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, Whamcloud, Inc.
  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
  67 /* fid_is_local() */
  68 #include <lustre_fid.h>
  69
  70 #include "osd_internal.h"
  71 #include "osd_igif.h"
  72
  73 /* llo_* api support */
  74 #include <md_object.h>
  75
  76 #ifdef HAVE_LDISKFS_PDO
  77 int ldiskfs_pdo = 1;
  78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  79                 "ldiskfs with parallel directory operations");
  80 #else
  81 int ldiskfs_pdo = 0;
  82 #endif
  83
  84 static const char dot[] = ".";
  85 static const char dotdot[] = "..";
  86 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  87
  88 static const struct lu_object_operations      osd_lu_obj_ops;
  89 static const struct dt_object_operations      osd_obj_ops;
  90 static const struct dt_object_operations      osd_obj_ea_ops;
  91 static const struct dt_index_operations       osd_index_iam_ops;
  92 static const struct dt_index_operations       osd_index_ea_ops;
  93
  94 static int osd_has_index(const struct osd_object *obj)
  95 {
  96         return obj->oo_dt.do_index_ops != NULL;
  97 }
  98
  99 static int osd_object_invariant(const struct lu_object *l)
 100 {
 101         return osd_invariant(osd_obj(l));
 102 }
 103
 104 #ifdef HAVE_QUOTA_SUPPORT
 105 static inline void
 106 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
 107 {
 108         struct md_ucred *uc = md_ucred(env);
 109         struct cred     *tc;
 110
 111         LASSERT(uc != NULL);
 112
 113         save->oc_uid = current_fsuid();
 114         save->oc_gid = current_fsgid();
 115         save->oc_cap = current_cap();
 116         if ((tc = prepare_creds())) {
 117                 tc->fsuid         = uc->mu_fsuid;
 118                 tc->fsgid         = uc->mu_fsgid;
 119                 commit_creds(tc);
 120         }
 121         /* XXX not suboptimal */
 122         cfs_curproc_cap_unpack(uc->mu_cap);
 123 }
 124
 125 static inline void
 126 osd_pop_ctxt(struct osd_ctxt *save)
 127 {
 128         struct cred *tc;
 129
 130         if ((tc = prepare_creds())) {
 131                 tc->fsuid         = save->oc_uid;
 132                 tc->fsgid         = save->oc_gid;
 133                 tc->cap_effective = save->oc_cap;
 134                 commit_creds(tc);
 135         }
 136 }
 137 #endif
 138
 139 /*
 140  * Concurrency: doesn't matter
 141  */
 142 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 143 {
 144         return osd_oti_get(env)->oti_r_locks > 0;
 145 }
 146
 147 /*
 148  * Concurrency: doesn't matter
 149  */
 150 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 151 {
 152         struct osd_thread_info *oti = osd_oti_get(env);
 153         return oti->oti_w_locks > 0 && o->oo_owner == env;
 154 }
 155
 156 /*
 157  * Concurrency: doesn't access mutable data
 158  */
 159 static int osd_root_get(const struct lu_env *env,
 160                         struct dt_device *dev, struct lu_fid *f)
 161 {
 162         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 163         return 0;
 164 }
 165
 166 static inline int osd_qid_type(struct osd_thandle *oh, int i)
 167 {
 168         return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
 169 }
 170
 171 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
 172 {
 173         oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
 174 }
 175
 176 void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
 177                      int type, uid_t id, struct inode *inode)
 178 {
 179 #ifdef CONFIG_QUOTA
 180         int i, allocated = 0;
 181         struct osd_object *obj;
 182
 183         LASSERT(dt != NULL);
 184         LASSERT(oh != NULL);
 185         LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
 186                  oh->ot_id_cnt);
 187
 188         /* id entry is allocated in the quota file */
 189         if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
 190                 allocated = 1;
 191
 192         for (i = 0; i < oh->ot_id_cnt; i++) {
 193                 if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
 194                         return;
 195         }
 196
 197         if (unlikely(i >= OSD_MAX_UGID_CNT)) {
 198                 CERROR("more than %d uid/gids for a transaction?\n", i);
 199                 return;
 200         }
 201
 202         oh->ot_id_array[i] = id;
 203         osd_qid_set_type(oh, i, type);
 204         oh->ot_id_cnt++;
 205         obj = osd_dt_obj(dt);
 206         oh->ot_credits += (allocated || id == 0) ?
 207                 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
 208 #endif
 209 }
 210
 211 /*
 212  * OSD object methods.
 213  */
 214
 215 /*
 216  * Concurrency: no concurrent access is possible that early in object
 217  * life-cycle.
 218  */
 219 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 220                                           const struct lu_object_header *hdr,
 221                                           struct lu_device *d)
 222 {
 223         struct osd_object *mo;
 224
 225         OBD_ALLOC_PTR(mo);
 226         if (mo != NULL) {
 227                 struct lu_object *l;
 228
 229                 l = &mo->oo_dt.do_lu;
 230                 dt_object_init(&mo->oo_dt, NULL, d);
 231                 if (osd_dev(d)->od_iop_mode)
 232                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 233                 else
 234                         mo->oo_dt.do_ops = &osd_obj_ops;
 235
 236                 l->lo_ops = &osd_lu_obj_ops;
 237                 cfs_init_rwsem(&mo->oo_sem);
 238                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 239                 cfs_spin_lock_init(&mo->oo_guard);
 240                 return l;
 241         } else {
 242                 return NULL;
 243         }
 244 }
 245
 246 static int osd_get_lma(struct inode *inode, struct dentry *dentry,
 247                        struct lustre_mdt_attrs *lma)
 248 {
 249         int rc;
 250
 251         dentry->d_inode = inode;
 252         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
 253                                    sizeof(*lma));
 254         if (rc > 0) {
 255                 /* Check LMA compatibility */
 256                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 257                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 258                               "%lx/%#x\n",
 259                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 260                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 261                                                         ~LMA_INCOMPAT_SUPP);
 262                         rc = -ENOSYS;
 263                 } else {
 264                         lustre_lma_swab(lma);
 265                         rc = 0;
 266                 }
 267         } else if (rc == 0) {
 268                 rc = -ENODATA;
 269         }
 270
 271         return rc;
 272 }
 273
 274 /*
 275  * retrieve object from backend ext fs.
 276  **/
 277 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 278                        struct osd_inode_id *id)
 279 {
 280         struct inode *inode = NULL;
 281
 282         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 283         if (IS_ERR(inode)) {
 284                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 285                        id->oii_ino, PTR_ERR(inode));
 286         } else if (id->oii_gen != OSD_OII_NOGEN &&
 287                    inode->i_generation != id->oii_gen) {
 288                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 289                        "gen1 = %u\n",
 290                        id->oii_ino, id->oii_gen, inode->i_generation);
 291                 iput(inode);
 292                 inode = ERR_PTR(-ESTALE);
 293         } else if (inode->i_nlink == 0) {
 294                 /* due to parallel readdir and unlink,
 295                 * we can have dead inode here. */
 296                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 297                 make_bad_inode(inode);
 298                 iput(inode);
 299                 inode = ERR_PTR(-ESTALE);
 300         } else if (is_bad_inode(inode)) {
 301                 CWARN("%s: bad inode: ino = %u\n",
 302                 dev->od_dt_dev.dd_lu_dev.ld_obd->obd_name, id->oii_ino);
 303                 iput(inode);
 304                 inode = ERR_PTR(-ENOENT);
 305         } else {
 306                 if (id->oii_gen == OSD_OII_NOGEN)
 307                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 308
 309                 /* Do not update file c/mtime in ldiskfs.
 310                  * NB: we don't have any lock to protect this because we don't
 311                  * have reference on osd_object now, but contention with
 312                  * another lookup + attr_set can't happen in the tiny window
 313                  * between if (...) and set S_NOCMTIME. */
 314                 if (!(inode->i_flags & S_NOCMTIME))
 315                         inode->i_flags |= S_NOCMTIME;
 316         }
 317         return inode;
 318 }
 319
 320 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 321                            struct osd_inode_id *id, struct lu_fid *fid)
 322 {
 323         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 324         struct inode            *inode;
 325         int                      rc;
 326
 327         inode = osd_iget(info, dev, id);
 328         if (IS_ERR(inode))
 329                 return inode;
 330
 331         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 332         if (rc == 0) {
 333                 *fid = lma->lma_self_fid;
 334         } else if (rc == -ENODATA) {
 335                 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
 336         } else {
 337                 iput(inode);
 338                 inode = ERR_PTR(rc);
 339         }
 340         return inode;
 341 }
 342
 343 static int osd_fid_lookup(const struct lu_env *env,
 344                           struct osd_object *obj, const struct lu_fid *fid)
 345 {
 346         struct osd_thread_info *info;
 347         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
 348         struct osd_device      *dev;
 349         struct osd_inode_id    *id;
 350         struct inode           *inode;
 351         int                     result;
 352
 353         LINVRNT(osd_invariant(obj));
 354         LASSERT(obj->oo_inode == NULL);
 355         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 356         /*
 357          * This assertion checks that osd layer sees only local
 358          * fids. Unfortunately it is somewhat expensive (does a
 359          * cache-lookup). Disabling it for production/acceptance-testing.
 360          */
 361         LASSERT(1 || fid_is_local(env, ldev->ld_site, fid));
 362
 363         ENTRY;
 364
 365         info = osd_oti_get(env);
 366         LASSERT(info);
 367         dev  = osd_dev(ldev);
 368         id   = &info->oti_id;
 369
 370         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 371                 RETURN(-ENOENT);
 372
 373         result = osd_oi_lookup(info, dev, fid, id);
 374         if (result != 0) {
 375                 if (result == -ENOENT)
 376                         result = 0;
 377                 GOTO(out, result);
 378         }
 379
 380         inode = osd_iget(info, dev, id);
 381         if (IS_ERR(inode)) {
 382                 /*
 383                  * If fid wasn't found in oi, inode-less object is
 384                  * created, for which lu_object_exists() returns
 385                  * false. This is used in a (frequent) case when
 386                  * objects are created as locking anchors or
 387                  * place holders for objects yet to be created.
 388                  */
 389                 result = PTR_ERR(inode);
 390                 GOTO(out, result);
 391         }
 392
 393         obj->oo_inode = inode;
 394         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 395         if (dev->od_iop_mode) {
 396                 obj->oo_compat_dot_created = 1;
 397                 obj->oo_compat_dotdot_created = 1;
 398         }
 399
 400         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 401                 GOTO(out, result = 0);
 402
 403         LASSERT(obj->oo_hl_head == NULL);
 404         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 405         if (obj->oo_hl_head == NULL) {
 406                 obj->oo_inode = NULL;
 407                 iput(inode);
 408                 GOTO(out, result = -ENOMEM);
 409         }
 410         GOTO(out, result = 0);
 411
 412 out:
 413         LINVRNT(osd_invariant(obj));
 414         return result;
 415 }
 416
 417 /*
 418  * Concurrency: shouldn't matter.
 419  */
 420 static void osd_object_init0(struct osd_object *obj)
 421 {
 422         LASSERT(obj->oo_inode != NULL);
 423         obj->oo_dt.do_body_ops = &osd_body_ops;
 424         obj->oo_dt.do_lu.lo_header->loh_attr |=
 425                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 426 }
 427
 428 /*
 429  * Concurrency: no concurrent access is possible that early in object
 430  * life-cycle.
 431  */
 432 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 433                            const struct lu_object_conf *unused)
 434 {
 435         struct osd_object *obj = osd_obj(l);
 436         int result;
 437
 438         LINVRNT(osd_invariant(obj));
 439
 440         result = osd_fid_lookup(env, obj, lu_object_fid(l));
 441         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 442         if (result == 0) {
 443                 if (obj->oo_inode != NULL)
 444                         osd_object_init0(obj);
 445         }
 446         LINVRNT(osd_invariant(obj));
 447         return result;
 448 }
 449
 450 /*
 451  * Concurrency: no concurrent access is possible that late in object
 452  * life-cycle.
 453  */
 454 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 455 {
 456         struct osd_object *obj = osd_obj(l);
 457
 458         LINVRNT(osd_invariant(obj));
 459
 460         dt_object_fini(&obj->oo_dt);
 461         if (obj->oo_hl_head != NULL)
 462                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 463         OBD_FREE_PTR(obj);
 464 }
 465
 466 /*
 467  * Concurrency: no concurrent access is possible that late in object
 468  * life-cycle.
 469  */
 470 static void osd_index_fini(struct osd_object *o)
 471 {
 472         struct iam_container *bag;
 473
 474         if (o->oo_dir != NULL) {
 475                 bag = &o->oo_dir->od_container;
 476                 if (o->oo_inode != NULL) {
 477                         if (bag->ic_object == o->oo_inode)
 478                                 iam_container_fini(bag);
 479                 }
 480                 OBD_FREE_PTR(o->oo_dir);
 481                 o->oo_dir = NULL;
 482         }
 483 }
 484
 485 /*
 486  * Concurrency: no concurrent access is possible that late in object
 487  * life-cycle (for all existing callers, that is. New callers have to provide
 488  * their own locking.)
 489  */
 490 static int osd_inode_unlinked(const struct inode *inode)
 491 {
 492         return inode->i_nlink == 0;
 493 }
 494
 495 enum {
 496         OSD_TXN_OI_DELETE_CREDITS    = 20,
 497         OSD_TXN_INODE_DELETE_CREDITS = 20
 498 };
 499
 500 /*
 501  * Journal
 502  */
 503
 504 #if OSD_THANDLE_STATS
 505 /**
 506  * Set time when the handle is allocated
 507  */
 508 static void osd_th_alloced(struct osd_thandle *oth)
 509 {
 510         oth->oth_alloced = cfs_time_current();
 511 }
 512
 513 /**
 514  * Set time when the handle started
 515  */
 516 static void osd_th_started(struct osd_thandle *oth)
 517 {
 518         oth->oth_started = cfs_time_current();
 519 }
 520
 521 /**
 522  * Helper function to convert time interval to microseconds packed in
 523  * long int (default time units for the counter in "stats" initialized
 524  * by lu_time_init() )
 525  */
 526 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 527 {
 528         struct timeval val;
 529
 530         cfs_duration_usec(cfs_time_sub(end, start), &val);
 531         return val.tv_sec * 1000000 + val.tv_usec;
 532 }
 533
 534 /**
 535  * Check whether the we deal with this handle for too long.
 536  */
 537 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 538                                 cfs_time_t alloced, cfs_time_t started,
 539                                 cfs_time_t closed)
 540 {
 541         cfs_time_t now = cfs_time_current();
 542
 543         LASSERT(dev != NULL);
 544
 545         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 546                             interval_to_usec(alloced, started));
 547         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 548                             interval_to_usec(started, closed));
 549         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 550                             interval_to_usec(closed, now));
 551
 552         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 553                 CWARN("transaction handle %p was open for too long: "
 554                       "now "CFS_TIME_T" ,"
 555                       "alloced "CFS_TIME_T" ,"
 556                       "started "CFS_TIME_T" ,"
 557                       "closed "CFS_TIME_T"\n",
 558                       oth, now, alloced, started, closed);
 559                 libcfs_debug_dumpstack(NULL);
 560         }
 561 }
 562
 563 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 564 {                                                                       \
 565         cfs_time_t __closed = cfs_time_current();                       \
 566         cfs_time_t __alloced = oth->oth_alloced;                        \
 567         cfs_time_t __started = oth->oth_started;                        \
 568                                                                         \
 569         expr;                                                           \
 570         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 571 }
 572
 573 #else /* OSD_THANDLE_STATS */
 574
 575 #define osd_th_alloced(h)                  do {} while(0)
 576 #define osd_th_started(h)                  do {} while(0)
 577 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 578
 579 #endif /* OSD_THANDLE_STATS */
 580
 581 /*
 582  * Concurrency: doesn't access mutable data.
 583  */
 584 static int osd_param_is_sane(const struct osd_device *dev,
 585                              const struct thandle *th)
 586 {
 587         struct osd_thandle *oh;
 588         oh = container_of0(th, struct osd_thandle, ot_super);
 589         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 590 }
 591
 592 /*
 593  * Concurrency: shouldn't matter.
