock         = osd_object_read_lock,
2972         .do_write_lock        = osd_object_write_lock,
2973         .do_read_unlock       = osd_object_read_unlock,
2974         .do_write_unlock      = osd_object_write_unlock,
2975         .do_write_locked      = osd_object_write_locked,
2976         .do_attr_get          = osd_attr_get,
2977         .do_declare_attr_set  = osd_declare_attr_set,
2978         .do_attr_set          = osd_attr_set,
2979         .do_ah_init           = osd_ah_init,
2980         .do_declare_create    = osd_declare_object_create,
2981         .do_create            = osd_object_create,
2982         .do_declare_destroy   = osd_declare_object_destroy,
2983         .do_destroy           = osd_object_destroy,
2984         .do_index_try         = osd_index_try,
2985         .do_declare_ref_add   = osd_declare_object_ref_add,
2986         .do_ref_add           = osd_object_ref_add,
2987         .do_declare_ref_del   = osd_declare_object_ref_del,
2988         .do_ref_del           = osd_object_ref_del,
2989         .do_xattr_get         = osd_xattr_get,
2990         .do_declare_xattr_set = osd_declare_xattr_set,
2991         .do_xattr_set         = osd_xattr_set,
2992         .do_declare_xattr_del = osd_declare_xattr_del,
2993         .do_xattr_del         = osd_xattr_del,
2994         .do_xattr_list        = osd_xattr_list,
2995         .do_capa_get          = osd_capa_get,
2996         .do_object_sync       = osd_object_sync,
2997         .do_data_get          = osd_data_get,
2998 };
2999
3000 /**
3001  * dt_object_operations for interoperability mode
3002  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3003  */
3004 static const struct dt_object_operations osd_obj_ea_ops = {
3005         .do_read_lock         = osd_object_read_lock,
3006         .do_write_lock        = osd_object_write_lock,
3007         .do_read_unlock       = osd_object_read_unlock,
3008         .do_write_unlock      = osd_object_write_unlock,
3009         .do_write_locked      = osd_object_write_locked,
3010         .do_attr_get          = osd_attr_get,
3011         .do_declare_attr_set  = osd_declare_attr_set,
3012         .do_attr_set          = osd_attr_set,
3013         .do_ah_init           = osd_ah_init,
3014         .do_declare_create    = osd_declare_object_create,
3015         .do_create            = osd_object_ea_create,
3016         .do_declare_destroy   = osd_declare_object_destroy,
3017         .do_destroy           = osd_object_destroy,
3018         .do_index_try         = osd_index_try,
3019         .do_declare_ref_add   = osd_declare_object_ref_add,
3020         .do_ref_add           = osd_object_ref_add,
3021         .do_declare_ref_del   = osd_declare_object_ref_del,
3022         .do_ref_del           = osd_object_ref_del,
3023         .do_xattr_get         = osd_xattr_get,
3024         .do_declare_xattr_set = osd_declare_xattr_set,
3025         .do_xattr_set         = osd_xattr_set,
3026         .do_declare_xattr_del = osd_declare_xattr_del,
3027         .do_xattr_del         = osd_xattr_del,
3028         .do_xattr_list        = osd_xattr_list,
3029         .do_capa_get          = osd_capa_get,
3030         .do_object_sync       = osd_object_sync,
3031         .do_data_get          = osd_data_get,
3032 };
3033
3034 static const struct dt_object_operations osd_obj_otable_it_ops = {
3035         .do_attr_get    = osd_otable_it_attr_get,
3036         .do_index_try   = osd_index_try,
3037 };
3038
3039 static int osd_index_declare_iam_delete(const struct lu_env *env,
3040                                         struct dt_object *dt,
3041                                         const struct dt_key *key,
3042                                         struct thandle *handle)
3043 {
3044         struct osd_thandle    *oh;
3045
3046         oh = container_of0(handle, struct osd_thandle, ot_super);
3047         LASSERT(oh->ot_handle == NULL);
3048
3049         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3050                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3051
3052         return 0;
3053 }
3054
3055 /**
3056  *      delete a (key, value) pair from index \a dt specified by \a key
3057  *
3058  *      \param  dt      osd index object
3059  *      \param  key     key for index
3060  *      \param  rec     record reference
3061  *      \param  handle  transaction handler
3062  *
3063  *      \retval  0  success
3064  *      \retval -ve   failure
3065  */
3066
3067 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3068                                 const struct dt_key *key,
3069                                 struct thandle *handle,
3070                                 struct lustre_capa *capa)
3071 {
3072         struct osd_thread_info *oti = osd_oti_get(env);
3073         struct osd_object      *obj = osd_dt_obj(dt);
3074         struct osd_thandle     *oh;
3075         struct iam_path_descr  *ipd;
3076         struct iam_container   *bag = &obj->oo_dir->od_container;
3077         int                     rc;
3078
3079         ENTRY;
3080
3081         LINVRNT(osd_invariant(obj));
3082         LASSERT(dt_object_exists(dt));
3083         LASSERT(bag->ic_object == obj->oo_inode);
3084         LASSERT(handle != NULL);
3085
3086         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3087                 RETURN(-EACCES);
3088
3089         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3090
3091         ipd = osd_idx_ipd_get(env, bag);
3092         if (unlikely(ipd == NULL))
3093                 RETURN(-ENOMEM);
3094
3095         oh = container_of0(handle, struct osd_thandle, ot_super);
3096         LASSERT(oh->ot_handle != NULL);
3097         LASSERT(oh->ot_handle->h_transaction != NULL);
3098
3099         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3100                 /* swab quota uid/gid provided by caller */
3101                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3102                 key = (const struct dt_key *)&oti->oti_quota_id;
3103         }
3104
3105         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3106         osd_ipd_put(env, bag, ipd);
3107         LINVRNT(osd_invariant(obj));
3108         RETURN(rc);
3109 }
3110
3111 static int osd_index_declare_ea_delete(const struct lu_env *env,
3112                                        struct dt_object *dt,
3113                                        const struct dt_key *key,
3114                                        struct thandle *handle)
3115 {
3116         struct osd_thandle *oh;
3117         struct inode       *inode;
3118         int                 rc;
3119         ENTRY;
3120
3121         LASSERT(dt_object_exists(dt));
3122         LASSERT(handle != NULL);
3123
3124         oh = container_of0(handle, struct osd_thandle, ot_super);
3125         LASSERT(oh->ot_handle == NULL);
3126
3127         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3128                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3129
3130         inode = osd_dt_obj(dt)->oo_inode;
3131         LASSERT(inode);
3132
3133         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3134                                    true, true, NULL, false);
3135         RETURN(rc);
3136 }
3137
3138 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3139                                           struct dt_rec *fid)
3140 {
3141         struct osd_fid_pack *rec;
3142         int                  rc = -ENODATA;
3143
3144         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3145                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3146                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3147         }
3148         RETURN(rc);
3149 }
3150
3151 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3152                           struct lu_fid *fid)
3153 {
3154         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
3155         struct seq_server_site  *ss = osd_seq_site(osd);
3156         int                     rc;
3157         ENTRY;
3158
3159         if ((!fid_is_norm(fid) && !fid_is_igif(fid)) || ss == NULL)
3160                 RETURN(0);
3161
3162         rc = osd_fld_lookup(env, osd, fid, range);
3163         if (rc != 0) {
3164                 CERROR("%s: Can not lookup fld for "DFID"\n",
3165                        osd_name(osd), PFID(fid));
3166                 RETURN(rc);
3167         }
3168
3169         RETURN(ss->ss_node_id != range->lsr_index);
3170 }
3171
3172 /**
3173  * Index delete function for interoperability mode (b11826).
3174  * It will remove the directory entry added by osd_index_ea_insert().
3175  * This entry is needed to maintain name->fid mapping.
3176  *
3177  * \param key,  key i.e. file entry to be deleted
3178  *
3179  * \retval   0, on success
3180  * \retval -ve, on error
3181  */
3182 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3183                                const struct dt_key *key,
3184                                struct thandle *handle,
3185                                struct lustre_capa *capa)
3186 {
3187         struct osd_object          *obj    = osd_dt_obj(dt);
3188         struct inode               *dir    = obj->oo_inode;
3189         struct dentry              *dentry;
3190         struct osd_thandle         *oh;
3191         struct ldiskfs_dir_entry_2 *de = NULL;
3192         struct buffer_head         *bh;
3193         struct htree_lock          *hlock = NULL;
3194         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3195         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3196         int                        rc;
3197         ENTRY;
3198
3199         LINVRNT(osd_invariant(obj));
3200         LASSERT(dt_object_exists(dt));
3201         LASSERT(handle != NULL);
3202
3203         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3204
3205         oh = container_of(handle, struct osd_thandle, ot_super);
3206         LASSERT(oh->ot_handle != NULL);
3207         LASSERT(oh->ot_handle->h_transaction != NULL);
3208
3209         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3210                 RETURN(-EACCES);
3211
3212         ll_vfs_dq_init(dir);
3213         dentry = osd_child_dentry_get(env, obj,
3214                                       (char *)key, strlen((char *)key));
3215
3216         if (obj->oo_hl_head != NULL) {
3217                 hlock = osd_oti_get(env)->oti_hlock;
3218                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3219                                    dir, LDISKFS_HLOCK_DEL);
3220         } else {
3221                 down_write(&obj->oo_ext_idx_sem);
3222         }
3223
3224         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3225         if (bh) {
3226                 rc = ldiskfs_delete_entry(oh->ot_handle,
3227                                           dir, de, bh);
3228                 brelse(bh);
3229         } else {
3230                 rc = -ENOENT;
3231         }
3232         if (hlock != NULL)
3233                 ldiskfs_htree_unlock(hlock);
3234         else
3235                 up_write(&obj->oo_ext_idx_sem);
3236
3237         if (rc != 0)
3238                 GOTO(out, rc);
3239
3240         LASSERT(de != NULL);
3241         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3242         if (rc == 0 && osd_remote_fid(env, osd, fid)) {
3243                 __u32 ino = le32_to_cpu(de->inode);
3244
3245                 rc = osd_delete_remote_inode(env, osd, fid, ino, oh);
3246                 if (rc != 0)
3247                         CERROR("%s: del local inode "DFID": rc = %d\n",
3248                                 osd_name(osd), PFID(fid), rc);
3249         } else {
3250                 if (rc == -ENODATA)
3251                         rc = 0;
3252         }
3253 out:
3254
3255         LASSERT(osd_invariant(obj));
3256         RETURN(rc);
3257 }
3258
3259 /**
3260  *      Lookup index for \a key and copy record to \a rec.
