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_object_operations      osd_obj_otable_it_ops;
  94 static const struct dt_index_operations       osd_index_iam_ops;
  95 static const struct dt_index_operations       osd_index_ea_ops;
  96
  97 static int osd_has_index(const struct osd_object *obj)
  98 {
  99         return obj->oo_dt.do_index_ops != NULL;
 100 }
 101
 102 static int osd_object_invariant(const struct lu_object *l)
 103 {
 104         return osd_invariant(osd_obj(l));
 105 }
 106
 107 #ifdef HAVE_QUOTA_SUPPORT
 108 static inline void
 109 osd_push_ctxt(const struct lu_env *env, struct osd_ctxt *save, bool is_md)
 110 {
 111         struct md_ucred *uc;
 112         struct cred     *tc;
 113
 114         if (!is_md)
 115                 /* OFD support */
 116                 return;
 117
 118         uc = md_ucred(env);
 119
 120         LASSERT(uc != NULL);
 121
 122         save->oc_uid = current_fsuid();
 123         save->oc_gid = current_fsgid();
 124         save->oc_cap = current_cap();
 125         if ((tc = prepare_creds())) {
 126                 tc->fsuid         = uc->mu_fsuid;
 127                 tc->fsgid         = uc->mu_fsgid;
 128                 commit_creds(tc);
 129         }
 130         /* XXX not suboptimal */
 131         cfs_curproc_cap_unpack(uc->mu_cap);
 132 }
 133
 134 static inline void
 135 osd_pop_ctxt(struct osd_ctxt *save, bool is_md)
 136 {
 137         struct cred *tc;
 138
 139         if (!is_md)
 140                 /* OFD support */
 141                 return;
 142
 143         if ((tc = prepare_creds())) {
 144                 tc->fsuid         = save->oc_uid;
 145                 tc->fsgid         = save->oc_gid;
 146                 tc->cap_effective = save->oc_cap;
 147                 commit_creds(tc);
 148         }
 149 }
 150 #endif
 151
 152 /*
 153  * Concurrency: doesn't matter
 154  */
 155 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 156 {
 157         return osd_oti_get(env)->oti_r_locks > 0;
 158 }
 159
 160 /*
 161  * Concurrency: doesn't matter
 162  */
 163 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 164 {
 165         struct osd_thread_info *oti = osd_oti_get(env);
 166         return oti->oti_w_locks > 0 && o->oo_owner == env;
 167 }
 168
 169 /*
 170  * Concurrency: doesn't access mutable data
 171  */
 172 static int osd_root_get(const struct lu_env *env,
 173                         struct dt_device *dev, struct lu_fid *f)
 174 {
 175         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
 176         return 0;
 177 }
 178
 179 static inline int osd_qid_type(struct osd_thandle *oh, int i)
 180 {
 181         return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
 182 }
 183
 184 static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
 185 {
 186         oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
 187 }
 188
 189 void osd_declare_qid(struct dt_object *dt, struct osd_thandle *oh,
 190                      int type, uid_t id, struct inode *inode)
 191 {
 192 #ifdef CONFIG_QUOTA
 193         int i, allocated = 0;
 194         struct osd_object *obj;
 195
 196         LASSERT(dt != NULL);
 197         LASSERT(oh != NULL);
 198         LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%u",
 199                  oh->ot_id_cnt);
 200
 201         /* id entry is allocated in the quota file */
 202         if (inode && inode->i_dquot[type] && inode->i_dquot[type]->dq_off)
 203                 allocated = 1;
 204
 205         for (i = 0; i < oh->ot_id_cnt; i++) {
 206                 if (oh->ot_id_array[i] == id && osd_qid_type(oh, i) == type)
 207                         return;
 208         }
 209
 210         if (unlikely(i >= OSD_MAX_UGID_CNT)) {
 211                 CERROR("more than %d uid/gids for a transaction?\n", i);
 212                 return;
 213         }
 214
 215         oh->ot_id_array[i] = id;
 216         osd_qid_set_type(oh, i, type);
 217         oh->ot_id_cnt++;
 218         obj = osd_dt_obj(dt);
 219         oh->ot_credits += (allocated || id == 0) ?
 220                 1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(osd_obj2dev(obj)));
 221 #endif
 222 }
 223
 224 /*
 225  * OSD object methods.
 226  */
 227
 228 /*
 229  * Concurrency: no concurrent access is possible that early in object
 230  * life-cycle.
 231  */
 232 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 233                                           const struct lu_object_header *hdr,
 234                                           struct lu_device *d)
 235 {
 236         struct osd_object *mo;
 237
 238         OBD_ALLOC_PTR(mo);
 239         if (mo != NULL) {
 240                 struct lu_object *l;
 241
 242                 l = &mo->oo_dt.do_lu;
 243                 dt_object_init(&mo->oo_dt, NULL, d);
 244                 if (osd_dev(d)->od_iop_mode)
 245                         mo->oo_dt.do_ops = &osd_obj_ea_ops;
 246                 else
 247                         mo->oo_dt.do_ops = &osd_obj_ops;
 248
 249                 l->lo_ops = &osd_lu_obj_ops;
 250                 cfs_init_rwsem(&mo->oo_sem);
 251                 cfs_init_rwsem(&mo->oo_ext_idx_sem);
 252                 cfs_spin_lock_init(&mo->oo_guard);
 253                 return l;
 254         } else {
 255                 return NULL;
 256         }
 257 }
 258
 259 static int osd_get_lma(struct inode *inode, struct dentry *dentry,
 260                        struct lustre_mdt_attrs *lma)
 261 {
 262         int rc;
 263
 264         dentry->d_inode = inode;
 265         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
 266                                    sizeof(*lma));
 267         if (rc > 0) {
 268                 /* Check LMA compatibility */
 269                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
 270                         CWARN("%.16s: unsupported incompat LMA feature(s) "
 271                               "%lx/%#x\n",
 272                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
 273                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
 274                                                         ~LMA_INCOMPAT_SUPP);
 275                         rc = -ENOSYS;
 276                 } else {
 277                         lustre_lma_swab(lma);
 278                         rc = 0;
 279                 }
 280         } else if (rc == 0) {
 281                 rc = -ENODATA;
 282         }
 283
 284         return rc;
 285 }
 286
 287 /*
 288  * retrieve object from backend ext fs.
 289  **/
 290 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
 291                        struct osd_inode_id *id)
 292 {
 293         struct inode *inode = NULL;
 294
 295         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
 296         if (IS_ERR(inode)) {
 297                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
 298                        id->oii_ino, PTR_ERR(inode));
 299         } else if (id->oii_gen != OSD_OII_NOGEN &&
 300                    inode->i_generation != id->oii_gen) {
 301                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
 302                        "gen1 = %u\n",
 303                        id->oii_ino, id->oii_gen, inode->i_generation);
 304                 iput(inode);
 305                 inode = ERR_PTR(-ESTALE);
 306         } else if (inode->i_nlink == 0) {
 307                 /* due to parallel readdir and unlink,
 308                 * we can have dead inode here. */
 309                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
 310                 make_bad_inode(inode);
 311                 iput(inode);
 312                 inode = ERR_PTR(-ESTALE);
 313         } else if (is_bad_inode(inode)) {
 314                 CWARN("%.16s: bad inode: ino = %u\n",
 315                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
 316                 iput(inode);
 317                 inode = ERR_PTR(-ENOENT);
 318         } else {
 319                 if (id->oii_gen == OSD_OII_NOGEN)
 320                         osd_id_gen(id, inode->i_ino, inode->i_generation);
 321
 322                 /* Do not update file c/mtime in ldiskfs.