 594  */
 595 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 596 static void osd_trans_commit_cb(struct super_block *sb,
 597                                 struct journal_callback *jcb, int error)
 598 #else
 599 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 600 #endif
 601 {
 602         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 603         struct thandle     *th  = &oh->ot_super;
 604         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 605         struct dt_txn_commit_cb *dcb, *tmp;
 606
 607         LASSERT(oh->ot_handle == NULL);
 608
 609         if (error)
 610                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 611
 612         dt_txn_hook_commit(th);
 613
 614         /* call per-transaction callbacks if any */
 615         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 616                 cfs_list_del_init(&dcb->dcb_linkage);
 617                 dcb->dcb_func(NULL, th, dcb, error);
 618         }
 619
 620         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 621         lu_device_put(lud);
 622         th->th_dev = NULL;
 623
 624         lu_context_exit(&th->th_ctx);
 625         lu_context_fini(&th->th_ctx);
 626         OBD_FREE_PTR(oh);
 627 }
 628
 629 static struct thandle *osd_trans_create(const struct lu_env *env,
 630                                         struct dt_device *d)
 631 {
 632         struct osd_thread_info *oti = osd_oti_get(env);
 633         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 634         struct osd_thandle     *oh;
 635         struct thandle         *th;
 636         ENTRY;
 637
 638         /* on pending IO in this thread should left from prev. request */
 639         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 640
 641         th = ERR_PTR(-ENOMEM);
 642         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 643         if (oh != NULL) {
 644                 th = &oh->ot_super;
 645                 th->th_dev = d;
 646                 th->th_result = 0;
 647                 th->th_tags = LCT_TX_HANDLE;
 648                 oh->ot_credits = 0;
 649                 oti->oti_dev = osd_dt_dev(d);
 650                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 651                 osd_th_alloced(oh);
 652         }
 653         RETURN(th);
 654 }
 655
 656 /*
 657  * Concurrency: shouldn't matter.
 658  */
 659 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 660                     struct thandle *th)
 661 {
 662         struct osd_thread_info *oti = osd_oti_get(env);
 663         struct osd_device  *dev = osd_dt_dev(d);
 664         handle_t           *jh;
 665         struct osd_thandle *oh;
 666         int rc;
 667
 668         ENTRY;
 669
 670         LASSERT(current->journal_info == NULL);
 671
 672         oh = container_of0(th, struct osd_thandle, ot_super);
 673         LASSERT(oh != NULL);
 674         LASSERT(oh->ot_handle == NULL);
 675
 676         rc = dt_txn_hook_start(env, d, th);
 677         if (rc != 0)
 678                 GOTO(out, rc);
 679
 680         if (!osd_param_is_sane(dev, th)) {
 681                 CWARN("%s: too many transaction credits (%d > %d)\n",
 682                       d->dd_lu_dev.ld_obd->obd_name, oh->ot_credits,
 683                       osd_journal(dev)->j_max_transaction_buffers);
 684                 /* XXX Limit the credits to 'max_transaction_buffers', and
 685                  *     let the underlying filesystem to catch the error if
 686                  *     we really need so many credits.
 687                  *
 688                  *     This should be removed when we can calculate the
 689                  *     credits precisely. */
 690                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 691 #ifdef OSD_TRACK_DECLARES
 692                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 693                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 694                        oh->ot_declare_xattr_set);
 695                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 696                        oh->ot_declare_create, oh->ot_declare_ref_add,
 697                        oh->ot_declare_ref_del, oh->ot_declare_write);
 698                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 699                        oh->ot_declare_insert, oh->ot_declare_delete,
 700                        oh->ot_declare_destroy);
 701 #endif
 702         }
 703
 704         /*
 705          * XXX temporary stuff. Some abstraction layer should
 706          * be used.
 707          */
 708         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 709         osd_th_started(oh);
 710         if (!IS_ERR(jh)) {
 711                 oh->ot_handle = jh;
 712                 LASSERT(oti->oti_txns == 0);
 713                 lu_context_init(&th->th_ctx, th->th_tags);
 714                 lu_context_enter(&th->th_ctx);
 715
 716                 lu_device_get(&d->dd_lu_dev);
 717                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 718                                              "osd-tx", th);
 719
 720                 /*
 721                  * XXX: current rule is that we first start tx,
 722                  *      then lock object(s), but we can't use
 723                  *      this rule for data (due to locking specifics
 724                  *      in ldiskfs). also in long-term we'd like to
 725                  *      use usually-used (locks;tx) ordering. so,
 726                  *      UGLY thing is that we'll use one ordering for
 727                  *      data (ofd) and reverse ordering for metadata
 728                  *      (mdd). then at some point we'll fix the latter
 729                  */
 730                 if (lu_device_is_md(&d->dd_lu_dev)) {
 731                         LASSERT(oti->oti_r_locks == 0);
 732                         LASSERT(oti->oti_w_locks == 0);
 733                 }
 734
 735                 oti->oti_txns++;
 736                 rc = 0;
 737         } else {
 738                 rc = PTR_ERR(jh);
 739         }
 740 out:
 741         RETURN(rc);
 742 }
 743
 744 /*
 745  * Concurrency: shouldn't matter.
 746  */
 747 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 748 {
 749         int                     rc = 0;
 750         struct osd_thandle     *oh;
 751         struct osd_thread_info *oti = osd_oti_get(env);
 752         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 753
 754         ENTRY;
 755
 756         oh = container_of0(th, struct osd_thandle, ot_super);
 757
 758         if (oh->ot_handle != NULL) {
 759                 handle_t *hdl = oh->ot_handle;
 760
 761                 hdl->h_sync = th->th_sync;
 762
 763                 /*
 764                  * add commit callback
 765                  * notice we don't do this in osd_trans_start()
 766                  * as underlying transaction can change during truncate
 767                  */
 768                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 769                                          &oh->ot_jcb);
 770
 771                 LASSERT(oti->oti_txns == 1);
 772                 oti->oti_txns--;
 773                 /*
 774                  * XXX: current rule is that we first start tx,
 775                  *      then lock object(s), but we can't use
 776                  *      this rule for data (due to locking specifics
 777                  *      in ldiskfs). also in long-term we'd like to
 778                  *      use usually-used (locks;tx) ordering. so,
 779                  *      UGLY thing is that we'll use one ordering for
 780                  *      data (ofd) and reverse ordering for metadata
 781                  *      (mdd). then at some point we'll fix the latter
 782                  */
 783                 if (lu_device_is_md(&th->th_dev->dd_lu_dev)) {
 784                         LASSERT(oti->oti_r_locks == 0);
 785                         LASSERT(oti->oti_w_locks == 0);
 786                 }
 787                 rc = dt_txn_hook_stop(env, th);
 788                 if (rc != 0)
 789                         CERROR("Failure in transaction hook: %d\n", rc);
 790                 oh->ot_handle = NULL;
 791                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 792                                   rc = ldiskfs_journal_stop(hdl));
 793                 if (rc != 0)
 794                         CERROR("Failure to stop transaction: %d\n", rc);
 795         } else {
 796                 OBD_FREE_PTR(oh);
 797         }
 798
 799         /* as we want IO to journal and data IO be concurrent, we don't block
 800          * awaiting data IO completion in osd_do_bio(), instead we wait here
 801          * once transaction is submitted to the journal. all reqular requests
 802          * don't do direct IO (except read/write), thus this wait_event becomes
 803          * no-op for them.
 804          *
 805          * IMPORTANT: we have to wait till any IO submited by the thread is
 806          * completed otherwise iobuf may be corrupted by different request
 807          */
 808         cfs_wait_event(iobuf->dr_wait,
 809                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 810         if (!rc)
 811                 rc = iobuf->dr_error;
 812
 813         RETURN(rc);
 814 }
 815
 816 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 817 {
 818         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 819                                                ot_super);
 820
 821         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 822
 823         return 0;
 824 }
 825
 826 /*
 827  * Called just before object is freed. Releases all resources except for
 828  * object itself (that is released by osd_object_free()).
 829  *
 830  * Concurrency: no concurrent access is possible that late in object
 831  * life-cycle.
 832  */
 833 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 834 {
 835         struct osd_object *obj   = osd_obj(l);
 836         struct inode      *inode = obj->oo_inode;
 837
 838         LINVRNT(osd_invariant(obj));
 839
 840         /*
 841          * If object is unlinked remove fid->ino mapping from object index.
 842          */
 843
 844         osd_index_fini(obj);
 845         if (inode != NULL) {
 846                 iput(inode);
 847                 obj->oo_inode = NULL;
 848         }
 849 }
 850
 851 /*
 852  * Concurrency: ->loo_object_release() is called under site spin-lock.
 853  */
 854 static void osd_object_release(const struct lu_env *env,
 855                                struct lu_object *l)
 856 {
 857 }
 858
 859 /*
 860  * Concurrency: shouldn't matter.
 861  */
 862 static int osd_object_print(const struct lu_env *env, void *cookie,
 863                             lu_printer_t p, const struct lu_object *l)
 864 {
 865         struct osd_object *o = osd_obj(l);
 866         struct iam_descr  *d;
 867
 868         if (o->oo_dir != NULL)
 869                 d = o->oo_dir->od_container.ic_descr;
 870         else
 871                 d = NULL;
 872         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 873                     o, o->oo_inode,
 874                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 875                     o->oo_inode ? o->oo_inode->i_generation : 0,
 876                     d ? d->id_ops->id_name : "plain");
 877 }
 878
 879 /*
 880  * Concurrency: shouldn't matter.
 881  */
 882 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 883                struct obd_statfs *sfs)
 884 {
 885         struct osd_device  *osd = osd_dt_dev(d);
 886         struct super_block *sb = osd_sb(osd);
 887         struct kstatfs     *ksfs;
 888         int result = 0;
 889
 890         /* osd_lproc.c call this without env, allocate ksfs for that case */
 891         if (unlikely(env == NULL)) {
 892                 OBD_ALLOC_PTR(ksfs);
 893                 if (ksfs == NULL)
 894                         return -ENOMEM;
 895         } else {
 896                 ksfs = &osd_oti_get(env)->oti_ksfs;
 897         }
 898
 899         cfs_spin_lock(&osd->od_osfs_lock);
 900         /* cache 1 second */
 901         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 902                 result = ll_do_statfs(sb, ksfs);
 903                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 904                         osd->od_osfs_age = cfs_time_current_64();
 905                         statfs_pack(&osd->od_statfs, ksfs);
 906                 }
 907         }
 908
 909         if (likely(result == 0))
 910                 *sfs = osd->od_statfs;
 911         cfs_spin_unlock(&osd->od_osfs_lock);
 912
 913         if (unlikely(env == NULL))
 914                 OBD_FREE_PTR(ksfs);
 915
 916         return result;
 917 }
 918
 919 /*
 920  * Concurrency: doesn't access mutable data.
 921  */
 922 static void osd_conf_get(const struct lu_env *env,
 923                          const struct dt_device *dev,
 924                          struct dt_device_param *param)
 925 {
 926         struct super_block *sb = osd_sb(osd_dt_dev(dev));
 927
 928         /*
 929          * XXX should be taken from not-yet-existing fs abstraction layer.
 930          */
 931         param->ddp_max_name_len = LDISKFS_NAME_LEN;
 932         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
 933         param->ddp_block_shift  = sb->s_blocksize_bits;
 934         param->ddp_mntopts      = 0;
 935         if (test_opt(sb, XATTR_USER))
 936                 param->ddp_mntopts |= MNTOPT_USERXATTR;
 937         if (test_opt(sb, POSIX_ACL))
 938                 param->ddp_mntopts |= MNTOPT_ACL;
 939
 940 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
 941         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
 942                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
 943         else
 944 #endif
 945                 param->ddp_max_ea_size = sb->s_blocksize;
 946
 947 }
 948
 949 /**
 950  * Helper function to get and fill the buffer with input values.
 951  */
 952 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
 953 {
 954         struct lu_buf *buf;
 955
 956         buf = &osd_oti_get(env)->oti_buf;
 957         buf->lb_buf = area;
 958         buf->lb_len = len;
 959         return buf;
 960 }
 961
 962 /*
 963  * Concurrency: shouldn't matter.
 964  */
 965 static int osd_sync(const struct lu_env *env, struct dt_device *d)
 966 {
 967         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
 968         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
 969 }
 970
 971 /**
 972  * Start commit for OSD device.
 973  *
 974  * An implementation of dt_commit_async method for OSD device.
 975  * Asychronously starts underlayng fs sync and thereby a transaction
 976  * commit.
 977  *
 978  * \param env environment
 979  * \param d dt device
 980  *
 981  * \see dt_device_operations
 982  */
 983 static int osd_commit_async(const struct lu_env *env,
 984                             struct dt_device *d)
 985 {
 986         struct super_block *s = osd_sb(osd_dt_dev(d));
 987         ENTRY;
 988
 989         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
 990         RETURN(s->s_op->sync_fs(s, 0));
 991 }
 992
 993 /*
 994  * Concurrency: shouldn't matter.
 995  */
 996
 997 static int osd_ro(const struct lu_env *env, struct dt_device *d)
 998 {
 999         struct super_block *sb = osd_sb(osd_dt_dev(d));
1000         int rc;
1001         ENTRY;
1002
1003         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
1004
1005         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1006         RETURN(rc);
1007 }
1008
1009 /*
1010  * Concurrency: serialization provided by callers.
1011  */
1012 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1013                               int mode, unsigned long timeout, __u32 alg,
1014                               struct lustre_capa_key *keys)
1015 {
1016         struct osd_device *dev = osd_dt_dev(d);
1017         ENTRY;
1018
1019         dev->od_fl_capa = mode;
1020         dev->od_capa_timeout = timeout;
1021         dev->od_capa_alg = alg;
1022         dev->od_capa_keys = keys;
1023         RETURN(0);
1024 }
1025
1026 /**
1027  * Concurrency: serialization provided by callers.
1028  */
1029 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1030                                struct dt_quota_ctxt *ctxt, void *data)
1031 {
1032         struct obd_device *obd = (void *)ctxt;
1033         struct vfsmount *mnt = (struct vfsmount *)data;
1034         ENTRY;
1035
1036         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1037         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1038         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1039         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1040         obd->obd_lvfs_ctxt.fs = get_ds();
1041
1042         EXIT;
1043 }
1044
1045 /**
1046  * Note: we do not count into QUOTA here.
1047  * If we mount with --data_journal we may need more.
1048  */
1049 const int osd_dto_credits_noquota[DTO_NR] = {
1050         /**
1051          * Insert/Delete.
1052          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1053          * SINGLEDATA_TRANS_BLOCKS(8)
1054          * XXX Note: maybe iam need more, since iam have more level than
1055          *           EXT3 htree.
1056          */
1057         [DTO_INDEX_INSERT]  = 16,
1058         [DTO_INDEX_DELETE]  = 16,
1059         /**
1060          * Used for OI scrub
1061          */
1062         [DTO_INDEX_UPDATE]  = 16,
1063         /**
1064          * Create a object. The same as create object in EXT3.
1065          * DATA_TRANS_BLOCKS(14) +
1066          * INDEX_EXTRA_BLOCKS(8) +
1067          * 3(inode bits, groups, GDT)
1068          */
1069         [DTO_OBJECT_CREATE] = 25,
1070         /**
1071          * XXX: real credits to be fixed
1072          */
1073         [DTO_OBJECT_DELETE] = 25,
1074         /**
1075          * Attr set credits (inode)
1076          */
1077         [DTO_ATTR_SET_BASE] = 1,
1078         /**
1079          * Xattr set. The same as xattr of EXT3.
1080          * DATA_TRANS_BLOCKS(14)
1081          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1082          * are also counted in. Do not know why?
1083          */
1084         [DTO_XATTR_SET]     = 14,
1085         [DTO_LOG_REC]       = 14,
1086         /**
1087          * credits for inode change during write.
1088          */
1089         [DTO_WRITE_BASE]    = 3,
1090         /**
1091          * credits for single block write.
1092          */
1093         [DTO_WRITE_BLOCK]   = 14,
1094         /**
1095          * Attr set credits for chown.