3261  *
3262  *      \param  dt      osd index object
3263  *      \param  key     key for index
3264  *      \param  rec     record reference
3265  *
3266  *      \retval  +ve  success : exact mach
3267  *      \retval  0    return record with key not greater than \a key
3268  *      \retval -ve   failure
3269  */
3270 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3271                                 struct dt_rec *rec, const struct dt_key *key,
3272                                 struct lustre_capa *capa)
3273 {
3274         struct osd_object      *obj = osd_dt_obj(dt);
3275         struct iam_path_descr  *ipd;
3276         struct iam_container   *bag = &obj->oo_dir->od_container;
3277         struct osd_thread_info *oti = osd_oti_get(env);
3278         struct iam_iterator    *it = &oti->oti_idx_it;
3279         struct iam_rec         *iam_rec;
3280         int                     rc;
3281
3282         ENTRY;
3283
3284         LASSERT(osd_invariant(obj));
3285         LASSERT(dt_object_exists(dt));
3286         LASSERT(bag->ic_object == obj->oo_inode);
3287
3288         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3289                 RETURN(-EACCES);
3290
3291         ipd = osd_idx_ipd_get(env, bag);
3292         if (IS_ERR(ipd))
3293                 RETURN(-ENOMEM);
3294
3295         /* got ipd now we can start iterator. */
3296         iam_it_init(it, bag, 0, ipd);
3297
3298         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3299                 /* swab quota uid/gid provided by caller */
3300                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3301                 key = (const struct dt_key *)&oti->oti_quota_id;
3302         }
3303
3304         rc = iam_it_get(it, (struct iam_key *)key);
3305         if (rc >= 0) {
3306                 if (S_ISDIR(obj->oo_inode->i_mode))
3307                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3308                 else
3309                         iam_rec = (struct iam_rec *) rec;
3310
3311                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3312
3313                 if (S_ISDIR(obj->oo_inode->i_mode))
3314                         osd_fid_unpack((struct lu_fid *) rec,
3315                                        (struct osd_fid_pack *)iam_rec);
3316                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3317                         osd_quota_unpack(obj, rec);
3318         }
3319
3320         iam_it_put(it);
3321         iam_it_fini(it);
3322         osd_ipd_put(env, bag, ipd);
3323
3324         LINVRNT(osd_invariant(obj));
3325
3326         RETURN(rc);
3327 }
3328
3329 static int osd_index_declare_iam_insert(const struct lu_env *env,
3330                                         struct dt_object *dt,
3331                                         const struct dt_rec *rec,
3332                                         const struct dt_key *key,
3333                                         struct thandle *handle)
3334 {
3335         struct osd_thandle *oh;
3336
3337         LASSERT(dt_object_exists(dt));
3338         LASSERT(handle != NULL);
3339
3340         oh = container_of0(handle, struct osd_thandle, ot_super);
3341         LASSERT(oh->ot_handle == NULL);
3342
3343         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3344                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3345
3346         return 0;
3347 }
3348
3349 /**
3350  *      Inserts (key, value) pair in \a dt index object.
3351  *
3352  *      \param  dt      osd index object
3353  *      \param  key     key for index
3354  *      \param  rec     record reference
3355  *      \param  th      transaction handler
3356  *
3357  *      \retval  0  success
3358  *      \retval -ve failure
3359  */
3360 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3361                                 const struct dt_rec *rec,
3362                                 const struct dt_key *key, struct thandle *th,
3363                                 struct lustre_capa *capa, int ignore_quota)
3364 {
3365         struct osd_object     *obj = osd_dt_obj(dt);
3366         struct iam_path_descr *ipd;
3367         struct osd_thandle    *oh;
3368         struct iam_container  *bag = &obj->oo_dir->od_container;
3369         struct osd_thread_info *oti = osd_oti_get(env);
3370         struct iam_rec         *iam_rec;
3371         int                     rc;
3372
3373         ENTRY;
3374
3375         LINVRNT(osd_invariant(obj));
3376         LASSERT(dt_object_exists(dt));
3377         LASSERT(bag->ic_object == obj->oo_inode);
3378         LASSERT(th != NULL);
3379
3380         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3381                 RETURN(-EACCES);
3382
3383         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3384
3385         ipd = osd_idx_ipd_get(env, bag);
3386         if (unlikely(ipd == NULL))
3387                 RETURN(-ENOMEM);
3388
3389         oh = container_of0(th, struct osd_thandle, ot_super);
3390         LASSERT(oh->ot_handle != NULL);
3391         LASSERT(oh->ot_handle->h_transaction != NULL);
3392         if (S_ISDIR(obj->oo_inode->i_mode)) {
3393                 iam_rec = (struct iam_rec *)oti->oti_ldp;
3394                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3395         } else if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3396                 /* pack quota uid/gid */
3397                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3398                 key = (const struct dt_key *)&oti->oti_quota_id;
3399                 /* pack quota record */
3400                 rec = osd_quota_pack(obj, rec, &oti->oti_quota_rec);
3401                 iam_rec = (struct iam_rec *)rec;
3402         } else {
3403                 iam_rec = (struct iam_rec *)rec;
3404         }
3405
3406         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3407                         iam_rec, ipd);
3408         osd_ipd_put(env, bag, ipd);
3409         LINVRNT(osd_invariant(obj));
3410         RETURN(rc);
3411 }
3412
3413 /**
3414  * Calls ldiskfs_add_entry() to add directory entry
3415  * into the directory. This is required for
3416  * interoperability mode (b11826)
3417  *
3418  * \retval   0, on success
3419  * \retval -ve, on error
3420  */
3421 static int __osd_ea_add_rec(struct osd_thread_info *info,
3422                             struct osd_object *pobj, struct inode  *cinode,
3423                             const char *name, const struct dt_rec *fid,
3424                             struct htree_lock *hlock, struct thandle *th)
3425 {
3426         struct ldiskfs_dentry_param *ldp;
3427         struct dentry               *child;
3428         struct osd_thandle          *oth;
3429         int                          rc;
3430
3431         oth = container_of(th, struct osd_thandle, ot_super);
3432         LASSERT(oth->ot_handle != NULL);
3433         LASSERT(oth->ot_handle->h_transaction != NULL);
3434         LASSERT(pobj->oo_inode);
3435
3436         ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3437         if (unlikely(pobj->oo_inode ==
3438                      osd_sb(osd_obj2dev(pobj))->s_root->d_inode))
3439                 ldp->edp_magic = 0;
3440         else
3441                 osd_get_ldiskfs_dirent_param(ldp, fid);
3442         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3443         child->d_fsdata = (void *)ldp;
3444         ll_vfs_dq_init(pobj->oo_inode);
3445         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3446
3447         RETURN(rc);
3448 }
3449
3450 /**
3451  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3452  * into the directory.Also sets flags into osd object to
3453  * indicate dot and dotdot are created. This is required for
3454  * interoperability mode (b11826)
3455  *
3456  * \param dir   directory for dot and dotdot fixup.
3457  * \param obj   child object for linking
3458  *
3459  * \retval   0, on success
3460  * \retval -ve, on error
3461  */
3462 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3463                               struct osd_object *dir,
3464                               struct inode  *parent_dir, const char *name,
3465                               const struct dt_rec *dot_fid,
3466                               const struct dt_rec *dot_dot_fid,
3467                               struct thandle *th)
3468 {
3469         struct inode                *inode = dir->oo_inode;
3470         struct osd_thandle          *oth;
3471         int result = 0;
3472
3473         oth = container_of(th, struct osd_thandle, ot_super);
3474         LASSERT(oth->ot_handle->h_transaction != NULL);
3475         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3476
3477         if (strcmp(name, dot) == 0) {
3478                 if (dir->oo_compat_dot_created) {
3479                         result = -EEXIST;
3480                 } else {
3481                         LASSERT(inode == parent_dir);
3482                         dir->oo_compat_dot_created = 1;
3483                         result = 0;
3484                 }
3485         } else if (strcmp(name, dotdot) == 0) {
3486                 if (!dir->oo_compat_dot_created)
3487                         return -EINVAL;
3488                 /* in case of rename, dotdot is already created */
3489                 if (dir->oo_compat_dotdot_created) {
3490                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3491                                                 dot_dot_fid, NULL, th);
3492                 }
3493
3494                 result = osd_add_dot_dotdot_internal(info, dir->oo_inode,
3495                                                 parent_dir, dot_fid,
3496                                                 dot_dot_fid, oth);
3497                 if (result == 0)
3498                         dir->oo_compat_dotdot_created = 1;
3499         }
3500
3501         return result;
3502 }
3503
3504
3505 /**
3506  * It will call the appropriate osd_add* function and return the
3507  * value, return by respective functions.
