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 /* dt_acct_features */
  76 #include <lquota.h>
  77
  78 #ifdef HAVE_LDISKFS_PDO
  79 int ldiskfs_pdo = 1;
  80 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
  81                 "ldiskfs with parallel directory operations");
  82 #else
  83 int ldiskfs_pdo = 0;
  84 #endif
  85
  86 static const char dot[] = ".";
  87 static const char dotdot[] = "..";
  88 static const char remote_obj_dir[] = "REM_OBJ_DIR";
  89
  90 static const struct lu_object_operations      osd_lu_obj_ops;
  91 static const struct dt_object_operations      osd_obj_ops;
  92 static const struct dt_object_operations      osd_obj_ea_ops;
  93 static const struct dt_index_operations       osd_index_iam_ops;
  94 static const struct dt_index_operations       osd_index_ea_ops;
  95
  96 static int osd_has_index(const struct osd_object *obj)
  97 {
  98         return obj->oo_dt.do_index_ops != NULL;
  99 }
 100
 101 static int osd_object_invariant(const struct lu_object *l)
 102 {
 103         return osd_invariant(osd_obj(l));
 104 }
 105
 106 #ifdef HAVE_QUOTA_SUPPORT
 107 static inline void
 108 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save)
 109 {
 110         struct md_ucred *uc = md_ucred(env);
 111         struct cred     *tc;
 112
 113         LASSERT(uc != NULL);
 114
 115         save->oc_uid = current_fsuid();
 116         save->oc_gid = current_fsgid();
 117         save->oc_cap = current_cap();
 118         if ((tc = prepare_creds())) {
 119                 tc->fsuid         = uc->mu_fsuid;
 120                 tc->fsgid         = uc->mu_fsgid;
 121                 commit_creds(tc);
 122         }
 123         /* XXX not suboptimal */
 124         cfs_curproc_cap_unpack(uc->mu_cap);
 125 }
 126
 127 static inline void
 128 osd_pop_ctxt(struct osd_ctxt *save)
 129 {
 130         struct cred *tc;
 131
 132         if ((tc = prepare_creds())) {
 133                 tc->fsuid         = save->oc_uid;
 134                 tc->fsgid         = save->oc_gid;
 135                 tc->cap_effective = save->oc_cap;
 136                 commit_creds(tc);
 137         }
 138 }
 139 #endif
 140
 141 /*
 142  * Concurrency: doesn't matter
 143  */
 144 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 145 {
 146         return osd_oti_get(env)->oti_r_locks > 0;
 147 }
 148
 149 /*
 150  * Concurrency: doesn't matter
 151  */
 152 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 153 {
 154         struct osd_thread_info *oti = osd_oti_get(env);
 155         return oti->oti_w_locks > 0 && o->oo_owner == env;
 156 }
 157
 158 /*
 159  * Concurrency: doesn't access mutable data
 160  */
 161 static int osd_root_get(const struct lu_env *env,
 162                         struct dt_device *dev, struct lu_fid *f)
 163 {
 164         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 165         return 0;
 166 }
 167
 168 static inline int osd_qid_type(struct osd_thandle *oh, int i)
 169 {
 170         return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
 171 }
 172
 173 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
 174 {
 175         oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
 176 }
 177
 178 void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
 179                      int type, uid_t id, struct inode *inode)
 180 {
 181 #ifdef CONFIG_QUOTA
 182         int i, allocated = 0;
 183         struct osd_object *obj;
 184
 185         LASSERT(dt != NULL);
 186         LASSERT(oh != NULL);
 187         LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
 188                  oh->ot_id_cnt);
 189
 190         /* id entry is allocated in the quota file */
 191         if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
 192                 allocated = 1;
 193
 194         for (i = 0; i < oh->ot_id_cnt; i++) {
 195                 if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
 196                         return;
 197         }
 198
 199         if (unlikely(i >= OSD_MAX_UGID_CNT)) {
 200                 CERROR("more than %d uid/gids for a transaction?\n", i);
 201                 return;
 202         }
 203
 204         oh->ot_id_array[i] = id;
 205         osd_qid_set_type(oh, i, type);
 206         oh->ot_id_cnt++;
 207         obj = osd_dt_obj(dt);
 208         oh->ot_credits += (allocated || id == 0) ?
 209                 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
 210 #endif
 211 }
 212
 213 /*
 214  * OSD object methods.
 215  */
 216
 217 /*
 218  * Concurrency: no concurrent access is possible that early in object
 219  * life-cycle.
 220  */
 221 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 222                                           const struct lu_object_header *hdr,
 223                                           struct lu_device *d)
 224 {
 225         struct osd_object *mo;
 226
 227         OBD_ALLOC_PTR(mo);
 228         if (mo != NULL) {
 229                 struct lu_object *l;
 230
 231                 l = &mo->oo_dt.do_lu;
 232                 dt_object_init(&mo->oo_dt, NULL, d);
 233                 if (osd_dev(d)->od_iop_mode)
 234                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 235                 else
 236                         mo->oo_dt.do_ops = &osd_obj_ops;
 237
 238                 l->lo_ops = &osd_lu_obj_ops;
 239                 cfs_init_rwsem(&mo->oo_sem);
 240                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 241                 cfs_spin_lock_init(&mo->oo_guard);
 242                 return l;
 243         } else {
 244                 return NULL;
 245         }
 246 }
 247
 248 static int osd_get_lma(struct inode *inode, struct dentry *dentry,
 249                        struct lustre_mdt_attrs *lma)
 250 {
 251         int rc;
 252
 253         dentry->d_inode = inode;
 254         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
 255                                    sizeof(*lma));
 256         if (rc > 0) {
 257                 /* Check LMA compatibility */
 258                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 259                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 260                               "%lx/%#x\n",
 261                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 262                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 263                                                         ~LMA_INCOMPAT_SUPP);
 264                         rc = -ENOSYS;
 265                 } else {
 266                         lustre_lma_swab(lma);
 267                         rc = 0;
 268                 }
 269         } else if (rc == 0) {
 270                 rc = -ENODATA;
 271         }
 272
 273         return rc;
 274 }
 275
 276 /*
 277  * retrieve object from backend ext fs.
 278  **/
 279 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 280                        struct osd_inode_id *id)
 281 {
 282         struct inode *inode = NULL;
 283
 284         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 285         if (IS_ERR(inode)) {
 286                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 287                        id->oii_ino, PTR_ERR(inode));
 288         } else if (id->oii_gen != OSD_OII_NOGEN &&
 289                    inode->i_generation != id->oii_gen) {
 290                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 291                        "gen1 = %u\n",
 292                        id->oii_ino, id->oii_gen, inode->i_generation);
 293                 iput(inode);
 294                 inode = ERR_PTR(-ESTALE);
 295         } else if (inode->i_nlink == 0) {
 296                 /* due to parallel readdir and unlink,
 297                 * we can have dead inode here. */
 298                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 299                 make_bad_inode(inode);
 300                 iput(inode);
 301                 inode = ERR_PTR(-ESTALE);
 302         } else if (is_bad_inode(inode)) {
 303                 CWARN("%.16s: bad inode: ino = %u\n",
 304                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 305                 iput(inode);
 306                 inode = ERR_PTR(-ENOENT);
 307         } else {
 308                 if (id->oii_gen == OSD_OII_NOGEN)
 309                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 310
 311                 /* Do not update file c/mtime in ldiskfs.
 312                  * NB: we don't have any lock to protect this because we don't
 313                  * have reference on osd_object now, but contention with
 314                  * another lookup + attr_set can't happen in the tiny window
 315                  * between if (...) and set S_NOCMTIME. */
 316                 if (!(inode->i_flags & S_NOCMTIME))
 317                         inode->i_flags |= S_NOCMTIME;
 318         }
 319         return inode;
 320 }
 321
 322 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 323                            struct osd_inode_id *id, struct lu_fid *fid)
 324 {
 325         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 326         struct inode            *inode;
 327         int                      rc;
 328
 329         inode = osd_iget(info, dev, id);
 330         if (IS_ERR(inode))
 331                 return inode;
 332
 333         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 334         if (rc == 0) {
 335                 *fid = lma->lma_self_fid;
 336         } else if (rc == -ENODATA) {
 337                 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
 338         } else {
 339                 iput(inode);
 340                 inode = ERR_PTR(rc);
 341         }
 342         return inode;
 343 }
 344
 345 static struct inode *
 346 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 347                 struct osd_inode_id *id, const struct lu_fid *fid)
 348 {
 349         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 350         struct inode            *inode;
 351         int                      rc;
 352
 353         inode = osd_iget(info, dev, id);
 354         if (IS_ERR(inode))
 355                 return inode;
 356
 357         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 358         if (rc != 0) {
 359                 if (rc == -ENODATA) {
 360                         CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
 361                                inode->i_ino, PFID(fid));
 362                         rc = -EREMCHG;
 363                 }
 364                 iput(inode);
 365                 return ERR_PTR(rc);
 366         }
 367
 368         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 369                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 370                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 371                 iput(inode);
 372                 return ERR_PTR(EREMCHG);
 373         }
 374         return inode;
 375 }
 376
 377 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 378                           const struct lu_fid *fid,
 379                           const struct lu_object_conf *conf)
 380 {
 381         struct osd_thread_info *info;
 382         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 383         struct osd_device      *dev;
 384         struct osd_idmap_cache *oic;
 385         struct osd_inode_id    *id;
 386         struct inode           *inode;
 387         struct osd_scrub       *scrub;
 388         struct scrub_file      *sf;
 389         int                     result;
 390         int                     verify = 0;
 391         ENTRY;
 392
 393         LINVRNT(osd_invariant(obj));
 394         LASSERT(obj->oo_inode == NULL);
 395         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 396
 397         dev = osd_dev(ldev);
 398         scrub = &dev->od_scrub;
 399         sf = &scrub->os_file;
 400         info = osd_oti_get(env);
 401         LASSERT(info);
 402         oic = &info->oti_cache;
 403         id  = &oic->oic_lid;
 404
 405         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 406                 RETURN(-ENOENT);
 407
 408         if (fid_is_norm(fid)) {
 409                 /* Search order: 1. per-thread cache. */
 410                 if (lu_fid_eq(fid, &oic->oic_fid)) {
 411                         goto iget;
 412                 } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 413                         /* Search order: 2. OI scrub pending list. */
 414                         result = osd_oii_lookup(dev, fid, id);
 415                         if (result == 0)
 416                                 goto iget;
 417                 }
 418
 419                 if (sf->sf_flags & SF_INCONSISTENT)
 420                         verify = 1;
 421         }
 422
 423         fid_zero(&oic->oic_fid);
 424         /* Search order: 3. OI files. */
 425         result = osd_oi_lookup(info, dev, fid, id);
 426         if (result != 0 && result != -ENOENT)
 427                 GOTO(out, result);
 428
 429         /* If fid wasn't found in oi, inode-less object is created,
 430          * for which lu_object_exists() returns false. This is used
 431          * in a (frequent) case when objects are created as locking
 432          * anchors or place holders for objects yet to be created. */
 433         if (conf != NULL && conf->loc_flags & LOC_F_NEW) {
 434                 if (unlikely(result == 0))
 435                         GOTO(out, result = -EEXIST);
 436                 else
 437                         GOTO(out, result = 0);
 438         }
 439
 440         if (result == -ENOENT) {
 441                 if (!fid_is_norm(fid) ||
 442                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 443                                       sf->sf_oi_bitmap))
 444                         GOTO(out, result = 0);
 445
 446                 goto trigger;
 447         }
 448
 449 iget:
 450         if (verify == 0)
 451                 inode = osd_iget(info, dev, id);
 452         else
 453                 inode = osd_iget_verify(info, dev, id, fid);
 454         if (IS_ERR(inode)) {
 455                 result = PTR_ERR(inode);
 456                 if (result == -ENOENT || result == -ESTALE) {
 457                         result = 0;
 458                 } else if (result == -EREMCHG) {
 459
 460 trigger:
 461                         if (thread_is_running(&scrub->os_thread)) {
 462                                 result = -EINPROGRESS;
 463                         } else if (!scrub->os_no_scrub) {
 464                                 result = osd_scrub_start(dev);
 465                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 466                                                "for "DFID", rc = %d [1]\n",
 467                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 468                                                s_volume_name,PFID(fid), result);
 469                                 if (result == 0 || result == -EALREADY)
 470                                         result = -EINPROGRESS;
 471                                 else
 472                                         result = -EREMCHG;
 473                         }
 474                 }
 475
 476                 GOTO(out, result);
 477         }
 478
 479         obj->oo_inode = inode;
 480         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 481         if (dev->od_iop_mode) {
 482                 obj->oo_compat_dot_created = 1;
 483                 obj->oo_compat_dotdot_created = 1;
 484         }
 485
 486         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 487                 GOTO(out, result = 0);
 488
 489         LASSERT(obj->oo_hl_head == NULL);
 490         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 491         if (obj->oo_hl_head == NULL) {
 492                 obj->oo_inode = NULL;
 493                 iput(inode);
 494                 GOTO(out, result = -ENOMEM);
 495         }
 496         GOTO(out, result = 0);
 497
 498 out:
 499         LINVRNT(osd_invariant(obj));
 500         return result;
 501 }
 502
 503 /*
 504  * Concurrency: shouldn't matter.
 505  */
 506 static void osd_object_init0(struct osd_object *obj)
 507 {
 508         LASSERT(obj->oo_inode != NULL);
 509         obj->oo_dt.do_body_ops = &osd_body_ops;
 510         obj->oo_dt.do_lu.lo_header->loh_attr |=
 511                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 512 }
 513
 514 /*
 515  * Concurrency: no concurrent access is possible that early in object
 516  * life-cycle.
 517  */
 518 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 519                            const struct lu_object_conf *conf)
 520 {
 521         struct osd_object *obj = osd_obj(l);
 522         int result;
 523
 524         LINVRNT(osd_invariant(obj));
 525
 526         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 527         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 528         if (result == 0) {
 529                 if (obj->oo_inode != NULL)
 530                         osd_object_init0(obj);
 531         }
 532         LINVRNT(osd_invariant(obj));
 533         return result;
 534 }
 535
 536 /*
 537  * Concurrency: no concurrent access is possible that late in object
 538  * life-cycle.
 539  */
 540 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 541 {
 542         struct osd_object *obj = osd_obj(l);
 543
 544         LINVRNT(osd_invariant(obj));
 545
 546         dt_object_fini(&obj->oo_dt);
 547         if (obj->oo_hl_head != NULL)
 548                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 549         OBD_FREE_PTR(obj);
 550 }
 551
 552 /*
 553  * Concurrency: no concurrent access is possible that late in object
 554  * life-cycle.
 555  */
 556 static void osd_index_fini(struct osd_object *o)
 557 {
 558         struct iam_container *bag;
 559
 560         if (o->oo_dir != NULL) {
 561                 bag = &o->oo_dir->od_container;
 562                 if (o->oo_inode != NULL) {
 563                         if (bag->ic_object == o->oo_inode)
 564                                 iam_container_fini(bag);
 565                 }
 566                 OBD_FREE_PTR(o->oo_dir);
 567                 o->oo_dir = NULL;
 568         }
 569 }
 570
 571 /*
 572  * Concurrency: no concurrent access is possible that late in object
 573  * life-cycle (for all existing callers, that is. New callers have to provide
 574  * their own locking.)
 575  */
 576 static int osd_inode_unlinked(const struct inode *inode)
 577 {
 578         return inode->i_nlink == 0;
 579 }
 580
 581 enum {
 582         OSD_TXN_OI_DELETE_CREDITS    = 20,
 583         OSD_TXN_INODE_DELETE_CREDITS = 20
 584 };
 585
 586 /*
 587  * Journal
 588  */
 589
 590 #if OSD_THANDLE_STATS
 591 /**
 592  * Set time when the handle is allocated
 593  */
 594 static void osd_th_alloced(struct osd_thandle *oth)
 595 {
 596         oth->oth_alloced = cfs_time_current();
 597 }
 598
 599 /**
 600  * Set time when the handle started
 601  */
 602 static void osd_th_started(struct osd_thandle *oth)
 603 {
 604         oth->oth_started = cfs_time_current();
 605 }
 606
 607 /**
 608  * Helper function to convert time interval to microseconds packed in
 609  * long int (default time units for the counter in "stats" initialized
 610  * by lu_time_init() )
 611  */
 612 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 613 {
 614         struct timeval val;
 615
 616         cfs_duration_usec(cfs_time_sub(end, start), &val);
 617         return val.tv_sec * 1000000 + val.tv_usec;
 618 }
 619
 620 /**
 621  * Check whether the we deal with this handle for too long.
 622  */
 623 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 624                                 cfs_time_t alloced, cfs_time_t started,
 625                                 cfs_time_t closed)
 626 {
 627         cfs_time_t now = cfs_time_current();
 628
 629         LASSERT(dev != NULL);
 630
 631         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 632                             interval_to_usec(alloced, started));
 633         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 634                             interval_to_usec(started, closed));
 635         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 636                             interval_to_usec(closed, now));
 637
 638         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 639                 CWARN("transaction handle %p was open for too long: "
 640                       "now "CFS_TIME_T" ,"
 641                       "alloced "CFS_TIME_T" ,"
 642                       "started "CFS_TIME_T" ,"
 643                       "closed "CFS_TIME_T"\n",
 644                       oth, now, alloced, started, closed);
 645                 libcfs_debug_dumpstack(NULL);
 646         }
 647 }
 648
 649 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 650 {                                                                       \
 651         cfs_time_t __closed = cfs_time_current();                       \
 652         cfs_time_t __alloced = oth->oth_alloced;                        \
 653         cfs_time_t __started = oth->oth_started;                        \
 654                                                                         \
 655         expr;                                                           \
 656         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 657 }
 658
 659 #else /* OSD_THANDLE_STATS */
 660
 661 #define osd_th_alloced(h)                  do {} while(0)
 662 #define osd_th_started(h)                  do {} while(0)
 663 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 664
 665 #endif /* OSD_THANDLE_STATS */
 666
 667 /*
 668  * Concurrency: doesn't access mutable data.