1096          * This is extra credits for setattr, and it is null without quota
1097          */
1098         [DTO_ATTR_SET_CHOWN]= 0
1099 };
1100
1101 static const struct dt_device_operations osd_dt_ops = {
1102         .dt_root_get       = osd_root_get,
1103         .dt_statfs         = osd_statfs,
1104         .dt_trans_create   = osd_trans_create,
1105         .dt_trans_start    = osd_trans_start,
1106         .dt_trans_stop     = osd_trans_stop,
1107         .dt_trans_cb_add   = osd_trans_cb_add,
1108         .dt_conf_get       = osd_conf_get,
1109         .dt_sync           = osd_sync,
1110         .dt_ro             = osd_ro,
1111         .dt_commit_async   = osd_commit_async,
1112         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1113         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1114 };
1115
1116 static void osd_object_read_lock(const struct lu_env *env,
1117                                  struct dt_object *dt, unsigned role)
1118 {
1119         struct osd_object *obj = osd_dt_obj(dt);
1120         struct osd_thread_info *oti = osd_oti_get(env);
1121
1122         LINVRNT(osd_invariant(obj));
1123
1124         LASSERT(obj->oo_owner != env);
1125         cfs_down_read_nested(&obj->oo_sem, role);
1126
1127         LASSERT(obj->oo_owner == NULL);
1128         oti->oti_r_locks++;
1129 }
1130
1131 static void osd_object_write_lock(const struct lu_env *env,
1132                                   struct dt_object *dt, unsigned role)
1133 {
1134         struct osd_object *obj = osd_dt_obj(dt);
1135         struct osd_thread_info *oti = osd_oti_get(env);
1136
1137         LINVRNT(osd_invariant(obj));
1138
1139         LASSERT(obj->oo_owner != env);
1140         cfs_down_write_nested(&obj->oo_sem, role);
1141
1142         LASSERT(obj->oo_owner == NULL);
1143         obj->oo_owner = env;
1144         oti->oti_w_locks++;
1145 }
1146
1147 static void osd_object_read_unlock(const struct lu_env *env,
1148                                    struct dt_object *dt)
1149 {
1150         struct osd_object *obj = osd_dt_obj(dt);
1151         struct osd_thread_info *oti = osd_oti_get(env);
1152
1153         LINVRNT(osd_invariant(obj));
1154
1155         LASSERT(oti->oti_r_locks > 0);
1156         oti->oti_r_locks--;
1157         cfs_up_read(&obj->oo_sem);
1158 }
1159
1160 static void osd_object_write_unlock(const struct lu_env *env,
1161                                     struct dt_object *dt)
1162 {
1163         struct osd_object *obj = osd_dt_obj(dt);
1164         struct osd_thread_info *oti = osd_oti_get(env);
1165
1166         LINVRNT(osd_invariant(obj));
1167
1168         LASSERT(obj->oo_owner == env);
1169         LASSERT(oti->oti_w_locks > 0);
1170         oti->oti_w_locks--;
1171         obj->oo_owner = NULL;
1172         cfs_up_write(&obj->oo_sem);
1173 }
1174
1175 static int osd_object_write_locked(const struct lu_env *env,
1176                                    struct dt_object *dt)
1177 {
1178         struct osd_object *obj = osd_dt_obj(dt);
1179
1180         LINVRNT(osd_invariant(obj));
1181
1182         return obj->oo_owner == env;
1183 }
1184
1185 static int capa_is_sane(const struct lu_env *env,
1186                         struct osd_device *dev,
1187                         struct lustre_capa *capa,
1188                         struct lustre_capa_key *keys)
1189 {
1190         struct osd_thread_info *oti = osd_oti_get(env);
1191         struct lustre_capa *tcapa = &oti->oti_capa;
1192         struct obd_capa *oc;
1193         int i, rc = 0;
1194         ENTRY;
1195
1196         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1197         if (oc) {
1198                 if (capa_is_expired(oc)) {
1199                         DEBUG_CAPA(D_ERROR, capa, "expired");
1200                         rc = -ESTALE;
1201                 }
1202                 capa_put(oc);
1203                 RETURN(rc);
1204         }
1205
1206         if (capa_is_expired_sec(capa)) {
1207                 DEBUG_CAPA(D_ERROR, capa, "expired");
1208                 RETURN(-ESTALE);
1209         }
1210
1211         cfs_spin_lock(&capa_lock);
1212         for (i = 0; i < 2; i++) {
1213                 if (keys[i].lk_keyid == capa->lc_keyid) {
1214                         oti->oti_capa_key = keys[i];
1215                         break;
1216                 }
1217         }
1218         cfs_spin_unlock(&capa_lock);
1219
1220         if (i == 2) {
1221                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1222                 RETURN(-ESTALE);
1223         }
1224
1225         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1226         if (rc)
1227                 RETURN(rc);
1228
1229         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1230                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1231                 RETURN(-EACCES);
1232         }
1233
1234         oc = capa_add(dev->od_capa_hash, capa);
1235         capa_put(oc);
1236
1237         RETURN(0);
1238 }
1239
1240 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1241                     struct lustre_capa *capa, __u64 opc)
1242 {
1243         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1244         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1245         struct md_capainfo *ci;
1246         int rc;
1247
1248         if (!dev->od_fl_capa)
1249                 return 0;
1250
1251         if (capa == BYPASS_CAPA)
1252                 return 0;
1253
1254         ci = md_capainfo(env);
1255         if (unlikely(!ci))
1256                 return 0;
1257
1258         if (ci->mc_auth == LC_ID_NONE)
1259                 return 0;
1260
1261         if (!capa) {
1262                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1263                 return -EACCES;
1264         }
1265
1266         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1267                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1268                            PFID(fid));
1269                 return -EACCES;
1270         }
1271
1272         if (!capa_opc_supported(capa, opc)) {
1273                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1274                 return -EACCES;
1275         }
1276
1277         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1278                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1279                 return -EACCES;
1280         }
1281
1282         return 0;
1283 }
1284
1285 static struct timespec *osd_inode_time(const struct lu_env *env,
1286                                        struct inode *inode, __u64 seconds)
1287 {
1288         struct osd_thread_info *oti = osd_oti_get(env);
1289         struct timespec        *t   = &oti->oti_time;
1290
1291         t->tv_sec  = seconds;
1292         t->tv_nsec = 0;
1293         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1294         return t;
1295 }
1296
1297
1298 static void osd_inode_getattr(const struct lu_env *env,
1299                               struct inode *inode, struct lu_attr *attr)
1300 {
1301         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1302                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1303                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1304
1305         attr->la_atime      = LTIME_S(inode->i_atime);
1306         attr->la_mtime      = LTIME_S(inode->i_mtime);
1307         attr->la_ctime      = LTIME_S(inode->i_ctime);
1308         attr->la_mode       = inode->i_mode;
1309         attr->la_size       = i_size_read(inode);
1310         attr->la_blocks     = inode->i_blocks;
1311         attr->la_uid        = inode->i_uid;
1312         attr->la_gid        = inode->i_gid;
1313         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1314         attr->la_nlink      = inode->i_nlink;
1315         attr->la_rdev       = inode->i_rdev;
1316         attr->la_blksize    = ll_inode_blksize(inode);
1317         attr->la_blkbits    = inode->i_blkbits;
1318 }
1319
1320 static int osd_attr_get(const struct lu_env *env,
1321                         struct dt_object *dt,
1322                         struct lu_attr *attr,
1323                         struct lustre_capa *capa)
1324 {
1325         struct osd_object *obj = osd_dt_obj(dt);
1326
1327         LASSERT(dt_object_exists(dt));
1328         LINVRNT(osd_invariant(obj));
1329
1330         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1331                 return -EACCES;
1332
1333         cfs_spin_lock(&obj->oo_guard);
1334         osd_inode_getattr(env, obj->oo_inode, attr);
1335         cfs_spin_unlock(&obj->oo_guard);
1336         return 0;
1337 }
1338
1339 static int osd_declare_attr_set(const struct lu_env *env,
1340                                 struct dt_object *dt,
1341                                 const struct lu_attr *attr,
1342                                 struct thandle *handle)
1343 {
1344         struct osd_thandle *oh;
1345         struct osd_object *obj;
1346
1347         LASSERT(dt != NULL);
1348         LASSERT(handle != NULL);
1349
1350         obj = osd_dt_obj(dt);
1351         LASSERT(osd_invariant(obj));
1352
1353         oh = container_of0(handle, struct osd_thandle, ot_super);
1354         LASSERT(oh->ot_handle == NULL);
1355
1356         OSD_DECLARE_OP(oh, attr_set);
1357         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1358
1359         if (attr && attr->la_valid & LA_UID) {
1360                 if (obj->oo_inode)
1361                         osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
1362                                         obj->oo_inode);
1363                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1364         }
1365         if (attr && attr->la_valid & LA_GID) {
1366                 if (obj->oo_inode)
1367                         osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
1368                                         obj->oo_inode);
1369                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1370         }
1371
1372         return 0;
1373 }
1374
1375 static int osd_inode_setattr(const struct lu_env *env,
1376                              struct inode *inode, const struct lu_attr *attr)
1377 {
1378         __u64 bits;
1379
1380         bits = attr->la_valid;
1381
1382         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1383
1384         if (bits & LA_ATIME)
1385                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1386         if (bits & LA_CTIME)
1387                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1388         if (bits & LA_MTIME)
1389                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1390         if (bits & LA_SIZE) {
1391                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1392                 i_size_write(inode, attr->la_size);
1393         }
1394
1395 #if 0
1396         /* OSD should not change "i_blocks" which is used by quota.
1397          * "i_blocks" should be changed by ldiskfs only. */
1398         if (bits & LA_BLOCKS)
1399                 inode->i_blocks = attr->la_blocks;
1400 #endif
1401         if (bits & LA_MODE)
1402                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1403                         (attr->la_mode & ~S_IFMT);
1404         if (bits & LA_UID)
1405                 inode->i_uid    = attr->la_uid;
1406         if (bits & LA_GID)
1407                 inode->i_gid    = attr->la_gid;
1408         if (bits & LA_NLINK)
1409                 inode->i_nlink  = attr->la_nlink;
1410         if (bits & LA_RDEV)
1411                 inode->i_rdev   = attr->la_rdev;
1412
1413         if (bits & LA_FLAGS) {
1414                 /* always keep S_NOCMTIME */
1415                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1416                                  S_NOCMTIME;
1417         }
1418         return 0;
1419 }
1420
1421 static int osd_attr_set(const struct lu_env *env,
1422                         struct dt_object *dt,
1423                         const struct lu_attr *attr,
1424                         struct thandle *handle,
1425                         struct lustre_capa *capa)
1426 {
1427         struct osd_object *obj = osd_dt_obj(dt);
1428         struct inode      *inode;
1429         int rc;
1430
1431         LASSERT(handle != NULL);
1432         LASSERT(dt_object_exists(dt));
1433         LASSERT(osd_invariant(obj));
1434
1435         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1436                 return -EACCES;
1437
1438         OSD_EXEC_OP(handle, attr_set);
1439
1440         inode = obj->oo_inode;
1441 #ifdef HAVE_QUOTA_SUPPORT
1442         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1443             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1444                 struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1445                 struct iattr iattr;
1446                 int rc;
1447
1448                 iattr.ia_valid = 0;
1449                 if (attr->la_valid & LA_UID)
1450                         iattr.ia_valid |= ATTR_UID;
1451                 if (attr->la_valid & LA_GID)
1452                         iattr.ia_valid |= ATTR_GID;
1453                 iattr.ia_uid = attr->la_uid;
1454                 iattr.ia_gid = attr->la_gid;
1455                 osd_push_ctxt(env, save);
1456                 rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1457                 osd_pop_ctxt(save);
1458                 if (rc != 0)
1459                         return rc;
1460         }
1461 #endif
1462
1463         cfs_spin_lock(&obj->oo_guard);
1464         rc = osd_inode_setattr(env, inode, attr);
1465         cfs_spin_unlock(&obj->oo_guard);
1466
1467         if (!rc)
1468                 inode->i_sb->s_op->dirty_inode(inode);
1469         return rc;
1470 }
1471
1472 /*
1473  * Object creation.
1474  *
1475  * XXX temporary solution.
1476  */
1477 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1478                           struct lu_attr *attr, struct thandle *th)
1479 {
1480         return 0;
1481 }
1482
1483 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1484                            struct lu_attr *attr, struct thandle *th)
1485 {
1486         osd_object_init0(obj);
1487         if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1488                 unlock_new_inode(obj->oo_inode);
1489         return 0;
1490 }
1491
1492 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1493                                     struct osd_object *obj,
1494                                     const char *name, const int namelen)
1495 {
1496         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1497 }
1498
1499 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1500                       cfs_umode_t mode,
1501                       struct dt_allocation_hint *hint,
1502                       struct thandle *th)
1503 {
1504         int result;
1505         struct osd_device  *osd = osd_obj2dev(obj);
1506         struct osd_thandle *oth;
1507         struct dt_object   *parent = NULL;
1508         struct inode       *inode;
1509 #ifdef HAVE_QUOTA_SUPPORT
1510         struct osd_ctxt    *save = &info->oti_ctxt;
1511 #endif
1512
1513         LINVRNT(osd_invariant(obj));
1514         LASSERT(obj->oo_inode == NULL);
1515         LASSERT(obj->oo_hl_head == NULL);
1516
1517         if (S_ISDIR(mode) && ldiskfs_pdo) {
1518                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1519                 if (obj->oo_hl_head == NULL)
1520                         return -ENOMEM;
1521         }
1522
1523         oth = container_of(th, struct osd_thandle, ot_super);
1524         LASSERT(oth->ot_handle->h_transaction != NULL);
1525
1526         if (hint && hint->dah_parent)
1527                 parent = hint->dah_parent;
1528
1529 #ifdef HAVE_QUOTA_SUPPORT
1530         osd_push_ctxt(info->oti_env, save);
1531 #endif
1532         inode = ldiskfs_create_inode(oth->ot_handle,
1533                                      parent ? osd_dt_obj(parent)->oo_inode :
1534                                               osd_sb(osd)->s_root->d_inode,
1535                                      mode);
1536 #ifdef HAVE_QUOTA_SUPPORT
1537         osd_pop_ctxt(save);
1538 #endif
1539         if (!IS_ERR(inode)) {
1540                 /* Do not update file c/mtime in ldiskfs.
1541                  * NB: don't need any lock because no contention at this
1542                  * early stage */
1543                 inode->i_flags |= S_NOCMTIME;
1544                 inode->i_state |= I_LUSTRE_NOSCRUB;
1545                 obj->oo_inode = inode;
1546                 result = 0;
1547         } else {
1548                 if (obj->oo_hl_head != NULL) {
1549                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1550                         obj->oo_hl_head = NULL;
1551                 }
1552                 result = PTR_ERR(inode);
1553         }
1554         LINVRNT(osd_invariant(obj));
1555         return result;
1556 }
1557
1558 enum {
1559         OSD_NAME_LEN = 255
1560 };
1561
1562 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1563                      struct lu_attr *attr,
1564                      struct dt_allocation_hint *hint,
1565                      struct dt_object_format *dof,
1566                      struct thandle *th)
1567 {
1568         int result;
1569         struct osd_thandle *oth;
1570         struct osd_device *osd = osd_obj2dev(obj);
1571         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1572
1573         LASSERT(S_ISDIR(attr->la_mode));
1574
1575         oth = container_of(th, struct osd_thandle, ot_super);
1576         LASSERT(oth->ot_handle->h_transaction != NULL);
1577         result = osd_mkfile(info, obj, mode, hint, th);
1578         if (result == 0 && osd->od_iop_mode == 0) {
1579                 LASSERT(obj->oo_inode != NULL);
1580                 /*
1581                  * XXX uh-oh... call low-level iam function directly.