3508  */
3509 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3510                           struct inode *cinode, const char *name,
3511                           const struct dt_rec *fid, struct thandle *th)
3512 {
3513         struct osd_thread_info *info   = osd_oti_get(env);
3514         struct htree_lock      *hlock;
3515         int                     rc;
3516
3517         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3518
3519         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3520                                                    name[2] =='\0'))) {
3521                 if (hlock != NULL) {
3522                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3523                                            pobj->oo_inode, 0);
3524                 } else {
3525                         down_write(&pobj->oo_ext_idx_sem);
3526                 }
3527                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3528                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3529                                         fid, th);
3530         } else {
3531                 if (hlock != NULL) {
3532                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3533                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3534                 } else {
3535                         down_write(&pobj->oo_ext_idx_sem);
3536                 }
3537
3538                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3539                                       hlock, th);
3540         }
3541         if (hlock != NULL)
3542                 ldiskfs_htree_unlock(hlock);
3543         else
3544                 up_write(&pobj->oo_ext_idx_sem);
3545
3546         return rc;
3547 }
3548
3549 static void
3550 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3551                       struct osd_idmap_cache *oic)
3552 {
3553         struct osd_scrub    *scrub = &dev->od_scrub;
3554         struct lu_fid       *fid   = &oic->oic_fid;
3555         struct osd_inode_id *id    = &oti->oti_id;
3556         int                  once  = 0;
3557         int                  rc;
3558         ENTRY;
3559
3560         if (!fid_is_norm(fid) && !fid_is_igif(fid))
3561                 RETURN_EXIT;
3562
3563 again:
3564         rc = osd_oi_lookup(oti, dev, fid, id, true);
3565         if (rc != 0 && rc != -ENOENT)
3566                 RETURN_EXIT;
3567
3568         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3569                 RETURN_EXIT;
3570
3571         if (thread_is_running(&scrub->os_thread)) {
3572                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3573                 /* There is race condition between osd_oi_lookup and OI scrub.
3574                  * The OI scrub finished just after osd_oi_lookup() failure.
3575                  * Under such case, it is unnecessary to trigger OI scrub again,
3576                  * but try to call osd_oi_lookup() again. */
3577                 if (unlikely(rc == -EAGAIN))
3578                         goto again;
3579
3580                 RETURN_EXIT;
3581         }
3582
3583         if (!dev->od_noscrub && ++once == 1) {
3584                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3585                        PFID(fid));
3586                 rc = osd_scrub_start(dev);
3587                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3588                                ", rc = %d [2]\n",
3589                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3590                                PFID(fid), rc);
3591                 if (rc == 0)
3592                         goto again;
3593         }
3594
3595         EXIT;
3596 }
3597
3598 /**
3599  * Calls ->lookup() to find dentry. From dentry get inode and
3600  * read inode's ea to get fid. This is required for  interoperability
3601  * mode (b11826)
3602  *
3603  * \retval   0, on success
3604  * \retval -ve, on error
3605  */
3606 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3607                              struct dt_rec *rec, const struct dt_key *key)
3608 {
3609         struct inode               *dir    = obj->oo_inode;
3610         struct dentry              *dentry;
3611         struct ldiskfs_dir_entry_2 *de;
3612         struct buffer_head         *bh;
3613         struct lu_fid              *fid = (struct lu_fid *) rec;
3614         struct htree_lock          *hlock = NULL;
3615         int                         ino;
3616         int                         rc;
3617         ENTRY;
3618
3619         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3620
3621         dentry = osd_child_dentry_get(env, obj,
3622                                       (char *)key, strlen((char *)key));
3623
3624         if (obj->oo_hl_head != NULL) {
3625                 hlock = osd_oti_get(env)->oti_hlock;
3626                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3627                                    dir, LDISKFS_HLOCK_LOOKUP);
3628         } else {
3629                 down_read(&obj->oo_ext_idx_sem);
3630         }
3631
3632         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3633         if (bh) {
3634                 struct osd_thread_info *oti = osd_oti_get(env);
3635                 struct osd_inode_id *id = &oti->oti_id;
3636                 struct osd_idmap_cache *oic = &oti->oti_cache;
3637                 struct osd_device *dev = osd_obj2dev(obj);
3638                 struct osd_scrub *scrub = &dev->od_scrub;
3639                 struct scrub_file *sf = &scrub->os_file;
3640
3641                 ino = le32_to_cpu(de->inode);
3642                 rc = osd_get_fid_from_dentry(de, rec);
3643
3644                 /* done with de, release bh */
3645                 brelse(bh);
3646                 if (rc != 0)
3647                         rc = osd_ea_fid_get(env, obj, ino, fid, id);
3648                 else
3649                         osd_id_gen(id, ino, OSD_OII_NOGEN);
3650                 if (rc != 0 || osd_remote_fid(env, dev, fid)) {
3651                         fid_zero(&oic->oic_fid);
3652                         GOTO(out, rc);
3653                 }
3654
3655                 oic->oic_lid = *id;
3656                 oic->oic_fid = *fid;
3657                 if ((scrub->os_pos_current <= ino) &&
3658                     ((sf->sf_flags & SF_INCONSISTENT) ||
3659                      (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
3660                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3661                                       sf->sf_oi_bitmap)))
3662                         osd_consistency_check(oti, dev, oic);
3663         } else {
3664                 rc = -ENOENT;
3665         }
3666
3667         GOTO(out, rc);
3668
3669 out:
3670         if (hlock != NULL)
3671                 ldiskfs_htree_unlock(hlock);
3672         else
3673                 up_read(&obj->oo_ext_idx_sem);
3674         return rc;
3675 }
3676
3677 /**
3678  * Find the osd object for given fid.
3679  *
3680  * \param fid need to find the osd object having this fid
3681  *
3682  * \retval osd_object on success
3683  * \retval        -ve on error
3684  */
3685 struct osd_object *osd_object_find(const struct lu_env *env,
3686                                    struct dt_object *dt,
3687                                    const struct lu_fid *fid)
3688 {
3689         struct lu_device  *ludev = dt->do_lu.lo_dev;
3690         struct osd_object *child = NULL;
3691         struct lu_object  *luch;
3692         struct lu_object  *lo;
3693
3694         /*
3695          * at this point topdev might not exist yet
3696          * (i.e. MGS is preparing profiles). so we can
3697          * not rely on topdev and instead lookup with
3698          * our device passed as topdev. this can't work
3699          * if the object isn't cached yet (as osd doesn't
3700          * allocate lu_header). IOW, the object must be
3701          * in the cache, otherwise lu_object_alloc() crashes
3702          * -bzzz
3703          */
3704         luch = lu_object_find_at(env, ludev, fid, NULL);
3705         if (!IS_ERR(luch)) {
3706                 if (lu_object_exists(luch)) {
3707                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3708                         if (lo != NULL)
3709                                 child = osd_obj(lo);
3710                         else
3711                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3712                                                 "lu_object can't be located"
3713                                                 DFID"\n", PFID(fid));
3714
3715                         if (child == NULL) {
3716                                 lu_object_put(env, luch);
3717                                 CERROR("Unable to get osd_object\n");
3718                                 child = ERR_PTR(-ENOENT);
3719                         }
3720                 } else {
3721                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3722                                         "lu_object does not exists "DFID"\n",
3723                                         PFID(fid));
3724                         lu_object_put(env, luch);
3725                         child = ERR_PTR(-ENOENT);
3726                 }
3727         } else
3728                 child = (void *)luch;
3729
3730         return child;
3731 }
3732
3733 /**
3734  * Put the osd object once done with it.
3735  *
3736  * \param obj osd object that needs to be put
3737  */
3738 static inline void osd_object_put(const struct lu_env *env,
3739                                   struct osd_object *obj)
3740 {
3741         lu_object_put(env, &obj->oo_dt.do_lu);
3742 }
3743
3744 static int osd_index_declare_ea_insert(const struct lu_env *env,
3745                                        struct dt_object *dt,
3746                                        const struct dt_rec *rec,
3747                                        const struct dt_key *key,
3748                                        struct thandle *handle)
3749 {
3750         struct osd_thandle      *oh;
3751         struct osd_device       *osd   = osd_dev(dt->do_lu.lo_dev);
3752         struct lu_fid           *fid = (struct lu_fid *)rec;
3753         int                     rc;
3754         ENTRY;
3755
3756         LASSERT(dt_object_exists(dt));
3757         LASSERT(handle != NULL);
3758
3759         oh = container_of0(handle, struct osd_thandle, ot_super);
3760         LASSERT(oh->ot_handle == NULL);
3761
3762         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3763                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3764
3765         if (osd_dt_obj(dt)->oo_inode == NULL) {
3766                 const char *name  = (const char *)key;
3767                 /* Object is not being created yet. Only happens when
3768                  *     1. declare directory create
3769                  *     2. declare insert .