 669  */
 670 static int osd_param_is_sane(const struct osd_device *dev,
 671                              const struct thandle *th)
 672 {
 673         struct osd_thandle *oh;
 674         oh = container_of0(th, struct osd_thandle, ot_super);
 675         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 676 }
 677
 678 /*
 679  * Concurrency: shouldn't matter.
 680  */
 681 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 682 static void osd_trans_commit_cb(struct super_block *sb,
 683                                 struct journal_callback *jcb, int error)
 684 #else
 685 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 686 #endif
 687 {
 688         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 689         struct thandle     *th  = &oh->ot_super;
 690         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 691         struct dt_txn_commit_cb *dcb, *tmp;
 692
 693         LASSERT(oh->ot_handle == NULL);
 694
 695         if (error)
 696                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 697
 698         dt_txn_hook_commit(th);
 699
 700         /* call per-transaction callbacks if any */
 701         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 702                 cfs_list_del_init(&dcb->dcb_linkage);
 703                 dcb->dcb_func(NULL, th, dcb, error);
 704         }
 705
 706         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 707         lu_device_put(lud);
 708         th->th_dev = NULL;
 709
 710         lu_context_exit(&th->th_ctx);
 711         lu_context_fini(&th->th_ctx);
 712         OBD_FREE_PTR(oh);
 713 }
 714
 715 static struct thandle *osd_trans_create(const struct lu_env *env,
 716                                         struct dt_device *d)
 717 {
 718         struct osd_thread_info *oti = osd_oti_get(env);
 719         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 720         struct osd_thandle     *oh;
 721         struct thandle         *th;
 722         ENTRY;
 723
 724         /* on pending IO in this thread should left from prev. request */
 725         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 726
 727         th = ERR_PTR(-ENOMEM);
 728         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 729         if (oh != NULL) {
 730                 th = &oh->ot_super;
 731                 th->th_dev = d;
 732                 th->th_result = 0;
 733                 th->th_tags = LCT_TX_HANDLE;
 734                 oh->ot_credits = 0;
 735                 oti->oti_dev = osd_dt_dev(d);
 736                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 737                 osd_th_alloced(oh);
 738         }
 739         RETURN(th);
 740 }
 741
 742 /*
 743  * Concurrency: shouldn't matter.
 744  */
 745 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 746                     struct thandle *th)
 747 {
 748         struct osd_thread_info *oti = osd_oti_get(env);
 749         struct osd_device  *dev = osd_dt_dev(d);
 750         handle_t           *jh;
 751         struct osd_thandle *oh;
 752         int rc;
 753
 754         ENTRY;
 755
 756         LASSERT(current->journal_info == NULL);
 757
 758         oh = container_of0(th, struct osd_thandle, ot_super);
 759         LASSERT(oh != NULL);
 760         LASSERT(oh->ot_handle == NULL);
 761
 762         rc = dt_txn_hook_start(env, d, th);
 763         if (rc != 0)
 764                 GOTO(out, rc);
 765
 766         if (!osd_param_is_sane(dev, th)) {
 767                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 768                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 769                       oh->ot_credits,
 770                       osd_journal(dev)->j_max_transaction_buffers);
 771                 /* XXX Limit the credits to 'max_transaction_buffers', and
 772                  *     let the underlying filesystem to catch the error if
 773                  *     we really need so many credits.
 774                  *
 775                  *     This should be removed when we can calculate the
 776                  *     credits precisely. */
 777                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 778 #ifdef OSD_TRACK_DECLARES
 779                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 780                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 781                        oh->ot_declare_xattr_set);
 782                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 783                        oh->ot_declare_create, oh->ot_declare_ref_add,
 784                        oh->ot_declare_ref_del, oh->ot_declare_write);
 785                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 786                        oh->ot_declare_insert, oh->ot_declare_delete,
 787                        oh->ot_declare_destroy);
 788 #endif
 789         }
 790
 791         /*
 792          * XXX temporary stuff. Some abstraction layer should
 793          * be used.
 794          */
 795         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 796         osd_th_started(oh);
 797         if (!IS_ERR(jh)) {
 798                 oh->ot_handle = jh;
 799                 LASSERT(oti->oti_txns == 0);
 800                 lu_context_init(&th->th_ctx, th->th_tags);
 801                 lu_context_enter(&th->th_ctx);
 802
 803                 lu_device_get(&d->dd_lu_dev);
 804                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 805                                              "osd-tx", th);
 806
 807                 /*
 808                  * XXX: current rule is that we first start tx,
 809                  *      then lock object(s), but we can't use
 810                  *      this rule for data (due to locking specifics
 811                  *      in ldiskfs). also in long-term we'd like to
 812                  *      use usually-used (locks;tx) ordering. so,
 813                  *      UGLY thing is that we'll use one ordering for
 814                  *      data (ofd) and reverse ordering for metadata
 815                  *      (mdd). then at some point we'll fix the latter
 816                  */
 817                 if (lu_device_is_md(&d->dd_lu_dev)) {
 818                         LASSERT(oti->oti_r_locks == 0);
 819                         LASSERT(oti->oti_w_locks == 0);
 820                 }
 821
 822                 oti->oti_txns++;
 823                 rc = 0;
 824         } else {
 825                 rc = PTR_ERR(jh);
 826         }
 827 out:
 828         RETURN(rc);
 829 }
 830
 831 /*
 832  * Concurrency: shouldn't matter.
 833  */
 834 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 835 {
 836         int                     rc = 0;
 837         struct osd_thandle     *oh;
 838         struct osd_thread_info *oti = osd_oti_get(env);
 839         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 840
 841         ENTRY;
 842
 843         oh = container_of0(th, struct osd_thandle, ot_super);
 844
 845         if (oh->ot_handle != NULL) {
 846                 handle_t *hdl = oh->ot_handle;
 847
 848                 hdl->h_sync = th->th_sync;
 849
 850                 /*
 851                  * add commit callback
 852                  * notice we don't do this in osd_trans_start()
 853                  * as underlying transaction can change during truncate
 854                  */
 855                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 856                                          &oh->ot_jcb);
 857
 858                 LASSERT(oti->oti_txns == 1);
 859                 oti->oti_txns--;
 860                 /*
 861                  * XXX: current rule is that we first start tx,
 862                  *      then lock object(s), but we can't use
 863                  *      this rule for data (due to locking specifics
 864                  *      in ldiskfs). also in long-term we'd like to
 865                  *      use usually-used (locks;tx) ordering. so,
 866                  *      UGLY thing is that we'll use one ordering for
 867                  *      data (ofd) and reverse ordering for metadata
 868                  *      (mdd). then at some point we'll fix the latter
 869                  */
 870                 if (lu_device_is_md(&th->th_dev->dd_lu_dev)) {
 871                         LASSERT(oti->oti_r_locks == 0);
 872                         LASSERT(oti->oti_w_locks == 0);
 873                 }
 874                 rc = dt_txn_hook_stop(env, th);
 875                 if (rc != 0)
 876                         CERROR("Failure in transaction hook: %d\n", rc);
 877                 oh->ot_handle = NULL;
 878                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 879                                   rc = ldiskfs_journal_stop(hdl));
 880                 if (rc != 0)
 881                         CERROR("Failure to stop transaction: %d\n", rc);
 882         } else {
 883                 OBD_FREE_PTR(oh);
 884         }
 885
 886         /* as we want IO to journal and data IO be concurrent, we don't block
 887          * awaiting data IO completion in osd_do_bio(), instead we wait here
 888          * once transaction is submitted to the journal. all reqular requests
 889          * don't do direct IO (except read/write), thus this wait_event becomes
 890          * no-op for them.
 891          *
 892          * IMPORTANT: we have to wait till any IO submited by the thread is
 893          * completed otherwise iobuf may be corrupted by different request
 894          */
 895         cfs_wait_event(iobuf->dr_wait,
 896                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 897         if (!rc)
 898                 rc = iobuf->dr_error;
 899
 900         RETURN(rc);
 901 }
 902
 903 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 904 {
 905         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 906                                                ot_super);
 907
 908         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 909
 910         return 0;
 911 }
 912
 913 /*
 914  * Called just before object is freed. Releases all resources except for
 915  * object itself (that is released by osd_object_free()).
 916  *
 917  * Concurrency: no concurrent access is possible that late in object
 918  * life-cycle.
 919  */
 920 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 921 {
 922         struct osd_object *obj   = osd_obj(l);
 923         struct inode      *inode = obj->oo_inode;
 924
 925         LINVRNT(osd_invariant(obj));
 926
 927         /*
 928          * If object is unlinked remove fid->ino mapping from object index.
 929          */
 930
 931         osd_index_fini(obj);
 932         if (inode != NULL) {
 933                 iput(inode);
 934                 obj->oo_inode = NULL;
 935         }
 936 }
 937
 938 /*
 939  * Concurrency: ->loo_object_release() is called under site spin-lock.
 940  */
 941 static void osd_object_release(const struct lu_env *env,
 942                                struct lu_object *l)
 943 {
 944 }
 945
 946 /*
 947  * Concurrency: shouldn't matter.
 948  */
 949 static int osd_object_print(const struct lu_env *env, void *cookie,
 950                             lu_printer_t p, const struct lu_object *l)
 951 {
 952         struct osd_object *o = osd_obj(l);
 953         struct iam_descr  *d;
 954
 955         if (o->oo_dir != NULL)
 956                 d = o->oo_dir->od_container.ic_descr;
 957         else
 958                 d = NULL;
 959         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 960                     o, o->oo_inode,
 961                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 962                     o->oo_inode ? o->oo_inode->i_generation : 0,
 963                     d ? d->id_ops->id_name : "plain");
 964 }
 965
 966 /*
 967  * Concurrency: shouldn't matter.
 968  */
 969 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 970                struct obd_statfs *sfs)
 971 {
 972         struct osd_device  *osd = osd_dt_dev(d);
 973         struct super_block *sb = osd_sb(osd);
 974         struct kstatfs     *ksfs;
 975         int result = 0;
 976
 977         /* osd_lproc.c call this without env, allocate ksfs for that case */
 978         if (unlikely(env == NULL)) {
 979                 OBD_ALLOC_PTR(ksfs);
 980                 if (ksfs == NULL)
 981                         return -ENOMEM;
 982         } else {
 983                 ksfs = &osd_oti_get(env)->oti_ksfs;
 984         }
 985
 986         cfs_spin_lock(&osd->od_osfs_lock);
 987         /* cache 1 second */
 988         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 989                 result = ll_do_statfs(sb, ksfs);
 990                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
 991                         osd->od_osfs_age = cfs_time_current_64();
 992                         statfs_pack(&osd->od_statfs, ksfs);
 993                 }
 994         }
 995
 996         if (likely(result == 0))
 997                 *sfs = osd->od_statfs;
 998         cfs_spin_unlock(&osd->od_osfs_lock);
 999
1000         if (unlikely(env == NULL))
1001                 OBD_FREE_PTR(ksfs);
1002
1003         return result;
1004 }
1005
1006 /*
1007  * Concurrency: doesn't access mutable data.
1008  */
1009 static void osd_conf_get(const struct lu_env *env,
1010                          const struct dt_device *dev,
1011                          struct dt_device_param *param)
1012 {
1013         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1014
1015         /*
1016          * XXX should be taken from not-yet-existing fs abstraction layer.
1017          */
1018         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1019         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1020         param->ddp_block_shift  = sb->s_blocksize_bits;
1021         param->ddp_mntopts      = 0;
1022         if (test_opt(sb, XATTR_USER))
1023                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1024         if (test_opt(sb, POSIX_ACL))
1025                 param->ddp_mntopts |= MNTOPT_ACL;
1026
1027 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1028         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1029                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1030         else
1031 #endif
1032                 param->ddp_max_ea_size = sb->s_blocksize;
1033
1034 }
1035
1036 /**
1037  * Helper function to get and fill the buffer with input values.
1038  */
1039 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1040 {
1041         struct lu_buf *buf;
1042
1043         buf = &osd_oti_get(env)->oti_buf;
1044         buf->lb_buf = area;
1045         buf->lb_len = len;
1046         return buf;
1047 }
1048
1049 /*
1050  * Concurrency: shouldn't matter.
1051  */
1052 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1053 {
1054         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
1055         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1056 }
1057
1058 /**
1059  * Start commit for OSD device.
1060  *
1061  * An implementation of dt_commit_async method for OSD device.
1062  * Asychronously starts underlayng fs sync and thereby a transaction
1063  * commit.
1064  *
1065  * \param env environment
1066  * \param d dt device
1067  *
1068  * \see dt_device_operations
1069  */
1070 static int osd_commit_async(const struct lu_env *env,
1071                             struct dt_device *d)
1072 {
1073         struct super_block *s = osd_sb(osd_dt_dev(d));
1074         ENTRY;
1075
1076         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
1077         RETURN(s->s_op->sync_fs(s, 0));
1078 }
1079
1080 /*
1081  * Concurrency: shouldn't matter.
1082  */
1083
1084 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1085 {
1086         struct super_block *sb = osd_sb(osd_dt_dev(d));
1087         int rc;
1088         ENTRY;
1089
1090         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
1091
1092         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1093         RETURN(rc);
1094 }
1095
1096 /*
1097  * Concurrency: serialization provided by callers.
1098  */
1099 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1100                               int mode, unsigned long timeout, __u32 alg,
1101                               struct lustre_capa_key *keys)
1102 {
1103         struct osd_device *dev = osd_dt_dev(d);
1104         ENTRY;
1105
1106         dev->od_fl_capa = mode;
1107         dev->od_capa_timeout = timeout;
1108         dev->od_capa_alg = alg;
1109         dev->od_capa_keys = keys;
1110         RETURN(0);
1111 }
1112
1113 /**
1114  * Concurrency: serialization provided by callers.
1115  */
1116 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1117                                struct dt_quota_ctxt *ctxt, void *data)
1118 {
1119         struct obd_device *obd = (void *)ctxt;
1120         struct vfsmount *mnt = (struct vfsmount *)data;
1121         ENTRY;
1122
1123         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1124         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1125         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1126         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1127         obd->obd_lvfs_ctxt.fs = get_ds();
1128
1129         EXIT;
1130 }
1131
1132 /**
1133  * Note: we do not count into QUOTA here.
1134  * If we mount with --data_journal we may need more.
1135  */
1136 const int osd_dto_credits_noquota[DTO_NR] = {
1137         /**
1138          * Insert/Delete.
1139          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1140          * SINGLEDATA_TRANS_BLOCKS(8)
1141          * XXX Note: maybe iam need more, since iam have more level than
1142          *           EXT3 htree.
1143          */
1144         [DTO_INDEX_INSERT]  = 16,
1145         [DTO_INDEX_DELETE]  = 16,
1146         /**
1147          * Used for OI scrub
1148          */
1149         [DTO_INDEX_UPDATE]  = 16,
1150         /**
1151          * Create a object. The same as create object in EXT3.
1152          * DATA_TRANS_BLOCKS(14) +
1153          * INDEX_EXTRA_BLOCKS(8) +
1154          * 3(inode bits, groups, GDT)
1155          */
1156         [DTO_OBJECT_CREATE] = 25,
1157         /**
1158          * XXX: real credits to be fixed
1159          */
1160         [DTO_OBJECT_DELETE] = 25,
1161         /**
1162          * Attr set credits (inode)
1163          */
1164         [DTO_ATTR_SET_BASE] = 1,
1165         /**
1166          * Xattr set. The same as xattr of EXT3.
1167          * DATA_TRANS_BLOCKS(14)
1168          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1169          * are also counted in. Do not know why?
1170          */
1171         [DTO_XATTR_SET]     = 14,
1172         [DTO_LOG_REC]       = 14,
1173         /**
1174          * credits for inode change during write.
1175          */
1176         [DTO_WRITE_BASE]    = 3,
1177         /**
1178          * credits for single block write.
1179          */
1180         [DTO_WRITE_BLOCK]   = 14,
1181         /**
1182          * Attr set credits for chown.