 323                  * NB: we don't have any lock to protect this because we don't
 324                  * have reference on osd_object now, but contention with
 325                  * another lookup + attr_set can't happen in the tiny window
 326                  * between if (...) and set S_NOCMTIME. */
 327                 if (!(inode->i_flags & S_NOCMTIME))
 328                         inode->i_flags |= S_NOCMTIME;
 329         }
 330         return inode;
 331 }
 332
 333 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
 334                            struct osd_inode_id *id, struct lu_fid *fid)
 335 {
 336         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 337         struct inode            *inode;
 338         int                      rc;
 339
 340         inode = osd_iget(info, dev, id);
 341         if (IS_ERR(inode))
 342                 return inode;
 343
 344         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 345         if (rc == 0) {
 346                 *fid = lma->lma_self_fid;
 347         } else if (rc == -ENODATA) {
 348                 LU_IGIF_BUILD(fid, inode->i_ino, inode->i_generation);
 349         } else {
 350                 iput(inode);
 351                 inode = ERR_PTR(rc);
 352         }
 353         return inode;
 354 }
 355
 356 static struct inode *
 357 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
 358                 struct osd_inode_id *id, const struct lu_fid *fid)
 359 {
 360         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
 361         struct inode            *inode;
 362         int                      rc;
 363
 364         inode = osd_iget(info, dev, id);
 365         if (IS_ERR(inode))
 366                 return inode;
 367
 368         rc = osd_get_lma(inode, &info->oti_obj_dentry, lma);
 369         if (rc != 0) {
 370                 if (rc == -ENODATA) {
 371                         CDEBUG(D_LFSCK, "inconsistent obj: NULL, %lu, "DFID"\n",
 372                                inode->i_ino, PFID(fid));
 373                         rc = -EREMCHG;
 374                 }
 375                 iput(inode);
 376                 return ERR_PTR(rc);
 377         }
 378
 379         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
 380                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
 381                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
 382                 iput(inode);
 383                 return ERR_PTR(EREMCHG);
 384         }
 385         return inode;
 386 }
 387
 388 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
 389                           const struct lu_fid *fid,
 390                           const struct lu_object_conf *conf)
 391 {
 392         struct osd_thread_info *info;
 393         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
 394         struct osd_device      *dev;
 395         struct osd_idmap_cache *oic;
 396         struct osd_inode_id    *id;
 397         struct inode           *inode;
 398         struct osd_scrub       *scrub;
 399         struct scrub_file      *sf;
 400         int                     result;
 401         int                     verify = 0;
 402         ENTRY;
 403
 404         LINVRNT(osd_invariant(obj));
 405         LASSERT(obj->oo_inode == NULL);
 406         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
 407
 408         dev = osd_dev(ldev);
 409         scrub = &dev->od_scrub;
 410         sf = &scrub->os_file;
 411         info = osd_oti_get(env);
 412         LASSERT(info);
 413         oic = &info->oti_cache;
 414         id  = &oic->oic_lid;
 415
 416         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
 417                 RETURN(-ENOENT);
 418
 419         if (fid_is_norm(fid)) {
 420                 /* Search order: 1. per-thread cache. */
 421                 if (lu_fid_eq(fid, &oic->oic_fid)) {
 422                         goto iget;
 423                 } else if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
 424                         /* Search order: 2. OI scrub pending list. */
 425                         result = osd_oii_lookup(dev, fid, id);
 426                         if (result == 0)
 427                                 goto iget;
 428                 }
 429
 430                 if (sf->sf_flags & SF_INCONSISTENT)
 431                         verify = 1;
 432         }
 433
 434         /*
 435          * Objects are created as locking anchors or place holders for objects
 436          * yet to be created. No need to osd_oi_lookup() at here because FID
 437          * shouldn't never be re-used, if it's really a duplicate FID from
 438          * unexpected reason, we should be able to detect it later by calling
 439          * do_create->osd_oi_insert()
 440          */
 441         if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
 442                 GOTO(out, result = 0);
 443
 444         /* Search order: 3. OI files. */
 445         result = osd_oi_lookup(info, dev, fid, id);
 446         if (result == -ENOENT) {
 447                 if (!fid_is_norm(fid) ||
 448                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
 449                                       sf->sf_oi_bitmap))
 450                         GOTO(out, result = 0);
 451
 452                 goto trigger;
 453         }
 454
 455         if (result != 0)
 456                 GOTO(out, result);
 457
 458 iget:
 459         if (verify == 0)
 460                 inode = osd_iget(info, dev, id);
 461         else
 462                 inode = osd_iget_verify(info, dev, id, fid);
 463         if (IS_ERR(inode)) {
 464                 result = PTR_ERR(inode);
 465                 if (result == -ENOENT || result == -ESTALE) {
 466                         fid_zero(&oic->oic_fid);
 467                         result = 0;
 468                 } else if (result == -EREMCHG) {
 469
 470 trigger:
 471                         if (thread_is_running(&scrub->os_thread)) {
 472                                 result = -EINPROGRESS;
 473                         } else if (!scrub->os_no_scrub) {
 474                                 result = osd_scrub_start(dev);
 475                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
 476                                                "for "DFID", rc = %d [1]\n",
 477                                                LDISKFS_SB(osd_sb(dev))->s_es->\
 478                                                s_volume_name,PFID(fid), result);
 479                                 if (result == 0 || result == -EALREADY)
 480                                         result = -EINPROGRESS;
 481                                 else
 482                                         result = -EREMCHG;
 483                         }
 484                 }
 485
 486                 GOTO(out, result);
 487         }
 488
 489         obj->oo_inode = inode;
 490         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
 491         if (dev->od_iop_mode) {
 492                 obj->oo_compat_dot_created = 1;
 493                 obj->oo_compat_dotdot_created = 1;
 494         }
 495
 496         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
 497                 GOTO(out, result = 0);
 498
 499         LASSERT(obj->oo_hl_head == NULL);
 500         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
 501         if (obj->oo_hl_head == NULL) {
 502                 obj->oo_inode = NULL;
 503                 iput(inode);
 504                 GOTO(out, result = -ENOMEM);
 505         }
 506         GOTO(out, result = 0);
 507
 508 out:
 509         LINVRNT(osd_invariant(obj));
 510         return result;
 511 }
 512
 513 /*
 514  * Concurrency: shouldn't matter.
 515  */
 516 static void osd_object_init0(struct osd_object *obj)
 517 {
 518         LASSERT(obj->oo_inode != NULL);
 519         obj->oo_dt.do_body_ops = &osd_body_ops;
 520         obj->oo_dt.do_lu.lo_header->loh_attr |=
 521                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 522 }
 523
 524 /*
 525  * Concurrency: no concurrent access is possible that early in object
 526  * life-cycle.
 527  */
 528 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
 529                            const struct lu_object_conf *conf)
 530 {
 531         struct osd_object *obj = osd_obj(l);
 532         int result;
 533
 534         LINVRNT(osd_invariant(obj));
 535
 536         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
 537         obj->oo_dt.do_body_ops = &osd_body_ops_new;
 538         if (result == 0) {
 539                 if (obj->oo_inode != NULL) {
 540                         osd_object_init0(obj);
 541                 } else if (fid_is_otable_it(&l->lo_header->loh_fid)) {
 542                         obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
 543                         /* LFSCK iterator object is special without inode */
 544                         l->lo_header->loh_attr |= LOHA_EXISTS;
 545                 }
 546         }
 547         LINVRNT(osd_invariant(obj));
 548         return result;
 549 }
 550
 551 /*
 552  * Concurrency: no concurrent access is possible that late in object
 553  * life-cycle.
 554  */
 555 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 556 {
 557         struct osd_object *obj = osd_obj(l);
 558
 559         LINVRNT(osd_invariant(obj));
 560
 561         dt_object_fini(&obj->oo_dt);
 562         if (obj->oo_hl_head != NULL)
 563                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
 564         OBD_FREE_PTR(obj);
 565 }
 566
 567 /*
 568  * Concurrency: no concurrent access is possible that late in object
 569  * life-cycle.
 570  */
 571 static void osd_index_fini(struct osd_object *o)
 572 {
 573         struct iam_container *bag;
 574
 575         if (o->oo_dir != NULL) {
 576                 bag = &o->oo_dir->od_container;
 577                 if (o->oo_inode != NULL) {
 578                         if (bag->ic_object == o->oo_inode)
 579                                 iam_container_fini(bag);
 580                 }
 581                 OBD_FREE_PTR(o->oo_dir);
 582                 o->oo_dir = NULL;
 583         }
 584 }
 585
 586 /*
 587  * Concurrency: no concurrent access is possible that late in object
 588  * life-cycle (for all existing callers, that is. New callers have to provide
 589  * their own locking.)
 590  */
 591 static int osd_inode_unlinked(const struct inode *inode)
 592 {
 593         return inode->i_nlink == 0;
 594 }
 595
 596 enum {
 597         OSD_TXN_OI_DELETE_CREDITS    = 20,
 598         OSD_TXN_INODE_DELETE_CREDITS = 20
 599 };
 600
 601 /*
 602  * Journal
 603  */
 604
 605 #if OSD_THANDLE_STATS
 606 /**
 607  * Set time when the handle is allocated
 608  */
 609 static void osd_th_alloced(struct osd_thandle *oth)
 610 {
 611         oth->oth_alloced = cfs_time_current();
 612 }
 613
 614 /**
 615  * Set time when the handle started
 616  */
 617 static void osd_th_started(struct osd_thandle *oth)
 618 {
 619         oth->oth_started = cfs_time_current();
 620 }
 621
 622 /**
 623  * Helper function to convert time interval to microseconds packed in
 624  * long int (default time units for the counter in "stats" initialized
 625  * by lu_time_init() )
 626  */
 627 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
 628 {
 629         struct timeval val;
 630
 631         cfs_duration_usec(cfs_time_sub(end, start), &val);
 632         return val.tv_sec * 1000000 + val.tv_usec;
 633 }
 634
 635 /**
 636  * Check whether the we deal with this handle for too long.
 637  */
 638 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
 639                                 cfs_time_t alloced, cfs_time_t started,
 640                                 cfs_time_t closed)
 641 {
 642         cfs_time_t now = cfs_time_current();
 643
 644         LASSERT(dev != NULL);
 645
 646         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
 647                             interval_to_usec(alloced, started));
 648         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
 649                             interval_to_usec(started, closed));
 650         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
 651                             interval_to_usec(closed, now));
 652
 653         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
 654                 CWARN("transaction handle %p was open for too long: "
 655                       "now "CFS_TIME_T" ,"
 656                       "alloced "CFS_TIME_T" ,"
 657                       "started "CFS_TIME_T" ,"
 658                       "closed "CFS_TIME_T"\n",
 659                       oth, now, alloced, started, closed);
 660                 libcfs_debug_dumpstack(NULL);
 661         }
 662 }
 663
 664 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
 665 {                                                                       \
 666         cfs_time_t __closed = cfs_time_current();                       \
 667         cfs_time_t __alloced = oth->oth_alloced;                        \
 668         cfs_time_t __started = oth->oth_started;                        \
 669                                                                         \
 670         expr;                                                           \
 671         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
 672 }
 673
 674 #else /* OSD_THANDLE_STATS */
 675
 676 #define osd_th_alloced(h)                  do {} while(0)
 677 #define osd_th_started(h)                  do {} while(0)
 678 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
 679
 680 #endif /* OSD_THANDLE_STATS */
 681
 682 /*
 683  * Concurrency: doesn't access mutable data.
 684  */
 685 static int osd_param_is_sane(const struct osd_device *dev,
 686                              const struct thandle *th)
 687 {
 688         struct osd_thandle *oh;
 689         oh = container_of0(th, struct osd_thandle, ot_super);
 690         return oh->ot_credits <= osd_journal(dev)->j_max_transaction_buffers;
 691 }
 692
 693 /*
 694  * Concurrency: shouldn't matter.
 695  */
 696 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
 697 static void osd_trans_commit_cb(struct super_block *sb,
 698                                 struct journal_callback *jcb, int error)
 699 #else
 700 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 701 #endif
 702 {
 703         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 704         struct thandle     *th  = &oh->ot_super;
 705         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
 706         struct dt_txn_commit_cb *dcb, *tmp;
 707
 708         LASSERT(oh->ot_handle == NULL);
 709
 710         if (error)
 711                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 712
 713         dt_txn_hook_commit(th);
 714
 715         /* call per-transaction callbacks if any */
 716         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
 717                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
 718                          "commit callback entry: magic=%x name='%s'\n",
 719                          dcb->dcb_magic, dcb->dcb_name);
 720                 cfs_list_del_init(&dcb->dcb_linkage);
 721                 dcb->dcb_func(NULL, th, dcb, error);
 722         }
 723
 724         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
 725         lu_device_put(lud);
 726         th->th_dev = NULL;
 727
 728         lu_context_exit(&th->th_ctx);
 729         lu_context_fini(&th->th_ctx);
 730         OBD_FREE_PTR(oh);
 731 }
 732
 733 static struct thandle *osd_trans_create(const struct lu_env *env,
 734                                         struct dt_device *d)
 735 {
 736         struct osd_thread_info *oti = osd_oti_get(env);
 737         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 738         struct osd_thandle     *oh;
 739         struct thandle         *th;
 740         ENTRY;
 741
 742         /* on pending IO in this thread should left from prev. request */
 743         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 744
 745         th = ERR_PTR(-ENOMEM);
 746         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 747         if (oh != NULL) {
 748                 th = &oh->ot_super;
 749                 th->th_dev = d;
 750                 th->th_result = 0;
 751                 th->th_tags = LCT_TX_HANDLE;
 752                 oh->ot_credits = 0;
 753                 oti->oti_dev = osd_dt_dev(d);
 754                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
 755                 osd_th_alloced(oh);
 756         }
 757         RETURN(th);
 758 }
 759
 760 /*
 761  * Concurrency: shouldn't matter.