3770                  *     3. declare insert ..
3771                  */
3772                 LASSERT(strcmp(name, dotdot) == 0 || strcmp(name, dot) == 0);
3773         } else {
3774                 struct inode *inode = osd_dt_obj(dt)->oo_inode;
3775
3776                 /* We ignore block quota on meta pool (MDTs), so needn't
3777                  * calculate how many blocks will be consumed by this index
3778                  * insert */
3779                 rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0,
3780                                            oh, true, true, NULL, false);
3781         }
3782
3783         if (fid == NULL)
3784                 RETURN(0);
3785
3786         rc = osd_remote_fid(env, osd, fid);
3787         if (rc <= 0)
3788                 RETURN(rc);
3789
3790         rc = 0;
3791
3792         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3793                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
3794         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3795                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3796         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3797                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3798
3799         RETURN(rc);
3800 }
3801
3802 /**
3803  * Index add function for interoperability mode (b11826).
3804  * It will add the directory entry.This entry is needed to
3805  * maintain name->fid mapping.
3806  *
3807  * \param key it is key i.e. file entry to be inserted
3808  * \param rec it is value of given key i.e. fid
3809  *
3810  * \retval   0, on success
3811  * \retval -ve, on error
3812  */
3813 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3814                                const struct dt_rec *rec,
3815                                const struct dt_key *key, struct thandle *th,
3816                                struct lustre_capa *capa, int ignore_quota)
3817 {
3818         struct osd_object       *obj = osd_dt_obj(dt);
3819         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
3820         struct lu_fid           *fid = (struct lu_fid *) rec;
3821         const char              *name = (const char *)key;
3822         struct osd_thread_info  *oti   = osd_oti_get(env);
3823         struct osd_inode_id     *id    = &oti->oti_id;
3824         struct inode            *child_inode = NULL;
3825         struct osd_object       *child = NULL;
3826         int                     rc;
3827         ENTRY;
3828
3829         LASSERT(osd_invariant(obj));
3830         LASSERT(dt_object_exists(dt));
3831         LASSERT(th != NULL);
3832
3833         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3834
3835         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3836                 RETURN(-EACCES);
3837
3838         LASSERTF(fid_is_sane(fid), "fid"DFID" is insane!", PFID(fid));
3839
3840         rc = osd_remote_fid(env, osd, fid);
3841         if (rc < 0) {
3842                 CERROR("%s: Can not find object "DFID" rc %d\n",
3843                        osd_name(osd), PFID(fid), rc);
3844                 RETURN(rc);
3845         }
3846
3847         if (rc == 1) {
3848                 /* Insert remote entry */
3849                 if (strcmp(name, dotdot) == 0 && strlen(name) == 2) {
3850                         struct osd_mdobj_map    *omm = osd->od_mdt_map;
3851                         struct osd_thandle      *oh;
3852
3853                         /* If parent on remote MDT, we need put this object
3854                          * under AGENT */
3855                         oh = container_of(th, typeof(*oh), ot_super);
3856                         rc = osd_create_agent_inode(env, osd, obj, oh);
3857                         if (rc != 0) {
3858                                 CERROR("%s: add agent "DFID" error: rc = %d\n",
3859                                        osd_name(osd),
3860                                        PFID(lu_object_fid(&dt->do_lu)), rc);
3861                                 RETURN(rc);
3862                         }
3863
3864                         child_inode = igrab(omm->omm_agent_dentry->d_inode);
3865                 } else {
3866                         child_inode = osd_create_remote_inode(env, osd, obj,
3867                                                               fid, th);
3868                         if (IS_ERR(child_inode))
3869                                 RETURN(PTR_ERR(child_inode));
3870                 }
3871         } else {
3872                 /* Insert local entry */
3873                 child = osd_object_find(env, dt, fid);
3874                 if (IS_ERR(child)) {
3875                         CERROR("%s: Can not find object "DFID"%u:%u: rc = %d\n",
3876                                osd_name(osd), PFID(fid),
3877                                id->oii_ino, id->oii_gen,
3878                                (int)PTR_ERR(child_inode));
3879                         RETURN(PTR_ERR(child_inode));
3880                 }
3881                 child_inode = igrab(child->oo_inode);
3882         }
3883
3884         rc = osd_ea_add_rec(env, obj, child_inode, name, rec, th);
3885
3886         iput(child_inode);
3887         if (child != NULL)
3888                 osd_object_put(env, child);
3889         LASSERT(osd_invariant(obj));
3890         RETURN(rc);
3891 }
3892
3893 /**
3894  *  Initialize osd Iterator for given osd index object.
3895  *
3896  *  \param  dt      osd index object
3897  */
3898
3899 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3900                                      struct dt_object *dt,
3901                                      __u32 unused,
3902                                      struct lustre_capa *capa)
3903 {
3904         struct osd_it_iam      *it;
3905         struct osd_thread_info *oti = osd_oti_get(env);
3906         struct osd_object      *obj = osd_dt_obj(dt);
3907         struct lu_object       *lo  = &dt->do_lu;
3908         struct iam_path_descr  *ipd;
3909         struct iam_container   *bag = &obj->oo_dir->od_container;
3910
3911         LASSERT(lu_object_exists(lo));
3912
3913         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3914                 return ERR_PTR(-EACCES);
3915
3916         it = &oti->oti_it;
3917         ipd = osd_it_ipd_get(env, bag);
3918         if (likely(ipd != NULL)) {
3919                 it->oi_obj = obj;
3920                 it->oi_ipd = ipd;
3921                 lu_object_get(lo);
3922                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3923                 return (struct dt_it *)it;
3924         }
3925         return ERR_PTR(-ENOMEM);
3926 }
3927
3928 /**
3929  * free given Iterator.
3930  */
3931
3932 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3933 {
3934         struct osd_it_iam *it = (struct osd_it_iam *)di;
3935         struct osd_object *obj = it->oi_obj;
3936
3937         iam_it_fini(&it->oi_it);
3938         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3939         lu_object_put(env, &obj->oo_dt.do_lu);
3940 }
3941
3942 /**
3943  *  Move Iterator to record specified by \a key
3944  *
3945  *  \param  di      osd iterator
3946  *  \param  key     key for index
3947  *
3948  *  \retval +ve  di points to record with least key not larger than key
3949  *  \retval  0   di points to exact matched key
3950  *  \retval -ve  failure
3951  */
3952
3953 static int osd_it_iam_get(const struct lu_env *env,
3954                           struct dt_it *di, const struct dt_key *key)
3955 {
3956         struct osd_thread_info  *oti = osd_oti_get(env);
3957         struct osd_it_iam       *it = (struct osd_it_iam *)di;
3958
3959         if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
3960                 /* swab quota uid/gid */
3961                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3962                 key = (struct dt_key *)&oti->oti_quota_id;
3963         }
3964
3965         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3966 }
3967
3968 /**
3969  *  Release Iterator
3970  *
3971  *  \param  di      osd iterator
3972  */
3973 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3974 {
3975         struct osd_it_iam *it = (struct osd_it_iam *)di;
3976
3977         iam_it_put(&it->oi_it);
3978 }
3979
3980 /**
3981  *  Move iterator by one record
3982  *
3983  *  \param  di      osd iterator
3984  *
3985  *  \retval +1   end of container reached
3986  *  \retval  0   success
3987  *  \retval -ve  failure
3988  */
3989
3990 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3991 {
3992         struct osd_it_iam *it = (struct osd_it_iam *)di;
3993
3994         return iam_it_next(&it->oi_it);
3995 }
3996
3997 /**
3998  * Return pointer to the key under iterator.
3999  */
4000
4001 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
4002                                  const struct dt_it *di)
4003 {
4004         struct osd_thread_info *oti = osd_oti_get(env);
4005         struct osd_it_iam      *it = (struct osd_it_iam *)di;
4006         struct osd_object      *obj = it->oi_obj;
4007         struct dt_key          *key;
4008
4009         key = (struct dt_key *)iam_it_key_get(&it->oi_it);
4010
4011         if (!IS_ERR(key) && fid_is_quota(lu_object_fid(&obj->oo_dt.do_lu))) {
4012                 /* swab quota uid/gid */
4013                 oti->oti_quota_id = le64_to_cpu(*((__u64 *)key));
4014                 key = (struct dt_key *)&oti->oti_quota_id;
4015         }
4016
4017         return key;
4018 }
4019
4020 /**
4021  * Return size of key under iterator (in bytes)
4022  */
4023
4024 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
4025 {
4026         struct osd_it_iam *it = (struct osd_it_iam *)di;
4027
4028         return iam_it_key_size(&it->oi_it);
4029 }
4030
4031 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
4032                                        int len, __u16 type)
4033 {
4034         struct luda_type *lt;
4035         const unsigned    align = sizeof(struct luda_type) - 1;
4036
4037         /* check if file type is required */
4038         if (attr & LUDA_TYPE) {
4039                         len = (len + align) & ~align;
4040
4041                         lt = (void *) ent->lde_name + len;
4042                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
4043                         ent->lde_attrs |= LUDA_TYPE;
4044         }
4045
4046         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
4047 }
4048
4049 /**
4050  * build lu direct from backend fs dirent.