1183          * This is extra credits for setattr, and it is null without quota
1184          */
1185         [DTO_ATTR_SET_CHOWN]= 0
1186 };
1187
1188 static const struct dt_device_operations osd_dt_ops = {
1189         .dt_root_get       = osd_root_get,
1190         .dt_statfs         = osd_statfs,
1191         .dt_trans_create   = osd_trans_create,
1192         .dt_trans_start    = osd_trans_start,
1193         .dt_trans_stop     = osd_trans_stop,
1194         .dt_trans_cb_add   = osd_trans_cb_add,
1195         .dt_conf_get       = osd_conf_get,
1196         .dt_sync           = osd_sync,
1197         .dt_ro             = osd_ro,
1198         .dt_commit_async   = osd_commit_async,
1199         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1200         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1201 };
1202
1203 static void osd_object_read_lock(const struct lu_env *env,
1204                                  struct dt_object *dt, unsigned role)
1205 {
1206         struct osd_object *obj = osd_dt_obj(dt);
1207         struct osd_thread_info *oti = osd_oti_get(env);
1208
1209         LINVRNT(osd_invariant(obj));
1210
1211         LASSERT(obj->oo_owner != env);
1212         cfs_down_read_nested(&obj->oo_sem, role);
1213
1214         LASSERT(obj->oo_owner == NULL);
1215         oti->oti_r_locks++;
1216 }
1217
1218 static void osd_object_write_lock(const struct lu_env *env,
1219                                   struct dt_object *dt, unsigned role)
1220 {
1221         struct osd_object *obj = osd_dt_obj(dt);
1222         struct osd_thread_info *oti = osd_oti_get(env);
1223
1224         LINVRNT(osd_invariant(obj));
1225
1226         LASSERT(obj->oo_owner != env);
1227         cfs_down_write_nested(&obj->oo_sem, role);
1228
1229         LASSERT(obj->oo_owner == NULL);
1230         obj->oo_owner = env;
1231         oti->oti_w_locks++;
1232 }
1233
1234 static void osd_object_read_unlock(const struct lu_env *env,
1235                                    struct dt_object *dt)
1236 {
1237         struct osd_object *obj = osd_dt_obj(dt);
1238         struct osd_thread_info *oti = osd_oti_get(env);
1239
1240         LINVRNT(osd_invariant(obj));
1241
1242         LASSERT(oti->oti_r_locks > 0);
1243         oti->oti_r_locks--;
1244         cfs_up_read(&obj->oo_sem);
1245 }
1246
1247 static void osd_object_write_unlock(const struct lu_env *env,
1248                                     struct dt_object *dt)
1249 {
1250         struct osd_object *obj = osd_dt_obj(dt);
1251         struct osd_thread_info *oti = osd_oti_get(env);
1252
1253         LINVRNT(osd_invariant(obj));
1254
1255         LASSERT(obj->oo_owner == env);
1256         LASSERT(oti->oti_w_locks > 0);
1257         oti->oti_w_locks--;
1258         obj->oo_owner = NULL;
1259         cfs_up_write(&obj->oo_sem);
1260 }
1261
1262 static int osd_object_write_locked(const struct lu_env *env,
1263                                    struct dt_object *dt)
1264 {
1265         struct osd_object *obj = osd_dt_obj(dt);
1266
1267         LINVRNT(osd_invariant(obj));
1268
1269         return obj->oo_owner == env;
1270 }
1271
1272 static int capa_is_sane(const struct lu_env *env,
1273                         struct osd_device *dev,
1274                         struct lustre_capa *capa,
1275                         struct lustre_capa_key *keys)
1276 {
1277         struct osd_thread_info *oti = osd_oti_get(env);
1278         struct lustre_capa *tcapa = &oti->oti_capa;
1279         struct obd_capa *oc;
1280         int i, rc = 0;
1281         ENTRY;
1282
1283         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1284         if (oc) {
1285                 if (capa_is_expired(oc)) {
1286                         DEBUG_CAPA(D_ERROR, capa, "expired");
1287                         rc = -ESTALE;
1288                 }
1289                 capa_put(oc);
1290                 RETURN(rc);
1291         }
1292
1293         if (capa_is_expired_sec(capa)) {
1294                 DEBUG_CAPA(D_ERROR, capa, "expired");
1295                 RETURN(-ESTALE);
1296         }
1297
1298         cfs_spin_lock(&capa_lock);
1299         for (i = 0; i < 2; i++) {
1300                 if (keys[i].lk_keyid == capa->lc_keyid) {
1301                         oti->oti_capa_key = keys[i];
1302                         break;
1303                 }
1304         }
1305         cfs_spin_unlock(&capa_lock);
1306
1307         if (i == 2) {
1308                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1309                 RETURN(-ESTALE);
1310         }
1311
1312         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1313         if (rc)
1314                 RETURN(rc);
1315
1316         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1317                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1318                 RETURN(-EACCES);
1319         }
1320
1321         oc = capa_add(dev->od_capa_hash, capa);
1322         capa_put(oc);
1323
1324         RETURN(0);
1325 }
1326
1327 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1328                     struct lustre_capa *capa, __u64 opc)
1329 {
1330         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1331         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1332         struct md_capainfo *ci;
1333         int rc;
1334
1335         if (!dev->od_fl_capa)
1336                 return 0;
1337
1338         if (capa == BYPASS_CAPA)
1339                 return 0;
1340
1341         ci = md_capainfo(env);
1342         if (unlikely(!ci))
1343                 return 0;
1344
1345         if (ci->mc_auth == LC_ID_NONE)
1346                 return 0;
1347
1348         if (!capa) {
1349                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1350                 return -EACCES;
1351         }
1352
1353         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1354                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1355                            PFID(fid));
1356                 return -EACCES;
1357         }
1358
1359         if (!capa_opc_supported(capa, opc)) {
1360                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1361                 return -EACCES;
1362         }
1363
1364         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1365                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1366                 return -EACCES;
1367         }
1368
1369         return 0;
1370 }
1371
1372 static struct timespec *osd_inode_time(const struct lu_env *env,
1373                                        struct inode *inode, __u64 seconds)
1374 {
1375         struct osd_thread_info *oti = osd_oti_get(env);
1376         struct timespec        *t   = &oti->oti_time;
1377
1378         t->tv_sec  = seconds;
1379         t->tv_nsec = 0;
1380         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1381         return t;
1382 }
1383
1384
1385 static void osd_inode_getattr(const struct lu_env *env,
1386                               struct inode *inode, struct lu_attr *attr)
1387 {
1388         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1389                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1390                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1391
1392         attr->la_atime      = LTIME_S(inode->i_atime);
1393         attr->la_mtime      = LTIME_S(inode->i_mtime);
1394         attr->la_ctime      = LTIME_S(inode->i_ctime);
1395         attr->la_mode       = inode->i_mode;
1396         attr->la_size       = i_size_read(inode);
1397         attr->la_blocks     = inode->i_blocks;
1398         attr->la_uid        = inode->i_uid;
1399         attr->la_gid        = inode->i_gid;
1400         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1401         attr->la_nlink      = inode->i_nlink;
1402         attr->la_rdev       = inode->i_rdev;
1403         attr->la_blksize    = ll_inode_blksize(inode);
1404         attr->la_blkbits    = inode->i_blkbits;
1405 }
1406
1407 static int osd_attr_get(const struct lu_env *env,
1408                         struct dt_object *dt,
1409                         struct lu_attr *attr,
1410                         struct lustre_capa *capa)
1411 {
1412         struct osd_object *obj = osd_dt_obj(dt);
1413
1414         LASSERT(dt_object_exists(dt));
1415         LINVRNT(osd_invariant(obj));
1416
1417         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1418                 return -EACCES;
1419
1420         cfs_spin_lock(&obj->oo_guard);
1421         osd_inode_getattr(env, obj->oo_inode, attr);
1422         cfs_spin_unlock(&obj->oo_guard);
1423         return 0;
1424 }
1425
1426 static int osd_declare_attr_set(const struct lu_env *env,
1427                                 struct dt_object *dt,
1428                                 const struct lu_attr *attr,
1429                                 struct thandle *handle)
1430 {
1431         struct osd_thandle *oh;
1432         struct osd_object *obj;
1433
1434         LASSERT(dt != NULL);
1435         LASSERT(handle != NULL);
1436
1437         obj = osd_dt_obj(dt);
1438         LASSERT(osd_invariant(obj));
1439
1440         oh = container_of0(handle, struct osd_thandle, ot_super);
1441         LASSERT(oh->ot_handle == NULL);
1442
1443         OSD_DECLARE_OP(oh, attr_set);
1444         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1445
1446         if (attr && attr->la_valid & LA_UID) {
1447                 if (obj->oo_inode)
1448                         osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
1449                                         obj->oo_inode);
1450                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1451         }
1452         if (attr && attr->la_valid & LA_GID) {
1453                 if (obj->oo_inode)
1454                         osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
1455                                         obj->oo_inode);
1456                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1457         }
1458
1459         return 0;
1460 }
1461
1462 static int osd_inode_setattr(const struct lu_env *env,
1463                              struct inode *inode, const struct lu_attr *attr)
1464 {
1465         __u64 bits;
1466
1467         bits = attr->la_valid;
1468
1469         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1470
1471         if (bits & LA_ATIME)
1472                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1473         if (bits & LA_CTIME)
1474                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1475         if (bits & LA_MTIME)
1476                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1477         if (bits & LA_SIZE) {
1478                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1479                 i_size_write(inode, attr->la_size);
1480         }
1481
1482 #if 0
1483         /* OSD should not change "i_blocks" which is used by quota.
1484          * "i_blocks" should be changed by ldiskfs only. */
1485         if (bits & LA_BLOCKS)
1486                 inode->i_blocks = attr->la_blocks;
1487 #endif
1488         if (bits & LA_MODE)
1489                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1490                         (attr->la_mode & ~S_IFMT);
1491         if (bits & LA_UID)
1492                 inode->i_uid    = attr->la_uid;
1493         if (bits & LA_GID)
1494                 inode->i_gid    = attr->la_gid;
1495         if (bits & LA_NLINK)
1496                 inode->i_nlink  = attr->la_nlink;
1497         if (bits & LA_RDEV)
1498                 inode->i_rdev   = attr->la_rdev;
1499
1500         if (bits & LA_FLAGS) {
1501                 /* always keep S_NOCMTIME */
1502                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1503                                  S_NOCMTIME;
1504         }
1505         return 0;
1506 }
1507
1508 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1509 {
1510         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1511             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1512                 struct iattr    iattr;
1513                 int             rc;
1514
1515                 iattr.ia_valid = 0;
1516                 if (attr->la_valid & LA_UID)
1517                         iattr.ia_valid |= ATTR_UID;
1518                 if (attr->la_valid & LA_GID)
1519                         iattr.ia_valid |= ATTR_GID;
1520                 iattr.ia_uid = attr->la_uid;
1521                 iattr.ia_gid = attr->la_gid;
1522
1523                 rc = ll_vfs_dq_transfer(inode, &iattr);
1524                 if (rc) {
1525                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1526                                "enforcement enabled on the ldiskfs filesystem?",
1527                                inode->i_sb->s_id, rc);
1528                         return rc;
1529                 }
1530         }
1531         return 0;
1532 }
1533
1534 static int osd_attr_set(const struct lu_env *env,
1535                         struct dt_object *dt,
1536                         const struct lu_attr *attr,
1537                         struct thandle *handle,
1538                         struct lustre_capa *capa)
1539 {
1540         struct osd_object *obj = osd_dt_obj(dt);
1541         struct inode      *inode;
1542         int rc;
1543
1544         LASSERT(handle != NULL);
1545         LASSERT(dt_object_exists(dt));
1546         LASSERT(osd_invariant(obj));
1547
1548         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1549                 return -EACCES;
1550
1551         OSD_EXEC_OP(handle, attr_set);
1552
1553         inode = obj->oo_inode;
1554         if (LDISKFS_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1555                                           LDISKFS_FEATURE_RO_COMPAT_QUOTA)) {
1556                 rc = osd_quota_transfer(inode, attr);
1557                 if (rc)
1558                         return rc;
1559         } else {
1560 #ifdef HAVE_QUOTA_SUPPORT
1561                 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1562                     (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1563                         struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1564                         struct           iattr iattr;
1565                         int              rc;
1566
1567                         iattr.ia_valid = 0;
1568                         if (attr->la_valid & LA_UID)
1569                                 iattr.ia_valid |= ATTR_UID;
1570                         if (attr->la_valid & LA_GID)
1571                                 iattr.ia_valid |= ATTR_GID;
1572                         iattr.ia_uid = attr->la_uid;
1573                         iattr.ia_gid = attr->la_gid;
1574                         osd_push_ctxt(env, save);
1575                         rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1576                         osd_pop_ctxt(save);
1577                         if (rc != 0)
1578                                 return rc;
1579                 }
1580 #endif
1581         }
1582         cfs_spin_lock(&obj->oo_guard);
1583         rc = osd_inode_setattr(env, inode, attr);
1584         cfs_spin_unlock(&obj->oo_guard);
1585
1586         if (!rc)
1587                 inode->i_sb->s_op->dirty_inode(inode);
1588         return rc;
1589 }
1590
1591 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1592                                     struct osd_object *obj,
1593                                     const char *name, const int namelen)
1594 {
1595         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1596 }
1597
1598 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1599                       cfs_umode_t mode,
1600                       struct dt_allocation_hint *hint,
1601                       struct thandle *th)
1602 {
1603         int result;
1604         struct osd_device  *osd = osd_obj2dev(obj);
1605         struct osd_thandle *oth;
1606         struct dt_object   *parent = NULL;
1607         struct inode       *inode;
1608 #ifdef HAVE_QUOTA_SUPPORT
1609         struct osd_ctxt    *save = &info->oti_ctxt;
1610 #endif
1611
1612         LINVRNT(osd_invariant(obj));
1613         LASSERT(obj->oo_inode == NULL);
1614         LASSERT(obj->oo_hl_head == NULL);
1615
1616         if (S_ISDIR(mode) && ldiskfs_pdo) {
1617                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1618                 if (obj->oo_hl_head == NULL)
1619                         return -ENOMEM;
1620         }
1621
1622         oth = container_of(th, struct osd_thandle, ot_super);
1623         LASSERT(oth->ot_handle->h_transaction != NULL);
1624
1625         if (hint && hint->dah_parent)
1626                 parent = hint->dah_parent;
1627
1628 #ifdef HAVE_QUOTA_SUPPORT
1629         osd_push_ctxt(info->oti_env, save);
1630 #endif
1631         inode = ldiskfs_create_inode(oth->ot_handle,
1632                                      parent ? osd_dt_obj(parent)->oo_inode :
1633                                               osd_sb(osd)->s_root->d_inode,
1634                                      mode);
1635 #ifdef HAVE_QUOTA_SUPPORT
1636         osd_pop_ctxt(save);
1637 #endif
1638         if (!IS_ERR(inode)) {
1639                 /* Do not update file c/mtime in ldiskfs.
1640                  * NB: don't need any lock because no contention at this
1641                  * early stage */
1642                 inode->i_flags |= S_NOCMTIME;
1643                 inode->i_state |= I_LUSTRE_NOSCRUB;
1644                 obj->oo_inode = inode;
1645                 result = 0;
1646         } else {
1647                 if (obj->oo_hl_head != NULL) {
1648                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1649                         obj->oo_hl_head = NULL;
1650                 }
1651                 result = PTR_ERR(inode);
1652         }
1653         LINVRNT(osd_invariant(obj));
1654         return result;
1655 }
1656
1657 enum {
1658         OSD_NAME_LEN = 255
1659 };
1660
1661 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1662                      struct lu_attr *attr,
1663                      struct dt_allocation_hint *hint,
1664                      struct dt_object_format *dof,
1665                      struct thandle *th)
1666 {
1667         int result;
1668         struct osd_thandle *oth;
1669         struct osd_device *osd = osd_obj2dev(obj);
1670         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1671
1672         LASSERT(S_ISDIR(attr->la_mode));
1673
1674         oth = container_of(th, struct osd_thandle, ot_super);
1675         LASSERT(oth->ot_handle->h_transaction != NULL);
1676         result = osd_mkfile(info, obj, mode, hint, th);
1677         if (result == 0 && osd->od_iop_mode == 0) {
1678                 LASSERT(obj->oo_inode != NULL);
1679                 /*
1680                  * XXX uh-oh... call low-level iam function directly.
1681                  */
1682
1683                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1684                                          sizeof (struct osd_fid_pack),
1685                                          oth->ot_handle);
1686         }
1687         return result;
1688 }
1689
1690 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1691                         struct lu_attr *attr,
1692                         struct dt_allocation_hint *hint,
1693                         struct dt_object_format *dof,
1694                         struct thandle *th)
1695 {
1696         int result;
1697         struct osd_thandle *oth;
1698         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1699
1700         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1701
1702         LASSERT(S_ISREG(attr->la_mode));
1703
1704         oth = container_of(th, struct osd_thandle, ot_super);
1705         LASSERT(oth->ot_handle->h_transaction != NULL);
1706
1707         result = osd_mkfile(info, obj, mode, hint, th);
1708         if (result == 0) {
1709                 LASSERT(obj->oo_inode != NULL);
1710                 if (feat->dif_flags & DT_IND_VARKEY)
1711                         result = iam_lvar_create(obj->oo_inode,
1712                                                  feat->dif_keysize_max,
1713                                                  feat->dif_ptrsize,
1714                                                  feat->dif_recsize_max,
1715                                                  oth->ot_handle);
1716                 else
1717                         result = iam_lfix_create(obj->oo_inode,
1718                                                  feat->dif_keysize_max,
1719                                                  feat->dif_ptrsize,
1720                                                  feat->dif_recsize_max,
1721                                                  oth->ot_handle);
1722
1723         }
1724         return result;
1725 }
1726
1727 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1728                      struct lu_attr *attr,
1729                      struct dt_allocation_hint *hint,
1730                      struct dt_object_format *dof,
1731                      struct thandle *th)
1732 {
1733         LASSERT(S_ISREG(attr->la_mode));
1734         return osd_mkfile(info, obj, (attr->la_mode &
1735                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1736 }
1737
1738 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1739                      struct lu_attr *attr,
1740                      struct dt_allocation_hint *hint,
1741                      struct dt_object_format *dof,
1742                      struct thandle *th)
1743 {
1744         LASSERT(S_ISLNK(attr->la_mode));
1745         return osd_mkfile(info, obj, (attr->la_mode &
1746                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1747 }
1748
1749 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1750                      struct lu_attr *attr,
1751                      struct dt_allocation_hint *hint,
1752                      struct dt_object_format *dof,
1753                      struct thandle *th)
1754 {
1755         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1756         int result;
1757
1758         LINVRNT(osd_invariant(obj));
1759         LASSERT(obj->oo_inode == NULL);
1760         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1761                 S_ISFIFO(mode) || S_ISSOCK(mode));
1762
1763         result = osd_mkfile(info, obj, mode, hint, th);
1764         if (result == 0) {
1765                 LASSERT(obj->oo_inode != NULL);
1766                 /*
1767                  * This inode should be marked dirty for i_rdev.  Currently
1768                  * that is done in the osd_attr_init().