 762  */
 763 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
 764                     struct thandle *th)
 765 {
 766         struct osd_thread_info *oti = osd_oti_get(env);
 767         struct osd_device  *dev = osd_dt_dev(d);
 768         handle_t           *jh;
 769         struct osd_thandle *oh;
 770         int rc;
 771
 772         ENTRY;
 773
 774         LASSERT(current->journal_info == NULL);
 775
 776         oh = container_of0(th, struct osd_thandle, ot_super);
 777         LASSERT(oh != NULL);
 778         LASSERT(oh->ot_handle == NULL);
 779
 780         rc = dt_txn_hook_start(env, d, th);
 781         if (rc != 0)
 782                 GOTO(out, rc);
 783
 784         if (!osd_param_is_sane(dev, th)) {
 785                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
 786                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
 787                       oh->ot_credits,
 788                       osd_journal(dev)->j_max_transaction_buffers);
 789                 /* XXX Limit the credits to 'max_transaction_buffers', and
 790                  *     let the underlying filesystem to catch the error if
 791                  *     we really need so many credits.
 792                  *
 793                  *     This should be removed when we can calculate the
 794                  *     credits precisely. */
 795                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
 796 #ifdef OSD_TRACK_DECLARES
 797                 CERROR("  attr_set: %d, punch: %d, xattr_set: %d,\n",
 798                        oh->ot_declare_attr_set, oh->ot_declare_punch,
 799                        oh->ot_declare_xattr_set);
 800                 CERROR("  create: %d, ref_add: %d, ref_del: %d, write: %d\n",
 801                        oh->ot_declare_create, oh->ot_declare_ref_add,
 802                        oh->ot_declare_ref_del, oh->ot_declare_write);
 803                 CERROR("  insert: %d, delete: %d, destroy: %d\n",
 804                        oh->ot_declare_insert, oh->ot_declare_delete,
 805                        oh->ot_declare_destroy);
 806 #endif
 807         }
 808
 809         /*
 810          * XXX temporary stuff. Some abstraction layer should
 811          * be used.
 812          */
 813         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
 814         osd_th_started(oh);
 815         if (!IS_ERR(jh)) {
 816                 oh->ot_handle = jh;
 817                 LASSERT(oti->oti_txns == 0);
 818                 lu_context_init(&th->th_ctx, th->th_tags);
 819                 lu_context_enter(&th->th_ctx);
 820
 821                 lu_device_get(&d->dd_lu_dev);
 822                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
 823                                              "osd-tx", th);
 824
 825                 /*
 826                  * XXX: current rule is that we first start tx,
 827                  *      then lock object(s), but we can't use
 828                  *      this rule for data (due to locking specifics
 829                  *      in ldiskfs). also in long-term we'd like to
 830                  *      use usually-used (locks;tx) ordering. so,
 831                  *      UGLY thing is that we'll use one ordering for
 832                  *      data (ofd) and reverse ordering for metadata
 833                  *      (mdd). then at some point we'll fix the latter
 834                  */
 835                 if (dev->od_is_md) {
 836                         LASSERT(oti->oti_r_locks == 0);
 837                         LASSERT(oti->oti_w_locks == 0);
 838                 }
 839
 840                 oti->oti_txns++;
 841                 rc = 0;
 842         } else {
 843                 rc = PTR_ERR(jh);
 844         }
 845 out:
 846         RETURN(rc);
 847 }
 848
 849 /*
 850  * Concurrency: shouldn't matter.
 851  */
 852 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
 853 {
 854         int                     rc = 0;
 855         struct osd_thandle     *oh;
 856         struct osd_thread_info *oti = osd_oti_get(env);
 857         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
 858
 859         ENTRY;
 860
 861         oh = container_of0(th, struct osd_thandle, ot_super);
 862
 863         if (oh->ot_handle != NULL) {
 864                 handle_t *hdl = oh->ot_handle;
 865
 866                 hdl->h_sync = th->th_sync;
 867
 868                 /*
 869                  * add commit callback
 870                  * notice we don't do this in osd_trans_start()
 871                  * as underlying transaction can change during truncate
 872                  */
 873                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
 874                                          &oh->ot_jcb);
 875
 876                 LASSERT(oti->oti_txns == 1);
 877                 oti->oti_txns--;
 878                 /*
 879                  * XXX: current rule is that we first start tx,
 880                  *      then lock object(s), but we can't use
 881                  *      this rule for data (due to locking specifics
 882                  *      in ldiskfs). also in long-term we'd like to
 883                  *      use usually-used (locks;tx) ordering. so,
 884                  *      UGLY thing is that we'll use one ordering for
 885                  *      data (ofd) and reverse ordering for metadata
 886                  *      (mdd). then at some point we'll fix the latter
 887                  */
 888                 if (osd_dt_dev(th->th_dev)->od_is_md) {
 889                         LASSERT(oti->oti_r_locks == 0);
 890                         LASSERT(oti->oti_w_locks == 0);
 891                 }
 892                 rc = dt_txn_hook_stop(env, th);
 893                 if (rc != 0)
 894                         CERROR("Failure in transaction hook: %d\n", rc);
 895                 oh->ot_handle = NULL;
 896                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
 897                                   rc = ldiskfs_journal_stop(hdl));
 898                 if (rc != 0)
 899                         CERROR("Failure to stop transaction: %d\n", rc);
 900         } else {
 901                 OBD_FREE_PTR(oh);
 902         }
 903
 904         /* as we want IO to journal and data IO be concurrent, we don't block
 905          * awaiting data IO completion in osd_do_bio(), instead we wait here
 906          * once transaction is submitted to the journal. all reqular requests
 907          * don't do direct IO (except read/write), thus this wait_event becomes
 908          * no-op for them.
 909          *
 910          * IMPORTANT: we have to wait till any IO submited by the thread is
 911          * completed otherwise iobuf may be corrupted by different request
 912          */
 913         cfs_wait_event(iobuf->dr_wait,
 914                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
 915         if (!rc)
 916                 rc = iobuf->dr_error;
 917
 918         RETURN(rc);
 919 }
 920
 921 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
 922 {
 923         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
 924                                                ot_super);
 925
 926         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
 927         LASSERT(&dcb->dcb_func != NULL);
 928         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
 929
 930         return 0;
 931 }
 932
 933 /*
 934  * Called just before object is freed. Releases all resources except for
 935  * object itself (that is released by osd_object_free()).
 936  *
 937  * Concurrency: no concurrent access is possible that late in object
 938  * life-cycle.
 939  */
 940 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 941 {
 942         struct osd_object *obj   = osd_obj(l);
 943         struct inode      *inode = obj->oo_inode;
 944
 945         LINVRNT(osd_invariant(obj));
 946
 947         /*
 948          * If object is unlinked remove fid->ino mapping from object index.
 949          */
 950
 951         osd_index_fini(obj);
 952         if (inode != NULL) {
 953                 iput(inode);
 954                 obj->oo_inode = NULL;
 955         }
 956 }
 957
 958 /*
 959  * Concurrency: ->loo_object_release() is called under site spin-lock.
 960  */
 961 static void osd_object_release(const struct lu_env *env,
 962                                struct lu_object *l)
 963 {
 964 }
 965
 966 /*
 967  * Concurrency: shouldn't matter.
 968  */
 969 static int osd_object_print(const struct lu_env *env, void *cookie,
 970                             lu_printer_t p, const struct lu_object *l)
 971 {
 972         struct osd_object *o = osd_obj(l);
 973         struct iam_descr  *d;
 974
 975         if (o->oo_dir != NULL)
 976                 d = o->oo_dir->od_container.ic_descr;
 977         else
 978                 d = NULL;
 979         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 980                     o, o->oo_inode,
 981                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 982                     o->oo_inode ? o->oo_inode->i_generation : 0,
 983                     d ? d->id_ops->id_name : "plain");
 984 }
 985
 986 /*
 987  * Concurrency: shouldn't matter.
 988  */
 989 int osd_statfs(const struct lu_env *env, struct dt_device *d,
 990                struct obd_statfs *sfs)
 991 {
 992         struct osd_device  *osd = osd_dt_dev(d);
 993         struct super_block *sb = osd_sb(osd);
 994         struct kstatfs     *ksfs;
 995         int result = 0;
 996
 997         if (unlikely(osd->od_mount == NULL))
 998                 return -EINPROGRESS;
 999
1000         /* osd_lproc.c call this without env, allocate ksfs for that case */
1001         if (unlikely(env == NULL)) {
1002                 OBD_ALLOC_PTR(ksfs);
1003                 if (ksfs == NULL)
1004                         return -ENOMEM;
1005         } else {
1006                 ksfs = &osd_oti_get(env)->oti_ksfs;
1007         }
1008
1009         cfs_spin_lock(&osd->od_osfs_lock);
1010         /* cache 1 second */
1011         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
1012                 result = sb->s_op->statfs(sb->s_root, ksfs);
1013                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
1014                         osd->od_osfs_age = cfs_time_current_64();
1015                         statfs_pack(&osd->od_statfs, ksfs);
1016                 }
1017         }
1018
1019         if (likely(result == 0))
1020                 *sfs = osd->od_statfs;
1021         cfs_spin_unlock(&osd->od_osfs_lock);
1022
1023         if (unlikely(env == NULL))
1024                 OBD_FREE_PTR(ksfs);
1025
1026         return result;
1027 }
1028
1029 /*
1030  * Concurrency: doesn't access mutable data.
1031  */
1032 static void osd_conf_get(const struct lu_env *env,
1033                          const struct dt_device *dev,
1034                          struct dt_device_param *param)
1035 {
1036         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1037
1038         /*
1039          * XXX should be taken from not-yet-existing fs abstraction layer.
1040          */
1041         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1042         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1043         param->ddp_block_shift  = sb->s_blocksize_bits;
1044         param->ddp_mntopts      = 0;
1045         if (test_opt(sb, XATTR_USER))
1046                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1047         if (test_opt(sb, POSIX_ACL))
1048                 param->ddp_mntopts |= MNTOPT_ACL;
1049
1050 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1051         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1052                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1053         else
1054 #endif
1055                 param->ddp_max_ea_size = sb->s_blocksize;
1056
1057 }
1058
1059 /**
1060  * Helper function to get and fill the buffer with input values.