4051  */
4052
4053 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
4054                                       struct lu_fid *fid, __u64 offset,
4055                                       char *name, __u16 namelen,
4056                                       __u16 type, __u32 attr)
4057 {
4058         fid_cpu_to_le(&ent->lde_fid, fid);
4059         ent->lde_attrs = LUDA_FID;
4060
4061         ent->lde_hash = cpu_to_le64(offset);
4062         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
4063
4064         strncpy(ent->lde_name, name, namelen);
4065         ent->lde_namelen = cpu_to_le16(namelen);
4066
4067         /* append lustre attributes */
4068         osd_it_append_attrs(ent, attr, namelen, type);
4069 }
4070
4071 /**
4072  * Return pointer to the record under iterator.
4073  */
4074 static int osd_it_iam_rec(const struct lu_env *env,
4075                           const struct dt_it *di,
4076                           struct dt_rec *dtrec, __u32 attr)
4077 {
4078         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
4079         struct osd_thread_info *info = osd_oti_get(env);
4080         ENTRY;
4081
4082         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
4083                 const struct osd_fid_pack *rec;
4084                 struct lu_fid             *fid = &info->oti_fid;
4085                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
4086                 char                      *name;
4087                 int                        namelen;
4088                 __u64                      hash;
4089                 int                        rc;
4090
4091                 name = (char *)iam_it_key_get(&it->oi_it);
4092                 if (IS_ERR(name))
4093                         RETURN(PTR_ERR(name));
4094
4095                 namelen = iam_it_key_size(&it->oi_it);
4096
4097                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
4098                 if (IS_ERR(rec))
4099                         RETURN(PTR_ERR(rec));
4100
4101                 rc = osd_fid_unpack(fid, rec);
4102                 if (rc)
4103                         RETURN(rc);
4104
4105                 hash = iam_it_store(&it->oi_it);
4106
4107                 /* IAM does not store object type in IAM index (dir) */
4108                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
4109                                    0, LUDA_FID);
4110         } else if (fid_is_quota(lu_object_fid(&it->oi_obj->oo_dt.do_lu))) {
4111                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4112                            (struct iam_rec *)dtrec);
4113                 osd_quota_unpack(it->oi_obj, dtrec);
4114         } else {
4115                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
4116                            (struct iam_rec *)dtrec);
4117         }
4118
4119         RETURN(0);
4120 }
4121
4122 /**
4123  * Returns cookie for current Iterator position.
4124  */
4125 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
4126 {
4127         struct osd_it_iam *it = (struct osd_it_iam *)di;
4128
4129         return iam_it_store(&it->oi_it);
4130 }
4131
4132 /**
4133  * Restore iterator from cookie.
4134  *
4135  * \param  di      osd iterator
4136  * \param  hash    Iterator location cookie
4137  *
4138  * \retval +ve  di points to record with least key not larger than key.
4139  * \retval  0   di points to exact matched key
4140  * \retval -ve  failure
4141  */
4142
4143 static int osd_it_iam_load(const struct lu_env *env,
4144                            const struct dt_it *di, __u64 hash)
4145 {
4146         struct osd_it_iam *it = (struct osd_it_iam *)di;
4147
4148         return iam_it_load(&it->oi_it, hash);
4149 }
4150
4151 static const struct dt_index_operations osd_index_iam_ops = {
4152         .dio_lookup         = osd_index_iam_lookup,
4153         .dio_declare_insert = osd_index_declare_iam_insert,
4154         .dio_insert         = osd_index_iam_insert,
4155         .dio_declare_delete = osd_index_declare_iam_delete,
4156         .dio_delete         = osd_index_iam_delete,
4157         .dio_it     = {
4158                 .init     = osd_it_iam_init,
4159                 .fini     = osd_it_iam_fini,
4160                 .get      = osd_it_iam_get,
4161                 .put      = osd_it_iam_put,
4162                 .next     = osd_it_iam_next,
4163                 .key      = osd_it_iam_key,
4164                 .key_size = osd_it_iam_key_size,
4165                 .rec      = osd_it_iam_rec,
4166                 .store    = osd_it_iam_store,
4167                 .load     = osd_it_iam_load
4168         }
4169 };
4170
4171
4172 /**
4173  * Creates or initializes iterator context.
4174  *
4175  * \retval struct osd_it_ea, iterator structure on success
4176  *
4177  */
4178 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
4179                                     struct dt_object *dt,
4180                                     __u32 attr,
4181                                     struct lustre_capa *capa)
4182 {
4183         struct osd_object       *obj  = osd_dt_obj(dt);
4184         struct osd_thread_info  *info = osd_oti_get(env);
4185         struct osd_it_ea        *it   = &info->oti_it_ea;
4186         struct lu_object        *lo   = &dt->do_lu;
4187         struct dentry           *obj_dentry = &info->oti_it_dentry;
4188         ENTRY;
4189         LASSERT(lu_object_exists(lo));
4190
4191         obj_dentry->d_inode = obj->oo_inode;
4192         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
4193         obj_dentry->d_name.hash = 0;
4194
4195         it->oie_rd_dirent       = 0;
4196         it->oie_it_dirent       = 0;
4197         it->oie_dirent          = NULL;
4198         it->oie_buf             = info->oti_it_ea_buf;
4199         it->oie_obj             = obj;
4200         it->oie_file.f_pos      = 0;
4201         it->oie_file.f_dentry   = obj_dentry;
4202         if (attr & LUDA_64BITHASH)
4203                 it->oie_file.f_mode |= FMODE_64BITHASH;
4204         else
4205                 it->oie_file.f_mode |= FMODE_32BITHASH;
4206         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
4207         it->oie_file.f_op         = obj->oo_inode->i_fop;
4208         it->oie_file.private_data = NULL;
4209         lu_object_get(lo);
4210         RETURN((struct dt_it *) it);
4211 }
4212
4213 /**
4214  * Destroy or finishes iterator context.
4215  *
4216  * \param di iterator structure to be destroyed
4217  */
4218 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
4219 {
4220         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4221         struct osd_object    *obj  = it->oie_obj;
4222         struct inode       *inode  = obj->oo_inode;
4223
4224         ENTRY;
4225         it->oie_file.f_op->release(inode, &it->oie_file);
4226         lu_object_put(env, &obj->oo_dt.do_lu);
4227         EXIT;
4228 }
4229
4230 /**
4231  * It position the iterator at given key, so that next lookup continues from
4232  * that key Or it is similar to dio_it->load() but based on a key,
4233  * rather than file position.
4234  *
4235  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
4236  * to the beginning.
4237  *
4238  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
4239  */
4240 static int osd_it_ea_get(const struct lu_env *env,
4241                          struct dt_it *di, const struct dt_key *key)
4242 {
4243         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
4244
4245         ENTRY;
4246         LASSERT(((const char *)key)[0] == '\0');
4247         it->oie_file.f_pos      = 0;
4248         it->oie_rd_dirent       = 0;
4249         it->oie_it_dirent       = 0;
4250         it->oie_dirent          = NULL;
4251
4252         RETURN(+1);
4253 }
4254
4255 /**
4256  * Does nothing
4257  */
4258 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
4259 {
4260 }
4261
4262 /**
4263  * It is called internally by ->readdir(). It fills the
4264  * iterator's in-memory data structure with required
4265  * information i.e. name, namelen, rec_size etc.
4266  *
4267  * \param buf in which information to be filled in.
4268  * \param name name of the file in given dir
4269  *
4270  * \retval 0 on success
4271  * \retval 1 on buffer full
4272  */
4273 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
4274                                loff_t offset, __u64 ino,
4275                                unsigned d_type)
4276 {
4277         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
4278         struct osd_object       *obj  = it->oie_obj;
4279         struct osd_it_ea_dirent *ent  = it->oie_dirent;
4280         struct lu_fid           *fid  = &ent->oied_fid;
4281         struct osd_fid_pack     *rec;
4282         ENTRY;
4283
4284         /* this should never happen */
4285         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
4286                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
4287                 RETURN(-EIO);
4288         }
4289
4290         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
4291             OSD_IT_EA_BUFSIZE)
4292                 RETURN(1);
4293
4294         /* "." is just the object itself. */
4295         if (namelen == 1 && name[0] == '.') {
4296                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4297         } else if (d_type & LDISKFS_DIRENT_LUFID) {
4298                 rec = (struct osd_fid_pack*) (name + namelen + 1);
4299                 if (osd_fid_unpack(fid, rec) != 0)
4300                         fid_zero(fid);
4301         } else {
4302                 fid_zero(fid);
4303         }
4304         d_type &= ~LDISKFS_DIRENT_LUFID;
4305
4306         /* NOT export local root. */
4307         if (unlikely(osd_sb(osd_obj2dev(obj))->s_root->d_inode->i_ino == ino)) {
4308                 ino = obj->oo_inode->i_ino;
4309                 *fid = obj->oo_dt.do_lu.lo_header->loh_fid;
4310         }
4311
4312         ent->oied_ino     = ino;
4313         ent->oied_off     = offset;
4314         ent->oied_namelen = namelen;
4315         ent->oied_type    = d_type;
4316
4317         memcpy(ent->oied_name, name, namelen);
4318
4319         it->oie_rd_dirent++;
4320         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
4321         RETURN(0);
4322 }
4323
4324 /**
4325  * Calls ->readdir() to load a directory entry at a time
4326  * and stored it in iterator's in-memory data structure.