1769                  */
1770                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1771         }
1772         LINVRNT(osd_invariant(obj));
1773         return result;
1774 }
1775
1776 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1777                               struct lu_attr *,
1778                               struct dt_allocation_hint *hint,
1779                               struct dt_object_format *dof,
1780                               struct thandle *);
1781
1782 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1783 {
1784         osd_obj_type_f result;
1785
1786         switch (type) {
1787         case DFT_DIR:
1788                 result = osd_mkdir;
1789                 break;
1790         case DFT_REGULAR:
1791                 result = osd_mkreg;
1792                 break;
1793         case DFT_SYM:
1794                 result = osd_mksym;
1795                 break;
1796         case DFT_NODE:
1797                 result = osd_mknod;
1798                 break;
1799         case DFT_INDEX:
1800                 result = osd_mk_index;
1801                 break;
1802
1803         default:
1804                 LBUG();
1805                 break;
1806         }
1807         return result;
1808 }
1809
1810
1811 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1812                         struct dt_object *parent, cfs_umode_t child_mode)
1813 {
1814         LASSERT(ah);
1815
1816         memset(ah, 0, sizeof(*ah));
1817         ah->dah_parent = parent;
1818         ah->dah_mode = child_mode;
1819 }
1820
1821 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1822                           struct lu_attr *attr, struct dt_object_format *dof)
1823 {
1824         struct inode   *inode = obj->oo_inode;
1825         __u64           valid = attr->la_valid;
1826         int             result;
1827
1828         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1829
1830         if (dof->dof_type != DFT_NODE)
1831                 attr->la_valid &= ~LA_RDEV;
1832         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1833                 attr->la_valid &= ~LA_ATIME;
1834         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1835                 attr->la_valid &= ~LA_CTIME;
1836         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1837                 attr->la_valid &= ~LA_MTIME;
1838
1839         if (LDISKFS_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1840                                           LDISKFS_FEATURE_RO_COMPAT_QUOTA)) {
1841                 result = osd_quota_transfer(inode, attr);
1842                 if (result)
1843                         return;
1844         } else {
1845 #ifdef HAVE_QUOTA_SUPPORT
1846                 attr->la_valid &= ~(LA_UID | LA_GID);
1847 #endif
1848         }
1849
1850         if (attr->la_valid != 0) {
1851                 result = osd_inode_setattr(info->oti_env, inode, attr);
1852                 /*
1853                  * The osd_inode_setattr() should always succeed here.  The
1854                  * only error that could be returned is EDQUOT when we are
1855                  * trying to change the UID or GID of the inode. However, this
1856                  * should not happen since quota enforcement is no longer
1857                  * enabled on ldiskfs (lquota takes care of it).
1858                  */
1859                 LASSERTF(result == 0, "%d", result);
1860                 inode->i_sb->s_op->dirty_inode(inode);
1861         }
1862
1863         attr->la_valid = valid;
1864 }
1865
1866 /**
1867  * Helper function for osd_object_create()
1868  *
1869  * \retval 0, on success
1870  */
1871 static int __osd_object_create(struct osd_thread_info *info,
1872                                struct osd_object *obj, struct lu_attr *attr,
1873                                struct dt_allocation_hint *hint,
1874                                struct dt_object_format *dof,
1875                                struct thandle *th)
1876 {
1877         int     result;
1878         __u32   umask;
1879
1880         /* we drop umask so that permissions we pass are not affected */
1881         umask = current->fs->umask;
1882         current->fs->umask = 0;
1883
1884         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1885                                                   th);
1886         if (result == 0) {
1887                 osd_attr_init(info, obj, attr, dof);
1888                 osd_object_init0(obj);
1889                 /* bz 24037 */
1890                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1891                         unlock_new_inode(obj->oo_inode);
1892         }
1893
1894         /* restore previous umask value */
1895         current->fs->umask = umask;
1896
1897         return result;
1898 }
1899
1900 /**
1901  * Helper function for osd_object_create()
1902  *
1903  * \retval 0, on success
1904  */
1905 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1906                            const struct lu_fid *fid, struct thandle *th)
1907 {
1908         struct osd_thread_info *info = osd_oti_get(env);
1909         struct osd_inode_id    *id   = &info->oti_id;
1910         struct osd_device      *osd  = osd_obj2dev(obj);
1911         struct md_ucred        *uc   = md_ucred(env);
1912
1913         LASSERT(obj->oo_inode != NULL);
1914         LASSERT(uc != NULL);
1915
1916         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1917         return osd_oi_insert(info, osd, fid, id, th);
1918 }
1919
1920 static int osd_declare_object_create(const struct lu_env *env,
1921                                      struct dt_object *dt,
1922                                      struct lu_attr *attr,
1923                                      struct dt_allocation_hint *hint,
1924                                      struct dt_object_format *dof,
1925                                      struct thandle *handle)
1926 {
1927         struct osd_thandle *oh;
1928
1929         LASSERT(handle != NULL);
1930
1931         oh = container_of0(handle, struct osd_thandle, ot_super);
1932         LASSERT(oh->ot_handle == NULL);
1933
1934         OSD_DECLARE_OP(oh, create);
1935         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1936         /* XXX: So far, only normal fid needs be inserted into the oi,
1937          *      things could be changed later. Revise following code then. */
1938         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1939                 OSD_DECLARE_OP(oh, insert);
1940                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1941         }
1942         /* If this is directory, then we expect . and .. to be inserted as
1943          * well. The one directory block always needs to be created for the
1944          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1945          * block), there is no danger of needing a tree for the first block.
1946          */
1947         if (attr && S_ISDIR(attr->la_mode)) {
1948                 OSD_DECLARE_OP(oh, insert);
1949                 OSD_DECLARE_OP(oh, insert);
1950                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1951         }
1952
1953         if (attr) {
1954                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1955                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1956         }
1957         return 0;
1958 }
1959
1960 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1961                              struct lu_attr *attr,
1962                              struct dt_allocation_hint *hint,
1963                              struct dt_object_format *dof,
1964                              struct thandle *th)
1965 {
1966         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1967         struct osd_object      *obj    = osd_dt_obj(dt);
1968         struct osd_thread_info *info   = osd_oti_get(env);
1969         int result;
1970
1971         ENTRY;
1972
1973         LINVRNT(osd_invariant(obj));
1974         LASSERT(!dt_object_exists(dt));
1975         LASSERT(osd_write_locked(env, obj));
1976         LASSERT(th != NULL);
1977
1978         if (unlikely(fid_is_acct(fid)))
1979                 /* Quota files can't be created from the kernel any more,
1980                  * 'tune2fs -O quota' will take care of creating them */
1981                 RETURN(-EPERM);
1982
1983         OSD_EXEC_OP(th, create);
1984
1985         result = __osd_object_create(info, obj, attr, hint, dof, th);
1986         if (result == 0)
1987                 result = __osd_oi_insert(env, obj, fid, th);
1988
1989         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1990         LASSERT(osd_invariant(obj));
1991         RETURN(result);
1992 }
1993
1994 /**
1995  * Called to destroy on-disk representation of the object
1996  *
1997  * Concurrency: must be locked
1998  */
1999 static int osd_declare_object_destroy(const struct lu_env *env,
2000                                       struct dt_object *dt,
2001                                       struct thandle *th)
2002 {
2003         struct osd_object  *obj = osd_dt_obj(dt);
2004         struct inode       *inode = obj->oo_inode;
2005         struct osd_thandle *oh;
2006
2007         ENTRY;
2008
2009         oh = container_of0(th, struct osd_thandle, ot_super);
2010         LASSERT(oh->ot_handle == NULL);
2011         LASSERT(inode);
2012
2013         OSD_DECLARE_OP(oh, destroy);
2014         OSD_DECLARE_OP(oh, delete);
2015         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
2016         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2017
2018         osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
2019         osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
2020
2021         RETURN(0);
2022 }
2023
2024 static int osd_object_destroy(const struct lu_env *env,
2025                               struct dt_object *dt,
2026                               struct thandle *th)
2027 {
2028         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2029         struct osd_object      *obj = osd_dt_obj(dt);
2030         struct inode           *inode = obj->oo_inode;
2031         struct osd_device      *osd = osd_obj2dev(obj);
2032         struct osd_thandle     *oh;
2033         int                     result;
2034         ENTRY;
2035
2036         oh = container_of0(th, struct osd_thandle, ot_super);
2037         LASSERT(oh->ot_handle);
2038         LASSERT(inode);
2039         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2040
2041         if (unlikely(fid_is_acct(fid)))
2042                 RETURN(-EPERM);
2043
2044         /* Parallel control for OI scrub. For most of cases, there is no
2045          * lock contention. So it will not affect unlink performance. */
2046         cfs_mutex_lock(&inode->i_mutex);
2047         if (S_ISDIR(inode->i_mode)) {
2048                 LASSERT(osd_inode_unlinked(inode) ||
2049                         inode->i_nlink == 1);
2050                 cfs_spin_lock(&obj->oo_guard);
2051                 inode->i_nlink = 0;
2052                 cfs_spin_unlock(&obj->oo_guard);
2053                 inode->i_sb->s_op->dirty_inode(inode);
2054         } else {
2055                 LASSERT(osd_inode_unlinked(inode));
2056         }
2057
2058         OSD_EXEC_OP(th, destroy);
2059
2060         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2061         cfs_mutex_unlock(&inode->i_mutex);
2062
2063         /* XXX: add to ext3 orphan list */
2064         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2065
2066         /* not needed in the cache anymore */
2067         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2068
2069         RETURN(0);
2070 }
2071
2072 /**
2073  * Helper function for osd_xattr_set()
2074  */
2075 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2076                            const struct lu_buf *buf, const char *name, int fl)
2077 {
2078         struct osd_object      *obj      = osd_dt_obj(dt);
2079         struct inode           *inode    = obj->oo_inode;
2080         struct osd_thread_info *info     = osd_oti_get(env);
2081         struct dentry          *dentry   = &info->oti_child_dentry;
2082         int                     fs_flags = 0;
2083         int                     rc;
2084
2085         LASSERT(dt_object_exists(dt));
2086         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2087
2088         if (fl & LU_XATTR_REPLACE)
2089                 fs_flags |= XATTR_REPLACE;
2090
2091         if (fl & LU_XATTR_CREATE)
2092                 fs_flags |= XATTR_CREATE;
2093
2094         dentry->d_inode = inode;
2095         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2096                                    buf->lb_len, fs_flags);
2097         return rc;
2098 }
2099
2100 /**
2101  * Put the fid into lustre_mdt_attrs, and then place the structure
2102  * inode's ea. This fid should not be altered during the life time
2103  * of the inode.
2104  *
2105  * \retval +ve, on success
2106  * \retval -ve, on error
2107  *
2108  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2109  */
2110 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2111                           const struct lu_fid *fid)
2112 {
2113         struct osd_thread_info  *info      = osd_oti_get(env);
2114         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2115
2116         lustre_lma_init(mdt_attrs, fid);
2117         lustre_lma_swab(mdt_attrs);
2118         return __osd_xattr_set(env, dt,
2119                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2120                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2121
2122 }
2123
2124 /**
2125  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2126  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2127  * To have compatilibility with 1.8 ldiskfs driver we need to have
2128  * magic number at start of fid data.
2129  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2130  * its inmemory API.
2131  */
2132 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2133                                   const struct dt_rec *fid)
2134 {
2135         param->edp_magic = LDISKFS_LUFID_MAGIC;
2136         param->edp_len =  sizeof(struct lu_fid) + 1;
2137
2138         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2139                       (struct lu_fid *)fid);
2140 }
2141
2142 /**
2143  * Try to read the fid from inode ea into dt_rec, if return value
2144  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2145  *
2146  * \param fid object fid.
2147  *
2148  * \retval 0 on success
2149  */
2150 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2151                           __u32 ino, struct lu_fid *fid,
2152                           struct osd_inode_id *id)
2153 {
2154         struct osd_thread_info *info  = osd_oti_get(env);
2155         struct inode           *inode;
2156         ENTRY;
2157
2158         osd_id_gen(id, ino, OSD_OII_NOGEN);
2159         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2160         if (IS_ERR(inode))
2161                 RETURN(PTR_ERR(inode));
2162
2163         iput(inode);
2164         RETURN(0);
2165 }
2166
2167 /**
2168  * OSD layer object create function for interoperability mode (b11826).
2169  * This is mostly similar to osd_object_create(). Only difference being, fid is
2170  * inserted into inode ea here.
2171  *
2172  * \retval   0, on success
2173  * \retval -ve, on error
2174  */
2175 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2176                                 struct lu_attr *attr,
2177                                 struct dt_allocation_hint *hint,
2178                                 struct dt_object_format *dof,
2179                                 struct thandle *th)
2180 {
2181         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2182         struct osd_object      *obj    = osd_dt_obj(dt);
2183         struct osd_thread_info *info   = osd_oti_get(env);
2184         int                     result;
2185
2186         ENTRY;
2187
2188         LASSERT(osd_invariant(obj));
2189         LASSERT(!dt_object_exists(dt));
2190         LASSERT(osd_write_locked(env, obj));
2191         LASSERT(th != NULL);
2192
2193         if (unlikely(fid_is_acct(fid)))
2194                 /* Quota files can't be created from the kernel any more,
2195                  * 'tune2fs -O quota' will take care of creating them */
2196                 RETURN(-EPERM);
2197
2198         OSD_EXEC_OP(th, create);
2199
2200         result = __osd_object_create(info, obj, attr, hint, dof, th);
2201         /* objects under osd root shld have igif fid, so dont add fid EA */
2202         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2203                 result = osd_ea_fid_set(env, dt, fid);
2204
2205         if (result == 0)
2206                 result = __osd_oi_insert(env, obj, fid, th);
2207
2208         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2209         LINVRNT(osd_invariant(obj));
2210         RETURN(result);
2211 }
2212
2213 static int osd_declare_object_ref_add(const struct lu_env *env,
2214                                       struct dt_object *dt,
2215                                       struct thandle *handle)
2216 {
2217         struct osd_thandle *oh;
2218
2219         /* it's possible that object doesn't exist yet */
2220         LASSERT(handle != NULL);
2221
2222         oh = container_of0(handle, struct osd_thandle, ot_super);
2223         LASSERT(oh->ot_handle == NULL);
2224
2225         OSD_DECLARE_OP(oh, ref_add);
2226         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2227
2228         return 0;
2229 }
2230
2231 /*
2232  * Concurrency: @dt is write locked.
2233  */
2234 static int osd_object_ref_add(const struct lu_env *env,
2235                               struct dt_object *dt, struct thandle *th)
2236 {
2237         struct osd_object *obj = osd_dt_obj(dt);
2238         struct inode      *inode = obj->oo_inode;
2239
2240         LINVRNT(osd_invariant(obj));
2241         LASSERT(dt_object_exists(dt));
2242         LASSERT(osd_write_locked(env, obj));
2243         LASSERT(th != NULL);
2244
2245         OSD_EXEC_OP(th, ref_add);
2246
2247         /*
2248          * DIR_NLINK feature is set for compatibility reasons if:
2249          * 1) nlinks > LDISKFS_LINK_MAX, or
2250          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2251          *
2252          * It is easier to always set this flag (rather than check and set),
2253          * since it has less overhead, and the superblock will be dirtied
2254          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2255          * do not actually care whether this flag is set or not.
2256          */
2257         cfs_spin_lock(&obj->oo_guard);
2258         inode->i_nlink++;
2259         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2260                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2261                     inode->i_nlink == 2)
2262                         inode->i_nlink = 1;
2263         }
2264         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2265         cfs_spin_unlock(&obj->oo_guard);
2266         inode->i_sb->s_op->dirty_inode(inode);
2267         LINVRNT(osd_invariant(obj));
2268
2269         return 0;
2270 }
2271
2272 static int osd_declare_object_ref_del(const struct lu_env *env,
2273                                       struct dt_object *dt,
2274                                       struct thandle *handle)
2275 {
2276         struct osd_thandle *oh;
2277
2278         LASSERT(dt_object_exists(dt));
2279         LASSERT(handle != NULL);
2280
2281         oh = container_of0(handle, struct osd_thandle, ot_super);
2282         LASSERT(oh->ot_handle == NULL);
2283
2284         OSD_DECLARE_OP(oh, ref_del);
2285         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2286
2287         return 0;
2288 }
2289
2290 /*
2291  * Concurrency: @dt is write locked.
2292  */
2293 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2294                               struct thandle *th)
2295 {
2296         struct osd_object *obj = osd_dt_obj(dt);
2297         struct inode      *inode = obj->oo_inode;
2298
2299         LINVRNT(osd_invariant(obj));
2300         LASSERT(dt_object_exists(dt));
2301         LASSERT(osd_write_locked(env, obj));
2302         LASSERT(th != NULL);
2303
2304         OSD_EXEC_OP(th, ref_del);
2305
2306         cfs_spin_lock(&obj->oo_guard);
2307         LASSERT(inode->i_nlink > 0);
2308         inode->i_nlink--;
2309         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2310          * then the nlink count is 1. Don't let it be set to 0 or the directory
2311          * inode will be deleted incorrectly. */
2312         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2313                 inode->i_nlink++;
2314         cfs_spin_unlock(&obj->oo_guard);
2315         inode->i_sb->s_op->dirty_inode(inode);
2316         LINVRNT(osd_invariant(obj));
2317
2318         return 0;
2319 }
2320
2321 /*
2322  * Get the 64-bit version for an inode.
2323  */
2324 static int osd_object_version_get(const struct lu_env *env,
2325                                   struct dt_object *dt, dt_obj_version_t *ver)
2326 {
2327         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2328
2329         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2330                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2331         *ver = LDISKFS_I(inode)->i_fs_version;
2332         return 0;
2333 }
2334
2335 /*
2336  * Concurrency: @dt is read locked.