1061  */
1062 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1063 {
1064         struct lu_buf *buf;
1065
1066         buf = &osd_oti_get(env)->oti_buf;
1067         buf->lb_buf = area;
1068         buf->lb_len = len;
1069         return buf;
1070 }
1071
1072 /*
1073  * Concurrency: shouldn't matter.
1074  */
1075 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1076 {
1077         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
1078         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1079 }
1080
1081 /**
1082  * Start commit for OSD device.
1083  *
1084  * An implementation of dt_commit_async method for OSD device.
1085  * Asychronously starts underlayng fs sync and thereby a transaction
1086  * commit.
1087  *
1088  * \param env environment
1089  * \param d dt device
1090  *
1091  * \see dt_device_operations
1092  */
1093 static int osd_commit_async(const struct lu_env *env,
1094                             struct dt_device *d)
1095 {
1096         struct super_block *s = osd_sb(osd_dt_dev(d));
1097         ENTRY;
1098
1099         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_NAME);
1100         RETURN(s->s_op->sync_fs(s, 0));
1101 }
1102
1103 /*
1104  * Concurrency: shouldn't matter.
1105  */
1106
1107 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1108 {
1109         struct super_block *sb = osd_sb(osd_dt_dev(d));
1110         int rc;
1111         ENTRY;
1112
1113         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
1114
1115         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1116         RETURN(rc);
1117 }
1118
1119 /*
1120  * Concurrency: serialization provided by callers.
1121  */
1122 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1123                               int mode, unsigned long timeout, __u32 alg,
1124                               struct lustre_capa_key *keys)
1125 {
1126         struct osd_device *dev = osd_dt_dev(d);
1127         ENTRY;
1128
1129         dev->od_fl_capa = mode;
1130         dev->od_capa_timeout = timeout;
1131         dev->od_capa_alg = alg;
1132         dev->od_capa_keys = keys;
1133         RETURN(0);
1134 }
1135
1136 /**
1137  * Concurrency: serialization provided by callers.
1138  */
1139 static void osd_init_quota_ctxt(const struct lu_env *env, struct dt_device *d,
1140                                struct dt_quota_ctxt *ctxt, void *data)
1141 {
1142         struct obd_device *obd = (void *)ctxt;
1143         struct vfsmount *mnt = (struct vfsmount *)data;
1144         ENTRY;
1145
1146         obd->u.obt.obt_sb = mnt->mnt_root->d_inode->i_sb;
1147         OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1148         obd->obd_lvfs_ctxt.pwdmnt = mnt;
1149         obd->obd_lvfs_ctxt.pwd = mnt->mnt_root;
1150         obd->obd_lvfs_ctxt.fs = get_ds();
1151
1152         EXIT;
1153 }
1154
1155 /**
1156  * Note: we do not count into QUOTA here.
1157  * If we mount with --data_journal we may need more.
1158  */
1159 const int osd_dto_credits_noquota[DTO_NR] = {
1160         /**
1161          * Insert/Delete.
1162          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1163          * SINGLEDATA_TRANS_BLOCKS(8)
1164          * XXX Note: maybe iam need more, since iam have more level than
1165          *           EXT3 htree.
1166          */
1167         [DTO_INDEX_INSERT]  = 16,
1168         [DTO_INDEX_DELETE]  = 16,
1169         /**
1170          * Used for OI scrub
1171          */
1172         [DTO_INDEX_UPDATE]  = 16,
1173         /**
1174          * Create a object. The same as create object in EXT3.
1175          * DATA_TRANS_BLOCKS(14) +
1176          * INDEX_EXTRA_BLOCKS(8) +
1177          * 3(inode bits, groups, GDT)
1178          */
1179         [DTO_OBJECT_CREATE] = 25,
1180         /**
1181          * XXX: real credits to be fixed
1182          */
1183         [DTO_OBJECT_DELETE] = 25,
1184         /**
1185          * Attr set credits (inode)
1186          */
1187         [DTO_ATTR_SET_BASE] = 1,
1188         /**
1189          * Xattr set. The same as xattr of EXT3.
1190          * DATA_TRANS_BLOCKS(14)
1191          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1192          * are also counted in. Do not know why?
1193          */
1194         [DTO_XATTR_SET]     = 14,
1195         [DTO_LOG_REC]       = 14,
1196         /**
1197          * credits for inode change during write.
1198          */
1199         [DTO_WRITE_BASE]    = 3,
1200         /**
1201          * credits for single block write.
1202          */
1203         [DTO_WRITE_BLOCK]   = 14,
1204         /**
1205          * Attr set credits for chown.
1206          * This is extra credits for setattr, and it is null without quota
1207          */
1208         [DTO_ATTR_SET_CHOWN]= 0
1209 };
1210
1211 static const struct dt_device_operations osd_dt_ops = {
1212         .dt_root_get       = osd_root_get,
1213         .dt_statfs         = osd_statfs,
1214         .dt_trans_create   = osd_trans_create,
1215         .dt_trans_start    = osd_trans_start,
1216         .dt_trans_stop     = osd_trans_stop,
1217         .dt_trans_cb_add   = osd_trans_cb_add,
1218         .dt_conf_get       = osd_conf_get,
1219         .dt_sync           = osd_sync,
1220         .dt_ro             = osd_ro,
1221         .dt_commit_async   = osd_commit_async,
1222         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1223         .dt_init_quota_ctxt= osd_init_quota_ctxt,
1224 };
1225
1226 static void osd_object_read_lock(const struct lu_env *env,
1227                                  struct dt_object *dt, unsigned role)
1228 {
1229         struct osd_object *obj = osd_dt_obj(dt);
1230         struct osd_thread_info *oti = osd_oti_get(env);
1231
1232         LINVRNT(osd_invariant(obj));
1233
1234         LASSERT(obj->oo_owner != env);
1235         cfs_down_read_nested(&obj->oo_sem, role);
1236
1237         LASSERT(obj->oo_owner == NULL);
1238         oti->oti_r_locks++;
1239 }
1240
1241 static void osd_object_write_lock(const struct lu_env *env,
1242                                   struct dt_object *dt, unsigned role)
1243 {
1244         struct osd_object *obj = osd_dt_obj(dt);
1245         struct osd_thread_info *oti = osd_oti_get(env);
1246
1247         LINVRNT(osd_invariant(obj));
1248
1249         LASSERT(obj->oo_owner != env);
1250         cfs_down_write_nested(&obj->oo_sem, role);
1251
1252         LASSERT(obj->oo_owner == NULL);
1253         obj->oo_owner = env;
1254         oti->oti_w_locks++;
1255 }
1256
1257 static void osd_object_read_unlock(const struct lu_env *env,
1258                                    struct dt_object *dt)
1259 {
1260         struct osd_object *obj = osd_dt_obj(dt);
1261         struct osd_thread_info *oti = osd_oti_get(env);
1262
1263         LINVRNT(osd_invariant(obj));
1264
1265         LASSERT(oti->oti_r_locks > 0);
1266         oti->oti_r_locks--;
1267         cfs_up_read(&obj->oo_sem);
1268 }
1269
1270 static void osd_object_write_unlock(const struct lu_env *env,
1271                                     struct dt_object *dt)
1272 {
1273         struct osd_object *obj = osd_dt_obj(dt);
1274         struct osd_thread_info *oti = osd_oti_get(env);
1275
1276         LINVRNT(osd_invariant(obj));
1277
1278         LASSERT(obj->oo_owner == env);
1279         LASSERT(oti->oti_w_locks > 0);
1280         oti->oti_w_locks--;
1281         obj->oo_owner = NULL;
1282         cfs_up_write(&obj->oo_sem);
1283 }
1284
1285 static int osd_object_write_locked(const struct lu_env *env,
1286                                    struct dt_object *dt)
1287 {
1288         struct osd_object *obj = osd_dt_obj(dt);
1289
1290         LINVRNT(osd_invariant(obj));
1291
1292         return obj->oo_owner == env;
1293 }
1294
1295 static int capa_is_sane(const struct lu_env *env,
1296                         struct osd_device *dev,
1297                         struct lustre_capa *capa,
1298                         struct lustre_capa_key *keys)
1299 {
1300         struct osd_thread_info *oti = osd_oti_get(env);
1301         struct lustre_capa *tcapa = &oti->oti_capa;
1302         struct obd_capa *oc;
1303         int i, rc = 0;
1304         ENTRY;
1305
1306         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1307         if (oc) {
1308                 if (capa_is_expired(oc)) {
1309                         DEBUG_CAPA(D_ERROR, capa, "expired");
1310                         rc = -ESTALE;
1311                 }
1312                 capa_put(oc);
1313                 RETURN(rc);
1314         }
1315
1316         if (capa_is_expired_sec(capa)) {
1317                 DEBUG_CAPA(D_ERROR, capa, "expired");
1318                 RETURN(-ESTALE);
1319         }
1320
1321         cfs_spin_lock(&capa_lock);
1322         for (i = 0; i < 2; i++) {
1323                 if (keys[i].lk_keyid == capa->lc_keyid) {
1324                         oti->oti_capa_key = keys[i];
1325                         break;
1326                 }
1327         }
1328         cfs_spin_unlock(&capa_lock);
1329
1330         if (i == 2) {
1331                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1332                 RETURN(-ESTALE);
1333         }
1334
1335         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1336         if (rc)
1337                 RETURN(rc);
1338
1339         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1340                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1341                 RETURN(-EACCES);
1342         }
1343
1344         oc = capa_add(dev->od_capa_hash, capa);
1345         capa_put(oc);
1346
1347         RETURN(0);
1348 }
1349
1350 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1351                     struct lustre_capa *capa, __u64 opc)
1352 {
1353         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1354         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1355         struct md_capainfo *ci;
1356         int rc;
1357
1358         if (!dev->od_fl_capa)
1359                 return 0;
1360
1361         if (capa == BYPASS_CAPA)
1362                 return 0;
1363
1364         ci = md_capainfo(env);
1365         if (unlikely(!ci))
1366                 return 0;
1367
1368         if (ci->mc_auth == LC_ID_NONE)
1369                 return 0;
1370
1371         if (!capa) {
1372                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1373                 return -EACCES;
1374         }
1375
1376         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1377                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1378                            PFID(fid));
1379                 return -EACCES;
1380         }
1381
1382         if (!capa_opc_supported(capa, opc)) {
1383                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1384                 return -EACCES;
1385         }
1386
1387         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1388                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1389                 return -EACCES;
1390         }
1391
1392         return 0;
1393 }
1394
1395 static struct timespec *osd_inode_time(const struct lu_env *env,
1396                                        struct inode *inode, __u64 seconds)
1397 {
1398         struct osd_thread_info  *oti = osd_oti_get(env);
1399         struct timespec         *t   = &oti->oti_time;
1400
1401         t->tv_sec = seconds;
1402         t->tv_nsec = 0;
1403         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1404         return t;
1405 }
1406
1407
1408 static void osd_inode_getattr(const struct lu_env *env,
1409                               struct inode *inode, struct lu_attr *attr)
1410 {
1411         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1412                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1413                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1414