1582                  */
1583
1584                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1585                                          sizeof (struct osd_fid_pack),
1586                                          oth->ot_handle);
1587         }
1588         return result;
1589 }
1590
1591 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1592                         struct lu_attr *attr,
1593                         struct dt_allocation_hint *hint,
1594                         struct dt_object_format *dof,
1595                         struct thandle *th)
1596 {
1597         int result;
1598         struct osd_thandle *oth;
1599         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1600
1601         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1602
1603         LASSERT(S_ISREG(attr->la_mode));
1604
1605         oth = container_of(th, struct osd_thandle, ot_super);
1606         LASSERT(oth->ot_handle->h_transaction != NULL);
1607
1608         result = osd_mkfile(info, obj, mode, hint, th);
1609         if (result == 0) {
1610                 LASSERT(obj->oo_inode != NULL);
1611                 if (feat->dif_flags & DT_IND_VARKEY)
1612                         result = iam_lvar_create(obj->oo_inode,
1613                                                  feat->dif_keysize_max,
1614                                                  feat->dif_ptrsize,
1615                                                  feat->dif_recsize_max,
1616                                                  oth->ot_handle);
1617                 else
1618                         result = iam_lfix_create(obj->oo_inode,
1619                                                  feat->dif_keysize_max,
1620                                                  feat->dif_ptrsize,
1621                                                  feat->dif_recsize_max,
1622                                                  oth->ot_handle);
1623
1624         }
1625         return result;
1626 }
1627
1628 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1629                      struct lu_attr *attr,
1630                      struct dt_allocation_hint *hint,
1631                      struct dt_object_format *dof,
1632                      struct thandle *th)
1633 {
1634         LASSERT(S_ISREG(attr->la_mode));
1635         return osd_mkfile(info, obj, (attr->la_mode &
1636                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1637 }
1638
1639 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1640                      struct lu_attr *attr,
1641                      struct dt_allocation_hint *hint,
1642                      struct dt_object_format *dof,
1643                      struct thandle *th)
1644 {
1645         LASSERT(S_ISLNK(attr->la_mode));
1646         return osd_mkfile(info, obj, (attr->la_mode &
1647                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1648 }
1649
1650 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1651                      struct lu_attr *attr,
1652                      struct dt_allocation_hint *hint,
1653                      struct dt_object_format *dof,
1654                      struct thandle *th)
1655 {
1656         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1657         int result;
1658
1659         LINVRNT(osd_invariant(obj));
1660         LASSERT(obj->oo_inode == NULL);
1661         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1662                 S_ISFIFO(mode) || S_ISSOCK(mode));
1663
1664         result = osd_mkfile(info, obj, mode, hint, th);
1665         if (result == 0) {
1666                 LASSERT(obj->oo_inode != NULL);
1667                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1668         }
1669         LINVRNT(osd_invariant(obj));
1670         return result;
1671 }
1672
1673 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1674                               struct lu_attr *,
1675                               struct dt_allocation_hint *hint,
1676                               struct dt_object_format *dof,
1677                               struct thandle *);
1678
1679 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1680 {
1681         osd_obj_type_f result;
1682
1683         switch (type) {
1684         case DFT_DIR:
1685                 result = osd_mkdir;
1686                 break;
1687         case DFT_REGULAR:
1688                 result = osd_mkreg;
1689                 break;
1690         case DFT_SYM:
1691                 result = osd_mksym;
1692                 break;
1693         case DFT_NODE:
1694                 result = osd_mknod;
1695                 break;
1696         case DFT_INDEX:
1697                 result = osd_mk_index;
1698                 break;
1699
1700         default:
1701                 LBUG();
1702                 break;
1703         }
1704         return result;
1705 }
1706
1707
1708 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1709                         struct dt_object *parent, cfs_umode_t child_mode)
1710 {
1711         LASSERT(ah);
1712
1713         memset(ah, 0, sizeof(*ah));
1714         ah->dah_parent = parent;
1715         ah->dah_mode = child_mode;
1716 }
1717
1718 /**
1719  * Helper function for osd_object_create()
1720  *
1721  * \retval 0, on success
1722  */
1723 static int __osd_object_create(struct osd_thread_info *info,
1724                                struct osd_object *obj, struct lu_attr *attr,
1725                                struct dt_allocation_hint *hint,
1726                                struct dt_object_format *dof,
1727                                struct thandle *th)
1728 {
1729
1730         int result;
1731
1732         result = osd_create_pre(info, obj, attr, th);
1733         if (result == 0) {
1734                 result = osd_create_type_f(dof->dof_type)(info, obj,
1735                                            attr, hint, dof, th);
1736                 if (result == 0)
1737                         result = osd_create_post(info, obj, attr, th);
1738         }
1739         return result;
1740 }
1741
1742 /**
1743  * Helper function for osd_object_create()
1744  *
1745  * \retval 0, on success
1746  */
1747 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1748                            const struct lu_fid *fid, struct thandle *th)
1749 {
1750         struct osd_thread_info *info = osd_oti_get(env);
1751         struct osd_inode_id    *id   = &info->oti_id;
1752         struct osd_device      *osd  = osd_obj2dev(obj);
1753         struct md_ucred        *uc   = md_ucred(env);
1754
1755         LASSERT(obj->oo_inode != NULL);
1756         LASSERT(uc != NULL);
1757
1758         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1759         return osd_oi_insert(info, osd, fid, id, th);
1760 }
1761
1762 static int osd_declare_object_create(const struct lu_env *env,
1763                                      struct dt_object *dt,
1764                                      struct lu_attr *attr,
1765                                      struct dt_allocation_hint *hint,
1766                                      struct dt_object_format *dof,
1767                                      struct thandle *handle)
1768 {
1769         struct osd_thandle *oh;
1770
1771         LASSERT(handle != NULL);
1772
1773         oh = container_of0(handle, struct osd_thandle, ot_super);
1774         LASSERT(oh->ot_handle == NULL);
1775
1776         OSD_DECLARE_OP(oh, create);
1777         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1778         /* XXX: So far, only normal fid needs be inserted into the oi,
1779          *      things could be changed later. Revise following code then. */
1780         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1781                 OSD_DECLARE_OP(oh, insert);
1782                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1783         }
1784         /* If this is directory, then we expect . and .. to be inserted as
1785          * well. The one directory block always needs to be created for the
1786          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1787          * block), there is no danger of needing a tree for the first block.
1788          */
1789         if (attr && S_ISDIR(attr->la_mode)) {
1790                 OSD_DECLARE_OP(oh, insert);
1791                 OSD_DECLARE_OP(oh, insert);
1792                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1793         }
1794
1795         if (attr) {
1796                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1797                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1798         }
1799         return 0;
1800 }
1801
1802 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1803                              struct lu_attr *attr,
1804                              struct dt_allocation_hint *hint,
1805                              struct dt_object_format *dof,
1806                              struct thandle *th)
1807 {
1808         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1809         struct osd_object      *obj    = osd_dt_obj(dt);
1810         struct osd_thread_info *info   = osd_oti_get(env);
1811         int result;
1812
1813         ENTRY;
1814
1815         LINVRNT(osd_invariant(obj));
1816         LASSERT(!dt_object_exists(dt));
1817         LASSERT(osd_write_locked(env, obj));
1818         LASSERT(th != NULL);
1819
1820         OSD_EXEC_OP(th, create);
1821
1822         result = __osd_object_create(info, obj, attr, hint, dof, th);
1823         if (result == 0)
1824                 result = __osd_oi_insert(env, obj, fid, th);
1825
1826         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1827         LASSERT(osd_invariant(obj));
1828         RETURN(result);
1829 }
1830
1831 /**
1832  * Called to destroy on-disk representation of the object
1833  *
1834  * Concurrency: must be locked
1835  */
1836 static int osd_declare_object_destroy(const struct lu_env *env,
1837                                       struct dt_object *dt,
1838                                       struct thandle *th)
1839 {
1840         struct osd_object  *obj = osd_dt_obj(dt);
1841         struct inode       *inode = obj->oo_inode;
1842         struct osd_thandle *oh;
1843
1844         ENTRY;
1845
1846         oh = container_of0(th, struct osd_thandle, ot_super);
1847         LASSERT(oh->ot_handle == NULL);
1848         LASSERT(inode);
1849
1850         OSD_DECLARE_OP(oh, destroy);
1851         OSD_DECLARE_OP(oh, delete);
1852         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
1853         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
1854
1855         osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
1856         osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
1857
1858         RETURN(0);
1859 }
1860
1861 static int osd_object_destroy(const struct lu_env *env,
1862                               struct dt_object *dt,
1863                               struct thandle *th)
1864 {
1865         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
1866         struct osd_object      *obj = osd_dt_obj(dt);
1867         struct inode           *inode = obj->oo_inode;
1868         struct osd_device      *osd = osd_obj2dev(obj);
1869         struct osd_thandle     *oh;
1870         int                     result;
1871         ENTRY;
1872
1873         oh = container_of0(th, struct osd_thandle, ot_super);
1874         LASSERT(oh->ot_handle);
1875         LASSERT(inode);
1876         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
1877
1878         /* Parallel control for OI scrub. For most of cases, there is no
1879          * lock contention. So it will not affect unlink performance. */
1880         cfs_mutex_lock(&inode->i_mutex);
1881         if (S_ISDIR(inode->i_mode)) {
1882                 LASSERT(osd_inode_unlinked(inode) ||
1883                         inode->i_nlink == 1);
1884                 cfs_spin_lock(&obj->oo_guard);
1885                 inode->i_nlink = 0;
1886                 cfs_spin_unlock(&obj->oo_guard);
1887                 inode->i_sb->s_op->dirty_inode(inode);
1888         } else {
1889                 LASSERT(osd_inode_unlinked(inode));
1890         }
1891
1892         OSD_EXEC_OP(th, destroy);
1893
1894         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
1895         cfs_mutex_unlock(&inode->i_mutex);
1896
1897         /* XXX: add to ext3 orphan list */
1898         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
1899
1900         /* not needed in the cache anymore */
1901         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
1902
1903         RETURN(0);
1904 }
1905
1906 /**
1907  * Helper function for osd_xattr_set()
1908  */
1909 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1910                            const struct lu_buf *buf, const char *name, int fl)
1911 {
1912         struct osd_object      *obj      = osd_dt_obj(dt);
1913         struct inode           *inode    = obj->oo_inode;
1914         struct osd_thread_info *info     = osd_oti_get(env);
1915         struct dentry          *dentry   = &info->oti_child_dentry;
1916         int                     fs_flags = 0;
1917         int                     rc;
1918
1919         LASSERT(dt_object_exists(dt));
1920         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1921
1922         if (fl & LU_XATTR_REPLACE)
1923                 fs_flags |= XATTR_REPLACE;
1924
1925         if (fl & LU_XATTR_CREATE)
1926                 fs_flags |= XATTR_CREATE;
1927
1928         dentry->d_inode = inode;
1929         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
1930                                    buf->lb_len, fs_flags);
1931         return rc;
1932 }
1933
1934 /**
1935  * Put the fid into lustre_mdt_attrs, and then place the structure
1936  * inode's ea. This fid should not be altered during the life time
1937  * of the inode.
1938  *
1939  * \retval +ve, on success
1940  * \retval -ve, on error
1941  *
1942  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
1943  */
1944 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
1945                           const struct lu_fid *fid)
1946 {
1947         struct osd_thread_info  *info      = osd_oti_get(env);
1948         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
1949
1950         lustre_lma_init(mdt_attrs, fid);
1951         lustre_lma_swab(mdt_attrs);
1952         return __osd_xattr_set(env, dt,
1953                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
1954                                XATTR_NAME_LMA, LU_XATTR_CREATE);
1955
1956 }
1957
1958 /**
1959  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
1960  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
1961  * To have compatilibility with 1.8 ldiskfs driver we need to have
1962  * magic number at start of fid data.
1963  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
1964  * its inmemory API.
1965  */
1966 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
1967                                   const struct dt_rec *fid)
1968 {
1969         param->edp_magic = LDISKFS_LUFID_MAGIC;
1970         param->edp_len =  sizeof(struct lu_fid) + 1;
1971
1972         fid_cpu_to_be((struct lu_fid *)param->edp_data,
1973                       (struct lu_fid *)fid);
1974 }
1975
1976 /**
1977  * Try to read the fid from inode ea into dt_rec, if return value
1978  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
1979  *
1980  * \param fid object fid.
1981  *
1982  * \retval 0 on success
1983  */
1984 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
1985                           __u32 ino, struct lu_fid *fid)
1986 {
1987         struct osd_thread_info  *info = osd_oti_get(env);
1988         struct osd_inode_id     *id = &info->oti_id;
1989         struct inode            *inode;
1990         ENTRY;
1991
1992         osd_id_gen(id, ino, OSD_OII_NOGEN);
1993         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
1994         if (IS_ERR(inode))
1995                 RETURN(PTR_ERR(inode));
1996
1997         iput(inode);
1998         RETURN(0);
1999 }
2000
2001 /**
2002  * OSD layer object create function for interoperability mode (b11826).
2003  * This is mostly similar to osd_object_create(). Only difference being, fid is
2004  * inserted into inode ea here.
2005  *
2006  * \retval   0, on success
2007  * \retval -ve, on error
2008  */
2009 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2010                                 struct lu_attr *attr,
2011                                 struct dt_allocation_hint *hint,
2012                                 struct dt_object_format *dof,
2013                                 struct thandle *th)
2014 {
2015         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2016         struct osd_object      *obj    = osd_dt_obj(dt);
2017         struct osd_thread_info *info   = osd_oti_get(env);
2018         int                     result;
2019
2020         ENTRY;
2021
2022         LASSERT(osd_invariant(obj));
2023         LASSERT(!dt_object_exists(dt));
2024         LASSERT(osd_write_locked(env, obj));
2025         LASSERT(th != NULL);
2026
2027         OSD_EXEC_OP(th, create);
2028
2029         result = __osd_object_create(info, obj, attr, hint, dof, th);
2030         /* objects under osd root shld have igif fid, so dont add fid EA */
2031         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2032                 result = osd_ea_fid_set(env, dt, fid);
2033
2034         if (result == 0)
2035                 result = __osd_oi_insert(env, obj, fid, th);
2036
2037         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2038         LINVRNT(osd_invariant(obj));
2039         RETURN(result);
2040 }
2041
2042 static int osd_declare_object_ref_add(const struct lu_env *env,
2043                                       struct dt_object *dt,
2044                                       struct thandle *handle)
2045 {
2046         struct osd_thandle *oh;
2047
2048         /* it's possible that object doesn't exist yet */
2049         LASSERT(handle != NULL);
2050
2051         oh = container_of0(handle, struct osd_thandle, ot_super);
2052         LASSERT(oh->ot_handle == NULL);
2053
2054         OSD_DECLARE_OP(oh, ref_add);
2055         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2056
2057         return 0;
2058 }
2059
2060 /*
2061  * Concurrency: @dt is write locked.
2062  */
2063 static int osd_object_ref_add(const struct lu_env *env,
2064                               struct dt_object *dt, struct thandle *th)
2065 {
2066         struct osd_object *obj = osd_dt_obj(dt);
2067         struct inode      *inode = obj->oo_inode;
2068
2069         LINVRNT(osd_invariant(obj));
2070         LASSERT(dt_object_exists(dt));
2071         LASSERT(osd_write_locked(env, obj));
2072         LASSERT(th != NULL);
2073
2074         OSD_EXEC_OP(th, ref_add);
2075
2076         /*
2077          * DIR_NLINK feature is set for compatibility reasons if:
2078          * 1) nlinks > LDISKFS_LINK_MAX, or
2079          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2080          *
2081          * It is easier to always set this flag (rather than check and set),
2082          * since it has less overhead, and the superblock will be dirtied
2083          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2084          * do not actually care whether this flag is set or not.
2085          */
2086         cfs_spin_lock(&obj->oo_guard);
2087         inode->i_nlink++;
2088         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2089                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2090                     inode->i_nlink == 2)
2091                         inode->i_nlink = 1;
2092         }
2093         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2094         cfs_spin_unlock(&obj->oo_guard);
2095         inode->i_sb->s_op->dirty_inode(inode);
2096         LINVRNT(osd_invariant(obj));
2097
2098         return 0;
2099 }
2100
2101 static int osd_declare_object_ref_del(const struct lu_env *env,
2102                                       struct dt_object *dt,
2103                                       struct thandle *handle)
2104 {
2105         struct osd_thandle *oh;
2106
2107         LASSERT(dt_object_exists(dt));
2108         LASSERT(handle != NULL);
2109
2110         oh = container_of0(handle, struct osd_thandle, ot_super);
2111         LASSERT(oh->ot_handle == NULL);
2112
2113         OSD_DECLARE_OP(oh, ref_del);
2114         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2115
2116         return 0;
2117 }
2118
2119 /*
2120  * Concurrency: @dt is write locked.
2121  */
2122 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2123                               struct thandle *th)
2124 {
2125         struct osd_object *obj = osd_dt_obj(dt);
2126         struct inode      *inode = obj->oo_inode;
2127
2128         LINVRNT(osd_invariant(obj));
2129         LASSERT(dt_object_exists(dt));
2130         LASSERT(osd_write_locked(env, obj));
2131         LASSERT(th != NULL);
2132
2133         OSD_EXEC_OP(th, ref_del);
2134
2135         cfs_spin_lock(&obj->oo_guard);
2136         LASSERT(inode->i_nlink > 0);
2137         inode->i_nlink--;
2138         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2139          * then the nlink count is 1. Don't let it be set to 0 or the directory
2140          * inode will be deleted incorrectly. */
2141         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2142                 inode->i_nlink++;
2143         cfs_spin_unlock(&obj->oo_guard);
2144         inode->i_sb->s_op->dirty_inode(inode);
2145         LINVRNT(osd_invariant(obj));
2146
2147         return 0;
2148 }
2149
2150 /*
2151  * Get the 64-bit version for an inode.
2152  */
2153 static int osd_object_version_get(const struct lu_env *env,
2154                                   struct dt_object *dt, dt_obj_version_t *ver)
2155 {
2156         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2157
2158         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2159                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2160         *ver = LDISKFS_I(inode)->i_fs_version;
2161         return 0;
2162 }
2163
2164 /*
2165  * Concurrency: @dt is read locked.
2166  */
2167 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2168                          struct lu_buf *buf, const char *name,
2169                          struct lustre_capa *capa)
2170 {
2171         struct osd_object      *obj    = osd_dt_obj(dt);
2172         struct inode           *inode  = obj->oo_inode;
2173         struct osd_thread_info *info   = osd_oti_get(env);
2174         struct dentry          *dentry = &info->oti_obj_dentry;
2175
2176         /* version get is not real XATTR but uses xattr API */
2177         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2178                 /* for version we are just using xattr API but change inode
2179                  * field instead */
2180                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2181                 osd_object_version_get(env, dt, buf->lb_buf);
2182                 return sizeof(dt_obj_version_t);
2183         }
2184
2185         LASSERT(dt_object_exists(dt));
2186         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2187         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2188
2189         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2190                 return -EACCES;
2191
2192         dentry->d_inode = inode;
2193         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2194 }
2195
2196
2197 static int osd_declare_xattr_set(const struct lu_env *env,
2198                                  struct dt_object *dt,
2199                                  const struct lu_buf *buf, const char *name,
2200                                  int fl, struct thandle *handle)
2201 {
2202         struct osd_thandle *oh;
2203
2204         LASSERT(handle != NULL);
2205
2206         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2207                 /* no credits for version */
2208                 return 0;
2209         }
2210
2211         oh = container_of0(handle, struct osd_thandle, ot_super);
2212         LASSERT(oh->ot_handle == NULL);
2213
2214         OSD_DECLARE_OP(oh, xattr_set);
2215         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2216
2217         return 0;
2218 }
2219
2220 /*
2221  * Set the 64-bit version for object
2222  */
2223 static void osd_object_version_set(const struct lu_env *env,
2224                                    struct dt_object *dt,
2225                                    dt_obj_version_t *new_version)
2226 {
2227         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2228
2229         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2230                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2231
2232         LDISKFS_I(inode)->i_fs_version = *new_version;
2233         /** Version is set after all inode operations are finished,
2234          *  so we should mark it dirty here */
2235         inode->i_sb->s_op->dirty_inode(inode);
2236 }
2237
2238 /*
2239  * Concurrency: @dt is write locked.