2337  */
2338 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2339                          struct lu_buf *buf, const char *name,
2340                          struct lustre_capa *capa)
2341 {
2342         struct osd_object      *obj    = osd_dt_obj(dt);
2343         struct inode           *inode  = obj->oo_inode;
2344         struct osd_thread_info *info   = osd_oti_get(env);
2345         struct dentry          *dentry = &info->oti_obj_dentry;
2346
2347         /* version get is not real XATTR but uses xattr API */
2348         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2349                 /* for version we are just using xattr API but change inode
2350                  * field instead */
2351                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2352                 osd_object_version_get(env, dt, buf->lb_buf);
2353                 return sizeof(dt_obj_version_t);
2354         }
2355
2356         LASSERT(dt_object_exists(dt));
2357         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2358         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2359
2360         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2361                 return -EACCES;
2362
2363         dentry->d_inode = inode;
2364         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2365 }
2366
2367
2368 static int osd_declare_xattr_set(const struct lu_env *env,
2369                                  struct dt_object *dt,
2370                                  const struct lu_buf *buf, const char *name,
2371                                  int fl, struct thandle *handle)
2372 {
2373         struct osd_thandle *oh;
2374
2375         LASSERT(handle != NULL);
2376
2377         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2378                 /* no credits for version */
2379                 return 0;
2380         }
2381
2382         oh = container_of0(handle, struct osd_thandle, ot_super);
2383         LASSERT(oh->ot_handle == NULL);
2384
2385         OSD_DECLARE_OP(oh, xattr_set);
2386         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2387
2388         return 0;
2389 }
2390
2391 /*
2392  * Set the 64-bit version for object
2393  */
2394 static void osd_object_version_set(const struct lu_env *env,
2395                                    struct dt_object *dt,
2396                                    dt_obj_version_t *new_version)
2397 {
2398         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2399
2400         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2401                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2402
2403         LDISKFS_I(inode)->i_fs_version = *new_version;
2404         /** Version is set after all inode operations are finished,
2405          *  so we should mark it dirty here */
2406         inode->i_sb->s_op->dirty_inode(inode);
2407 }
2408
2409 /*
2410  * Concurrency: @dt is write locked.
2411  */
2412 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2413                          const struct lu_buf *buf, const char *name, int fl,
2414                          struct thandle *handle, struct lustre_capa *capa)
2415 {
2416         LASSERT(handle != NULL);
2417
2418         /* version set is not real XATTR */
2419         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2420                 /* for version we are just using xattr API but change inode
2421                  * field instead */
2422                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2423                 osd_object_version_set(env, dt, buf->lb_buf);
2424                 return sizeof(dt_obj_version_t);
2425         }
2426
2427         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2428                 return -EACCES;
2429
2430         OSD_EXEC_OP(handle, xattr_set);
2431         return __osd_xattr_set(env, dt, buf, name, fl);
2432 }
2433
2434 /*
2435  * Concurrency: @dt is read locked.
2436  */
2437 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2438                           struct lu_buf *buf, struct lustre_capa *capa)
2439 {
2440         struct osd_object      *obj    = osd_dt_obj(dt);
2441         struct inode           *inode  = obj->oo_inode;
2442         struct osd_thread_info *info   = osd_oti_get(env);
2443         struct dentry          *dentry = &info->oti_obj_dentry;
2444
2445         LASSERT(dt_object_exists(dt));
2446         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2447         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2448
2449         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2450                 return -EACCES;
2451
2452         dentry->d_inode = inode;
2453         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2454 }
2455
2456 static int osd_declare_xattr_del(const struct lu_env *env,
2457                                  struct dt_object *dt, const char *name,
2458                                  struct thandle *handle)
2459 {
2460         struct osd_thandle *oh;
2461
2462         LASSERT(dt_object_exists(dt));
2463         LASSERT(handle != NULL);
2464
2465         oh = container_of0(handle, struct osd_thandle, ot_super);
2466         LASSERT(oh->ot_handle == NULL);
2467
2468         OSD_DECLARE_OP(oh, xattr_set);
2469         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2470
2471         return 0;
2472 }
2473
2474 /*
2475  * Concurrency: @dt is write locked.
2476  */
2477 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2478                          const char *name, struct thandle *handle,
2479                          struct lustre_capa *capa)
2480 {
2481         struct osd_object      *obj    = osd_dt_obj(dt);
2482         struct inode           *inode  = obj->oo_inode;
2483         struct osd_thread_info *info   = osd_oti_get(env);
2484         struct dentry          *dentry = &info->oti_obj_dentry;
2485         int                     rc;
2486
2487         LASSERT(dt_object_exists(dt));
2488         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2489         LASSERT(osd_write_locked(env, obj));
2490         LASSERT(handle != NULL);
2491
2492         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2493                 return -EACCES;
2494
2495         OSD_EXEC_OP(handle, xattr_set);
2496
2497         dentry->d_inode = inode;
2498         rc = inode->i_op->removexattr(dentry, name);
2499         return rc;
2500 }
2501
2502 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2503                                      struct dt_object *dt,
2504                                      struct lustre_capa *old,
2505                                      __u64 opc)
2506 {
2507         struct osd_thread_info *info = osd_oti_get(env);
2508         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2509         struct osd_object *obj = osd_dt_obj(dt);
2510         struct osd_device *dev = osd_obj2dev(obj);
2511         struct lustre_capa_key *key = &info->oti_capa_key;
2512         struct lustre_capa *capa = &info->oti_capa;
2513         struct obd_capa *oc;
2514         struct md_capainfo *ci;
2515         int rc;
2516         ENTRY;
2517
2518         if (!dev->od_fl_capa)
2519                 RETURN(ERR_PTR(-ENOENT));
2520
2521         LASSERT(dt_object_exists(dt));
2522         LINVRNT(osd_invariant(obj));
2523
2524         /* renewal sanity check */
2525         if (old && osd_object_auth(env, dt, old, opc))
2526                 RETURN(ERR_PTR(-EACCES));
2527
2528         ci = md_capainfo(env);
2529         if (unlikely(!ci))
2530                 RETURN(ERR_PTR(-ENOENT));
2531
2532         switch (ci->mc_auth) {
2533         case LC_ID_NONE:
2534                 RETURN(NULL);
2535         case LC_ID_PLAIN:
2536                 capa->lc_uid = obj->oo_inode->i_uid;
2537                 capa->lc_gid = obj->oo_inode->i_gid;
2538                 capa->lc_flags = LC_ID_PLAIN;
2539                 break;
2540         case LC_ID_CONVERT: {
2541                 __u32 d[4], s[4];
2542
2543                 s[0] = obj->oo_inode->i_uid;
2544                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2545                 s[2] = obj->oo_inode->i_gid;
2546                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2547                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2548                 if (unlikely(rc))
2549                         RETURN(ERR_PTR(rc));
2550
2551                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2552                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2553                 capa->lc_flags = LC_ID_CONVERT;
2554                 break;
2555         }
2556         default:
2557                 RETURN(ERR_PTR(-EINVAL));
2558         }
2559
2560         capa->lc_fid = *fid;
2561         capa->lc_opc = opc;
2562         capa->lc_flags |= dev->od_capa_alg << 24;
2563         capa->lc_timeout = dev->od_capa_timeout;
2564         capa->lc_expiry = 0;
2565
2566         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2567         if (oc) {
2568                 LASSERT(!capa_is_expired(oc));
2569                 RETURN(oc);
2570         }
2571
2572         cfs_spin_lock(&capa_lock);
2573         *key = dev->od_capa_keys[1];
2574         cfs_spin_unlock(&capa_lock);
2575
2576         capa->lc_keyid = key->lk_keyid;
2577         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2578
2579         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2580         if (rc) {
2581                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2582                 RETURN(ERR_PTR(rc));
2583         }
2584
2585         oc = capa_add(dev->od_capa_hash, capa);
2586         RETURN(oc);
2587 }
2588
2589 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2590 {
2591         struct osd_object      *obj    = osd_dt_obj(dt);
2592         struct inode           *inode  = obj->oo_inode;
2593         struct osd_thread_info *info   = osd_oti_get(env);
2594         struct dentry          *dentry = &info->oti_obj_dentry;
2595         struct file            *file   = &info->oti_file;
2596         int                     rc;
2597
2598         ENTRY;
2599
2600         dentry->d_inode = inode;
2601         file->f_dentry = dentry;
2602         file->f_mapping = inode->i_mapping;
2603         file->f_op = inode->i_fop;
2604         LOCK_INODE_MUTEX(inode);
2605         rc = file->f_op->fsync(file, dentry, 0);
2606         UNLOCK_INODE_MUTEX(inode);
2607         RETURN(rc);
2608 }
2609
2610 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2611                         void **data)
2612 {
2613         struct osd_object *obj = osd_dt_obj(dt);
2614         ENTRY;
2615
2616         *data = (void *)obj->oo_inode;
2617         RETURN(0);
2618 }
2619
2620 /*
2621  * Index operations.
2622  */
2623
2624 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2625                            const struct dt_index_features *feat)
2626 {
2627         struct iam_descr *descr;
2628
2629         if (osd_object_is_root(o))
2630                 return feat == &dt_directory_features;
2631
2632         LASSERT(o->oo_dir != NULL);
2633
2634         descr = o->oo_dir->od_container.ic_descr;
2635         if (feat == &dt_directory_features) {
2636                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2637                         return 1;
2638                 else
2639                         return 0;
2640         } else {
2641                 return
2642                         feat->dif_keysize_min <= descr->id_key_size &&
2643                         descr->id_key_size <= feat->dif_keysize_max &&
2644                         feat->dif_recsize_min <= descr->id_rec_size &&
2645                         descr->id_rec_size <= feat->dif_recsize_max &&
2646                         !(feat->dif_flags & (DT_IND_VARKEY |
2647                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2648                         ergo(feat->dif_flags & DT_IND_UPDATE,
2649                              1 /* XXX check that object (and file system) is
2650                                 * writable */);
2651         }
2652 }
2653
2654 static int osd_iam_container_init(const struct lu_env *env,
2655                                   struct osd_object *obj,
2656                                   struct osd_directory *dir)
2657 {
2658         struct iam_container *bag = &dir->od_container;
2659         int result;
2660
2661         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2662         if (result != 0)
2663                 return result;
2664
2665         result = iam_container_setup(bag);
2666         if (result != 0)
2667                 goto out;
2668
2669         if (osd_obj2dev(obj)->od_iop_mode) {
2670                 u32 ptr = bag->ic_descr->id_ops->id_root_ptr(bag);
2671
2672                 bag->ic_root_bh = ldiskfs_bread(NULL, obj->oo_inode,
2673                                                 ptr, 0, &result);
2674         }
2675
2676  out:
2677         if (result == 0)
2678                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2679         else
2680                 iam_container_fini(bag);
2681
2682         return result;
2683 }
2684
2685
2686 /*
2687  * Concurrency: no external locking is necessary.
2688  */
2689 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2690                          const struct dt_index_features *feat)
2691 {
2692         int                      result;
2693         int                      skip_iam = 0;
2694         struct osd_object       *obj = osd_dt_obj(dt);
2695         struct osd_device       *osd = osd_obj2dev(obj);
2696
2697         LINVRNT(osd_invariant(obj));
2698         LASSERT(dt_object_exists(dt));
2699
2700         if (osd_object_is_root(obj)) {
2701                 dt->do_index_ops = &osd_index_ea_ops;
2702                 result = 0;
2703         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2704                 dt->do_index_ops = &osd_index_ea_ops;
2705                 if (S_ISDIR(obj->oo_inode->i_mode))
2706                         result = 0;
2707                 else
2708                         result = -ENOTDIR;
2709                 skip_iam = 1;
2710         } else if (unlikely(feat == &dt_otable_features)) {
2711                 dt->do_index_ops = &osd_otable_ops;
2712                 return 0;
2713         } else if (feat == &dt_acct_features) {
2714                 dt->do_index_ops = &osd_acct_index_ops;
2715                 result = 0;
2716                 skip_iam = 1;
2717         } else if (!osd_has_index(obj)) {
2718                 struct osd_directory *dir;
2719
2720                 OBD_ALLOC_PTR(dir);
2721                 if (dir != NULL) {
2722
2723                         cfs_spin_lock(&obj->oo_guard);
2724                         if (obj->oo_dir == NULL)
2725                                 obj->oo_dir = dir;
2726                         else
2727                                 /*
2728                                  * Concurrent thread allocated container data.
2729                                  */
2730                                 OBD_FREE_PTR(dir);
2731                         cfs_spin_unlock(&obj->oo_guard);
2732                         /*
2733                          * Now, that we have container data, serialize its
2734                          * initialization.
2735                          */
2736                         cfs_down_write(&obj->oo_ext_idx_sem);
2737                         /*
2738                          * recheck under lock.
2739                          */
2740                         if (!osd_has_index(obj))
2741                                 result = osd_iam_container_init(env, obj, dir);
2742                         else
2743                                 result = 0;
2744                         cfs_up_write(&obj->oo_ext_idx_sem);
2745                 } else {
2746                         result = -ENOMEM;
2747                 }
2748         } else {
2749                 result = 0;
2750         }
2751
2752         if (result == 0 && skip_iam == 0) {
2753                 if (!osd_iam_index_probe(env, obj, feat))
2754                         result = -ENOTDIR;
2755         }
2756         LINVRNT(osd_invariant(obj));
2757
2758         return result;
2759 }
2760
2761 static const struct dt_object_operations osd_obj_ops = {
2762         .do_read_lock         = osd_object_read_lock,
2763         .do_write_lock        = osd_object_write_lock,
2764         .do_read_unlock       = osd_object_read_unlock,
2765         .do_write_unlock      = osd_object_write_unlock,
2766         .do_write_locked      = osd_object_write_locked,
2767         .do_attr_get          = osd_attr_get,
2768         .do_declare_attr_set  = osd_declare_attr_set,
2769         .do_attr_set          = osd_attr_set,
2770         .do_ah_init           = osd_ah_init,
2771         .do_declare_create    = osd_declare_object_create,
2772         .do_create            = osd_object_create,
2773         .do_declare_destroy   = osd_declare_object_destroy,
2774         .do_destroy           = osd_object_destroy,
2775         .do_index_try         = osd_index_try,
2776         .do_declare_ref_add   = osd_declare_object_ref_add,
2777         .do_ref_add           = osd_object_ref_add,
2778         .do_declare_ref_del   = osd_declare_object_ref_del,
2779         .do_ref_del           = osd_object_ref_del,
2780         .do_xattr_get         = osd_xattr_get,
2781         .do_declare_xattr_set = osd_declare_xattr_set,
2782         .do_xattr_set         = osd_xattr_set,
2783         .do_declare_xattr_del = osd_declare_xattr_del,
2784         .do_xattr_del         = osd_xattr_del,
2785         .do_xattr_list        = osd_xattr_list,
2786         .do_capa_get          = osd_capa_get,
2787         .do_object_sync       = osd_object_sync,
2788         .do_data_get          = osd_data_get,
2789 };
2790
2791 /**
2792  * dt_object_operations for interoperability mode
2793  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2794  */
2795 static const struct dt_object_operations osd_obj_ea_ops = {
2796         .do_read_lock         = osd_object_read_lock,
2797         .do_write_lock        = osd_object_write_lock,
2798         .do_read_unlock       = osd_object_read_unlock,
2799         .do_write_unlock      = osd_object_write_unlock,
2800         .do_write_locked      = osd_object_write_locked,
2801         .do_attr_get          = osd_attr_get,
2802         .do_declare_attr_set  = osd_declare_attr_set,
2803         .do_attr_set          = osd_attr_set,
2804         .do_ah_init           = osd_ah_init,
2805         .do_declare_create    = osd_declare_object_create,
2806         .do_create            = osd_object_ea_create,
2807         .do_declare_destroy   = osd_declare_object_destroy,
2808         .do_destroy           = osd_object_destroy,
2809         .do_index_try         = osd_index_try,
2810         .do_declare_ref_add   = osd_declare_object_ref_add,
2811         .do_ref_add           = osd_object_ref_add,
2812         .do_declare_ref_del   = osd_declare_object_ref_del,
2813         .do_ref_del           = osd_object_ref_del,
2814         .do_xattr_get         = osd_xattr_get,
2815         .do_declare_xattr_set = osd_declare_xattr_set,
2816         .do_xattr_set         = osd_xattr_set,
2817         .do_declare_xattr_del = osd_declare_xattr_del,
2818         .do_xattr_del         = osd_xattr_del,
2819         .do_xattr_list        = osd_xattr_list,
2820         .do_capa_get          = osd_capa_get,
2821         .do_object_sync       = osd_object_sync,
2822         .do_data_get          = osd_data_get,
2823 };
2824
2825 static int osd_index_declare_iam_delete(const struct lu_env *env,
2826                                         struct dt_object *dt,
2827                                         const struct dt_key *key,
2828                                         struct thandle *handle)
2829 {
2830         struct osd_thandle    *oh;
2831
2832         oh = container_of0(handle, struct osd_thandle, ot_super);
2833         LASSERT(oh->ot_handle == NULL);
2834
2835         OSD_DECLARE_OP(oh, delete);
2836         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2837
2838         return 0;
2839 }
2840
2841 /**
2842  *      delete a (key, value) pair from index \a dt specified by \a key
2843  *
2844  *      \param  dt      osd index object
2845  *      \param  key     key for index
2846  *      \param  rec     record reference
2847  *      \param  handle  transaction handler
2848  *
2849  *      \retval  0  success
2850  *      \retval -ve   failure
2851  */
2852
2853 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2854                                 const struct dt_key *key,
2855                                 struct thandle *handle,
2856                                 struct lustre_capa *capa)
2857 {
2858         struct osd_object     *obj = osd_dt_obj(dt);
2859         struct osd_thandle    *oh;
2860         struct iam_path_descr *ipd;
2861         struct iam_container  *bag = &obj->oo_dir->od_container;
2862         int                    rc;
2863
2864         ENTRY;
2865
2866         LINVRNT(osd_invariant(obj));
2867         LASSERT(dt_object_exists(dt));
2868         LASSERT(bag->ic_object == obj->oo_inode);
2869         LASSERT(handle != NULL);
2870
2871         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2872                 RETURN(-EACCES);
2873
2874         OSD_EXEC_OP(handle, delete);
2875
2876         ipd = osd_idx_ipd_get(env, bag);
2877         if (unlikely(ipd == NULL))
2878                 RETURN(-ENOMEM);
2879
2880         oh = container_of0(handle, struct osd_thandle, ot_super);
2881         LASSERT(oh->ot_handle != NULL);
2882         LASSERT(oh->ot_handle->h_transaction != NULL);
2883
2884         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2885         osd_ipd_put(env, bag, ipd);
2886         LINVRNT(osd_invariant(obj));
2887         RETURN(rc);
2888 }
2889
2890 static int osd_index_declare_ea_delete(const struct lu_env *env,
2891                                        struct dt_object *dt,
2892                                        const struct dt_key *key,
2893                                        struct thandle *handle)
2894 {
2895         struct osd_thandle *oh;
2896
2897         LASSERT(dt_object_exists(dt));
2898         LASSERT(handle != NULL);
2899
2900         oh = container_of0(handle, struct osd_thandle, ot_super);
2901         LASSERT(oh->ot_handle == NULL);
2902
2903         OSD_DECLARE_OP(oh, delete);
2904         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2905
2906         LASSERT(osd_dt_obj(dt)->oo_inode);
2907         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
2908                         osd_dt_obj(dt)->oo_inode);
2909         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
2910                         osd_dt_obj(dt)->oo_inode);
2911
2912         return 0;
2913 }
2914
2915 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2916                                           struct dt_rec *fid)
2917 {
2918         struct osd_fid_pack *rec;
2919         int                  rc = -ENODATA;
2920
2921         if (de->file_type & LDISKFS_DIRENT_LUFID) {
2922                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2923                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2924         }
2925         RETURN(rc);
2926 }
2927
2928 /**
2929  * Index delete function for interoperability mode (b11826).