1415         attr->la_atime      = LTIME_S(inode->i_atime);
1416         attr->la_mtime      = LTIME_S(inode->i_mtime);
1417         attr->la_ctime      = LTIME_S(inode->i_ctime);
1418         attr->la_mode       = inode->i_mode;
1419         attr->la_size       = i_size_read(inode);
1420         attr->la_blocks     = inode->i_blocks;
1421         attr->la_uid        = inode->i_uid;
1422         attr->la_gid        = inode->i_gid;
1423         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1424         attr->la_nlink      = inode->i_nlink;
1425         attr->la_rdev       = inode->i_rdev;
1426         attr->la_blksize    = 1 << inode->i_blkbits;
1427         attr->la_blkbits    = inode->i_blkbits;
1428 }
1429
1430 static int osd_attr_get(const struct lu_env *env,
1431                         struct dt_object *dt,
1432                         struct lu_attr *attr,
1433                         struct lustre_capa *capa)
1434 {
1435         struct osd_object *obj = osd_dt_obj(dt);
1436
1437         LASSERT(dt_object_exists(dt));
1438         LINVRNT(osd_invariant(obj));
1439
1440         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1441                 return -EACCES;
1442
1443         cfs_spin_lock(&obj->oo_guard);
1444         osd_inode_getattr(env, obj->oo_inode, attr);
1445         cfs_spin_unlock(&obj->oo_guard);
1446         return 0;
1447 }
1448
1449 static int osd_declare_attr_set(const struct lu_env *env,
1450                                 struct dt_object *dt,
1451                                 const struct lu_attr *attr,
1452                                 struct thandle *handle)
1453 {
1454         struct osd_thandle *oh;
1455         struct osd_object *obj;
1456
1457         LASSERT(dt != NULL);
1458         LASSERT(handle != NULL);
1459
1460         obj = osd_dt_obj(dt);
1461         LASSERT(osd_invariant(obj));
1462
1463         oh = container_of0(handle, struct osd_thandle, ot_super);
1464         LASSERT(oh->ot_handle == NULL);
1465
1466         OSD_DECLARE_OP(oh, attr_set);
1467         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
1468
1469         if (attr && attr->la_valid & LA_UID) {
1470                 if (obj->oo_inode)
1471                         osd_declare_qid(dt, oh, USRQUOTA, obj->oo_inode->i_uid,
1472                                         obj->oo_inode);
1473                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1474         }
1475         if (attr && attr->la_valid & LA_GID) {
1476                 if (obj->oo_inode)
1477                         osd_declare_qid(dt, oh, GRPQUOTA, obj->oo_inode->i_gid,
1478                                         obj->oo_inode);
1479                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1480         }
1481
1482         return 0;
1483 }
1484
1485 static int osd_inode_setattr(const struct lu_env *env,
1486                              struct inode *inode, const struct lu_attr *attr)
1487 {
1488         __u64 bits;
1489
1490         bits = attr->la_valid;
1491
1492         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1493
1494         if (bits & LA_ATIME)
1495                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1496         if (bits & LA_CTIME)
1497                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1498         if (bits & LA_MTIME)
1499                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1500         if (bits & LA_SIZE) {
1501                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1502                 i_size_write(inode, attr->la_size);
1503         }
1504
1505 #if 0
1506         /* OSD should not change "i_blocks" which is used by quota.
1507          * "i_blocks" should be changed by ldiskfs only. */
1508         if (bits & LA_BLOCKS)
1509                 inode->i_blocks = attr->la_blocks;
1510 #endif
1511         if (bits & LA_MODE)
1512                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1513                         (attr->la_mode & ~S_IFMT);
1514         if (bits & LA_UID)
1515                 inode->i_uid    = attr->la_uid;
1516         if (bits & LA_GID)
1517                 inode->i_gid    = attr->la_gid;
1518         if (bits & LA_NLINK)
1519                 inode->i_nlink  = attr->la_nlink;
1520         if (bits & LA_RDEV)
1521                 inode->i_rdev   = attr->la_rdev;
1522
1523         if (bits & LA_FLAGS) {
1524                 /* always keep S_NOCMTIME */
1525                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1526                                  S_NOCMTIME;
1527         }
1528         return 0;
1529 }
1530
1531 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1532 {
1533         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1534             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1535                 struct iattr    iattr;
1536                 int             rc;
1537
1538                 iattr.ia_valid = 0;
1539                 if (attr->la_valid & LA_UID)
1540                         iattr.ia_valid |= ATTR_UID;
1541                 if (attr->la_valid & LA_GID)
1542                         iattr.ia_valid |= ATTR_GID;
1543                 iattr.ia_uid = attr->la_uid;
1544                 iattr.ia_gid = attr->la_gid;
1545
1546                 rc = ll_vfs_dq_transfer(inode, &iattr);
1547                 if (rc) {
1548                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1549                                "enforcement enabled on the ldiskfs filesystem?",
1550                                inode->i_sb->s_id, rc);
1551                         return rc;
1552                 }
1553         }
1554         return 0;
1555 }
1556
1557 static int osd_attr_set(const struct lu_env *env,
1558                         struct dt_object *dt,
1559                         const struct lu_attr *attr,
1560                         struct thandle *handle,
1561                         struct lustre_capa *capa)
1562 {
1563         struct osd_object *obj = osd_dt_obj(dt);
1564         struct inode      *inode;
1565         int rc;
1566
1567         LASSERT(handle != NULL);
1568         LASSERT(dt_object_exists(dt));
1569         LASSERT(osd_invariant(obj));
1570
1571         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1572                 return -EACCES;
1573
1574         OSD_EXEC_OP(handle, attr_set);
1575
1576         inode = obj->oo_inode;
1577         if (!osd_dt_dev(handle->th_dev)->od_is_md) {
1578                 /* OFD support */
1579                 rc = osd_quota_transfer(inode, attr);
1580                 if (rc)
1581                         return rc;
1582         } else {
1583 #ifdef HAVE_QUOTA_SUPPORT
1584                 if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1585                     (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1586                         struct osd_ctxt *save = &osd_oti_get(env)->oti_ctxt;
1587                         struct           iattr iattr;
1588                         int              rc;
1589
1590                         iattr.ia_valid = 0;
1591                         if (attr->la_valid & LA_UID)
1592                                 iattr.ia_valid |= ATTR_UID;
1593                         if (attr->la_valid & LA_GID)
1594                                 iattr.ia_valid |= ATTR_GID;
1595                         iattr.ia_uid = attr->la_uid;
1596                         iattr.ia_gid = attr->la_gid;
1597                         osd_push_ctxt(env, save, 1);
1598                         rc = ll_vfs_dq_transfer(inode, &iattr) ? -EDQUOT : 0;
1599                         osd_pop_ctxt(save, 1);
1600                         if (rc != 0)
1601                                 return rc;
1602                 }
1603 #endif
1604         }
1605         cfs_spin_lock(&obj->oo_guard);
1606         rc = osd_inode_setattr(env, inode, attr);
1607         cfs_spin_unlock(&obj->oo_guard);
1608
1609         if (!rc)
1610                 inode->i_sb->s_op->dirty_inode(inode);
1611         return rc;
1612 }
1613
1614 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1615                                     struct osd_object *obj,
1616                                     const char *name, const int namelen)
1617 {
1618         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1619 }
1620
1621 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1622                       cfs_umode_t mode,
1623                       struct dt_allocation_hint *hint,
1624                       struct thandle *th)
1625 {
1626         int result;
1627         struct osd_device  *osd = osd_obj2dev(obj);
1628         struct osd_thandle *oth;
1629         struct dt_object   *parent = NULL;
1630         struct inode       *inode;
1631 #ifdef HAVE_QUOTA_SUPPORT
1632         struct osd_ctxt    *save = &info->oti_ctxt;
1633 #endif
1634
1635         LINVRNT(osd_invariant(obj));
1636         LASSERT(obj->oo_inode == NULL);
1637         LASSERT(obj->oo_hl_head == NULL);
1638
1639         if (S_ISDIR(mode) && ldiskfs_pdo) {
1640                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1641                 if (obj->oo_hl_head == NULL)
1642                         return -ENOMEM;
1643         }
1644
1645         oth = container_of(th, struct osd_thandle, ot_super);
1646         LASSERT(oth->ot_handle->h_transaction != NULL);
1647
1648         if (hint && hint->dah_parent)
1649                 parent = hint->dah_parent;
1650
1651 #ifdef HAVE_QUOTA_SUPPORT
1652         osd_push_ctxt(info->oti_env, save, osd_dt_dev(th->th_dev)->od_is_md);
1653 #endif
1654         inode = ldiskfs_create_inode(oth->ot_handle,
1655                                      parent ? osd_dt_obj(parent)->oo_inode :
1656                                               osd_sb(osd)->s_root->d_inode,
1657                                      mode);
1658 #ifdef HAVE_QUOTA_SUPPORT
1659         osd_pop_ctxt(save, osd_dt_dev(th->th_dev)->od_is_md);
1660 #endif
1661         if (!IS_ERR(inode)) {
1662                 /* Do not update file c/mtime in ldiskfs.
1663                  * NB: don't need any lock because no contention at this
1664                  * early stage */
1665                 inode->i_flags |= S_NOCMTIME;
1666                 inode->i_state |= I_LUSTRE_NOSCRUB;
1667                 obj->oo_inode = inode;
1668                 result = 0;
1669         } else {
1670                 if (obj->oo_hl_head != NULL) {
1671                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1672                         obj->oo_hl_head = NULL;
1673                 }
1674                 result = PTR_ERR(inode);
1675         }
1676         LINVRNT(osd_invariant(obj));
1677         return result;
1678 }
1679
1680 enum {
1681         OSD_NAME_LEN = 255
1682 };
1683
1684 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1685                      struct lu_attr *attr,
1686                      struct dt_allocation_hint *hint,
1687                      struct dt_object_format *dof,
1688                      struct thandle *th)
1689 {
1690         int result;
1691         struct osd_thandle *oth;
1692         struct osd_device *osd = osd_obj2dev(obj);
1693         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1694
1695         LASSERT(S_ISDIR(attr->la_mode));
1696
1697         oth = container_of(th, struct osd_thandle, ot_super);
1698         LASSERT(oth->ot_handle->h_transaction != NULL);
1699         result = osd_mkfile(info, obj, mode, hint, th);
1700         if (result == 0 && osd->od_iop_mode == 0) {
1701                 LASSERT(obj->oo_inode != NULL);
1702                 /*
1703                  * XXX uh-oh... call low-level iam function directly.