2240  */
2241 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2242                          const struct lu_buf *buf, const char *name, int fl,
2243                          struct thandle *handle, struct lustre_capa *capa)
2244 {
2245         LASSERT(handle != NULL);
2246
2247         /* version set is not real XATTR */
2248         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2249                 /* for version we are just using xattr API but change inode
2250                  * field instead */
2251                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2252                 osd_object_version_set(env, dt, buf->lb_buf);
2253                 return sizeof(dt_obj_version_t);
2254         }
2255
2256         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2257                 return -EACCES;
2258
2259         OSD_EXEC_OP(handle, xattr_set);
2260         return __osd_xattr_set(env, dt, buf, name, fl);
2261 }
2262
2263 /*
2264  * Concurrency: @dt is read locked.
2265  */
2266 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2267                           struct lu_buf *buf, struct lustre_capa *capa)
2268 {
2269         struct osd_object      *obj    = osd_dt_obj(dt);
2270         struct inode           *inode  = obj->oo_inode;
2271         struct osd_thread_info *info   = osd_oti_get(env);
2272         struct dentry          *dentry = &info->oti_obj_dentry;
2273
2274         LASSERT(dt_object_exists(dt));
2275         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2276         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2277
2278         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2279                 return -EACCES;
2280
2281         dentry->d_inode = inode;
2282         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2283 }
2284
2285 static int osd_declare_xattr_del(const struct lu_env *env,
2286                                  struct dt_object *dt, const char *name,
2287                                  struct thandle *handle)
2288 {
2289         struct osd_thandle *oh;
2290
2291         LASSERT(dt_object_exists(dt));
2292         LASSERT(handle != NULL);
2293
2294         oh = container_of0(handle, struct osd_thandle, ot_super);
2295         LASSERT(oh->ot_handle == NULL);
2296
2297         OSD_DECLARE_OP(oh, xattr_set);
2298         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2299
2300         return 0;
2301 }
2302
2303 /*
2304  * Concurrency: @dt is write locked.
2305  */
2306 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2307                          const char *name, struct thandle *handle,
2308                          struct lustre_capa *capa)
2309 {
2310         struct osd_object      *obj    = osd_dt_obj(dt);
2311         struct inode           *inode  = obj->oo_inode;
2312         struct osd_thread_info *info   = osd_oti_get(env);
2313         struct dentry          *dentry = &info->oti_obj_dentry;
2314         int                     rc;
2315
2316         LASSERT(dt_object_exists(dt));
2317         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2318         LASSERT(osd_write_locked(env, obj));
2319         LASSERT(handle != NULL);
2320
2321         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2322                 return -EACCES;
2323
2324         OSD_EXEC_OP(handle, xattr_set);
2325
2326         dentry->d_inode = inode;
2327         rc = inode->i_op->removexattr(dentry, name);
2328         return rc;
2329 }
2330
2331 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2332                                      struct dt_object *dt,
2333                                      struct lustre_capa *old,
2334                                      __u64 opc)
2335 {
2336         struct osd_thread_info *info = osd_oti_get(env);
2337         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2338         struct osd_object *obj = osd_dt_obj(dt);
2339         struct osd_device *dev = osd_obj2dev(obj);
2340         struct lustre_capa_key *key = &info->oti_capa_key;
2341         struct lustre_capa *capa = &info->oti_capa;
2342         struct obd_capa *oc;
2343         struct md_capainfo *ci;
2344         int rc;
2345         ENTRY;
2346
2347         if (!dev->od_fl_capa)
2348                 RETURN(ERR_PTR(-ENOENT));
2349
2350         LASSERT(dt_object_exists(dt));
2351         LINVRNT(osd_invariant(obj));
2352
2353         /* renewal sanity check */
2354         if (old && osd_object_auth(env, dt, old, opc))
2355                 RETURN(ERR_PTR(-EACCES));
2356
2357         ci = md_capainfo(env);
2358         if (unlikely(!ci))
2359                 RETURN(ERR_PTR(-ENOENT));
2360
2361         switch (ci->mc_auth) {
2362         case LC_ID_NONE:
2363                 RETURN(NULL);
2364         case LC_ID_PLAIN:
2365                 capa->lc_uid = obj->oo_inode->i_uid;
2366                 capa->lc_gid = obj->oo_inode->i_gid;
2367                 capa->lc_flags = LC_ID_PLAIN;
2368                 break;
2369         case LC_ID_CONVERT: {
2370                 __u32 d[4], s[4];
2371
2372                 s[0] = obj->oo_inode->i_uid;
2373                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2374                 s[2] = obj->oo_inode->i_gid;
2375                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2376                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2377                 if (unlikely(rc))
2378                         RETURN(ERR_PTR(rc));
2379
2380                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2381                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2382                 capa->lc_flags = LC_ID_CONVERT;
2383                 break;
2384         }
2385         default:
2386                 RETURN(ERR_PTR(-EINVAL));
2387         }
2388
2389         capa->lc_fid = *fid;
2390         capa->lc_opc = opc;
2391         capa->lc_flags |= dev->od_capa_alg << 24;
2392         capa->lc_timeout = dev->od_capa_timeout;
2393         capa->lc_expiry = 0;
2394
2395         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2396         if (oc) {
2397                 LASSERT(!capa_is_expired(oc));
2398                 RETURN(oc);
2399         }
2400
2401         cfs_spin_lock(&capa_lock);
2402         *key = dev->od_capa_keys[1];
2403         cfs_spin_unlock(&capa_lock);
2404
2405         capa->lc_keyid = key->lk_keyid;
2406         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2407
2408         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2409         if (rc) {
2410                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2411                 RETURN(ERR_PTR(rc));
2412         }
2413
2414         oc = capa_add(dev->od_capa_hash, capa);
2415         RETURN(oc);
2416 }
2417
2418 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2419 {
2420         struct osd_object      *obj    = osd_dt_obj(dt);
2421         struct inode           *inode  = obj->oo_inode;
2422         struct osd_thread_info *info   = osd_oti_get(env);
2423         struct dentry          *dentry = &info->oti_obj_dentry;
2424         struct file            *file   = &info->oti_file;
2425         int                     rc;
2426
2427         ENTRY;
2428
2429         dentry->d_inode = inode;
2430         file->f_dentry = dentry;
2431         file->f_mapping = inode->i_mapping;
2432         file->f_op = inode->i_fop;
2433         LOCK_INODE_MUTEX(inode);
2434         rc = file->f_op->fsync(file, dentry, 0);
2435         UNLOCK_INODE_MUTEX(inode);
2436         RETURN(rc);
2437 }
2438
2439 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2440                         void **data)
2441 {
2442         struct osd_object *obj = osd_dt_obj(dt);
2443         ENTRY;
2444
2445         *data = (void *)obj->oo_inode;
2446         RETURN(0);
2447 }
2448
2449 /*
2450  * Index operations.
2451  */
2452
2453 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2454                            const struct dt_index_features *feat)
2455 {
2456         struct iam_descr *descr;
2457
2458         if (osd_object_is_root(o))
2459                 return feat == &dt_directory_features;
2460
2461         LASSERT(o->oo_dir != NULL);
2462
2463         descr = o->oo_dir->od_container.ic_descr;
2464         if (feat == &dt_directory_features) {
2465                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2466                         return 1;
2467                 else
2468                         return 0;
2469         } else {
2470                 return
2471                         feat->dif_keysize_min <= descr->id_key_size &&
2472                         descr->id_key_size <= feat->dif_keysize_max &&
2473                         feat->dif_recsize_min <= descr->id_rec_size &&
2474                         descr->id_rec_size <= feat->dif_recsize_max &&
2475                         !(feat->dif_flags & (DT_IND_VARKEY |
2476                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2477                         ergo(feat->dif_flags & DT_IND_UPDATE,
2478                              1 /* XXX check that object (and file system) is
2479                                 * writable */);
2480         }
2481 }
2482
2483 static int osd_iam_container_init(const struct lu_env *env,
2484                                   struct osd_object *obj,
2485                                   struct osd_directory *dir)
2486 {
2487         struct iam_container *bag = &dir->od_container;
2488         int result;
2489
2490         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2491         if (result != 0)
2492                 return result;
2493
2494         result = iam_container_setup(bag);
2495         if (result != 0)
2496                 goto out;
2497
2498         if (osd_obj2dev(obj)->od_iop_mode) {
2499                 u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag);
2500
2501                 bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode,
2502                                                 ptr, 0, &result);
2503         }
2504
2505  out:
2506         if (result == 0)
2507                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2508         else
2509                 iam_container_fini(bag);
2510
2511         return result;
2512 }
2513
2514
2515 /*
2516  * Concurrency: no external locking is necessary.
2517  */
2518 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2519                          const struct dt_index_features *feat)
2520 {
2521         int result;
2522         int ea_dir = 0;
2523         struct osd_object *obj = osd_dt_obj(dt);
2524         struct osd_device *osd = osd_obj2dev(obj);
2525
2526         LINVRNT(osd_invariant(obj));
2527         LASSERT(dt_object_exists(dt));
2528
2529         if (osd_object_is_root(obj)) {
2530                 dt->do_index_ops = &osd_index_ea_ops;
2531                 result = 0;
2532         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2533                 dt->do_index_ops = &osd_index_ea_ops;
2534                 if (S_ISDIR(obj->oo_inode->i_mode))
2535                         result = 0;
2536                 else
2537                         result = -ENOTDIR;
2538                 ea_dir = 1;
2539         } else if (!osd_has_index(obj)) {
2540                 struct osd_directory *dir;
2541
2542                 OBD_ALLOC_PTR(dir);
2543                 if (dir != NULL) {
2544
2545                         cfs_spin_lock(&obj->oo_guard);
2546                         if (obj->oo_dir == NULL)
2547                                 obj->oo_dir = dir;
2548                         else
2549                                 /*
2550                                  * Concurrent thread allocated container data.
2551                                  */
2552                                 OBD_FREE_PTR(dir);
2553                         cfs_spin_unlock(&obj->oo_guard);
2554                         /*
2555                          * Now, that we have container data, serialize its
2556                          * initialization.
2557                          */
2558                         cfs_down_write(&obj->oo_ext_idx_sem);
2559                         /*
2560                          * recheck under lock.
2561                          */
2562                         if (!osd_has_index(obj))
2563                                 result = osd_iam_container_init(env, obj, dir);
2564                         else
2565                                 result = 0;
2566                         cfs_up_write(&obj->oo_ext_idx_sem);
2567                 } else {
2568                         result = -ENOMEM;
2569                 }
2570         } else {
2571                 result = 0;
2572         }
2573
2574         if (result == 0 && ea_dir == 0) {
2575                 if (!osd_iam_index_probe(env, obj, feat))
2576                         result = -ENOTDIR;
2577         }
2578         LINVRNT(osd_invariant(obj));
2579
2580         return result;
2581 }
2582
2583 static const struct dt_object_operations osd_obj_ops = {
2584         .do_read_lock         = osd_object_read_lock,
2585         .do_write_lock        = osd_object_write_lock,
2586         .do_read_unlock       = osd_object_read_unlock,
2587         .do_write_unlock      = osd_object_write_unlock,
2588         .do_write_locked      = osd_object_write_locked,
2589         .do_attr_get          = osd_attr_get,
2590         .do_declare_attr_set  = osd_declare_attr_set,
2591         .do_attr_set          = osd_attr_set,
2592         .do_ah_init           = osd_ah_init,
2593         .do_declare_create    = osd_declare_object_create,
2594         .do_create            = osd_object_create,
2595         .do_declare_destroy   = osd_declare_object_destroy,
2596         .do_destroy           = osd_object_destroy,
2597         .do_index_try         = osd_index_try,
2598         .do_declare_ref_add   = osd_declare_object_ref_add,
2599         .do_ref_add           = osd_object_ref_add,
2600         .do_declare_ref_del   = osd_declare_object_ref_del,
2601         .do_ref_del           = osd_object_ref_del,
2602         .do_xattr_get         = osd_xattr_get,
2603         .do_declare_xattr_set = osd_declare_xattr_set,
2604         .do_xattr_set         = osd_xattr_set,
2605         .do_declare_xattr_del = osd_declare_xattr_del,
2606         .do_xattr_del         = osd_xattr_del,
2607         .do_xattr_list        = osd_xattr_list,
2608         .do_capa_get          = osd_capa_get,
2609         .do_object_sync       = osd_object_sync,
2610         .do_data_get          = osd_data_get,
2611 };
2612
2613 /**
2614  * dt_object_operations for interoperability mode
2615  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2616  */
2617 static const struct dt_object_operations osd_obj_ea_ops = {
2618         .do_read_lock         = osd_object_read_lock,
2619         .do_write_lock        = osd_object_write_lock,
2620         .do_read_unlock       = osd_object_read_unlock,
2621         .do_write_unlock      = osd_object_write_unlock,
2622         .do_write_locked      = osd_object_write_locked,
2623         .do_attr_get          = osd_attr_get,
2624         .do_declare_attr_set  = osd_declare_attr_set,
2625         .do_attr_set          = osd_attr_set,
2626         .do_ah_init           = osd_ah_init,
2627         .do_declare_create    = osd_declare_object_create,
2628         .do_create            = osd_object_ea_create,
2629         .do_declare_destroy   = osd_declare_object_destroy,
2630         .do_destroy           = osd_object_destroy,
2631         .do_index_try         = osd_index_try,
2632         .do_declare_ref_add   = osd_declare_object_ref_add,
2633         .do_ref_add           = osd_object_ref_add,
2634         .do_declare_ref_del   = osd_declare_object_ref_del,
2635         .do_ref_del           = osd_object_ref_del,
2636         .do_xattr_get         = osd_xattr_get,
2637         .do_declare_xattr_set = osd_declare_xattr_set,
2638         .do_xattr_set         = osd_xattr_set,
2639         .do_declare_xattr_del = osd_declare_xattr_del,
2640         .do_xattr_del         = osd_xattr_del,
2641         .do_xattr_list        = osd_xattr_list,
2642         .do_capa_get          = osd_capa_get,
2643         .do_object_sync       = osd_object_sync,
2644         .do_data_get          = osd_data_get,
2645 };
2646
2647 static int osd_index_declare_iam_delete(const struct lu_env *env,
2648                                         struct dt_object *dt,
2649                                         const struct dt_key *key,
2650                                         struct thandle *handle)
2651 {
2652         struct osd_thandle    *oh;
2653
2654         oh = container_of0(handle, struct osd_thandle, ot_super);
2655         LASSERT(oh->ot_handle == NULL);
2656
2657         OSD_DECLARE_OP(oh, delete);
2658         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2659
2660         return 0;
2661 }
2662
2663 /**
2664  *      delete a (key, value) pair from index \a dt specified by \a key
2665  *
2666  *      \param  dt      osd index object
2667  *      \param  key     key for index
2668  *      \param  rec     record reference
2669  *      \param  handle  transaction handler
2670  *
2671  *      \retval  0  success
2672  *      \retval -ve   failure
2673  */
2674
2675 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2676                                 const struct dt_key *key,
2677                                 struct thandle *handle,
2678                                 struct lustre_capa *capa)
2679 {
2680         struct osd_object     *obj = osd_dt_obj(dt);
2681         struct osd_thandle    *oh;
2682         struct iam_path_descr *ipd;
2683         struct iam_container  *bag = &obj->oo_dir->od_container;
2684         int                    rc;
2685
2686         ENTRY;
2687
2688         LINVRNT(osd_invariant(obj));
2689         LASSERT(dt_object_exists(dt));
2690         LASSERT(bag->ic_object == obj->oo_inode);
2691         LASSERT(handle != NULL);
2692
2693         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2694                 RETURN(-EACCES);
2695
2696         OSD_EXEC_OP(handle, delete);
2697
2698         ipd = osd_idx_ipd_get(env, bag);
2699         if (unlikely(ipd == NULL))
2700                 RETURN(-ENOMEM);
2701
2702         oh = container_of0(handle, struct osd_thandle, ot_super);
2703         LASSERT(oh->ot_handle != NULL);
2704         LASSERT(oh->ot_handle->h_transaction != NULL);
2705
2706         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2707         osd_ipd_put(env, bag, ipd);
2708         LINVRNT(osd_invariant(obj));
2709         RETURN(rc);
2710 }
2711
2712 static int osd_index_declare_ea_delete(const struct lu_env *env,
2713                                        struct dt_object *dt,
2714                                        const struct dt_key *key,
2715                                        struct thandle *handle)
2716 {
2717         struct osd_thandle *oh;
2718
2719         LASSERT(dt_object_exists(dt));
2720         LASSERT(handle != NULL);
2721
2722         oh = container_of0(handle, struct osd_thandle, ot_super);
2723         LASSERT(oh->ot_handle == NULL);
2724
2725         OSD_DECLARE_OP(oh, delete);
2726         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2727
2728         LASSERT(osd_dt_obj(dt)->oo_inode);
2729         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
2730                         osd_dt_obj(dt)->oo_inode);
2731         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
2732                         osd_dt_obj(dt)->oo_inode);
2733
2734         return 0;
2735 }
2736
2737 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2738                                           struct dt_rec *fid)
2739 {
2740         struct osd_fid_pack *rec;
2741         int                  rc = -ENODATA;
2742
2743         if (de->file_type & LDISKFS_DIRENT_LUFID) {
2744                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2745                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2746         }
2747         RETURN(rc);
2748 }
2749
2750 /**
2751  * Index delete function for interoperability mode (b11826).