4327  *
4328  * \param di iterator's in memory structure
4329  *
4330  * \retval   0 on success
4331  * \retval -ve on error
4332  */
4333 static int osd_ldiskfs_it_fill(const struct lu_env *env,
4334                                const struct dt_it *di)
4335 {
4336         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
4337         struct osd_object  *obj   = it->oie_obj;
4338         struct inode       *inode = obj->oo_inode;
4339         struct htree_lock  *hlock = NULL;
4340         int                 result = 0;
4341
4342         ENTRY;
4343         it->oie_dirent = it->oie_buf;
4344         it->oie_rd_dirent = 0;
4345
4346         if (obj->oo_hl_head != NULL) {
4347                 hlock = osd_oti_get(env)->oti_hlock;
4348                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4349                                    inode, LDISKFS_HLOCK_READDIR);
4350         } else {
4351                 down_read(&obj->oo_ext_idx_sem);
4352         }
4353
4354         result = inode->i_fop->readdir(&it->oie_file, it,
4355                                        (filldir_t) osd_ldiskfs_filldir);
4356
4357         if (hlock != NULL)
4358                 ldiskfs_htree_unlock(hlock);
4359         else
4360                 up_read(&obj->oo_ext_idx_sem);
4361
4362         if (it->oie_rd_dirent == 0) {
4363                 result = -EIO;
4364         } else {
4365                 it->oie_dirent = it->oie_buf;
4366                 it->oie_it_dirent = 1;
4367         }
4368
4369         RETURN(result);
4370 }
4371
4372 /**
4373  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4374  * to load a directory entry at a time and stored it in
4375  * iterator's in-memory data structure.
4376  *
4377  * \param di iterator's in memory structure
4378  *
4379  * \retval +ve iterator reached to end
4380  * \retval   0 iterator not reached to end
4381  * \retval -ve on error
4382  */
4383 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4384 {
4385         struct osd_it_ea *it = (struct osd_it_ea *)di;
4386         int rc;
4387
4388         ENTRY;
4389
4390         if (it->oie_it_dirent < it->oie_rd_dirent) {
4391                 it->oie_dirent =
4392                         (void *) it->oie_dirent +
4393                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4394                                        it->oie_dirent->oied_namelen);
4395                 it->oie_it_dirent++;
4396                 RETURN(0);
4397         } else {
4398                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4399                         rc = +1;
4400                 else
4401                         rc = osd_ldiskfs_it_fill(env, di);
4402         }
4403
4404         RETURN(rc);
4405 }
4406
4407 /**
4408  * Returns the key at current position from iterator's in memory structure.
4409  *
4410  * \param di iterator's in memory structure
4411  *
4412  * \retval key i.e. struct dt_key on success
4413  */
4414 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4415                                     const struct dt_it *di)
4416 {
4417         struct osd_it_ea *it = (struct osd_it_ea *)di;
4418
4419         return (struct dt_key *)it->oie_dirent->oied_name;
4420 }
4421
4422 /**
4423  * Returns the key's size at current position from iterator's in memory structure.
4424  *
4425  * \param di iterator's in memory structure
4426  *
4427  * \retval key_size i.e. struct dt_key on success
4428  */
4429 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4430 {
4431         struct osd_it_ea *it = (struct osd_it_ea *)di;
4432
4433         return it->oie_dirent->oied_namelen;
4434 }
4435
4436
4437 /**
4438  * Returns the value (i.e. fid/igif) at current position from iterator's
4439  * in memory structure.
4440  *
4441  * \param di struct osd_it_ea, iterator's in memory structure
4442  * \param attr attr requested for dirent.
4443  * \param lde lustre dirent
4444  *
4445  * \retval   0 no error and \param lde has correct lustre dirent.
4446  * \retval -ve on error
4447  */
4448 static inline int osd_it_ea_rec(const struct lu_env *env,
4449                                 const struct dt_it *di,
4450                                 struct dt_rec *dtrec, __u32 attr)
4451 {
4452         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4453         struct osd_object      *obj   = it->oie_obj;
4454         struct osd_device      *dev   = osd_obj2dev(obj);
4455         struct osd_scrub       *scrub = &dev->od_scrub;
4456         struct scrub_file      *sf    = &scrub->os_file;
4457         struct osd_thread_info *oti   = osd_oti_get(env);
4458         struct osd_inode_id    *id    = &oti->oti_id;
4459         struct osd_idmap_cache *oic   = &oti->oti_cache;
4460         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4461         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4462         __u32                   ino   = it->oie_dirent->oied_ino;
4463         int                     rc    = 0;
4464         ENTRY;
4465
4466         if (!fid_is_sane(fid)) {
4467                 rc = osd_ea_fid_get(env, obj, ino, fid, id);
4468                 if (rc != 0) {
4469                         fid_zero(&oic->oic_fid);
4470                         RETURN(rc);
4471                 }
4472         } else {
4473                 osd_id_gen(id, ino, OSD_OII_NOGEN);
4474         }
4475
4476         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4477                            it->oie_dirent->oied_name,
4478                            it->oie_dirent->oied_namelen,
4479                            it->oie_dirent->oied_type, attr);
4480
4481         if (osd_remote_fid(env, dev, fid))
4482                 RETURN(0);
4483
4484         oic->oic_lid = *id;
4485         oic->oic_fid = *fid;
4486         if ((scrub->os_pos_current <= ino) &&
4487             ((sf->sf_flags & SF_INCONSISTENT) ||
4488              (sf->sf_flags & SF_UPGRADE && fid_is_igif(fid)) ||
4489              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4490                 osd_consistency_check(oti, dev, oic);
4491
4492         RETURN(rc);
4493 }
4494
4495 /**
4496  * Returns a cookie for current position of the iterator head, so that
4497  * user can use this cookie to load/start the iterator next time.
4498  *
4499  * \param di iterator's in memory structure
4500  *
4501  * \retval cookie for current position, on success
4502  */
4503 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4504 {
4505         struct osd_it_ea *it = (struct osd_it_ea *)di;
4506
4507         return it->oie_dirent->oied_off;
4508 }
4509
4510 /**
4511  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4512  * to load a directory entry at a time and stored it i inn,
4513  * in iterator's in-memory data structure.
4514  *
4515  * \param di struct osd_it_ea, iterator's in memory structure
4516  *
4517  * \retval +ve on success
4518  * \retval -ve on error
4519  */
4520 static int osd_it_ea_load(const struct lu_env *env,
4521                           const struct dt_it *di, __u64 hash)
4522 {
4523         struct osd_it_ea *it = (struct osd_it_ea *)di;
4524         int rc;
4525
4526         ENTRY;
4527         it->oie_file.f_pos = hash;
4528
4529         rc =  osd_ldiskfs_it_fill(env, di);
4530         if (rc == 0)
4531                 rc = +1;
4532
4533         RETURN(rc);
4534 }
4535
4536 /**
4537  * Index lookup function for interoperability mode (b11826).