1704                  */
1705
1706                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1707                                          sizeof (struct osd_fid_pack),
1708                                          oth->ot_handle);
1709         }
1710         return result;
1711 }
1712
1713 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1714                         struct lu_attr *attr,
1715                         struct dt_allocation_hint *hint,
1716                         struct dt_object_format *dof,
1717                         struct thandle *th)
1718 {
1719         int result;
1720         struct osd_thandle *oth;
1721         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1722
1723         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1724
1725         LASSERT(S_ISREG(attr->la_mode));
1726
1727         oth = container_of(th, struct osd_thandle, ot_super);
1728         LASSERT(oth->ot_handle->h_transaction != NULL);
1729
1730         result = osd_mkfile(info, obj, mode, hint, th);
1731         if (result == 0) {
1732                 LASSERT(obj->oo_inode != NULL);
1733                 if (feat->dif_flags & DT_IND_VARKEY)
1734                         result = iam_lvar_create(obj->oo_inode,
1735                                                  feat->dif_keysize_max,
1736                                                  feat->dif_ptrsize,
1737                                                  feat->dif_recsize_max,
1738                                                  oth->ot_handle);
1739                 else
1740                         result = iam_lfix_create(obj->oo_inode,
1741                                                  feat->dif_keysize_max,
1742                                                  feat->dif_ptrsize,
1743                                                  feat->dif_recsize_max,
1744                                                  oth->ot_handle);
1745
1746         }
1747         return result;
1748 }
1749
1750 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1751                      struct lu_attr *attr,
1752                      struct dt_allocation_hint *hint,
1753                      struct dt_object_format *dof,
1754                      struct thandle *th)
1755 {
1756         LASSERT(S_ISREG(attr->la_mode));
1757         return osd_mkfile(info, obj, (attr->la_mode &
1758                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1759 }
1760
1761 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1762                      struct lu_attr *attr,
1763                      struct dt_allocation_hint *hint,
1764                      struct dt_object_format *dof,
1765                      struct thandle *th)
1766 {
1767         LASSERT(S_ISLNK(attr->la_mode));
1768         return osd_mkfile(info, obj, (attr->la_mode &
1769                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1770 }
1771
1772 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1773                      struct lu_attr *attr,
1774                      struct dt_allocation_hint *hint,
1775                      struct dt_object_format *dof,
1776                      struct thandle *th)
1777 {
1778         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1779         int result;
1780
1781         LINVRNT(osd_invariant(obj));
1782         LASSERT(obj->oo_inode == NULL);
1783         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1784                 S_ISFIFO(mode) || S_ISSOCK(mode));
1785
1786         result = osd_mkfile(info, obj, mode, hint, th);
1787         if (result == 0) {
1788                 LASSERT(obj->oo_inode != NULL);
1789                 /*
1790                  * This inode should be marked dirty for i_rdev.  Currently
1791                  * that is done in the osd_attr_init().
1792                  */
1793                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1794         }
1795         LINVRNT(osd_invariant(obj));
1796         return result;
1797 }
1798
1799 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1800                               struct lu_attr *,
1801                               struct dt_allocation_hint *hint,
1802                               struct dt_object_format *dof,
1803                               struct thandle *);
1804
1805 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1806 {
1807         osd_obj_type_f result;
1808
1809         switch (type) {
1810         case DFT_DIR:
1811                 result = osd_mkdir;
1812                 break;
1813         case DFT_REGULAR:
1814                 result = osd_mkreg;
1815                 break;
1816         case DFT_SYM:
1817                 result = osd_mksym;
1818                 break;
1819         case DFT_NODE:
1820                 result = osd_mknod;
1821                 break;
1822         case DFT_INDEX:
1823                 result = osd_mk_index;
1824                 break;
1825
1826         default:
1827                 LBUG();
1828                 break;
1829         }
1830         return result;
1831 }
1832
1833
1834 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1835                         struct dt_object *parent, cfs_umode_t child_mode)
1836 {
1837         LASSERT(ah);
1838
1839         memset(ah, 0, sizeof(*ah));
1840         ah->dah_parent = parent;
1841         ah->dah_mode = child_mode;
1842 }
1843
1844 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1845                           struct lu_attr *attr, struct dt_object_format *dof)
1846 {
1847         struct inode   *inode = obj->oo_inode;
1848         __u64           valid = attr->la_valid;
1849         int             result;
1850
1851         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1852
1853         if (dof->dof_type != DFT_NODE)
1854                 attr->la_valid &= ~LA_RDEV;
1855         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1856                 attr->la_valid &= ~LA_ATIME;
1857         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1858                 attr->la_valid &= ~LA_CTIME;
1859         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1860                 attr->la_valid &= ~LA_MTIME;
1861
1862         if (!osd_obj2dev(obj)->od_is_md) {
1863                 /* OFD support */
1864                 result = osd_quota_transfer(inode, attr);
1865                 if (result)
1866                         return;
1867         } else {
1868 #ifdef HAVE_QUOTA_SUPPORT
1869                 attr->la_valid &= ~(LA_UID | LA_GID);
1870 #endif
1871         }
1872
1873         if (attr->la_valid != 0) {
1874                 result = osd_inode_setattr(info->oti_env, inode, attr);
1875                 /*
1876                  * The osd_inode_setattr() should always succeed here.  The
1877                  * only error that could be returned is EDQUOT when we are
1878                  * trying to change the UID or GID of the inode. However, this
1879                  * should not happen since quota enforcement is no longer
1880                  * enabled on ldiskfs (lquota takes care of it).
1881                  */
1882                 LASSERTF(result == 0, "%d", result);
1883                 inode->i_sb->s_op->dirty_inode(inode);
1884         }
1885
1886         attr->la_valid = valid;
1887 }
1888
1889 /**
1890  * Helper function for osd_object_create()
1891  *
1892  * \retval 0, on success
1893  */
1894 static int __osd_object_create(struct osd_thread_info *info,
1895                                struct osd_object *obj, struct lu_attr *attr,
1896                                struct dt_allocation_hint *hint,
1897                                struct dt_object_format *dof,
1898                                struct thandle *th)
1899 {
1900         int     result;
1901         __u32   umask;
1902
1903         /* we drop umask so that permissions we pass are not affected */
1904         umask = current->fs->umask;
1905         current->fs->umask = 0;
1906
1907         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1908                                                   th);
1909         if (result == 0) {
1910                 osd_attr_init(info, obj, attr, dof);
1911                 osd_object_init0(obj);
1912                 /* bz 24037 */
1913                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1914                         unlock_new_inode(obj->oo_inode);
1915         }
1916
1917         /* restore previous umask value */
1918         current->fs->umask = umask;
1919
1920         return result;
1921 }
1922
1923 /**
1924  * Helper function for osd_object_create()
1925  *
1926  * \retval 0, on success
1927  */
1928 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1929                            const struct lu_fid *fid, struct thandle *th)
1930 {
1931         struct osd_thread_info *info = osd_oti_get(env);
1932         struct osd_inode_id    *id   = &info->oti_id;
1933         struct osd_device      *osd  = osd_obj2dev(obj);
1934
1935         LASSERT(obj->oo_inode != NULL);
1936
1937         if (osd->od_is_md) {
1938                 struct md_ucred *uc = md_ucred(env);
1939                 LASSERT(uc != NULL);
1940         }
1941
1942         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1943         return osd_oi_insert(info, osd, fid, id, th);
1944 }
1945
1946 static int osd_declare_object_create(const struct lu_env *env,
1947                                      struct dt_object *dt,
1948                                      struct lu_attr *attr,
1949                                      struct dt_allocation_hint *hint,
1950                                      struct dt_object_format *dof,
1951                                      struct thandle *handle)
1952 {
1953         struct osd_thandle *oh;
1954
1955         LASSERT(handle != NULL);
1956
1957         oh = container_of0(handle, struct osd_thandle, ot_super);
1958         LASSERT(oh->ot_handle == NULL);
1959
1960         OSD_DECLARE_OP(oh, create);
1961         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_CREATE];
1962         /* XXX: So far, only normal fid needs be inserted into the oi,
1963          *      things could be changed later. Revise following code then. */
1964         if (fid_is_norm(lu_object_fid(&dt->do_lu))) {
1965                 OSD_DECLARE_OP(oh, insert);
1966                 oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
1967         }
1968         /* If this is directory, then we expect . and .. to be inserted as
1969          * well. The one directory block always needs to be created for the
1970          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
1971          * block), there is no danger of needing a tree for the first block.