2752  * It will remove the directory entry added by osd_index_ea_insert().
2753  * This entry is needed to maintain name->fid mapping.
2754  *
2755  * \param key,  key i.e. file entry to be deleted
2756  *
2757  * \retval   0, on success
2758  * \retval -ve, on error
2759  */
2760 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2761                                const struct dt_key *key,
2762                                struct thandle *handle,
2763                                struct lustre_capa *capa)
2764 {
2765         struct osd_object          *obj    = osd_dt_obj(dt);
2766         struct inode               *dir    = obj->oo_inode;
2767         struct dentry              *dentry;
2768         struct osd_thandle         *oh;
2769         struct ldiskfs_dir_entry_2 *de;
2770         struct buffer_head         *bh;
2771         struct htree_lock          *hlock = NULL;
2772         int                         rc;
2773
2774         ENTRY;
2775
2776         LINVRNT(osd_invariant(obj));
2777         LASSERT(dt_object_exists(dt));
2778         LASSERT(handle != NULL);
2779
2780         OSD_EXEC_OP(handle, delete);
2781
2782         oh = container_of(handle, struct osd_thandle, ot_super);
2783         LASSERT(oh->ot_handle != NULL);
2784         LASSERT(oh->ot_handle->h_transaction != NULL);
2785
2786         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2787                 RETURN(-EACCES);
2788
2789         dentry = osd_child_dentry_get(env, obj,
2790                                       (char *)key, strlen((char *)key));
2791
2792         if (obj->oo_hl_head != NULL) {
2793                 hlock = osd_oti_get(env)->oti_hlock;
2794                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
2795                                    dir, LDISKFS_HLOCK_DEL);
2796         } else {
2797                 cfs_down_write(&obj->oo_ext_idx_sem);
2798         }
2799
2800         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
2801         if (bh) {
2802                 rc = ldiskfs_delete_entry(oh->ot_handle,
2803                                           dir, de, bh);
2804                 brelse(bh);
2805         } else {
2806                 rc = -ENOENT;
2807         }
2808         if (hlock != NULL)
2809                 ldiskfs_htree_unlock(hlock);
2810         else
2811                 cfs_up_write(&obj->oo_ext_idx_sem);
2812
2813         LASSERT(osd_invariant(obj));
2814         RETURN(rc);
2815 }
2816
2817 /**
2818  *      Lookup index for \a key and copy record to \a rec.
2819  *
2820  *      \param  dt      osd index object
2821  *      \param  key     key for index
2822  *      \param  rec     record reference
2823  *
2824  *      \retval  +ve  success : exact mach
2825  *      \retval  0    return record with key not greater than \a key
2826  *      \retval -ve   failure
2827  */
2828 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
2829                                 struct dt_rec *rec, const struct dt_key *key,
2830                                 struct lustre_capa *capa)
2831 {
2832         struct osd_object      *obj = osd_dt_obj(dt);
2833         struct iam_path_descr  *ipd;
2834         struct iam_container   *bag = &obj->oo_dir->od_container;
2835         struct osd_thread_info *oti = osd_oti_get(env);
2836         struct iam_iterator    *it = &oti->oti_idx_it;
2837         struct iam_rec         *iam_rec;
2838         int                     rc;
2839
2840         ENTRY;
2841
2842         LASSERT(osd_invariant(obj));
2843         LASSERT(dt_object_exists(dt));
2844         LASSERT(bag->ic_object == obj->oo_inode);
2845
2846         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2847                 RETURN(-EACCES);
2848
2849         ipd = osd_idx_ipd_get(env, bag);
2850         if (IS_ERR(ipd))
2851                 RETURN(-ENOMEM);
2852
2853         /* got ipd now we can start iterator. */
2854         iam_it_init(it, bag, 0, ipd);
2855
2856         rc = iam_it_get(it, (struct iam_key *)key);
2857         if (rc >= 0) {
2858                 if (S_ISDIR(obj->oo_inode->i_mode))
2859                         iam_rec = (struct iam_rec *)oti->oti_ldp;
2860                 else
2861                         iam_rec = (struct iam_rec *) rec;
2862
2863                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
2864                 if (S_ISDIR(obj->oo_inode->i_mode))
2865                         osd_fid_unpack((struct lu_fid *) rec,
2866                                        (struct osd_fid_pack *)iam_rec);
2867         }
2868         iam_it_put(it);
2869         iam_it_fini(it);
2870         osd_ipd_put(env, bag, ipd);
2871
2872         LINVRNT(osd_invariant(obj));
2873
2874         RETURN(rc);
2875 }
2876
2877 static int osd_index_declare_iam_insert(const struct lu_env *env,
2878                                         struct dt_object *dt,
2879                                         const struct dt_rec *rec,
2880                                         const struct dt_key *key,
2881                                         struct thandle *handle)
2882 {
2883         struct osd_thandle *oh;
2884
2885         LASSERT(dt_object_exists(dt));
2886         LASSERT(handle != NULL);
2887
2888         oh = container_of0(handle, struct osd_thandle, ot_super);
2889         LASSERT(oh->ot_handle == NULL);
2890
2891         OSD_DECLARE_OP(oh, insert);
2892         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
2893
2894         return 0;
2895 }
2896
2897 /**
2898  *      Inserts (key, value) pair in \a dt index object.
2899  *
2900  *      \param  dt      osd index object
2901  *      \param  key     key for index
2902  *      \param  rec     record reference
2903  *      \param  th      transaction handler
2904  *
2905  *      \retval  0  success
2906  *      \retval -ve failure
2907  */
2908 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
2909                                 const struct dt_rec *rec,
2910                                 const struct dt_key *key, struct thandle *th,
2911                                 struct lustre_capa *capa, int ignore_quota)
2912 {
2913         struct osd_object     *obj = osd_dt_obj(dt);
2914         struct iam_path_descr *ipd;
2915         struct osd_thandle    *oh;
2916         struct iam_container  *bag = &obj->oo_dir->od_container;
2917 #ifdef HAVE_QUOTA_SUPPORT
2918         cfs_cap_t              save = cfs_curproc_cap_pack();
2919 #endif
2920         struct osd_thread_info *oti = osd_oti_get(env);
2921         struct iam_rec         *iam_rec = (struct iam_rec *)oti->oti_ldp;
2922         int                     rc;
2923
2924         ENTRY;
2925
2926         LINVRNT(osd_invariant(obj));
2927         LASSERT(dt_object_exists(dt));
2928         LASSERT(bag->ic_object == obj->oo_inode);
2929         LASSERT(th != NULL);
2930
2931         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2932                 RETURN(-EACCES);
2933
2934         OSD_EXEC_OP(th, insert);
2935
2936         ipd = osd_idx_ipd_get(env, bag);
2937         if (unlikely(ipd == NULL))
2938                 RETURN(-ENOMEM);
2939
2940         oh = container_of0(th, struct osd_thandle, ot_super);
2941         LASSERT(oh->ot_handle != NULL);
2942         LASSERT(oh->ot_handle->h_transaction != NULL);
2943 #ifdef HAVE_QUOTA_SUPPORT
2944         if (ignore_quota)
2945                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
2946         else
2947                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
2948 #endif
2949         if (S_ISDIR(obj->oo_inode->i_mode))
2950                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
2951         else
2952                 iam_rec = (struct iam_rec *) rec;
2953         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
2954                         iam_rec, ipd);
2955 #ifdef HAVE_QUOTA_SUPPORT
2956         cfs_curproc_cap_unpack(save);
2957 #endif
2958         osd_ipd_put(env, bag, ipd);
2959         LINVRNT(osd_invariant(obj));
2960         RETURN(rc);
2961 }
2962
2963 /**
2964  * Calls ldiskfs_add_entry() to add directory entry
2965  * into the directory. This is required for
2966  * interoperability mode (b11826)
2967  *
2968  * \retval   0, on success
2969  * \retval -ve, on error
2970  */
2971 static int __osd_ea_add_rec(struct osd_thread_info *info,
2972                             struct osd_object *pobj, struct inode  *cinode,
2973                             const char *name, const struct dt_rec *fid,
2974                             struct htree_lock *hlock, struct thandle *th)
2975 {
2976         struct ldiskfs_dentry_param *ldp;
2977         struct dentry               *child;
2978         struct osd_thandle          *oth;
2979         int                          rc;
2980
2981         oth = container_of(th, struct osd_thandle, ot_super);
2982         LASSERT(oth->ot_handle != NULL);
2983         LASSERT(oth->ot_handle->h_transaction != NULL);
2984
2985         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
2986
2987         /* XXX: remove fid_is_igif() check here.
2988          * IGIF check is just to handle insertion of .. when it is 'ROOT',
2989          * it is IGIF now but needs FID in dir entry as well for readdir
2990          * to work.
2991          * LU-838 should fix that and remove fid_is_igif() check */
2992         if (fid_is_igif((struct lu_fid *)fid) ||
2993             fid_is_norm((struct lu_fid *)fid)) {
2994                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2995                 osd_get_ldiskfs_dirent_param(ldp, fid);
2996                 child->d_fsdata = (void *)ldp;
2997         } else {
2998                 child->d_fsdata = NULL;
2999         }
3000         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3001
3002         RETURN(rc);
3003 }
3004
3005 /**
3006  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3007  * into the directory.Also sets flags into osd object to
3008  * indicate dot and dotdot are created. This is required for
3009  * interoperability mode (b11826)
3010  *
3011  * \param dir   directory for dot and dotdot fixup.
3012  * \param obj   child object for linking
3013  *
3014  * \retval   0, on success
3015  * \retval -ve, on error
3016  */
3017 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3018                               struct osd_object *dir,
3019                               struct inode  *parent_dir, const char *name,
3020                               const struct dt_rec *dot_fid,
3021                               const struct dt_rec *dot_dot_fid,
3022                               struct thandle *th)
3023 {
3024         struct inode                *inode = dir->oo_inode;
3025         struct ldiskfs_dentry_param *dot_ldp;
3026         struct ldiskfs_dentry_param *dot_dot_ldp;
3027         struct osd_thandle          *oth;
3028         int result = 0;
3029
3030         oth = container_of(th, struct osd_thandle, ot_super);
3031         LASSERT(oth->ot_handle->h_transaction != NULL);
3032         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3033
3034         if (strcmp(name, dot) == 0) {
3035                 if (dir->oo_compat_dot_created) {
3036                         result = -EEXIST;
3037                 } else {
3038                         LASSERT(inode == parent_dir);
3039                         dir->oo_compat_dot_created = 1;
3040                         result = 0;
3041                 }
3042         } else if(strcmp(name, dotdot) == 0) {
3043                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3044                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3045
3046                 if (!dir->oo_compat_dot_created)
3047                         return -EINVAL;
3048                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3049                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3050                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3051                 } else {
3052                         dot_ldp = NULL;
3053                         dot_dot_ldp = NULL;
3054                 }
3055                 /* in case of rename, dotdot is already created */
3056                 if (dir->oo_compat_dotdot_created) {
3057                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3058                                                 dot_dot_fid, NULL, th);
3059                 }
3060
3061                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3062                                                 inode, dot_ldp, dot_dot_ldp);
3063                 if (result == 0)
3064                        dir->oo_compat_dotdot_created = 1;
3065         }
3066
3067         return result;
3068 }
3069
3070
3071 /**
3072  * It will call the appropriate osd_add* function and return the
3073  * value, return by respective functions.
3074  */
3075 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3076                           struct inode *cinode, const char *name,
3077                           const struct dt_rec *fid, struct thandle *th)
3078 {
3079         struct osd_thread_info *info   = osd_oti_get(env);
3080         struct htree_lock      *hlock;
3081         int                     rc;
3082
3083         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3084
3085         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3086                                                    name[2] =='\0'))) {
3087                 if (hlock != NULL) {
3088                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3089                                            pobj->oo_inode, 0);
3090                 } else {
3091                         cfs_down_write(&pobj->oo_ext_idx_sem);
3092                 }
3093                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3094                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3095                                         fid, th);
3096         } else {
3097                 if (hlock != NULL) {
3098                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3099                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3100                 } else {
3101                         cfs_down_write(&pobj->oo_ext_idx_sem);
3102                 }
3103
3104                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3105                                       hlock, th);
3106         }
3107         if (hlock != NULL)
3108                 ldiskfs_htree_unlock(hlock);
3109         else
3110                 cfs_up_write(&pobj->oo_ext_idx_sem);
3111
3112         return rc;
3113 }
3114
3115 /**
3116  * Calls ->lookup() to find dentry. From dentry get inode and
3117  * read inode's ea to get fid. This is required for  interoperability
3118  * mode (b11826)
3119  *
3120  * \retval   0, on success
3121  * \retval -ve, on error
3122  */
3123 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3124                              struct dt_rec *rec, const struct dt_key *key)
3125 {
3126         struct inode               *dir    = obj->oo_inode;
3127         struct dentry              *dentry;
3128         struct ldiskfs_dir_entry_2 *de;
3129         struct buffer_head         *bh;
3130         struct lu_fid              *fid = (struct lu_fid *) rec;
3131         struct htree_lock          *hlock = NULL;
3132         int                         ino;
3133         int                         rc;
3134
3135         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3136
3137         dentry = osd_child_dentry_get(env, obj,
3138                                       (char *)key, strlen((char *)key));
3139
3140         if (obj->oo_hl_head != NULL) {
3141                 hlock = osd_oti_get(env)->oti_hlock;
3142                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3143                                    dir, LDISKFS_HLOCK_LOOKUP);
3144         } else {
3145                 cfs_down_read(&obj->oo_ext_idx_sem);
3146         }
3147
3148         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3149         if (bh) {
3150                 ino = le32_to_cpu(de->inode);
3151                 rc = osd_get_fid_from_dentry(de, rec);
3152
3153                 /* done with de, release bh */
3154                 brelse(bh);
3155                 if (rc != 0)
3156                         rc = osd_ea_fid_get(env, obj, ino, fid);
3157         } else {
3158                 rc = -ENOENT;
3159         }
3160
3161         if (hlock != NULL)
3162                 ldiskfs_htree_unlock(hlock);
3163         else
3164                 cfs_up_read(&obj->oo_ext_idx_sem);
3165         RETURN (rc);
3166 }
3167
3168 /**
3169  * Find the osd object for given fid.
3170  *
3171  * \param fid need to find the osd object having this fid
3172  *
3173  * \retval osd_object on success
3174  * \retval        -ve on error
3175  */
3176 struct osd_object *osd_object_find(const struct lu_env *env,
3177                                    struct dt_object *dt,
3178                                    const struct lu_fid *fid)
3179 {
3180         struct lu_device  *ludev = dt->do_lu.lo_dev;
3181         struct osd_object *child = NULL;
3182         struct lu_object  *luch;
3183         struct lu_object  *lo;
3184
3185         luch = lu_object_find(env, ludev, fid, NULL);
3186         if (!IS_ERR(luch)) {
3187                 if (lu_object_exists(luch)) {
3188                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3189                         if (lo != NULL)
3190                                 child = osd_obj(lo);
3191                         else
3192                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3193                                                 "lu_object can't be located"
3194                                                 DFID"\n", PFID(fid));
3195
3196                         if (child == NULL) {
3197                                 lu_object_put(env, luch);
3198                                 CERROR("Unable to get osd_object\n");
3199                                 child = ERR_PTR(-ENOENT);
3200                         }
3201                 } else {
3202                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3203                                         "lu_object does not exists "DFID"\n",
3204                                         PFID(fid));
3205                         lu_object_put(env, luch);
3206                         child = ERR_PTR(-ENOENT);
3207                 }
3208         } else
3209                 child = (void *)luch;
3210
3211         return child;
3212 }
3213
3214 /**
3215  * Put the osd object once done with it.