2930  * It will remove the directory entry added by osd_index_ea_insert().
2931  * This entry is needed to maintain name->fid mapping.
2932  *
2933  * \param key,  key i.e. file entry to be deleted
2934  *
2935  * \retval   0, on success
2936  * \retval -ve, on error
2937  */
2938 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2939                                const struct dt_key *key,
2940                                struct thandle *handle,
2941                                struct lustre_capa *capa)
2942 {
2943         struct osd_object          *obj    = osd_dt_obj(dt);
2944         struct inode               *dir    = obj->oo_inode;
2945         struct dentry              *dentry;
2946         struct osd_thandle         *oh;
2947         struct ldiskfs_dir_entry_2 *de;
2948         struct buffer_head         *bh;
2949         struct htree_lock          *hlock = NULL;
2950         int                         rc;
2951
2952         ENTRY;
2953
2954         LINVRNT(osd_invariant(obj));
2955         LASSERT(dt_object_exists(dt));
2956         LASSERT(handle != NULL);
2957
2958         OSD_EXEC_OP(handle, delete);
2959
2960         oh = container_of(handle, struct osd_thandle, ot_super);
2961         LASSERT(oh->ot_handle != NULL);
2962         LASSERT(oh->ot_handle->h_transaction != NULL);
2963
2964         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2965                 RETURN(-EACCES);
2966
2967         dentry = osd_child_dentry_get(env, obj,
2968                                       (char *)key, strlen((char *)key));
2969
2970         if (obj->oo_hl_head != NULL) {
2971                 hlock = osd_oti_get(env)->oti_hlock;
2972                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
2973                                    dir, LDISKFS_HLOCK_DEL);
2974         } else {
2975                 cfs_down_write(&obj->oo_ext_idx_sem);
2976         }
2977
2978         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
2979         if (bh) {
2980                 rc = ldiskfs_delete_entry(oh->ot_handle,
2981                                           dir, de, bh);
2982                 brelse(bh);
2983         } else {
2984                 rc = -ENOENT;
2985         }
2986         if (hlock != NULL)
2987                 ldiskfs_htree_unlock(hlock);
2988         else
2989                 cfs_up_write(&obj->oo_ext_idx_sem);
2990
2991         LASSERT(osd_invariant(obj));
2992         RETURN(rc);
2993 }
2994
2995 /**
2996  *      Lookup index for \a key and copy record to \a rec.
2997  *
2998  *      \param  dt      osd index object
2999  *      \param  key     key for index
3000  *      \param  rec     record reference
3001  *
3002  *      \retval  +ve  success : exact mach
3003  *      \retval  0    return record with key not greater than \a key
3004  *      \retval -ve   failure
3005  */
3006 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3007                                 struct dt_rec *rec, const struct dt_key *key,
3008                                 struct lustre_capa *capa)
3009 {
3010         struct osd_object      *obj = osd_dt_obj(dt);
3011         struct iam_path_descr  *ipd;
3012         struct iam_container   *bag = &obj->oo_dir->od_container;
3013         struct osd_thread_info *oti = osd_oti_get(env);
3014         struct iam_iterator    *it = &oti->oti_idx_it;
3015         struct iam_rec         *iam_rec;
3016         int                     rc;
3017
3018         ENTRY;
3019
3020         LASSERT(osd_invariant(obj));
3021         LASSERT(dt_object_exists(dt));
3022         LASSERT(bag->ic_object == obj->oo_inode);
3023
3024         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3025                 RETURN(-EACCES);
3026
3027         ipd = osd_idx_ipd_get(env, bag);
3028         if (IS_ERR(ipd))
3029                 RETURN(-ENOMEM);
3030
3031         /* got ipd now we can start iterator. */
3032         iam_it_init(it, bag, 0, ipd);
3033
3034         rc = iam_it_get(it, (struct iam_key *)key);
3035         if (rc >= 0) {
3036                 if (S_ISDIR(obj->oo_inode->i_mode))
3037                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3038                 else
3039                         iam_rec = (struct iam_rec *) rec;
3040
3041                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3042                 if (S_ISDIR(obj->oo_inode->i_mode))
3043                         osd_fid_unpack((struct lu_fid *) rec,
3044                                        (struct osd_fid_pack *)iam_rec);
3045         }
3046         iam_it_put(it);
3047         iam_it_fini(it);
3048         osd_ipd_put(env, bag, ipd);
3049
3050         LINVRNT(osd_invariant(obj));
3051
3052         RETURN(rc);
3053 }
3054
3055 static int osd_index_declare_iam_insert(const struct lu_env *env,
3056                                         struct dt_object *dt,
3057                                         const struct dt_rec *rec,
3058                                         const struct dt_key *key,
3059                                         struct thandle *handle)
3060 {
3061         struct osd_thandle *oh;
3062
3063         LASSERT(dt_object_exists(dt));
3064         LASSERT(handle != NULL);
3065
3066         oh = container_of0(handle, struct osd_thandle, ot_super);
3067         LASSERT(oh->ot_handle == NULL);
3068
3069         OSD_DECLARE_OP(oh, insert);
3070         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3071
3072         return 0;
3073 }
3074
3075 /**
3076  *      Inserts (key, value) pair in \a dt index object.
3077  *
3078  *      \param  dt      osd index object
3079  *      \param  key     key for index
3080  *      \param  rec     record reference
3081  *      \param  th      transaction handler
3082  *
3083  *      \retval  0  success
3084  *      \retval -ve failure
3085  */
3086 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3087                                 const struct dt_rec *rec,
3088                                 const struct dt_key *key, struct thandle *th,
3089                                 struct lustre_capa *capa, int ignore_quota)
3090 {
3091         struct osd_object     *obj = osd_dt_obj(dt);
3092         struct iam_path_descr *ipd;
3093         struct osd_thandle    *oh;
3094         struct iam_container  *bag = &obj->oo_dir->od_container;
3095 #ifdef HAVE_QUOTA_SUPPORT
3096         cfs_cap_t              save = cfs_curproc_cap_pack();
3097 #endif
3098         struct osd_thread_info *oti = osd_oti_get(env);
3099         struct iam_rec         *iam_rec = (struct iam_rec *)oti->oti_ldp;
3100         int                     rc;
3101
3102         ENTRY;
3103
3104         LINVRNT(osd_invariant(obj));
3105         LASSERT(dt_object_exists(dt));
3106         LASSERT(bag->ic_object == obj->oo_inode);
3107         LASSERT(th != NULL);
3108
3109         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3110                 RETURN(-EACCES);
3111
3112         OSD_EXEC_OP(th, insert);
3113
3114         ipd = osd_idx_ipd_get(env, bag);
3115         if (unlikely(ipd == NULL))
3116                 RETURN(-ENOMEM);
3117
3118         oh = container_of0(th, struct osd_thandle, ot_super);
3119         LASSERT(oh->ot_handle != NULL);
3120         LASSERT(oh->ot_handle->h_transaction != NULL);
3121 #ifdef HAVE_QUOTA_SUPPORT
3122         if (ignore_quota)
3123                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3124         else
3125                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3126 #endif
3127         if (S_ISDIR(obj->oo_inode->i_mode))
3128                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3129         else
3130                 iam_rec = (struct iam_rec *) rec;
3131         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3132                         iam_rec, ipd);
3133 #ifdef HAVE_QUOTA_SUPPORT
3134         cfs_curproc_cap_unpack(save);
3135 #endif
3136         osd_ipd_put(env, bag, ipd);
3137         LINVRNT(osd_invariant(obj));
3138         RETURN(rc);
3139 }
3140
3141 /**
3142  * Calls ldiskfs_add_entry() to add directory entry
3143  * into the directory. This is required for
3144  * interoperability mode (b11826)
3145  *
3146  * \retval   0, on success
3147  * \retval -ve, on error
3148  */
3149 static int __osd_ea_add_rec(struct osd_thread_info *info,
3150                             struct osd_object *pobj, struct inode  *cinode,
3151                             const char *name, const struct dt_rec *fid,
3152                             struct htree_lock *hlock, struct thandle *th)
3153 {
3154         struct ldiskfs_dentry_param *ldp;
3155         struct dentry               *child;
3156         struct osd_thandle          *oth;
3157         int                          rc;
3158
3159         oth = container_of(th, struct osd_thandle, ot_super);
3160         LASSERT(oth->ot_handle != NULL);
3161         LASSERT(oth->ot_handle->h_transaction != NULL);
3162
3163         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3164
3165         /* XXX: remove fid_is_igif() check here.
3166          * IGIF check is just to handle insertion of .. when it is 'ROOT',
3167          * it is IGIF now but needs FID in dir entry as well for readdir
3168          * to work.
3169          * LU-838 should fix that and remove fid_is_igif() check */
3170         if (fid_is_igif((struct lu_fid *)fid) ||
3171             fid_is_norm((struct lu_fid *)fid)) {
3172                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3173                 osd_get_ldiskfs_dirent_param(ldp, fid);
3174                 child->d_fsdata = (void *)ldp;
3175         } else {
3176                 child->d_fsdata = NULL;
3177         }
3178         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3179
3180         RETURN(rc);
3181 }
3182
3183 /**
3184  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3185  * into the directory.Also sets flags into osd object to
3186  * indicate dot and dotdot are created. This is required for
3187  * interoperability mode (b11826)
3188  *
3189  * \param dir   directory for dot and dotdot fixup.
3190  * \param obj   child object for linking
3191  *
3192  * \retval   0, on success
3193  * \retval -ve, on error
3194  */
3195 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3196                               struct osd_object *dir,
3197                               struct inode  *parent_dir, const char *name,
3198                               const struct dt_rec *dot_fid,
3199                               const struct dt_rec *dot_dot_fid,
3200                               struct thandle *th)
3201 {
3202         struct inode                *inode = dir->oo_inode;
3203         struct ldiskfs_dentry_param *dot_ldp;
3204         struct ldiskfs_dentry_param *dot_dot_ldp;
3205         struct osd_thandle          *oth;
3206         int result = 0;
3207
3208         oth = container_of(th, struct osd_thandle, ot_super);
3209         LASSERT(oth->ot_handle->h_transaction != NULL);
3210         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3211
3212         if (strcmp(name, dot) == 0) {
3213                 if (dir->oo_compat_dot_created) {
3214                         result = -EEXIST;
3215                 } else {
3216                         LASSERT(inode == parent_dir);
3217                         dir->oo_compat_dot_created = 1;
3218                         result = 0;
3219                 }
3220         } else if(strcmp(name, dotdot) == 0) {
3221                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3222                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3223
3224                 if (!dir->oo_compat_dot_created)
3225                         return -EINVAL;
3226                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3227                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3228                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3229                 } else {
3230                         dot_ldp = NULL;
3231                         dot_dot_ldp = NULL;
3232                 }
3233                 /* in case of rename, dotdot is already created */
3234                 if (dir->oo_compat_dotdot_created) {
3235                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3236                                                 dot_dot_fid, NULL, th);
3237                 }
3238
3239                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3240                                                 inode, dot_ldp, dot_dot_ldp);
3241                 if (result == 0)
3242                        dir->oo_compat_dotdot_created = 1;
3243         }
3244
3245         return result;
3246 }
3247
3248
3249 /**
3250  * It will call the appropriate osd_add* function and return the
3251  * value, return by respective functions.
3252  */
3253 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3254                           struct inode *cinode, const char *name,
3255                           const struct dt_rec *fid, struct thandle *th)
3256 {
3257         struct osd_thread_info *info   = osd_oti_get(env);
3258         struct htree_lock      *hlock;
3259         int                     rc;
3260
3261         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3262
3263         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3264                                                    name[2] =='\0'))) {
3265                 if (hlock != NULL) {
3266                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3267                                            pobj->oo_inode, 0);
3268                 } else {
3269                         cfs_down_write(&pobj->oo_ext_idx_sem);
3270                 }
3271                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3272                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3273                                         fid, th);
3274         } else {
3275                 if (hlock != NULL) {
3276                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3277                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3278                 } else {
3279                         cfs_down_write(&pobj->oo_ext_idx_sem);
3280                 }
3281
3282                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3283                                       hlock, th);
3284         }
3285         if (hlock != NULL)
3286                 ldiskfs_htree_unlock(hlock);
3287         else
3288                 cfs_up_write(&pobj->oo_ext_idx_sem);
3289
3290         return rc;
3291 }
3292
3293 static int
3294 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3295                       struct osd_idmap_cache *oic)
3296 {
3297         struct osd_scrub    *scrub = &dev->od_scrub;
3298         struct lu_fid       *fid   = &oic->oic_fid;
3299         struct osd_inode_id *id    = &oti->oti_id;
3300         int                  once  = 0;
3301         int                  rc;
3302         ENTRY;
3303
3304 again:
3305         rc = osd_oi_lookup(oti, dev, fid, id);
3306         if (rc != 0 && rc != -ENOENT)
3307                 RETURN(rc);
3308
3309         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3310                 RETURN(0);
3311
3312         if (thread_is_running(&scrub->os_thread)) {
3313                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3314                 /* There is race condition between osd_oi_lookup and OI scrub.
3315                  * The OI scrub finished just after osd_oi_lookup() failure.
3316                  * Under such case, it is unnecessary to trigger OI scrub again,
3317                  * but try to call osd_oi_lookup() again. */
3318                 if (unlikely(rc == -EAGAIN))
3319                         goto again;
3320
3321                 RETURN(rc);
3322         }
3323
3324         if (!scrub->os_no_scrub && ++once == 1) {
3325                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3326                        PFID(fid));
3327                 rc = osd_scrub_start(dev);
3328                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3329                                ", rc = %d [2]\n",
3330                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3331                                PFID(fid), rc);
3332                 if (rc == 0)
3333                         goto again;
3334         }
3335
3336         RETURN(rc = -EREMCHG);
3337 }
3338
3339 /**
3340  * Calls ->lookup() to find dentry. From dentry get inode and
3341  * read inode's ea to get fid. This is required for  interoperability
3342  * mode (b11826)
3343  *
3344  * \retval   0, on success
3345  * \retval -ve, on error
3346  */
3347 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3348                              struct dt_rec *rec, const struct dt_key *key)
3349 {
3350         struct inode               *dir    = obj->oo_inode;
3351         struct dentry              *dentry;
3352         struct ldiskfs_dir_entry_2 *de;
3353         struct buffer_head         *bh;
3354         struct lu_fid              *fid = (struct lu_fid *) rec;
3355         struct htree_lock          *hlock = NULL;
3356         int                         ino;
3357         int                         rc;
3358
3359         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3360
3361         dentry = osd_child_dentry_get(env, obj,
3362                                       (char *)key, strlen((char *)key));
3363
3364         if (obj->oo_hl_head != NULL) {
3365                 hlock = osd_oti_get(env)->oti_hlock;
3366                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3367                                    dir, LDISKFS_HLOCK_LOOKUP);
3368         } else {
3369                 cfs_down_read(&obj->oo_ext_idx_sem);
3370         }
3371
3372         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3373         if (bh) {
3374                 struct osd_thread_info *oti = osd_oti_get(env);
3375                 struct osd_idmap_cache *oic = &oti->oti_cache;
3376                 struct osd_device *dev = osd_obj2dev(obj);
3377                 struct osd_scrub *scrub = &dev->od_scrub;
3378                 struct scrub_file *sf = &scrub->os_file;
3379
3380                 ino = le32_to_cpu(de->inode);
3381                 rc = osd_get_fid_from_dentry(de, rec);
3382
3383                 /* done with de, release bh */
3384                 brelse(bh);
3385                 if (rc != 0)
3386                         rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3387                 else
3388                         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3389
3390                 if (rc != 0 || !fid_is_norm(fid))
3391                         GOTO(out, rc);
3392
3393                 oic->oic_fid = *fid;
3394                 if ((scrub->os_pos_current <= ino) &&
3395                     (sf->sf_flags & SF_INCONSISTENT ||
3396                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3397                                       sf->sf_oi_bitmap)))
3398                         rc = osd_consistency_check(oti, dev, oic);
3399         } else {
3400                 rc = -ENOENT;
3401         }
3402
3403         GOTO(out, rc);
3404
3405 out:
3406         if (hlock != NULL)
3407                 ldiskfs_htree_unlock(hlock);
3408         else
3409                 cfs_up_read(&obj->oo_ext_idx_sem);
3410         return rc;
3411 }
3412
3413 /**
3414  * Find the osd object for given fid.