1972          */
1973         if (attr && S_ISDIR(attr->la_mode)) {
1974                 OSD_DECLARE_OP(oh, insert);
1975                 OSD_DECLARE_OP(oh, insert);
1976                 oh->ot_credits += osd_dto_credits_noquota[DTO_WRITE_BASE];
1977         }
1978
1979         if (attr) {
1980                 osd_declare_qid(dt, oh, USRQUOTA, attr->la_uid, NULL);
1981                 osd_declare_qid(dt, oh, GRPQUOTA, attr->la_gid, NULL);
1982         }
1983         return 0;
1984 }
1985
1986 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1987                              struct lu_attr *attr,
1988                              struct dt_allocation_hint *hint,
1989                              struct dt_object_format *dof,
1990                              struct thandle *th)
1991 {
1992         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
1993         struct osd_object      *obj    = osd_dt_obj(dt);
1994         struct osd_thread_info *info   = osd_oti_get(env);
1995         int result;
1996
1997         ENTRY;
1998
1999         LINVRNT(osd_invariant(obj));
2000         LASSERT(!dt_object_exists(dt));
2001         LASSERT(osd_write_locked(env, obj));
2002         LASSERT(th != NULL);
2003
2004         if (unlikely(fid_is_acct(fid)))
2005                 /* Quota files can't be created from the kernel any more,
2006                  * 'tune2fs -O quota' will take care of creating them */
2007                 RETURN(-EPERM);
2008
2009         OSD_EXEC_OP(th, create);
2010
2011         result = __osd_object_create(info, obj, attr, hint, dof, th);
2012         if (result == 0)
2013                 result = __osd_oi_insert(env, obj, fid, th);
2014
2015         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2016         LASSERT(osd_invariant(obj));
2017         RETURN(result);
2018 }
2019
2020 /**
2021  * Called to destroy on-disk representation of the object
2022  *
2023  * Concurrency: must be locked
2024  */
2025 static int osd_declare_object_destroy(const struct lu_env *env,
2026                                       struct dt_object *dt,
2027                                       struct thandle *th)
2028 {
2029         struct osd_object  *obj = osd_dt_obj(dt);
2030         struct inode       *inode = obj->oo_inode;
2031         struct osd_thandle *oh;
2032
2033         ENTRY;
2034
2035         oh = container_of0(th, struct osd_thandle, ot_super);
2036         LASSERT(oh->ot_handle == NULL);
2037         LASSERT(inode);
2038
2039         OSD_DECLARE_OP(oh, destroy);
2040         OSD_DECLARE_OP(oh, delete);
2041         oh->ot_credits += osd_dto_credits_noquota[DTO_OBJECT_DELETE];
2042         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2043
2044         osd_declare_qid(dt, oh, USRQUOTA, inode->i_uid, inode);
2045         osd_declare_qid(dt, oh, GRPQUOTA, inode->i_gid, inode);
2046
2047         RETURN(0);
2048 }
2049
2050 static int osd_object_destroy(const struct lu_env *env,
2051                               struct dt_object *dt,
2052                               struct thandle *th)
2053 {
2054         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2055         struct osd_object      *obj = osd_dt_obj(dt);
2056         struct inode           *inode = obj->oo_inode;
2057         struct osd_device      *osd = osd_obj2dev(obj);
2058         struct osd_thandle     *oh;
2059         int                     result;
2060         ENTRY;
2061
2062         oh = container_of0(th, struct osd_thandle, ot_super);
2063         LASSERT(oh->ot_handle);
2064         LASSERT(inode);
2065         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2066
2067         if (unlikely(fid_is_acct(fid)))
2068                 RETURN(-EPERM);
2069
2070         /* Parallel control for OI scrub. For most of cases, there is no
2071          * lock contention. So it will not affect unlink performance. */
2072         cfs_mutex_lock(&inode->i_mutex);
2073         if (S_ISDIR(inode->i_mode)) {
2074                 LASSERT(osd_inode_unlinked(inode) ||
2075                         inode->i_nlink == 1);
2076                 cfs_spin_lock(&obj->oo_guard);
2077                 inode->i_nlink = 0;
2078                 cfs_spin_unlock(&obj->oo_guard);
2079                 inode->i_sb->s_op->dirty_inode(inode);
2080         } else {
2081                 LASSERT(osd_inode_unlinked(inode));
2082         }
2083
2084         OSD_EXEC_OP(th, destroy);
2085
2086         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2087         cfs_mutex_unlock(&inode->i_mutex);
2088
2089         /* XXX: add to ext3 orphan list */
2090         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2091
2092         /* not needed in the cache anymore */
2093         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2094
2095         RETURN(0);
2096 }
2097
2098 /**
2099  * Helper function for osd_xattr_set()
2100  */
2101 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2102                            const struct lu_buf *buf, const char *name, int fl)
2103 {
2104         struct osd_object      *obj      = osd_dt_obj(dt);
2105         struct inode           *inode    = obj->oo_inode;
2106         struct osd_thread_info *info     = osd_oti_get(env);
2107         struct dentry          *dentry   = &info->oti_child_dentry;
2108         int                     fs_flags = 0;
2109         int                     rc;
2110
2111         LASSERT(dt_object_exists(dt));
2112         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2113
2114         if (fl & LU_XATTR_REPLACE)
2115                 fs_flags |= XATTR_REPLACE;
2116
2117         if (fl & LU_XATTR_CREATE)
2118                 fs_flags |= XATTR_CREATE;
2119
2120         dentry->d_inode = inode;
2121         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2122                                    buf->lb_len, fs_flags);
2123         return rc;
2124 }
2125
2126 /**
2127  * Put the fid into lustre_mdt_attrs, and then place the structure
2128  * inode's ea. This fid should not be altered during the life time
2129  * of the inode.
2130  *
2131  * \retval +ve, on success
2132  * \retval -ve, on error
2133  *
2134  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2135  */
2136 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2137                           const struct lu_fid *fid)
2138 {
2139         struct osd_thread_info  *info      = osd_oti_get(env);
2140         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2141
2142         lustre_lma_init(mdt_attrs, fid);
2143         lustre_lma_swab(mdt_attrs);
2144         return __osd_xattr_set(env, dt,
2145                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2146                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2147
2148 }
2149
2150 /**
2151  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2152  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2153  * To have compatilibility with 1.8 ldiskfs driver we need to have
2154  * magic number at start of fid data.
2155  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2156  * its inmemory API.
2157  */
2158 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2159                                   const struct dt_rec *fid)
2160 {
2161         param->edp_magic = LDISKFS_LUFID_MAGIC;
2162         param->edp_len =  sizeof(struct lu_fid) + 1;
2163
2164         fid_cpu_to_be((struct lu_fid *)param->edp_data,
2165                       (struct lu_fid *)fid);
2166 }
2167
2168 /**
2169  * Try to read the fid from inode ea into dt_rec, if return value
2170  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2171  *
2172  * \param fid object fid.
2173  *
2174  * \retval 0 on success
2175  */
2176 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2177                           __u32 ino, struct lu_fid *fid,
2178                           struct osd_inode_id *id)
2179 {
2180         struct osd_thread_info *info  = osd_oti_get(env);
2181         struct inode           *inode;
2182         ENTRY;
2183
2184         osd_id_gen(id, ino, OSD_OII_NOGEN);
2185         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2186         if (IS_ERR(inode))
2187                 RETURN(PTR_ERR(inode));
2188
2189         iput(inode);
2190         RETURN(0);
2191 }
2192
2193 /**
2194  * OSD layer object create function for interoperability mode (b11826).
2195  * This is mostly similar to osd_object_create(). Only difference being, fid is
2196  * inserted into inode ea here.
2197  *
2198  * \retval   0, on success
2199  * \retval -ve, on error
2200  */
2201 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2202                                 struct lu_attr *attr,
2203                                 struct dt_allocation_hint *hint,
2204                                 struct dt_object_format *dof,
2205                                 struct thandle *th)
2206 {
2207         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2208         struct osd_object      *obj    = osd_dt_obj(dt);
2209         struct osd_thread_info *info   = osd_oti_get(env);
2210         int                     result;
2211
2212         ENTRY;
2213
2214         LASSERT(osd_invariant(obj));
2215         LASSERT(!dt_object_exists(dt));
2216         LASSERT(osd_write_locked(env, obj));
2217         LASSERT(th != NULL);
2218
2219         if (unlikely(fid_is_acct(fid)))
2220                 /* Quota files can't be created from the kernel any more,
2221                  * 'tune2fs -O quota' will take care of creating them */
2222                 RETURN(-EPERM);
2223
2224         OSD_EXEC_OP(th, create);
2225
2226         result = __osd_object_create(info, obj, attr, hint, dof, th);
2227         /* objects under osd root shld have igif fid, so dont add fid EA */
2228         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL)
2229                 result = osd_ea_fid_set(env, dt, fid);
2230
2231         if (result == 0)
2232                 result = __osd_oi_insert(env, obj, fid, th);
2233
2234         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2235         LINVRNT(osd_invariant(obj));
2236         RETURN(result);
2237 }
2238
2239 static int osd_declare_object_ref_add(const struct lu_env *env,
2240                                       struct dt_object *dt,
2241                                       struct thandle *handle)
2242 {
2243         struct osd_thandle *oh;
2244
2245         /* it's possible that object doesn't exist yet */
2246         LASSERT(handle != NULL);
2247
2248         oh = container_of0(handle, struct osd_thandle, ot_super);
2249         LASSERT(oh->ot_handle == NULL);
2250
2251         OSD_DECLARE_OP(oh, ref_add);
2252         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2253
2254         return 0;
2255 }
2256
2257 /*
2258  * Concurrency: @dt is write locked.
2259  */
2260 static int osd_object_ref_add(const struct lu_env *env,
2261                               struct dt_object *dt, struct thandle *th)
2262 {
2263         struct osd_object *obj = osd_dt_obj(dt);
2264         struct inode      *inode = obj->oo_inode;
2265
2266         LINVRNT(osd_invariant(obj));
2267         LASSERT(dt_object_exists(dt));
2268         LASSERT(osd_write_locked(env, obj));
2269         LASSERT(th != NULL);
2270
2271         OSD_EXEC_OP(th, ref_add);
2272
2273         /*
2274          * DIR_NLINK feature is set for compatibility reasons if:
2275          * 1) nlinks > LDISKFS_LINK_MAX, or
2276          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2277          *
2278          * It is easier to always set this flag (rather than check and set),
2279          * since it has less overhead, and the superblock will be dirtied
2280          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2281          * do not actually care whether this flag is set or not.
2282          */
2283         cfs_spin_lock(&obj->oo_guard);
2284         inode->i_nlink++;
2285         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2286                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2287                     inode->i_nlink == 2)
2288                         inode->i_nlink = 1;
2289         }
2290         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2291         cfs_spin_unlock(&obj->oo_guard);
2292         inode->i_sb->s_op->dirty_inode(inode);
2293         LINVRNT(osd_invariant(obj));
2294
2295         return 0;
2296 }
2297
2298 static int osd_declare_object_ref_del(const struct lu_env *env,
2299                                       struct dt_object *dt,
2300                                       struct thandle *handle)
2301 {
2302         struct osd_thandle *oh;
2303
2304         LASSERT(dt_object_exists(dt));
2305         LASSERT(handle != NULL);
2306
2307         oh = container_of0(handle, struct osd_thandle, ot_super);
2308         LASSERT(oh->ot_handle == NULL);
2309
2310         OSD_DECLARE_OP(oh, ref_del);
2311         oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2312
2313         return 0;
2314 }
2315
2316 /*
2317  * Concurrency: @dt is write locked.
2318  */
2319 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2320                               struct thandle *th)
2321 {
2322         struct osd_object *obj = osd_dt_obj(dt);
2323         struct inode      *inode = obj->oo_inode;
2324
2325         LINVRNT(osd_invariant(obj));
2326         LASSERT(dt_object_exists(dt));
2327         LASSERT(osd_write_locked(env, obj));
2328         LASSERT(th != NULL);
2329
2330         OSD_EXEC_OP(th, ref_del);
2331
2332         cfs_spin_lock(&obj->oo_guard);
2333         LASSERT(inode->i_nlink > 0);
2334         inode->i_nlink--;
2335         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2336          * then the nlink count is 1. Don't let it be set to 0 or the directory
2337          * inode will be deleted incorrectly. */
2338         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2339                 inode->i_nlink++;
2340         cfs_spin_unlock(&obj->oo_guard);
2341         inode->i_sb->s_op->dirty_inode(inode);
2342         LINVRNT(osd_invariant(obj));
2343
2344         return 0;
2345 }
2346
2347 /*
2348  * Get the 64-bit version for an inode.