3216  *
3217  * \param obj osd object that needs to be put
3218  */
3219 static inline void osd_object_put(const struct lu_env *env,
3220                                   struct osd_object *obj)
3221 {
3222         lu_object_put(env, &obj->oo_dt.do_lu);
3223 }
3224
3225 static int osd_index_declare_ea_insert(const struct lu_env *env,
3226                                        struct dt_object *dt,
3227                                        const struct dt_rec *rec,
3228                                        const struct dt_key *key,
3229                                        struct thandle *handle)
3230 {
3231         struct osd_thandle *oh;
3232
3233         LASSERT(dt_object_exists(dt));
3234         LASSERT(handle != NULL);
3235
3236         oh = container_of0(handle, struct osd_thandle, ot_super);
3237         LASSERT(oh->ot_handle == NULL);
3238
3239         OSD_DECLARE_OP(oh, insert);
3240         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3241
3242         LASSERT(osd_dt_obj(dt)->oo_inode);
3243         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3244                         osd_dt_obj(dt)->oo_inode);
3245         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3246                         osd_dt_obj(dt)->oo_inode);
3247
3248         return 0;
3249 }
3250
3251 /**
3252  * Index add function for interoperability mode (b11826).
3253  * It will add the directory entry.This entry is needed to
3254  * maintain name->fid mapping.
3255  *
3256  * \param key it is key i.e. file entry to be inserted
3257  * \param rec it is value of given key i.e. fid
3258  *
3259  * \retval   0, on success
3260  * \retval -ve, on error
3261  */
3262 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3263                                const struct dt_rec *rec,
3264                                const struct dt_key *key, struct thandle *th,
3265                                struct lustre_capa *capa, int ignore_quota)
3266 {
3267         struct osd_object *obj   = osd_dt_obj(dt);
3268         struct lu_fid     *fid   = (struct lu_fid *) rec;
3269         const char        *name  = (const char *)key;
3270         struct osd_object *child;
3271 #ifdef HAVE_QUOTA_SUPPORT
3272         cfs_cap_t          save  = cfs_curproc_cap_pack();
3273 #endif
3274         int                rc;
3275
3276         ENTRY;
3277
3278         LASSERT(osd_invariant(obj));
3279         LASSERT(dt_object_exists(dt));
3280         LASSERT(th != NULL);
3281
3282         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3283                 RETURN(-EACCES);
3284
3285         child = osd_object_find(env, dt, fid);
3286         if (!IS_ERR(child)) {
3287 #ifdef HAVE_QUOTA_SUPPORT
3288                 if (ignore_quota)
3289                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3290                 else
3291                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3292 #endif
3293                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3294 #ifdef HAVE_QUOTA_SUPPORT
3295                 cfs_curproc_cap_unpack(save);
3296 #endif
3297                 osd_object_put(env, child);
3298         } else {
3299                 rc = PTR_ERR(child);
3300         }
3301
3302         LASSERT(osd_invariant(obj));
3303         RETURN(rc);
3304 }
3305
3306 /**
3307  *  Initialize osd Iterator for given osd index object.
3308  *
3309  *  \param  dt      osd index object
3310  */
3311
3312 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3313                                      struct dt_object *dt,
3314                                      __u32 unused,
3315                                      struct lustre_capa *capa)
3316 {
3317         struct osd_it_iam      *it;
3318         struct osd_thread_info *oti = osd_oti_get(env);
3319         struct osd_object      *obj = osd_dt_obj(dt);
3320         struct lu_object       *lo  = &dt->do_lu;
3321         struct iam_path_descr  *ipd;
3322         struct iam_container   *bag = &obj->oo_dir->od_container;
3323
3324         LASSERT(lu_object_exists(lo));
3325
3326         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3327                 return ERR_PTR(-EACCES);
3328
3329         it = &oti->oti_it;
3330         ipd = osd_it_ipd_get(env, bag);
3331         if (likely(ipd != NULL)) {
3332                 it->oi_obj = obj;
3333                 it->oi_ipd = ipd;
3334                 lu_object_get(lo);
3335                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3336                 return (struct dt_it *)it;
3337         }
3338         return ERR_PTR(-ENOMEM);
3339 }
3340
3341 /**
3342  * free given Iterator.
3343  */
3344
3345 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3346 {
3347         struct osd_it_iam *it = (struct osd_it_iam *)di;
3348         struct osd_object *obj = it->oi_obj;
3349
3350         iam_it_fini(&it->oi_it);
3351         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3352         lu_object_put(env, &obj->oo_dt.do_lu);
3353 }
3354
3355 /**
3356  *  Move Iterator to record specified by \a key
3357  *
3358  *  \param  di      osd iterator
3359  *  \param  key     key for index
3360  *
3361  *  \retval +ve  di points to record with least key not larger than key
3362  *  \retval  0   di points to exact matched key
3363  *  \retval -ve  failure
3364  */
3365
3366 static int osd_it_iam_get(const struct lu_env *env,
3367                           struct dt_it *di, const struct dt_key *key)
3368 {
3369         struct osd_it_iam *it = (struct osd_it_iam *)di;
3370
3371         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3372 }
3373
3374 /**
3375  *  Release Iterator
3376  *
3377  *  \param  di      osd iterator
3378  */
3379
3380 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3381 {
3382         struct osd_it_iam *it = (struct osd_it_iam *)di;
3383
3384         iam_it_put(&it->oi_it);
3385 }
3386
3387 /**
3388  *  Move iterator by one record
3389  *
3390  *  \param  di      osd iterator
3391  *
3392  *  \retval +1   end of container reached
3393  *  \retval  0   success
3394  *  \retval -ve  failure
3395  */
3396
3397 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3398 {
3399         struct osd_it_iam *it = (struct osd_it_iam *)di;
3400
3401         return iam_it_next(&it->oi_it);
3402 }
3403
3404 /**
3405  * Return pointer to the key under iterator.
3406  */
3407
3408 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3409                                  const struct dt_it *di)
3410 {
3411         struct osd_it_iam *it = (struct osd_it_iam *)di;
3412
3413         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3414 }
3415
3416 /**
3417  * Return size of key under iterator (in bytes)
3418  */
3419
3420 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3421 {
3422         struct osd_it_iam *it = (struct osd_it_iam *)di;
3423
3424         return iam_it_key_size(&it->oi_it);
3425 }
3426
3427 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3428                                        int len, __u16 type)
3429 {
3430         struct luda_type *lt;
3431         const unsigned    align = sizeof(struct luda_type) - 1;
3432
3433         /* check if file type is required */
3434         if (attr & LUDA_TYPE) {
3435                         len = (len + align) & ~align;
3436
3437                         lt = (void *) ent->lde_name + len;
3438                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3439                         ent->lde_attrs |= LUDA_TYPE;
3440         }
3441
3442         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3443 }
3444
3445 /**
3446  * build lu direct from backend fs dirent.
3447  */
3448
3449 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3450                                       struct lu_fid *fid, __u64 offset,
3451                                       char *name, __u16 namelen,
3452                                       __u16 type, __u32 attr)
3453 {
3454         fid_cpu_to_le(&ent->lde_fid, fid);
3455         ent->lde_attrs = LUDA_FID;
3456
3457         ent->lde_hash = cpu_to_le64(offset);
3458         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3459
3460         strncpy(ent->lde_name, name, namelen);
3461         ent->lde_namelen = cpu_to_le16(namelen);
3462
3463         /* append lustre attributes */
3464         osd_it_append_attrs(ent, attr, namelen, type);
3465 }
3466
3467 /**
3468  * Return pointer to the record under iterator.
3469  */
3470 static int osd_it_iam_rec(const struct lu_env *env,
3471                           const struct dt_it *di,
3472                           struct dt_rec *dtrec, __u32 attr)
3473 {
3474         struct osd_it_iam *it        = (struct osd_it_iam *)di;
3475         struct osd_thread_info *info = osd_oti_get(env);
3476         struct lu_fid     *fid       = &info->oti_fid;
3477         const struct osd_fid_pack *rec;
3478         struct lu_dirent *lde = (struct lu_dirent *)dtrec;
3479         char *name;
3480         int namelen;
3481         __u64 hash;
3482         int rc;
3483
3484         name = (char *)iam_it_key_get(&it->oi_it);
3485         if (IS_ERR(name))
3486                 RETURN(PTR_ERR(name));
3487
3488         namelen = iam_it_key_size(&it->oi_it);
3489
3490         rec = (const struct osd_fid_pack *) iam_it_rec_get(&it->oi_it);
3491         if (IS_ERR(rec))
3492                 RETURN(PTR_ERR(rec));
3493
3494         rc = osd_fid_unpack(fid, rec);
3495         if (rc)
3496                 RETURN(rc);
3497
3498         hash = iam_it_store(&it->oi_it);
3499
3500         /* IAM does not store object type in IAM index (dir) */
3501         osd_it_pack_dirent(lde, fid, hash, name, namelen,
3502                            0, LUDA_FID);
3503
3504         return 0;
3505 }
3506
3507 /**
3508  * Returns cookie for current Iterator position.
3509  */
3510 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3511 {
3512         struct osd_it_iam *it = (struct osd_it_iam *)di;
3513
3514         return iam_it_store(&it->oi_it);
3515 }
3516
3517 /**
3518  * Restore iterator from cookie.
3519  *
3520  * \param  di      osd iterator
3521  * \param  hash    Iterator location cookie
3522  *
3523  * \retval +ve  di points to record with least key not larger than key.
3524  * \retval  0   di points to exact matched key
3525  * \retval -ve  failure
3526  */
3527
3528 static int osd_it_iam_load(const struct lu_env *env,
3529                            const struct dt_it *di, __u64 hash)
3530 {
3531         struct osd_it_iam *it = (struct osd_it_iam *)di;
3532
3533         return iam_it_load(&it->oi_it, hash);
3534 }
3535
3536 static const struct dt_index_operations osd_index_iam_ops = {
3537         .dio_lookup         = osd_index_iam_lookup,
3538         .dio_declare_insert = osd_index_declare_iam_insert,
3539         .dio_insert         = osd_index_iam_insert,
3540         .dio_declare_delete = osd_index_declare_iam_delete,
3541         .dio_delete         = osd_index_iam_delete,
3542         .dio_it     = {
3543                 .init     = osd_it_iam_init,
3544                 .fini     = osd_it_iam_fini,
3545                 .get      = osd_it_iam_get,
3546                 .put      = osd_it_iam_put,
3547                 .next     = osd_it_iam_next,
3548                 .key      = osd_it_iam_key,
3549                 .key_size = osd_it_iam_key_size,
3550                 .rec      = osd_it_iam_rec,
3551                 .store    = osd_it_iam_store,
3552                 .load     = osd_it_iam_load
3553         }
3554 };
3555
3556 /**
3557  * Creates or initializes iterator context.
3558  *
3559  * \retval struct osd_it_ea, iterator structure on success
3560  *
3561  */
3562 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3563                                     struct dt_object *dt,
3564                                     __u32 attr,
3565                                     struct lustre_capa *capa)
3566 {
3567         struct osd_object       *obj  = osd_dt_obj(dt);
3568         struct osd_thread_info  *info = osd_oti_get(env);
3569         struct osd_it_ea        *it   = &info->oti_it_ea;
3570         struct lu_object        *lo   = &dt->do_lu;
3571         struct dentry           *obj_dentry = &info->oti_it_dentry;
3572         ENTRY;
3573         LASSERT(lu_object_exists(lo));
3574
3575         obj_dentry->d_inode = obj->oo_inode;
3576         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3577         obj_dentry->d_name.hash = 0;
3578
3579         it->oie_rd_dirent       = 0;
3580         it->oie_it_dirent       = 0;
3581         it->oie_dirent          = NULL;
3582         it->oie_buf             = info->oti_it_ea_buf;
3583         it->oie_obj             = obj;
3584         it->oie_file.f_pos      = 0;
3585         it->oie_file.f_dentry   = obj_dentry;
3586         if (attr & LUDA_64BITHASH)
3587                 it->oie_file.f_flags = O_64BITHASH;
3588         else
3589                 it->oie_file.f_flags = O_32BITHASH;
3590         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
3591         it->oie_file.f_op         = obj->oo_inode->i_fop;
3592         it->oie_file.private_data = NULL;
3593         lu_object_get(lo);
3594         RETURN((struct dt_it *) it);
3595 }
3596
3597 /**
3598  * Destroy or finishes iterator context.
3599  *
3600  * \param di iterator structure to be destroyed
3601  */
3602 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3603 {
3604         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3605         struct osd_object    *obj  = it->oie_obj;
3606         struct inode       *inode  = obj->oo_inode;
3607
3608         ENTRY;
3609         it->oie_file.f_op->release(inode, &it->oie_file);
3610         lu_object_put(env, &obj->oo_dt.do_lu);
3611         EXIT;
3612 }
3613
3614 /**
3615  * It position the iterator at given key, so that next lookup continues from
3616  * that key Or it is similar to dio_it->load() but based on a key,
3617  * rather than file position.
3618  *
3619  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3620  * to the beginning.
3621  *
3622  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3623  */
3624 static int osd_it_ea_get(const struct lu_env *env,
3625                          struct dt_it *di, const struct dt_key *key)
3626 {
3627         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3628
3629         ENTRY;
3630         LASSERT(((const char *)key)[0] == '\0');
3631         it->oie_file.f_pos      = 0;
3632         it->oie_rd_dirent       = 0;
3633         it->oie_it_dirent       = 0;
3634         it->oie_dirent          = NULL;
3635
3636         RETURN(+1);
3637 }
3638
3639 /**
3640  * Does nothing
3641  */
3642 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3643 {
3644 }
3645
3646 /**
3647  * It is called internally by ->readdir(). It fills the
3648  * iterator's in-memory data structure with required
3649  * information i.e. name, namelen, rec_size etc.
3650  *
3651  * \param buf in which information to be filled in.
3652  * \param name name of the file in given dir
3653  *
3654  * \retval 0 on success
3655  * \retval 1 on buffer full
3656  */
3657 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3658                                loff_t offset, __u64 ino,
3659                                unsigned d_type)
3660 {
3661         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
3662         struct osd_it_ea_dirent *ent  = it->oie_dirent;
3663         struct lu_fid           *fid  = &ent->oied_fid;
3664         struct osd_fid_pack     *rec;
3665         ENTRY;
3666
3667         /* this should never happen */
3668         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3669                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3670                 RETURN(-EIO);
3671         }
3672
3673         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3674             OSD_IT_EA_BUFSIZE)
3675                 RETURN(1);
3676
3677         if (d_type & LDISKFS_DIRENT_LUFID) {
3678                 rec = (struct osd_fid_pack*) (name + namelen + 1);
3679
3680                 if (osd_fid_unpack(fid, rec) != 0)
3681                         fid_zero(fid);
3682
3683                 d_type &= ~LDISKFS_DIRENT_LUFID;
3684         } else {
3685                 fid_zero(fid);
3686         }
3687
3688         ent->oied_ino     = ino;
3689         ent->oied_off     = offset;
3690         ent->oied_namelen = namelen;
3691         ent->oied_type    = d_type;
3692
3693         memcpy(ent->oied_name, name, namelen);
3694
3695         it->oie_rd_dirent++;
3696         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
3697         RETURN(0);
3698 }
3699
3700 /**
3701  * Calls ->readdir() to load a directory entry at a time
3702  * and stored it in iterator's in-memory data structure.
3703  *
3704  * \param di iterator's in memory structure
3705  *
3706  * \retval   0 on success
3707  * \retval -ve on error
3708  */
3709 static int osd_ldiskfs_it_fill(const struct lu_env *env,
3710                                const struct dt_it *di)
3711 {
3712         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
3713         struct osd_object  *obj   = it->oie_obj;
3714         struct inode       *inode = obj->oo_inode;
3715         struct htree_lock  *hlock = NULL;
3716         int                 result = 0;
3717
3718         ENTRY;
3719         it->oie_dirent = it->oie_buf;
3720         it->oie_rd_dirent = 0;
3721
3722         if (obj->oo_hl_head != NULL) {
3723                 hlock = osd_oti_get(env)->oti_hlock;
3724                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3725                                    inode, LDISKFS_HLOCK_READDIR);
3726         } else {
3727                 cfs_down_read(&obj->oo_ext_idx_sem);
3728         }
3729
3730         result = inode->i_fop->readdir(&it->oie_file, it,
3731                                        (filldir_t) osd_ldiskfs_filldir);
3732
3733         if (hlock != NULL)
3734                 ldiskfs_htree_unlock(hlock);
3735         else
3736                 cfs_up_read(&obj->oo_ext_idx_sem);
3737
3738         if (it->oie_rd_dirent == 0) {
3739                 result = -EIO;
3740         } else {
3741                 it->oie_dirent = it->oie_buf;
3742                 it->oie_it_dirent = 1;
3743         }
3744
3745         RETURN(result);
3746 }
3747
3748 /**
3749  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3750  * to load a directory entry at a time and stored it in
3751  * iterator's in-memory data structure.