4538  *
4539  * \param key,  key i.e. file name to be searched
4540  *
4541  * \retval +ve, on success
4542  * \retval -ve, on error
4543  */
4544 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4545                                struct dt_rec *rec, const struct dt_key *key,
4546                                struct lustre_capa *capa)
4547 {
4548         struct osd_object *obj = osd_dt_obj(dt);
4549         int rc = 0;
4550
4551         ENTRY;
4552
4553         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4554         LINVRNT(osd_invariant(obj));
4555
4556         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4557                 return -EACCES;
4558
4559         rc = osd_ea_lookup_rec(env, obj, rec, key);
4560         if (rc == 0)
4561                 rc = +1;
4562         RETURN(rc);
4563 }
4564
4565 /**
4566  * Index and Iterator operations for interoperability
4567  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4568  */
4569 static const struct dt_index_operations osd_index_ea_ops = {
4570         .dio_lookup         = osd_index_ea_lookup,
4571         .dio_declare_insert = osd_index_declare_ea_insert,
4572         .dio_insert         = osd_index_ea_insert,
4573         .dio_declare_delete = osd_index_declare_ea_delete,
4574         .dio_delete         = osd_index_ea_delete,
4575         .dio_it     = {
4576                 .init     = osd_it_ea_init,
4577                 .fini     = osd_it_ea_fini,
4578                 .get      = osd_it_ea_get,
4579                 .put      = osd_it_ea_put,
4580                 .next     = osd_it_ea_next,
4581                 .key      = osd_it_ea_key,
4582                 .key_size = osd_it_ea_key_size,
4583                 .rec      = osd_it_ea_rec,
4584                 .store    = osd_it_ea_store,
4585                 .load     = osd_it_ea_load
4586         }
4587 };
4588
4589 static void *osd_key_init(const struct lu_context *ctx,
4590                           struct lu_context_key *key)
4591 {
4592         struct osd_thread_info *info;
4593
4594         OBD_ALLOC_PTR(info);
4595         if (info == NULL)
4596                 return ERR_PTR(-ENOMEM);
4597
4598         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4599         if (info->oti_it_ea_buf == NULL)
4600                 goto out_free_info;
4601
4602         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4603
4604         info->oti_hlock = ldiskfs_htree_lock_alloc();
4605         if (info->oti_hlock == NULL)
4606                 goto out_free_ea;
4607
4608         return info;
4609
4610  out_free_ea:
4611         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4612  out_free_info:
4613         OBD_FREE_PTR(info);
4614         return ERR_PTR(-ENOMEM);
4615 }
4616
4617 static void osd_key_fini(const struct lu_context *ctx,
4618                          struct lu_context_key *key, void* data)
4619 {
4620         struct osd_thread_info *info = data;
4621
4622         if (info->oti_hlock != NULL)
4623                 ldiskfs_htree_lock_free(info->oti_hlock);
4624         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4625         OBD_FREE_PTR(info);
4626 }
4627
4628 static void osd_key_exit(const struct lu_context *ctx,
4629                          struct lu_context_key *key, void *data)
4630 {
4631         struct osd_thread_info *info = data;
4632
4633         LASSERT(info->oti_r_locks == 0);
4634         LASSERT(info->oti_w_locks == 0);
4635         LASSERT(info->oti_txns    == 0);
4636 }
4637
4638 /* type constructor/destructor: osd_type_init, osd_type_fini */
4639 LU_TYPE_INIT_FINI(osd, &osd_key);
4640
4641 struct lu_context_key osd_key = {
4642         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4643         .lct_init = osd_key_init,
4644         .lct_fini = osd_key_fini,
4645         .lct_exit = osd_key_exit
4646 };
4647
4648
4649 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4650                            const char *name, struct lu_device *next)
4651 {
4652         struct osd_device *osd = osd_dev(d);
4653
4654         strncpy(osd->od_svname, name, MAX_OBD_NAME);
4655         return osd_procfs_init(osd, name);
4656 }
4657
4658 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4659 {
4660         ENTRY;
4661
4662         osd_scrub_cleanup(env, o);
4663
4664         if (o->od_fsops) {
4665                 fsfilt_put_ops(o->od_fsops);
4666                 o->od_fsops = NULL;
4667         }
4668
4669         /* shutdown quota slave instance associated with the device */
4670         if (o->od_quota_slave != NULL) {
4671                 qsd_fini(env, o->od_quota_slave);
4672                 o->od_quota_slave = NULL;
4673         }
4674
4675         RETURN(0);
4676 }
4677
4678 static int osd_mount(const struct lu_env *env,
4679                      struct osd_device *o, struct lustre_cfg *cfg)
4680 {
4681         const char              *name  = lustre_cfg_string(cfg, 0);
4682         const char              *dev  = lustre_cfg_string(cfg, 1);
4683         const char              *opts;
4684         unsigned long            page, s_flags, lmd_flags = 0;
4685         struct page             *__page;
4686         struct file_system_type *type;
4687         char                    *options = NULL;
4688         char                    *str;
4689         int                       rc = 0;
4690         ENTRY;
4691
4692         if (o->od_mnt != NULL)
4693                 RETURN(0);
4694
4695         if (strlen(dev) >= sizeof(o->od_mntdev))
4696                 RETURN(-E2BIG);
4697         strcpy(o->od_mntdev, dev);
4698
4699         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4700         if (o->od_fsops == NULL) {
4701                 CERROR("Can't find fsfilt_ldiskfs\n");
4702                 RETURN(-ENOTSUPP);
4703         }
4704
4705         OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
4706         if (__page == NULL)
4707                 GOTO(out, rc = -ENOMEM);
4708
4709         str = lustre_cfg_string(cfg, 2);
4710         s_flags = simple_strtoul(str, NULL, 0);
4711         str = strstr(str, ":");
4712         if (str)
4713                 lmd_flags = simple_strtoul(str + 1, NULL, 0);
4714         opts = lustre_cfg_string(cfg, 3);
4715         page = (unsigned long)cfs_page_address(__page);
4716         options = (char *)page;
4717         *options = '\0';
4718         if (opts == NULL)
4719                 strcat(options, "user_xattr,acl");
4720         else
4721                 strcat(options, opts);
4722
4723         /* Glom up mount options */
4724         if (*options != '\0')
4725                 strcat(options, ",");
4726         strlcat(options, "no_mbcache", CFS_PAGE_SIZE);
4727
4728         type = get_fs_type("ldiskfs");
4729         if (!type) {
4730                 CERROR("%s: cannot find ldiskfs module\n", name);
4731                 GOTO(out, rc = -ENODEV);
4732         }
4733
4734         o->od_mnt = vfs_kern_mount(type, s_flags, dev, options);
4735         cfs_module_put(type->owner);
4736
4737         if (IS_ERR(o->od_mnt)) {
4738                 rc = PTR_ERR(o->od_mnt);
4739                 CERROR("%s: can't mount %s: %d\n", name, dev, rc);
4740                 o->od_mnt = NULL;
4741                 GOTO(out, rc);
4742         }
4743
4744         if (lvfs_check_rdonly(o->od_mnt->mnt_sb->s_bdev)) {
4745                 CERROR("%s: underlying device %s is marked as read-only. "
4746                        "Setup failed\n", name, dev);
4747                 mntput(o->od_mnt);
4748                 o->od_mnt = NULL;
4749                 GOTO(out, rc = -EROFS);
4750         }
4751
4752         if (!LDISKFS_HAS_COMPAT_FEATURE(o->od_mnt->mnt_sb,
4753             LDISKFS_FEATURE_COMPAT_HAS_JOURNAL)) {
4754                 CERROR("%s: device %s is mounted w/o journal\n", name, dev);
4755                 mntput(o->od_mnt);
4756                 o->od_mnt = NULL;
4757                 GOTO(out, rc = -EINVAL);
4758         }
4759
4760         ldiskfs_set_inode_state(osd_sb(o)->s_root->d_inode,
4761                                 LDISKFS_STATE_LUSTRE_NO_OI);
4762         if (lmd_flags & LMD_FLG_NOSCRUB)
4763                 o->od_noscrub = 1;
4764
4765 out:
4766         if (__page)
4767                 OBD_PAGE_FREE(__page);
4768         if (rc)
4769                 fsfilt_put_ops(o->od_fsops);
4770
4771         RETURN(rc);
4772 }
4773
4774 static struct lu_device *osd_device_fini(const struct lu_env *env,
4775                                          struct lu_device *d)
4776 {
4777         int rc;
4778         ENTRY;
4779
4780         rc = osd_shutdown(env, osd_dev(d));
4781
4782         osd_obj_map_fini(osd_dev(d));
4783
4784         shrink_dcache_sb(osd_sb(osd_dev(d)));
4785         osd_sync(env, lu2dt_dev(d));
4786
4787         rc = osd_procfs_fini(osd_dev(d));
4788         if (rc) {
4789                 CERROR("proc fini error %d \n", rc);
4790                 RETURN (ERR_PTR(rc));
4791         }
4792
4793         if (osd_dev(d)->od_mnt) {
4794                 mntput(osd_dev(d)->od_mnt);
4795                 osd_dev(d)->od_mnt = NULL;
4796         }
4797
4798         RETURN(NULL);
4799 }
4800
4801 static int osd_device_init0(const struct lu_env *env,
4802                             struct osd_device *o,
4803                             struct lustre_cfg *cfg)
4804 {
4805         struct lu_device        *l = osd2lu_dev(o);
4806         struct osd_thread_info *info;
4807         int                     rc;
4808
4809         /* if the module was re-loaded, env can loose its keys */
4810         rc = lu_env_refill((struct lu_env *) env);
4811         if (rc)
4812                 GOTO(out, rc);
4813         info = osd_oti_get(env);
4814         LASSERT(info);
4815
4816         l->ld_ops = &osd_lu_ops;
4817         o->od_dt_dev.dd_ops = &osd_dt_ops;
4818
4819         spin_lock_init(&o->od_osfs_lock);
4820         mutex_init(&o->od_otable_mutex);
4821         o->od_osfs_age = cfs_time_shift_64(-1000);
4822
4823         o->od_capa_hash = init_capa_hash();
4824         if (o->od_capa_hash == NULL)
4825                 GOTO(out, rc = -ENOMEM);
4826
4827         o->od_read_cache = 1;
4828         o->od_writethrough_cache = 1;