3415  *
3416  * \param fid need to find the osd object having this fid
3417  *
3418  * \retval osd_object on success
3419  * \retval        -ve on error
3420  */
3421 struct osd_object *osd_object_find(const struct lu_env *env,
3422                                    struct dt_object *dt,
3423                                    const struct lu_fid *fid)
3424 {
3425         struct lu_device  *ludev = dt->do_lu.lo_dev;
3426         struct osd_object *child = NULL;
3427         struct lu_object  *luch;
3428         struct lu_object  *lo;
3429
3430         luch = lu_object_find(env, ludev, fid, NULL);
3431         if (!IS_ERR(luch)) {
3432                 if (lu_object_exists(luch)) {
3433                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3434                         if (lo != NULL)
3435                                 child = osd_obj(lo);
3436                         else
3437                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3438                                                 "lu_object can't be located"
3439                                                 DFID"\n", PFID(fid));
3440
3441                         if (child == NULL) {
3442                                 lu_object_put(env, luch);
3443                                 CERROR("Unable to get osd_object\n");
3444                                 child = ERR_PTR(-ENOENT);
3445                         }
3446                 } else {
3447                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3448                                         "lu_object does not exists "DFID"\n",
3449                                         PFID(fid));
3450                         lu_object_put(env, luch);
3451                         child = ERR_PTR(-ENOENT);
3452                 }
3453         } else
3454                 child = (void *)luch;
3455
3456         return child;
3457 }
3458
3459 /**
3460  * Put the osd object once done with it.
3461  *
3462  * \param obj osd object that needs to be put
3463  */
3464 static inline void osd_object_put(const struct lu_env *env,
3465                                   struct osd_object *obj)
3466 {
3467         lu_object_put(env, &obj->oo_dt.do_lu);
3468 }
3469
3470 static int osd_index_declare_ea_insert(const struct lu_env *env,
3471                                        struct dt_object *dt,
3472                                        const struct dt_rec *rec,
3473                                        const struct dt_key *key,
3474                                        struct thandle *handle)
3475 {
3476         struct osd_thandle *oh;
3477
3478         LASSERT(dt_object_exists(dt));
3479         LASSERT(handle != NULL);
3480
3481         oh = container_of0(handle, struct osd_thandle, ot_super);
3482         LASSERT(oh->ot_handle == NULL);
3483
3484         OSD_DECLARE_OP(oh, insert);
3485         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3486
3487         LASSERT(osd_dt_obj(dt)->oo_inode);
3488         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3489                         osd_dt_obj(dt)->oo_inode);
3490         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3491                         osd_dt_obj(dt)->oo_inode);
3492
3493         return 0;
3494 }
3495
3496 /**
3497  * Index add function for interoperability mode (b11826).
3498  * It will add the directory entry.This entry is needed to
3499  * maintain name->fid mapping.
3500  *
3501  * \param key it is key i.e. file entry to be inserted
3502  * \param rec it is value of given key i.e. fid
3503  *
3504  * \retval   0, on success
3505  * \retval -ve, on error
3506  */
3507 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3508                                const struct dt_rec *rec,
3509                                const struct dt_key *key, struct thandle *th,
3510                                struct lustre_capa *capa, int ignore_quota)
3511 {
3512         struct osd_object *obj   = osd_dt_obj(dt);
3513         struct lu_fid     *fid   = (struct lu_fid *) rec;
3514         const char        *name  = (const char *)key;
3515         struct osd_object *child;
3516 #ifdef HAVE_QUOTA_SUPPORT
3517         cfs_cap_t          save  = cfs_curproc_cap_pack();
3518 #endif
3519         int                rc;
3520
3521         ENTRY;
3522
3523         LASSERT(osd_invariant(obj));
3524         LASSERT(dt_object_exists(dt));
3525         LASSERT(th != NULL);
3526
3527         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3528                 RETURN(-EACCES);
3529
3530         child = osd_object_find(env, dt, fid);
3531         if (!IS_ERR(child)) {
3532 #ifdef HAVE_QUOTA_SUPPORT
3533                 if (ignore_quota)
3534                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3535                 else
3536                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3537 #endif
3538                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3539 #ifdef HAVE_QUOTA_SUPPORT
3540                 cfs_curproc_cap_unpack(save);
3541 #endif
3542                 osd_object_put(env, child);
3543         } else {
3544                 rc = PTR_ERR(child);
3545         }
3546
3547         LASSERT(osd_invariant(obj));
3548         RETURN(rc);
3549 }
3550
3551 /**
3552  *  Initialize osd Iterator for given osd index object.
3553  *
3554  *  \param  dt      osd index object
3555  */
3556
3557 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3558                                      struct dt_object *dt,
3559                                      __u32 unused,
3560                                      struct lustre_capa *capa)
3561 {
3562         struct osd_it_iam      *it;
3563         struct osd_thread_info *oti = osd_oti_get(env);
3564         struct osd_object      *obj = osd_dt_obj(dt);
3565         struct lu_object       *lo  = &dt->do_lu;
3566         struct iam_path_descr  *ipd;
3567         struct iam_container   *bag = &obj->oo_dir->od_container;
3568
3569         LASSERT(lu_object_exists(lo));
3570
3571         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3572                 return ERR_PTR(-EACCES);
3573
3574         it = &oti->oti_it;
3575         ipd = osd_it_ipd_get(env, bag);
3576         if (likely(ipd != NULL)) {
3577                 it->oi_obj = obj;
3578                 it->oi_ipd = ipd;
3579                 lu_object_get(lo);
3580                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3581                 return (struct dt_it *)it;
3582         }
3583         return ERR_PTR(-ENOMEM);
3584 }
3585
3586 /**
3587  * free given Iterator.
3588  */
3589
3590 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3591 {
3592         struct osd_it_iam *it = (struct osd_it_iam *)di;
3593         struct osd_object *obj = it->oi_obj;
3594
3595         iam_it_fini(&it->oi_it);
3596         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3597         lu_object_put(env, &obj->oo_dt.do_lu);
3598 }
3599
3600 /**
3601  *  Move Iterator to record specified by \a key
3602  *
3603  *  \param  di      osd iterator
3604  *  \param  key     key for index
3605  *
3606  *  \retval +ve  di points to record with least key not larger than key
3607  *  \retval  0   di points to exact matched key
3608  *  \retval -ve  failure
3609  */
3610
3611 static int osd_it_iam_get(const struct lu_env *env,
3612                           struct dt_it *di, const struct dt_key *key)
3613 {
3614         struct osd_it_iam *it = (struct osd_it_iam *)di;
3615
3616         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3617 }
3618
3619 /**
3620  *  Release Iterator
3621  *
3622  *  \param  di      osd iterator
3623  */
3624
3625 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3626 {
3627         struct osd_it_iam *it = (struct osd_it_iam *)di;
3628
3629         iam_it_put(&it->oi_it);
3630 }
3631
3632 /**
3633  *  Move iterator by one record
3634  *
3635  *  \param  di      osd iterator
3636  *
3637  *  \retval +1   end of container reached
3638  *  \retval  0   success
3639  *  \retval -ve  failure
3640  */
3641
3642 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3643 {
3644         struct osd_it_iam *it = (struct osd_it_iam *)di;
3645
3646         return iam_it_next(&it->oi_it);
3647 }
3648
3649 /**
3650  * Return pointer to the key under iterator.
3651  */
3652
3653 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3654                                  const struct dt_it *di)
3655 {
3656         struct osd_it_iam *it = (struct osd_it_iam *)di;
3657
3658         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3659 }
3660
3661 /**
3662  * Return size of key under iterator (in bytes)
3663  */
3664
3665 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3666 {
3667         struct osd_it_iam *it = (struct osd_it_iam *)di;
3668
3669         return iam_it_key_size(&it->oi_it);
3670 }
3671
3672 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3673                                        int len, __u16 type)
3674 {
3675         struct luda_type *lt;
3676         const unsigned    align = sizeof(struct luda_type) - 1;
3677
3678         /* check if file type is required */
3679         if (attr & LUDA_TYPE) {
3680                         len = (len + align) & ~align;
3681
3682                         lt = (void *) ent->lde_name + len;
3683                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3684                         ent->lde_attrs |= LUDA_TYPE;
3685         }
3686
3687         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3688 }
3689
3690 /**
3691  * build lu direct from backend fs dirent.
3692  */
3693
3694 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3695                                       struct lu_fid *fid, __u64 offset,
3696                                       char *name, __u16 namelen,
3697                                       __u16 type, __u32 attr)
3698 {
3699         fid_cpu_to_le(&ent->lde_fid, fid);
3700         ent->lde_attrs = LUDA_FID;
3701
3702         ent->lde_hash = cpu_to_le64(offset);
3703         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3704
3705         strncpy(ent->lde_name, name, namelen);
3706         ent->lde_namelen = cpu_to_le16(namelen);
3707
3708         /* append lustre attributes */
3709         osd_it_append_attrs(ent, attr, namelen, type);
3710 }
3711
3712 /**
3713  * Return pointer to the record under iterator.
3714  */
3715 static int osd_it_iam_rec(const struct lu_env *env,
3716                           const struct dt_it *di,
3717                           struct dt_rec *dtrec, __u32 attr)
3718 {
3719         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
3720         struct osd_thread_info *info = osd_oti_get(env);
3721         ENTRY;
3722
3723         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3724                 const struct osd_fid_pack *rec;
3725                 struct lu_fid             *fid = &info->oti_fid;
3726                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
3727                 char                      *name;
3728                 int                        namelen;
3729                 __u64                      hash;
3730                 int                        rc;
3731
3732                 name = (char *)iam_it_key_get(&it->oi_it);
3733                 if (IS_ERR(name))
3734                         RETURN(PTR_ERR(name));
3735
3736                 namelen = iam_it_key_size(&it->oi_it);
3737
3738                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3739                 if (IS_ERR(rec))
3740                         RETURN(PTR_ERR(rec));
3741
3742                 rc = osd_fid_unpack(fid, rec);
3743                 if (rc)
3744                         RETURN(rc);
3745
3746                 hash = iam_it_store(&it->oi_it);
3747
3748                 /* IAM does not store object type in IAM index (dir) */
3749                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3750                                    0, LUDA_FID);
3751         } else {
3752                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3753                            (struct iam_rec *)dtrec);
3754         }
3755
3756         RETURN(0);
3757 }
3758
3759 /**
3760  * Returns cookie for current Iterator position.
3761  */
3762 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3763 {
3764         struct osd_it_iam *it = (struct osd_it_iam *)di;
3765
3766         return iam_it_store(&it->oi_it);
3767 }
3768
3769 /**
3770  * Restore iterator from cookie.
3771  *
3772  * \param  di      osd iterator
3773  * \param  hash    Iterator location cookie
3774  *
3775  * \retval +ve  di points to record with least key not larger than key.
3776  * \retval  0   di points to exact matched key
3777  * \retval -ve  failure
3778  */
3779
3780 static int osd_it_iam_load(const struct lu_env *env,
3781                            const struct dt_it *di, __u64 hash)
3782 {
3783         struct osd_it_iam *it = (struct osd_it_iam *)di;
3784
3785         return iam_it_load(&it->oi_it, hash);
3786 }
3787
3788 static const struct dt_index_operations osd_index_iam_ops = {
3789         .dio_lookup         = osd_index_iam_lookup,
3790         .dio_declare_insert = osd_index_declare_iam_insert,
3791         .dio_insert         = osd_index_iam_insert,
3792         .dio_declare_delete = osd_index_declare_iam_delete,
3793         .dio_delete         = osd_index_iam_delete,
3794         .dio_it     = {
3795                 .init     = osd_it_iam_init,
3796                 .fini     = osd_it_iam_fini,
3797                 .get      = osd_it_iam_get,
3798                 .put      = osd_it_iam_put,
3799                 .next     = osd_it_iam_next,
3800                 .key      = osd_it_iam_key,
3801                 .key_size = osd_it_iam_key_size,
3802                 .rec      = osd_it_iam_rec,
3803                 .store    = osd_it_iam_store,
3804                 .load     = osd_it_iam_load
3805         }
3806 };
3807
3808 /**
3809  * Creates or initializes iterator context.
3810  *
3811  * \retval struct osd_it_ea, iterator structure on success
3812  *
3813  */
3814 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3815                                     struct dt_object *dt,
3816                                     __u32 attr,
3817                                     struct lustre_capa *capa)
3818 {
3819         struct osd_object       *obj  = osd_dt_obj(dt);
3820         struct osd_thread_info  *info = osd_oti_get(env);
3821         struct osd_it_ea        *it   = &info->oti_it_ea;
3822         struct lu_object        *lo   = &dt->do_lu;
3823         struct dentry           *obj_dentry = &info->oti_it_dentry;
3824         ENTRY;
3825         LASSERT(lu_object_exists(lo));
3826
3827         obj_dentry->d_inode = obj->oo_inode;
3828         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3829         obj_dentry->d_name.hash = 0;
3830
3831         it->oie_rd_dirent       = 0;
3832         it->oie_it_dirent       = 0;
3833         it->oie_dirent          = NULL;
3834         it->oie_buf             = info->oti_it_ea_buf;
3835         it->oie_obj             = obj;
3836         it->oie_file.f_pos      = 0;
3837         it->oie_file.f_dentry   = obj_dentry;
3838         if (attr & LUDA_64BITHASH)
3839                 it->oie_file.f_flags = O_64BITHASH;
3840         else
3841                 it->oie_file.f_flags = O_32BITHASH;
3842         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
3843         it->oie_file.f_op         = obj->oo_inode->i_fop;
3844         it->oie_file.private_data = NULL;
3845         lu_object_get(lo);
3846         RETURN((struct dt_it *) it);
3847 }
3848
3849 /**
3850  * Destroy or finishes iterator context.
3851  *
3852  * \param di iterator structure to be destroyed
3853  */
3854 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3855 {
3856         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3857         struct osd_object    *obj  = it->oie_obj;
3858         struct inode       *inode  = obj->oo_inode;
3859
3860         ENTRY;
3861         it->oie_file.f_op->release(inode, &it->oie_file);
3862         lu_object_put(env, &obj->oo_dt.do_lu);
3863         EXIT;
3864 }
3865
3866 /**
3867  * It position the iterator at given key, so that next lookup continues from
3868  * that key Or it is similar to dio_it->load() but based on a key,
3869  * rather than file position.
3870  *
3871  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3872  * to the beginning.
3873  *
3874  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3875  */
3876 static int osd_it_ea_get(const struct lu_env *env,
3877                          struct dt_it *di, const struct dt_key *key)
3878 {
3879         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3880
3881         ENTRY;
3882         LASSERT(((const char *)key)[0] == '\0');
3883         it->oie_file.f_pos      = 0;
3884         it->oie_rd_dirent       = 0;
3885         it->oie_it_dirent       = 0;
3886         it->oie_dirent          = NULL;
3887
3888         RETURN(+1);
3889 }
3890
3891 /**
3892  * Does nothing
3893  */
3894 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3895 {
3896 }
3897
3898 /**
3899  * It is called internally by ->readdir(). It fills the
3900  * iterator's in-memory data structure with required
3901  * information i.e. name, namelen, rec_size etc.
3902  *
3903  * \param buf in which information to be filled in.
3904  * \param name name of the file in given dir
3905  *
3906  * \retval 0 on success
3907  * \retval 1 on buffer full
3908  */
3909 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3910                                loff_t offset, __u64 ino,
3911                                unsigned d_type)
3912 {
3913         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
3914         struct osd_it_ea_dirent *ent  = it->oie_dirent;
3915         struct lu_fid           *fid  = &ent->oied_fid;
3916         struct osd_fid_pack     *rec;
3917         ENTRY;
3918
3919         /* this should never happen */
3920         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3921                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3922                 RETURN(-EIO);
3923         }
3924
3925         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3926             OSD_IT_EA_BUFSIZE)
3927                 RETURN(1);
3928
3929         if (d_type & LDISKFS_DIRENT_LUFID) {
3930                 rec = (struct osd_fid_pack*) (name + namelen + 1);
3931
3932                 if (osd_fid_unpack(fid, rec) != 0)
3933                         fid_zero(fid);
3934
3935                 d_type &= ~LDISKFS_DIRENT_LUFID;
3936         } else {
3937                 fid_zero(fid);
3938         }
3939
3940         ent->oied_ino     = ino;
3941         ent->oied_off     = offset;
3942         ent->oied_namelen = namelen;
3943         ent->oied_type    = d_type;
3944
3945         memcpy(ent->oied_name, name, namelen);
3946
3947         it->oie_rd_dirent++;
3948         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
3949         RETURN(0);
3950 }
3951
3952 /**
3953  * Calls ->readdir() to load a directory entry at a time
3954  * and stored it in iterator's in-memory data structure.
3955  *
3956  * \param di iterator's in memory structure
3957  *
3958  * \retval   0 on success
3959  * \retval -ve on error
3960  */
3961 static int osd_ldiskfs_it_fill(const struct lu_env *env,
3962                                const struct dt_it *di)
3963 {
3964         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
3965         struct osd_object  *obj   = it->oie_obj;
3966         struct inode       *inode = obj->oo_inode;
3967         struct htree_lock  *hlock = NULL;
3968         int                 result = 0;
3969
3970         ENTRY;
3971         it->oie_dirent = it->oie_buf;
3972         it->oie_rd_dirent = 0;
3973
3974         if (obj->oo_hl_head != NULL) {
3975                 hlock = osd_oti_get(env)->oti_hlock;
3976                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3977                                    inode, LDISKFS_HLOCK_READDIR);
3978         } else {
3979                 cfs_down_read(&obj->oo_ext_idx_sem);
3980         }
3981
3982         result = inode->i_fop->readdir(&it->oie_file, it,
3983                                        (filldir_t) osd_ldiskfs_filldir);
3984
3985         if (hlock != NULL)
3986                 ldiskfs_htree_unlock(hlock);
3987         else
3988                 cfs_up_read(&obj->oo_ext_idx_sem);
3989
3990         if (it->oie_rd_dirent == 0) {
3991                 result = -EIO;
3992         } else {
3993                 it->oie_dirent = it->oie_buf;
3994                 it->oie_it_dirent = 1;
3995         }
3996
3997         RETURN(result);
3998 }
3999
4000 /**
4001  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4002  * to load a directory entry at a time and stored it in
4003  * iterator's in-memory data structure.