3752  *
3753  * \param di iterator's in memory structure
3754  *
3755  * \retval +ve iterator reached to end
3756  * \retval   0 iterator not reached to end
3757  * \retval -ve on error
3758  */
3759 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
3760 {
3761         struct osd_it_ea *it = (struct osd_it_ea *)di;
3762         int rc;
3763
3764         ENTRY;
3765
3766         if (it->oie_it_dirent < it->oie_rd_dirent) {
3767                 it->oie_dirent =
3768                         (void *) it->oie_dirent +
3769                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
3770                                        it->oie_dirent->oied_namelen);
3771                 it->oie_it_dirent++;
3772                 RETURN(0);
3773         } else {
3774                 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
3775                         rc = +1;
3776                 else
3777                         rc = osd_ldiskfs_it_fill(env, di);
3778         }
3779
3780         RETURN(rc);
3781 }
3782
3783 /**
3784  * Returns the key at current position from iterator's in memory structure.
3785  *
3786  * \param di iterator's in memory structure
3787  *
3788  * \retval key i.e. struct dt_key on success
3789  */
3790 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
3791                                     const struct dt_it *di)
3792 {
3793         struct osd_it_ea *it = (struct osd_it_ea *)di;
3794
3795         return (struct dt_key *)it->oie_dirent->oied_name;
3796 }
3797
3798 /**
3799  * Returns the key's size at current position from iterator's in memory structure.
3800  *
3801  * \param di iterator's in memory structure
3802  *
3803  * \retval key_size i.e. struct dt_key on success
3804  */
3805 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
3806 {
3807         struct osd_it_ea *it = (struct osd_it_ea *)di;
3808
3809         return it->oie_dirent->oied_namelen;
3810 }
3811
3812
3813 /**
3814  * Returns the value (i.e. fid/igif) at current position from iterator's
3815  * in memory structure.
3816  *
3817  * \param di struct osd_it_ea, iterator's in memory structure
3818  * \param attr attr requested for dirent.
3819  * \param lde lustre dirent
3820  *
3821  * \retval   0 no error and \param lde has correct lustre dirent.
3822  * \retval -ve on error
3823  */
3824 static inline int osd_it_ea_rec(const struct lu_env *env,
3825                                 const struct dt_it *di,
3826                                 struct dt_rec *dtrec, __u32 attr)
3827 {
3828         struct osd_it_ea        *it     = (struct osd_it_ea *)di;
3829         struct osd_object       *obj    = it->oie_obj;
3830         struct lu_fid           *fid    = &it->oie_dirent->oied_fid;
3831         struct lu_dirent        *lde    = (struct lu_dirent *)dtrec;
3832         int    rc = 0;
3833
3834         ENTRY;
3835
3836         if (!fid_is_sane(fid))
3837                 rc = osd_ea_fid_get(env, obj, it->oie_dirent->oied_ino, fid);
3838
3839         if (rc == 0)
3840                 osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
3841                                    it->oie_dirent->oied_name,
3842                                    it->oie_dirent->oied_namelen,
3843                                    it->oie_dirent->oied_type,
3844                                    attr);
3845         RETURN(rc);
3846 }
3847
3848 /**
3849  * Returns a cookie for current position of the iterator head, so that
3850  * user can use this cookie to load/start the iterator next time.
3851  *
3852  * \param di iterator's in memory structure
3853  *
3854  * \retval cookie for current position, on success
3855  */
3856 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
3857 {
3858         struct osd_it_ea *it = (struct osd_it_ea *)di;
3859
3860         return it->oie_dirent->oied_off;
3861 }
3862
3863 /**
3864  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
3865  * to load a directory entry at a time and stored it i inn,
3866  * in iterator's in-memory data structure.
3867  *
3868  * \param di struct osd_it_ea, iterator's in memory structure
3869  *
3870  * \retval +ve on success
3871  * \retval -ve on error
3872  */
3873 static int osd_it_ea_load(const struct lu_env *env,
3874                           const struct dt_it *di, __u64 hash)
3875 {
3876         struct osd_it_ea *it = (struct osd_it_ea *)di;
3877         int rc;
3878
3879         ENTRY;
3880         it->oie_file.f_pos = hash;
3881
3882         rc =  osd_ldiskfs_it_fill(env, di);
3883         if (rc == 0)
3884                 rc = +1;
3885
3886         RETURN(rc);
3887 }
3888
3889 /**
3890  * Index lookup function for interoperability mode (b11826).
3891  *
3892  * \param key,  key i.e. file name to be searched
3893  *
3894  * \retval +ve, on success
3895  * \retval -ve, on error
3896  */
3897 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
3898                                struct dt_rec *rec, const struct dt_key *key,
3899                                struct lustre_capa *capa)
3900 {
3901         struct osd_object *obj = osd_dt_obj(dt);
3902         int rc = 0;
3903
3904         ENTRY;
3905
3906         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
3907         LINVRNT(osd_invariant(obj));
3908
3909         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3910                 return -EACCES;
3911
3912         rc = osd_ea_lookup_rec(env, obj, rec, key);
3913
3914         if (rc == 0)
3915                 rc = +1;
3916         RETURN(rc);
3917 }
3918
3919 /**
3920  * Index and Iterator operations for interoperability
3921  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3922  */
3923 static const struct dt_index_operations osd_index_ea_ops = {
3924         .dio_lookup         = osd_index_ea_lookup,
3925         .dio_declare_insert = osd_index_declare_ea_insert,
3926         .dio_insert         = osd_index_ea_insert,
3927         .dio_declare_delete = osd_index_declare_ea_delete,
3928         .dio_delete         = osd_index_ea_delete,
3929         .dio_it     = {
3930                 .init     = osd_it_ea_init,
3931                 .fini     = osd_it_ea_fini,
3932                 .get      = osd_it_ea_get,
3933                 .put      = osd_it_ea_put,
3934                 .next     = osd_it_ea_next,
3935                 .key      = osd_it_ea_key,
3936                 .key_size = osd_it_ea_key_size,
3937                 .rec      = osd_it_ea_rec,
3938                 .store    = osd_it_ea_store,
3939                 .load     = osd_it_ea_load
3940         }
3941 };
3942
3943 static void *osd_key_init(const struct lu_context *ctx,
3944                           struct lu_context_key *key)
3945 {
3946         struct osd_thread_info *info;
3947
3948         OBD_ALLOC_PTR(info);
3949         if (info == NULL)
3950                 return ERR_PTR(-ENOMEM);
3951
3952         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3953         if (info->oti_it_ea_buf == NULL)
3954                 goto out_free_info;
3955
3956         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
3957
3958         info->oti_hlock = ldiskfs_htree_lock_alloc();
3959         if (info->oti_hlock == NULL)
3960                 goto out_free_ea;
3961
3962         return info;
3963
3964  out_free_ea:
3965         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3966  out_free_info:
3967         OBD_FREE_PTR(info);
3968         return ERR_PTR(-ENOMEM);
3969 }
3970
3971 static void osd_key_fini(const struct lu_context *ctx,
3972                          struct lu_context_key *key, void* data)
3973 {
3974         struct osd_thread_info *info = data;
3975
3976         if (info->oti_hlock != NULL)
3977                 ldiskfs_htree_lock_free(info->oti_hlock);
3978         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
3979         OBD_FREE_PTR(info);
3980 }
3981
3982 static void osd_key_exit(const struct lu_context *ctx,
3983                          struct lu_context_key *key, void *data)
3984 {
3985         struct osd_thread_info *info = data;
3986
3987         LASSERT(info->oti_r_locks == 0);
3988         LASSERT(info->oti_w_locks == 0);
3989         LASSERT(info->oti_txns    == 0);
3990 }
3991
3992 /* type constructor/destructor: osd_type_init, osd_type_fini */
3993 LU_TYPE_INIT_FINI(osd, &osd_key);
3994
3995 struct lu_context_key osd_key = {
3996         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
3997         .lct_init = osd_key_init,
3998         .lct_fini = osd_key_fini,
3999         .lct_exit = osd_key_exit
4000 };
4001
4002
4003 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4004                            const char *name, struct lu_device *next)
4005 {
4006         return osd_procfs_init(osd_dev(d), name);
4007 }
4008
4009 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4010 {
4011         ENTRY;
4012
4013         osd_scrub_cleanup(env, o);
4014
4015         if (o->od_fsops) {
4016                 fsfilt_put_ops(o->od_fsops);
4017         o->od_fsops = NULL;
4018         }
4019
4020         RETURN(0);
4021 }
4022
4023 static int osd_mount(const struct lu_env *env,
4024                      struct osd_device *o, struct lustre_cfg *cfg)
4025 {
4026         struct lustre_mount_info *lmi;
4027         const char               *dev  = lustre_cfg_string(cfg, 0);
4028         struct lustre_disk_data  *ldd;
4029         struct lustre_sb_info    *lsi;
4030         int                       rc = 0;
4031
4032         ENTRY;
4033
4034         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4035         if (o->od_fsops == NULL) {
4036                 CERROR("Can't find fsfilt_ldiskfs\n");
4037                 RETURN(-ENOTSUPP);
4038         }
4039
4040         if (o->od_mount != NULL) {
4041                 CERROR("Already mounted (%s)\n", dev);
4042                 RETURN(-EEXIST);
4043         }
4044
4045         /* get mount */
4046         lmi = server_get_mount(dev);
4047         if (lmi == NULL) {
4048                 CERROR("Cannot get mount info for %s!\n", dev);
4049                 RETURN(-EFAULT);
4050         }
4051
4052         LASSERT(lmi != NULL);
4053         /* save lustre_mount_info in dt_device */
4054         o->od_mount = lmi;
4055         o->od_mnt = lmi->lmi_mnt;
4056
4057         lsi = s2lsi(lmi->lmi_sb);
4058         ldd = lsi->lsi_ldd;
4059
4060         if (ldd->ldd_flags & LDD_F_IAM_DIR) {
4061                 o->od_iop_mode = 0;
4062                 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", dev);
4063         } else
4064                 o->od_iop_mode = 1;
4065
4066         if (ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
4067                 rc = osd_compat_init(o);
4068                 if (rc)
4069                         CERROR("%s: can't initialize compats: %d\n", dev, rc);
4070         }
4071
4072         RETURN(rc);
4073 }
4074
4075 static struct lu_device *osd_device_fini(const struct lu_env *env,
4076                                          struct lu_device *d)
4077 {
4078         int rc;
4079         ENTRY;
4080
4081         osd_compat_fini(osd_dev(d));
4082
4083         shrink_dcache_sb(osd_sb(osd_dev(d)));
4084         osd_sync(env, lu2dt_dev(d));
4085
4086         rc = osd_procfs_fini(osd_dev(d));
4087         if (rc) {
4088                 CERROR("proc fini error %d \n", rc);
4089                 RETURN (ERR_PTR(rc));
4090         }
4091
4092         if (osd_dev(d)->od_mount)
4093                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
4094                                  osd_dev(d)->od_mount->lmi_mnt);
4095         osd_dev(d)->od_mount = NULL;
4096
4097         RETURN(NULL);
4098 }
4099
4100 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4101                                           struct lu_device_type *t,
4102                                           struct lustre_cfg *cfg)
4103 {
4104         struct lu_device  *l;
4105         struct osd_device *o;
4106
4107         OBD_ALLOC_PTR(o);
4108         if (o != NULL) {
4109                 int result;
4110
4111                 result = dt_device_init(&o->od_dt_dev, t);
4112                 if (result == 0) {
4113                         l = osd2lu_dev(o);
4114                         l->ld_ops = &osd_lu_ops;
4115                         o->od_dt_dev.dd_ops = &osd_dt_ops;
4116                         cfs_spin_lock_init(&o->od_osfs_lock);
4117                         cfs_mutex_init(&o->od_otable_mutex);
4118                         o->od_osfs_age = cfs_time_shift_64(-1000);
4119                         o->od_capa_hash = init_capa_hash();
4120                         if (o->od_capa_hash == NULL) {
4121                                 dt_device_fini(&o->od_dt_dev);
4122                                 l = ERR_PTR(-ENOMEM);
4123                         }
4124                 } else
4125                         l = ERR_PTR(result);
4126
4127                 if (IS_ERR(l))
4128                         OBD_FREE_PTR(o);
4129         } else
4130                 l = ERR_PTR(-ENOMEM);
4131         return l;
4132 }
4133
4134 static struct lu_device *osd_device_free(const struct lu_env *env,
4135                                          struct lu_device *d)
4136 {
4137         struct osd_device *o = osd_dev(d);
4138         ENTRY;
4139
4140         cleanup_capa_hash(o->od_capa_hash);
4141         dt_device_fini(&o->od_dt_dev);
4142         OBD_FREE_PTR(o);
4143         RETURN(NULL);
4144 }
4145
4146 static int osd_process_config(const struct lu_env *env,
4147                               struct lu_device *d, struct lustre_cfg *cfg)
4148 {
4149         struct osd_device *o = osd_dev(d);
4150         int err;
4151         ENTRY;
4152
4153         switch(cfg->lcfg_command) {
4154         case LCFG_SETUP:
4155                 err = osd_mount(env, o, cfg);
4156                 break;
4157         case LCFG_CLEANUP:
4158                 err = osd_shutdown(env, o);
4159                 break;
4160         default:
4161                 err = -ENOSYS;
4162         }
4163
4164         RETURN(err);
4165 }
4166
4167 static int osd_recovery_complete(const struct lu_env *env,
4168                                  struct lu_device *d)
4169 {
4170         RETURN(0);
4171 }
4172
4173 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4174                        struct lu_device *dev)
4175 {
4176         struct osd_device *osd = osd_dev(dev);
4177         int                result;
4178         ENTRY;
4179
4180         /* 1. setup scrub, including OI files initialization */
4181         result = osd_scrub_setup(env, osd);
4182         if (result < 0)
4183                 RETURN(result);
4184
4185         if (!lu_device_is_md(pdev))
4186                 RETURN(0);
4187
4188         /* 2. setup local objects */
4189         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4190         RETURN(result);
4191 }
4192
4193 static const struct lu_object_operations osd_lu_obj_ops = {
4194         .loo_object_init      = osd_object_init,
4195         .loo_object_delete    = osd_object_delete,
4196         .loo_object_release   = osd_object_release,
4197         .loo_object_free      = osd_object_free,
4198         .loo_object_print     = osd_object_print,
4199         .loo_object_invariant = osd_object_invariant
4200 };
4201
4202 const struct lu_device_operations osd_lu_ops = {
4203         .ldo_object_alloc      = osd_object_alloc,
4204         .ldo_process_config    = osd_process_config,
4205         .ldo_recovery_complete = osd_recovery_complete,
4206         .ldo_prepare           = osd_prepare,
4207 };
4208
4209 static const struct lu_device_type_operations osd_device_type_ops = {
4210         .ldto_init = osd_type_init,
4211         .ldto_fini = osd_type_fini,
4212
4213         .ldto_start = osd_type_start,
4214         .ldto_stop  = osd_type_stop,
4215
4216         .ldto_device_alloc = osd_device_alloc,
4217         .ldto_device_free  = osd_device_free,
4218
4219         .ldto_device_init    = osd_device_init,
4220         .ldto_device_fini    = osd_device_fini
4221 };
4222
4223 static struct lu_device_type osd_device_type = {
4224         .ldt_tags     = LU_DEVICE_DT,
4225         .ldt_name     = LUSTRE_OSD_NAME,
4226         .ldt_ops      = &osd_device_type_ops,
4227         .ldt_ctx_tags = LCT_LOCAL,
4228 };
4229
4230 /*
4231  * lprocfs legacy support.
4232  */
4233 static struct obd_ops osd_obd_device_ops = {
4234         .o_owner = THIS_MODULE
4235 };
4236
4237 static int __init osd_mod_init(void)
4238 {
4239         struct lprocfs_static_vars lvars;
4240
4241         osd_oi_mod_init();
4242         lprocfs_osd_init_vars(&lvars);
4243         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4244                                    LUSTRE_OSD_NAME, &osd_device_type);
4245 }
4246
4247 static void __exit osd_mod_exit(void)
4248 {
4249         class_unregister_type(LUSTRE_OSD_NAME);
4250 }
4251
4252 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4253 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4254 MODULE_LICENSE("GPL");
4255
4256 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);