4829         o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
4830
4831         rc = osd_mount(env, o, cfg);
4832         if (rc)
4833                 GOTO(out_capa, rc);
4834
4835         CFS_INIT_LIST_HEAD(&o->od_ios_list);
4836         /* setup scrub, including OI files initialization */
4837         rc = osd_scrub_setup(env, o);
4838         if (rc < 0)
4839                 GOTO(out_mnt, rc);
4840
4841         strncpy(o->od_svname, lustre_cfg_string(cfg, 4),
4842                         sizeof(o->od_svname) - 1);
4843
4844         rc = osd_obj_map_init(o);
4845         if (rc != 0)
4846                 GOTO(out_scrub, rc);
4847
4848         rc = lu_site_init(&o->od_site, l);
4849         if (rc)
4850                 GOTO(out_compat, rc);
4851         o->od_site.ls_bottom_dev = l;
4852
4853         rc = lu_site_init_finish(&o->od_site);
4854         if (rc)
4855                 GOTO(out_site, rc);
4856
4857         rc = osd_procfs_init(o, o->od_svname);
4858         if (rc != 0) {
4859                 CERROR("%s: can't initialize procfs: rc = %d\n",
4860                        o->od_svname, rc);
4861                 GOTO(out_site, rc);
4862         }
4863
4864         LASSERT(l->ld_site->ls_linkage.next && l->ld_site->ls_linkage.prev);
4865
4866         /* initialize quota slave instance */
4867         o->od_quota_slave = qsd_init(env, o->od_svname, &o->od_dt_dev,
4868                                      o->od_proc_entry);
4869         if (IS_ERR(o->od_quota_slave)) {
4870                 rc = PTR_ERR(o->od_quota_slave);
4871                 o->od_quota_slave = NULL;
4872                 GOTO(out_procfs, rc);
4873         }
4874
4875         RETURN(0);
4876 out_procfs:
4877         osd_procfs_fini(o);
4878 out_site:
4879         lu_site_fini(&o->od_site);
4880 out_compat:
4881         osd_obj_map_fini(o);
4882 out_scrub:
4883         osd_scrub_cleanup(env, o);
4884 out_mnt:
4885         osd_oi_fini(info, o);
4886         osd_shutdown(env, o);
4887         mntput(o->od_mnt);
4888         o->od_mnt = NULL;
4889 out_capa:
4890         cleanup_capa_hash(o->od_capa_hash);
4891 out:
4892         RETURN(rc);
4893 }
4894
4895 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4896                                           struct lu_device_type *t,
4897                                           struct lustre_cfg *cfg)
4898 {
4899         struct osd_device *o;
4900         int                rc;
4901
4902         OBD_ALLOC_PTR(o);
4903         if (o == NULL)
4904                 return ERR_PTR(-ENOMEM);
4905
4906         rc = dt_device_init(&o->od_dt_dev, t);
4907         if (rc == 0) {
4908                 /* Because the ctx might be revived in dt_device_init,
4909                  * refill the env here */
4910                 lu_env_refill((struct lu_env *)env);
4911                 rc = osd_device_init0(env, o, cfg);
4912                 if (rc)
4913                         dt_device_fini(&o->od_dt_dev);
4914         }
4915
4916         if (unlikely(rc != 0))
4917                 OBD_FREE_PTR(o);
4918
4919         return rc == 0 ? osd2lu_dev(o) : ERR_PTR(rc);
4920 }
4921
4922 static struct lu_device *osd_device_free(const struct lu_env *env,
4923                                          struct lu_device *d)
4924 {
4925         struct osd_device *o = osd_dev(d);
4926         ENTRY;
4927
4928         cleanup_capa_hash(o->od_capa_hash);
4929         /* XXX: make osd top device in order to release reference */
4930         d->ld_site->ls_top_dev = d;
4931         lu_site_purge(env, d->ld_site, -1);
4932         if (!cfs_hash_is_empty(d->ld_site->ls_obj_hash)) {
4933                 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_ERROR, NULL);
4934                 lu_site_print(env, d->ld_site, &msgdata, lu_cdebug_printer);
4935         }
4936         lu_site_fini(&o->od_site);
4937         dt_device_fini(&o->od_dt_dev);
4938         OBD_FREE_PTR(o);
4939         RETURN(NULL);
4940 }
4941
4942 static int osd_process_config(const struct lu_env *env,
4943                               struct lu_device *d, struct lustre_cfg *cfg)
4944 {
4945         struct osd_device *o = osd_dev(d);
4946         int err;
4947         ENTRY;
4948
4949         switch(cfg->lcfg_command) {
4950         case LCFG_SETUP:
4951                 err = osd_mount(env, o, cfg);
4952                 break;
4953         case LCFG_CLEANUP:
4954                 lu_dev_del_linkage(d->ld_site, d);
4955                 err = osd_shutdown(env, o);
4956                 break;
4957         default:
4958                 err = -ENOSYS;
4959         }
4960
4961         RETURN(err);
4962 }
4963
4964 static int osd_recovery_complete(const struct lu_env *env,
4965                                  struct lu_device *d)
4966 {
4967         struct osd_device       *osd = osd_dev(d);
4968         int                      rc = 0;
4969         ENTRY;
4970
4971         if (osd->od_quota_slave == NULL)
4972                 RETURN(0);
4973
4974         /* start qsd instance on recovery completion, this notifies the quota
4975          * slave code that we are about to process new requests now */
4976         rc = qsd_start(env, osd->od_quota_slave);
4977         RETURN(rc);
4978 }
4979
4980 /*
4981  * we use exports to track all osd users
4982  */
4983 static int osd_obd_connect(const struct lu_env *env, struct obd_export **exp,
4984                            struct obd_device *obd, struct obd_uuid *cluuid,
4985                            struct obd_connect_data *data, void *localdata)
4986 {
4987         struct osd_device    *osd = osd_dev(obd->obd_lu_dev);
4988         struct lustre_handle  conn;
4989         int                   rc;
4990         ENTRY;
4991
4992         CDEBUG(D_CONFIG, "connect #%d\n", osd->od_connects);
4993
4994         rc = class_connect(&conn, obd, cluuid);
4995         if (rc)
4996                 RETURN(rc);
4997
4998         *exp = class_conn2export(&conn);
4999
5000         spin_lock(&osd->od_osfs_lock);
5001         osd->od_connects++;
5002         spin_unlock(&osd->od_osfs_lock);
5003
5004         RETURN(0);
5005 }
5006
5007 /*
5008  * once last export (we don't count self-export) disappeared
5009  * osd can be released
5010  */
5011 static int osd_obd_disconnect(struct obd_export *exp)
5012 {
5013         struct obd_device *obd = exp->exp_obd;
5014         struct osd_device *osd = osd_dev(obd->obd_lu_dev);
5015         int                rc, release = 0;
5016         ENTRY;
5017
5018         /* Only disconnect the underlying layers on the final disconnect. */
5019         spin_lock(&osd->od_osfs_lock);
5020         osd->od_connects--;
5021         if (osd->od_connects == 0)
5022                 release = 1;
5023         spin_unlock(&osd->od_osfs_lock);
5024
5025         rc = class_disconnect(exp); /* bz 9811 */
5026
5027         if (rc == 0 && release)
5028                 class_manual_cleanup(obd);
5029         RETURN(rc);
5030 }
5031
5032 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
5033                        struct lu_device *dev)
5034 {
5035         struct osd_device *osd = osd_dev(dev);
5036         int                result = 0;
5037         ENTRY;
5038
5039         if (dev->ld_site && lu_device_is_md(dev->ld_site->ls_top_dev)) {
5040                 /* MDT/MDD still use old infrastructure to create
5041                  * special files */
5042                 result = llo_local_objects_setup(env, lu2md_dev(pdev),
5043                                                  lu2dt_dev(dev));
5044                 if (result)
5045                         RETURN(result);
5046         }
5047
5048         if (osd->od_quota_slave != NULL)
5049                 /* set up quota slave objects */
5050                 result = qsd_prepare(env, osd->od_quota_slave);
5051
5052         RETURN(result);
5053 }
5054
5055 static const struct lu_object_operations osd_lu_obj_ops = {
5056         .loo_object_init      = osd_object_init,
5057         .loo_object_delete    = osd_object_delete,
5058         .loo_object_release   = osd_object_release,
5059         .loo_object_free      = osd_object_free,
5060         .loo_object_print     = osd_object_print,
5061         .loo_object_invariant = osd_object_invariant
5062 };
5063
5064 const struct lu_device_operations osd_lu_ops = {
5065         .ldo_object_alloc      = osd_object_alloc,
5066         .ldo_process_config    = osd_process_config,
5067         .ldo_recovery_complete = osd_recovery_complete,
5068         .ldo_prepare           = osd_prepare,
5069 };
5070
5071 static const struct lu_device_type_operations osd_device_type_ops = {
5072         .ldto_init = osd_type_init,
5073         .ldto_fini = osd_type_fini,
5074
5075         .ldto_start = osd_type_start,
5076         .ldto_stop  = osd_type_stop,
5077
5078         .ldto_device_alloc = osd_device_alloc,
5079         .ldto_device_free  = osd_device_free,
5080
5081         .ldto_device_init    = osd_device_init,
5082         .ldto_device_fini    = osd_device_fini
5083 };
5084
5085 struct lu_device_type osd_device_type = {
5086         .ldt_tags     = LU_DEVICE_DT,
5087         .ldt_name     = LUSTRE_OSD_LDISKFS_NAME,
5088         .ldt_ops      = &osd_device_type_ops,
5089         .ldt_ctx_tags = LCT_LOCAL,
5090 };
5091
5092 /*
5093  * lprocfs legacy support.
5094  */
5095 static struct obd_ops osd_obd_device_ops = {
5096         .o_owner = THIS_MODULE,
5097         .o_connect      = osd_obd_connect,
5098         .o_disconnect   = osd_obd_disconnect
5099 };
5100
5101 static int __init osd_mod_init(void)
5102 {
5103         struct lprocfs_static_vars lvars;
5104
5105         osd_oi_mod_init();
5106         lprocfs_osd_init_vars(&lvars);
5107         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
5108                                    LUSTRE_OSD_LDISKFS_NAME, &osd_device_type);
5109 }
5110
5111 static void __exit osd_mod_exit(void)
5112 {
5113         class_unregister_type(LUSTRE_OSD_LDISKFS_NAME);
5114 }
5115
5116 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
5117 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_LDISKFS_NAME")");
5118 MODULE_LICENSE("GPL");
5119
5120 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);