4004  *
4005  * \param di iterator's in memory structure
4006  *
4007  * \retval +ve iterator reached to end
4008  * \retval   0 iterator not reached to end
4009  * \retval -ve on error
4010  */
4011 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4012 {
4013         struct osd_it_ea *it = (struct osd_it_ea *)di;
4014         int rc;
4015
4016         ENTRY;
4017
4018         if (it->oie_it_dirent < it->oie_rd_dirent) {
4019                 it->oie_dirent =
4020                         (void *) it->oie_dirent +
4021                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4022                                        it->oie_dirent->oied_namelen);
4023                 it->oie_it_dirent++;
4024                 RETURN(0);
4025         } else {
4026                 if (it->oie_file.f_pos == LDISKFS_HTREE_EOF)
4027                         rc = +1;
4028                 else
4029                         rc = osd_ldiskfs_it_fill(env, di);
4030         }
4031
4032         RETURN(rc);
4033 }
4034
4035 /**
4036  * Returns the key at current position from iterator's in memory structure.
4037  *
4038  * \param di iterator's in memory structure
4039  *
4040  * \retval key i.e. struct dt_key on success
4041  */
4042 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4043                                     const struct dt_it *di)
4044 {
4045         struct osd_it_ea *it = (struct osd_it_ea *)di;
4046
4047         return (struct dt_key *)it->oie_dirent->oied_name;
4048 }
4049
4050 /**
4051  * Returns the key's size at current position from iterator's in memory structure.
4052  *
4053  * \param di iterator's in memory structure
4054  *
4055  * \retval key_size i.e. struct dt_key on success
4056  */
4057 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4058 {
4059         struct osd_it_ea *it = (struct osd_it_ea *)di;
4060
4061         return it->oie_dirent->oied_namelen;
4062 }
4063
4064
4065 /**
4066  * Returns the value (i.e. fid/igif) at current position from iterator's
4067  * in memory structure.
4068  *
4069  * \param di struct osd_it_ea, iterator's in memory structure
4070  * \param attr attr requested for dirent.
4071  * \param lde lustre dirent
4072  *
4073  * \retval   0 no error and \param lde has correct lustre dirent.
4074  * \retval -ve on error
4075  */
4076 static inline int osd_it_ea_rec(const struct lu_env *env,
4077                                 const struct dt_it *di,
4078                                 struct dt_rec *dtrec, __u32 attr)
4079 {
4080         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4081         struct osd_object      *obj   = it->oie_obj;
4082         struct osd_device      *dev   = osd_obj2dev(obj);
4083         struct osd_scrub       *scrub = &dev->od_scrub;
4084         struct scrub_file      *sf    = &scrub->os_file;
4085         struct osd_thread_info *oti   = osd_oti_get(env);
4086         struct osd_idmap_cache *oic   = &oti->oti_cache;
4087         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4088         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4089         __u32                   ino   = it->oie_dirent->oied_ino;
4090         int                     rc    = 0;
4091         ENTRY;
4092
4093         if (!fid_is_sane(fid)) {
4094                 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4095                 if (rc != 0)
4096                         RETURN(rc);
4097         } else {
4098                 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4099         }
4100
4101         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4102                            it->oie_dirent->oied_name,
4103                            it->oie_dirent->oied_namelen,
4104                            it->oie_dirent->oied_type, attr);
4105
4106         if (!fid_is_norm(fid))
4107                 RETURN(0);
4108
4109         oic->oic_fid = *fid;
4110         if ((scrub->os_pos_current <= ino) &&
4111             (sf->sf_flags & SF_INCONSISTENT ||
4112              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4113                 rc = osd_consistency_check(oti, dev, oic);
4114
4115         RETURN(rc);
4116 }
4117
4118 /**
4119  * Returns a cookie for current position of the iterator head, so that
4120  * user can use this cookie to load/start the iterator next time.
4121  *
4122  * \param di iterator's in memory structure
4123  *
4124  * \retval cookie for current position, on success
4125  */
4126 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4127 {
4128         struct osd_it_ea *it = (struct osd_it_ea *)di;
4129
4130         return it->oie_dirent->oied_off;
4131 }
4132
4133 /**
4134  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4135  * to load a directory entry at a time and stored it i inn,
4136  * in iterator's in-memory data structure.
4137  *
4138  * \param di struct osd_it_ea, iterator's in memory structure
4139  *
4140  * \retval +ve on success
4141  * \retval -ve on error
4142  */
4143 static int osd_it_ea_load(const struct lu_env *env,
4144                           const struct dt_it *di, __u64 hash)
4145 {
4146         struct osd_it_ea *it = (struct osd_it_ea *)di;
4147         int rc;
4148
4149         ENTRY;
4150         it->oie_file.f_pos = hash;
4151
4152         rc =  osd_ldiskfs_it_fill(env, di);
4153         if (rc == 0)
4154                 rc = +1;
4155
4156         RETURN(rc);
4157 }
4158
4159 /**
4160  * Index lookup function for interoperability mode (b11826).
4161  *
4162  * \param key,  key i.e. file name to be searched
4163  *
4164  * \retval +ve, on success
4165  * \retval -ve, on error
4166  */
4167 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4168                                struct dt_rec *rec, const struct dt_key *key,
4169                                struct lustre_capa *capa)
4170 {
4171         struct osd_object *obj = osd_dt_obj(dt);
4172         int rc = 0;
4173
4174         ENTRY;
4175
4176         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4177         LINVRNT(osd_invariant(obj));
4178
4179         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4180                 return -EACCES;
4181
4182         rc = osd_ea_lookup_rec(env, obj, rec, key);
4183
4184         if (rc == 0)
4185                 rc = +1;
4186         RETURN(rc);
4187 }
4188
4189 /**
4190  * Index and Iterator operations for interoperability
4191  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4192  */
4193 static const struct dt_index_operations osd_index_ea_ops = {
4194         .dio_lookup         = osd_index_ea_lookup,
4195         .dio_declare_insert = osd_index_declare_ea_insert,
4196         .dio_insert         = osd_index_ea_insert,
4197         .dio_declare_delete = osd_index_declare_ea_delete,
4198         .dio_delete         = osd_index_ea_delete,
4199         .dio_it     = {
4200                 .init     = osd_it_ea_init,
4201                 .fini     = osd_it_ea_fini,
4202                 .get      = osd_it_ea_get,
4203                 .put      = osd_it_ea_put,
4204                 .next     = osd_it_ea_next,
4205                 .key      = osd_it_ea_key,
4206                 .key_size = osd_it_ea_key_size,
4207                 .rec      = osd_it_ea_rec,
4208                 .store    = osd_it_ea_store,
4209                 .load     = osd_it_ea_load
4210         }
4211 };
4212
4213 static void *osd_key_init(const struct lu_context *ctx,
4214                           struct lu_context_key *key)
4215 {
4216         struct osd_thread_info *info;
4217
4218         OBD_ALLOC_PTR(info);
4219         if (info == NULL)
4220                 return ERR_PTR(-ENOMEM);
4221
4222         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4223         if (info->oti_it_ea_buf == NULL)
4224                 goto out_free_info;
4225
4226         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4227
4228         info->oti_hlock = ldiskfs_htree_lock_alloc();
4229         if (info->oti_hlock == NULL)
4230                 goto out_free_ea;
4231
4232         return info;
4233
4234  out_free_ea:
4235         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4236  out_free_info:
4237         OBD_FREE_PTR(info);
4238         return ERR_PTR(-ENOMEM);
4239 }
4240
4241 static void osd_key_fini(const struct lu_context *ctx,
4242                          struct lu_context_key *key, void* data)
4243 {
4244         struct osd_thread_info *info = data;
4245
4246         if (info->oti_hlock != NULL)
4247                 ldiskfs_htree_lock_free(info->oti_hlock);
4248         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4249         OBD_FREE_PTR(info);
4250 }
4251
4252 static void osd_key_exit(const struct lu_context *ctx,
4253                          struct lu_context_key *key, void *data)
4254 {
4255         struct osd_thread_info *info = data;
4256
4257         LASSERT(info->oti_r_locks == 0);
4258         LASSERT(info->oti_w_locks == 0);
4259         LASSERT(info->oti_txns    == 0);
4260 }
4261
4262 /* type constructor/destructor: osd_type_init, osd_type_fini */
4263 LU_TYPE_INIT_FINI(osd, &osd_key);
4264
4265 struct lu_context_key osd_key = {
4266         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4267         .lct_init = osd_key_init,
4268         .lct_fini = osd_key_fini,
4269         .lct_exit = osd_key_exit
4270 };
4271
4272
4273 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4274                            const char *name, struct lu_device *next)
4275 {
4276         struct osd_device *osd = osd_dev(d);
4277
4278         strncpy(osd->od_svname, name, MAX_OBD_NAME);
4279         return osd_procfs_init(osd, name);
4280 }
4281
4282 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4283 {
4284         ENTRY;
4285
4286         osd_scrub_cleanup(env, o);
4287
4288         if (o->od_fsops) {
4289                 fsfilt_put_ops(o->od_fsops);
4290                 o->od_fsops = NULL;
4291         }
4292
4293         /* shutdown quota slave instance associated with the device */
4294         if (o->od_quota_slave != NULL) {
4295                 qsd_fini(env, o->od_quota_slave);
4296                 o->od_quota_slave = NULL;
4297         }
4298
4299         RETURN(0);
4300 }
4301
4302 static int osd_mount(const struct lu_env *env,
4303                      struct osd_device *o, struct lustre_cfg *cfg)
4304 {
4305         struct lustre_mount_info *lmi;
4306         const char               *dev  = lustre_cfg_string(cfg, 0);
4307         struct lustre_disk_data  *ldd;
4308         struct lustre_sb_info    *lsi;
4309         int                       rc = 0;
4310
4311         ENTRY;
4312
4313         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4314         if (o->od_fsops == NULL) {
4315                 CERROR("Can't find fsfilt_ldiskfs\n");
4316                 RETURN(-ENOTSUPP);
4317         }
4318
4319         if (o->od_mount != NULL) {
4320                 CERROR("Already mounted (%s)\n", dev);
4321                 RETURN(-EEXIST);
4322         }
4323
4324         /* get mount */
4325         lmi = server_get_mount(dev);
4326         if (lmi == NULL) {
4327                 CERROR("Cannot get mount info for %s!\n", dev);
4328                 RETURN(-EFAULT);
4329         }
4330
4331         LASSERT(lmi != NULL);
4332         /* save lustre_mount_info in dt_device */
4333         o->od_mount = lmi;
4334         o->od_mnt = lmi->lmi_mnt;
4335
4336         lsi = s2lsi(lmi->lmi_sb);
4337         ldd = lsi->lsi_ldd;
4338
4339         if (ldd->ldd_flags & LDD_F_IAM_DIR) {
4340                 o->od_iop_mode = 0;
4341                 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", dev);
4342         } else
4343                 o->od_iop_mode = 1;
4344
4345         if (ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
4346                 rc = osd_compat_init(o);
4347                 if (rc)
4348                         CERROR("%s: can't initialize compats: %d\n", dev, rc);
4349         }
4350
4351         RETURN(rc);
4352 }
4353
4354 static struct lu_device *osd_device_fini(const struct lu_env *env,
4355                                          struct lu_device *d)
4356 {
4357         int rc;
4358         ENTRY;
4359
4360         osd_compat_fini(osd_dev(d));
4361
4362         shrink_dcache_sb(osd_sb(osd_dev(d)));
4363         osd_sync(env, lu2dt_dev(d));
4364
4365         rc = osd_procfs_fini(osd_dev(d));
4366         if (rc) {
4367                 CERROR("proc fini error %d \n", rc);
4368                 RETURN (ERR_PTR(rc));
4369         }
4370
4371         if (osd_dev(d)->od_mount)
4372                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
4373                                  osd_dev(d)->od_mount->lmi_mnt);
4374         osd_dev(d)->od_mount = NULL;
4375
4376         RETURN(NULL);
4377 }
4378
4379 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4380                                           struct lu_device_type *t,
4381                                           struct lustre_cfg *cfg)
4382 {
4383         struct lu_device  *l;
4384         struct osd_device *o;
4385
4386         OBD_ALLOC_PTR(o);
4387         if (o != NULL) {
4388                 int result;
4389
4390                 result = dt_device_init(&o->od_dt_dev, t);
4391                 if (result == 0) {
4392                         l = osd2lu_dev(o);
4393                         l->ld_ops = &osd_lu_ops;
4394                         o->od_dt_dev.dd_ops = &osd_dt_ops;
4395                         cfs_spin_lock_init(&o->od_osfs_lock);
4396                         cfs_mutex_init(&o->od_otable_mutex);
4397                         o->od_osfs_age = cfs_time_shift_64(-1000);
4398                         o->od_capa_hash = init_capa_hash();
4399                         if (o->od_capa_hash == NULL) {
4400                                 dt_device_fini(&o->od_dt_dev);
4401                                 l = ERR_PTR(-ENOMEM);
4402                         }
4403                 } else
4404                         l = ERR_PTR(result);
4405
4406                 if (IS_ERR(l))
4407                         OBD_FREE_PTR(o);
4408         } else
4409                 l = ERR_PTR(-ENOMEM);
4410         return l;
4411 }
4412
4413 static struct lu_device *osd_device_free(const struct lu_env *env,
4414                                          struct lu_device *d)
4415 {
4416         struct osd_device *o = osd_dev(d);
4417         ENTRY;
4418
4419         cleanup_capa_hash(o->od_capa_hash);
4420         dt_device_fini(&o->od_dt_dev);
4421         OBD_FREE_PTR(o);
4422         RETURN(NULL);
4423 }
4424
4425 static int osd_process_config(const struct lu_env *env,
4426                               struct lu_device *d, struct lustre_cfg *cfg)
4427 {
4428         struct osd_device *o = osd_dev(d);
4429         int err;
4430         ENTRY;
4431
4432         switch(cfg->lcfg_command) {
4433         case LCFG_SETUP:
4434                 err = osd_mount(env, o, cfg);
4435                 break;
4436         case LCFG_CLEANUP:
4437                 err = osd_shutdown(env, o);
4438                 break;
4439         default:
4440                 err = -ENOSYS;
4441         }
4442
4443         RETURN(err);
4444 }
4445
4446 static int osd_recovery_complete(const struct lu_env *env,
4447                                  struct lu_device *d)
4448 {
4449         RETURN(0);
4450 }
4451
4452 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4453                        struct lu_device *dev)
4454 {
4455         struct osd_device *osd = osd_dev(dev);
4456         int                result;
4457         ENTRY;
4458
4459         /* 1. setup scrub, including OI files initialization */
4460         result = osd_scrub_setup(env, osd);
4461         if (result < 0)
4462                 RETURN(result);
4463
4464         /* 2. setup quota slave instance */
4465         osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
4466                                        osd->od_proc_entry);
4467         if (IS_ERR(osd->od_quota_slave)) {
4468                 result = PTR_ERR(osd->od_quota_slave);
4469                 osd->od_quota_slave = NULL;
4470                 RETURN(result);
4471         }
4472
4473         if (!lu_device_is_md(pdev))
4474                 RETURN(0);
4475
4476         /* 3. setup local objects */
4477         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4478         RETURN(result);
4479 }
4480
4481 static const struct lu_object_operations osd_lu_obj_ops = {
4482         .loo_object_init      = osd_object_init,
4483         .loo_object_delete    = osd_object_delete,
4484         .loo_object_release   = osd_object_release,
4485         .loo_object_free      = osd_object_free,
4486         .loo_object_print     = osd_object_print,
4487         .loo_object_invariant = osd_object_invariant
4488 };
4489
4490 const struct lu_device_operations osd_lu_ops = {
4491         .ldo_object_alloc      = osd_object_alloc,
4492         .ldo_process_config    = osd_process_config,
4493         .ldo_recovery_complete = osd_recovery_complete,
4494         .ldo_prepare           = osd_prepare,
4495 };
4496
4497 static const struct lu_device_type_operations osd_device_type_ops = {
4498         .ldto_init = osd_type_init,
4499         .ldto_fini = osd_type_fini,
4500
4501         .ldto_start = osd_type_start,
4502         .ldto_stop  = osd_type_stop,
4503
4504         .ldto_device_alloc = osd_device_alloc,
4505         .ldto_device_free  = osd_device_free,
4506
4507         .ldto_device_init    = osd_device_init,
4508         .ldto_device_fini    = osd_device_fini
4509 };
4510
4511 static struct lu_device_type osd_device_type = {
4512         .ldt_tags     = LU_DEVICE_DT,
4513         .ldt_name     = LUSTRE_OSD_NAME,
4514         .ldt_ops      = &osd_device_type_ops,
4515         .ldt_ctx_tags = LCT_LOCAL,
4516 };
4517
4518 /*
4519  * lprocfs legacy support.
4520  */
4521 static struct obd_ops osd_obd_device_ops = {
4522         .o_owner = THIS_MODULE
4523 };
4524
4525 static int __init osd_mod_init(void)
4526 {
4527         struct lprocfs_static_vars lvars;
4528
4529         osd_oi_mod_init();
4530         lprocfs_osd_init_vars(&lvars);
4531         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4532                                    LUSTRE_OSD_NAME, &osd_device_type);
4533 }
4534
4535 static void __exit osd_mod_exit(void)
4536 {
4537         class_unregister_type(LUSTRE_OSD_NAME);
4538 }
4539
4540 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4541 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4542 MODULE_LICENSE("GPL");
4543
4544 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);