2349  */
2350 static int osd_object_version_get(const struct lu_env *env,
2351                                   struct dt_object *dt, dt_obj_version_t *ver)
2352 {
2353         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2354
2355         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2356                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2357         *ver = LDISKFS_I(inode)->i_fs_version;
2358         return 0;
2359 }
2360
2361 /*
2362  * Concurrency: @dt is read locked.
2363  */
2364 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2365                          struct lu_buf *buf, const char *name,
2366                          struct lustre_capa *capa)
2367 {
2368         struct osd_object      *obj    = osd_dt_obj(dt);
2369         struct inode           *inode  = obj->oo_inode;
2370         struct osd_thread_info *info   = osd_oti_get(env);
2371         struct dentry          *dentry = &info->oti_obj_dentry;
2372
2373         /* version get is not real XATTR but uses xattr API */
2374         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2375                 /* for version we are just using xattr API but change inode
2376                  * field instead */
2377                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2378                 osd_object_version_get(env, dt, buf->lb_buf);
2379                 return sizeof(dt_obj_version_t);
2380         }
2381
2382         LASSERT(dt_object_exists(dt));
2383         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2384         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2385
2386         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2387                 return -EACCES;
2388
2389         dentry->d_inode = inode;
2390         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2391 }
2392
2393
2394 static int osd_declare_xattr_set(const struct lu_env *env,
2395                                  struct dt_object *dt,
2396                                  const struct lu_buf *buf, const char *name,
2397                                  int fl, struct thandle *handle)
2398 {
2399         struct osd_thandle *oh;
2400
2401         LASSERT(handle != NULL);
2402
2403         oh = container_of0(handle, struct osd_thandle, ot_super);
2404         LASSERT(oh->ot_handle == NULL);
2405
2406         OSD_DECLARE_OP(oh, xattr_set);
2407         if (strcmp(name, XATTR_NAME_VERSION) == 0)
2408                 oh->ot_credits += osd_dto_credits_noquota[DTO_ATTR_SET_BASE];
2409         else
2410                 oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2411
2412         return 0;
2413 }
2414
2415 /*
2416  * Set the 64-bit version for object
2417  */
2418 static void osd_object_version_set(const struct lu_env *env,
2419                                    struct dt_object *dt,
2420                                    dt_obj_version_t *new_version)
2421 {
2422         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2423
2424         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2425                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2426
2427         LDISKFS_I(inode)->i_fs_version = *new_version;
2428         /** Version is set after all inode operations are finished,
2429          *  so we should mark it dirty here */
2430         inode->i_sb->s_op->dirty_inode(inode);
2431 }
2432
2433 /*
2434  * Concurrency: @dt is write locked.
2435  */
2436 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2437                          const struct lu_buf *buf, const char *name, int fl,
2438                          struct thandle *handle, struct lustre_capa *capa)
2439 {
2440         LASSERT(handle != NULL);
2441
2442         /* version set is not real XATTR */
2443         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2444                 /* for version we are just using xattr API but change inode
2445                  * field instead */
2446                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2447                 osd_object_version_set(env, dt, buf->lb_buf);
2448                 return sizeof(dt_obj_version_t);
2449         }
2450
2451         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2452                 return -EACCES;
2453
2454         OSD_EXEC_OP(handle, xattr_set);
2455         return __osd_xattr_set(env, dt, buf, name, fl);
2456 }
2457
2458 /*
2459  * Concurrency: @dt is read locked.
2460  */
2461 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2462                           struct lu_buf *buf, struct lustre_capa *capa)
2463 {
2464         struct osd_object      *obj    = osd_dt_obj(dt);
2465         struct inode           *inode  = obj->oo_inode;
2466         struct osd_thread_info *info   = osd_oti_get(env);
2467         struct dentry          *dentry = &info->oti_obj_dentry;
2468
2469         LASSERT(dt_object_exists(dt));
2470         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2471         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2472
2473         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2474                 return -EACCES;
2475
2476         dentry->d_inode = inode;
2477         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2478 }
2479
2480 static int osd_declare_xattr_del(const struct lu_env *env,
2481                                  struct dt_object *dt, const char *name,
2482                                  struct thandle *handle)
2483 {
2484         struct osd_thandle *oh;
2485
2486         LASSERT(dt_object_exists(dt));
2487         LASSERT(handle != NULL);
2488
2489         oh = container_of0(handle, struct osd_thandle, ot_super);
2490         LASSERT(oh->ot_handle == NULL);
2491
2492         OSD_DECLARE_OP(oh, xattr_set);
2493         oh->ot_credits += osd_dto_credits_noquota[DTO_XATTR_SET];
2494
2495         return 0;
2496 }
2497
2498 /*
2499  * Concurrency: @dt is write locked.
2500  */
2501 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2502                          const char *name, struct thandle *handle,
2503                          struct lustre_capa *capa)
2504 {
2505         struct osd_object      *obj    = osd_dt_obj(dt);
2506         struct inode           *inode  = obj->oo_inode;
2507         struct osd_thread_info *info   = osd_oti_get(env);
2508         struct dentry          *dentry = &info->oti_obj_dentry;
2509         int                     rc;
2510
2511         LASSERT(dt_object_exists(dt));
2512         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2513         LASSERT(osd_write_locked(env, obj));
2514         LASSERT(handle != NULL);
2515
2516         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2517                 return -EACCES;
2518
2519         OSD_EXEC_OP(handle, xattr_set);
2520
2521         dentry->d_inode = inode;
2522         rc = inode->i_op->removexattr(dentry, name);
2523         return rc;
2524 }
2525
2526 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2527                                      struct dt_object *dt,
2528                                      struct lustre_capa *old,
2529                                      __u64 opc)
2530 {
2531         struct osd_thread_info *info = osd_oti_get(env);
2532         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2533         struct osd_object *obj = osd_dt_obj(dt);
2534         struct osd_device *dev = osd_obj2dev(obj);
2535         struct lustre_capa_key *key = &info->oti_capa_key;
2536         struct lustre_capa *capa = &info->oti_capa;
2537         struct obd_capa *oc;
2538         struct md_capainfo *ci;
2539         int rc;
2540         ENTRY;
2541
2542         if (!dev->od_fl_capa)
2543                 RETURN(ERR_PTR(-ENOENT));
2544
2545         LASSERT(dt_object_exists(dt));
2546         LINVRNT(osd_invariant(obj));
2547
2548         /* renewal sanity check */
2549         if (old && osd_object_auth(env, dt, old, opc))
2550                 RETURN(ERR_PTR(-EACCES));
2551
2552         ci = md_capainfo(env);
2553         if (unlikely(!ci))
2554                 RETURN(ERR_PTR(-ENOENT));
2555
2556         switch (ci->mc_auth) {
2557         case LC_ID_NONE:
2558                 RETURN(NULL);
2559         case LC_ID_PLAIN:
2560                 capa->lc_uid = obj->oo_inode->i_uid;
2561                 capa->lc_gid = obj->oo_inode->i_gid;
2562                 capa->lc_flags = LC_ID_PLAIN;
2563                 break;
2564         case LC_ID_CONVERT: {
2565                 __u32 d[4], s[4];
2566
2567                 s[0] = obj->oo_inode->i_uid;
2568                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2569                 s[2] = obj->oo_inode->i_gid;
2570                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2571                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2572                 if (unlikely(rc))
2573                         RETURN(ERR_PTR(rc));
2574
2575                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2576                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2577                 capa->lc_flags = LC_ID_CONVERT;
2578                 break;
2579         }
2580         default:
2581                 RETURN(ERR_PTR(-EINVAL));
2582         }
2583
2584         capa->lc_fid = *fid;
2585         capa->lc_opc = opc;
2586         capa->lc_flags |= dev->od_capa_alg << 24;
2587         capa->lc_timeout = dev->od_capa_timeout;
2588         capa->lc_expiry = 0;
2589
2590         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2591         if (oc) {
2592                 LASSERT(!capa_is_expired(oc));
2593                 RETURN(oc);
2594         }
2595
2596         cfs_spin_lock(&capa_lock);
2597         *key = dev->od_capa_keys[1];
2598         cfs_spin_unlock(&capa_lock);
2599
2600         capa->lc_keyid = key->lk_keyid;
2601         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2602
2603         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2604         if (rc) {
2605                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2606                 RETURN(ERR_PTR(rc));
2607         }
2608
2609         oc = capa_add(dev->od_capa_hash, capa);
2610         RETURN(oc);
2611 }
2612
2613 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2614 {
2615         struct osd_object       *obj    = osd_dt_obj(dt);
2616         struct inode            *inode  = obj->oo_inode;
2617         struct osd_thread_info  *info   = osd_oti_get(env);
2618         struct dentry           *dentry = &info->oti_obj_dentry;
2619         struct file             *file   = &info->oti_file;
2620         int                     rc;
2621
2622         ENTRY;
2623
2624         dentry->d_inode = inode;
2625         file->f_dentry = dentry;
2626         file->f_mapping = inode->i_mapping;
2627         file->f_op = inode->i_fop;
2628         mutex_lock(&inode->i_mutex);
2629         rc = file->f_op->fsync(file, dentry, 0);
2630         mutex_unlock(&inode->i_mutex);
2631         RETURN(rc);
2632 }
2633
2634 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2635                         void **data)
2636 {
2637         struct osd_object *obj = osd_dt_obj(dt);
2638         ENTRY;
2639
2640         *data = (void *)obj->oo_inode;
2641         RETURN(0);
2642 }
2643
2644 /*
2645  * Index operations.
2646  */
2647
2648 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2649                            const struct dt_index_features *feat)
2650 {
2651         struct iam_descr *descr;
2652
2653         if (osd_object_is_root(o))
2654                 return feat == &dt_directory_features;
2655
2656         LASSERT(o->oo_dir != NULL);
2657
2658         descr = o->oo_dir->od_container.ic_descr;
2659         if (feat == &dt_directory_features) {
2660                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2661                         return 1;
2662                 else
2663                         return 0;
2664         } else {
2665                 return
2666                         feat->dif_keysize_min <= descr->id_key_size &&
2667                         descr->id_key_size <= feat->dif_keysize_max &&
2668                         feat->dif_recsize_min <= descr->id_rec_size &&
2669                         descr->id_rec_size <= feat->dif_recsize_max &&
2670                         !(feat->dif_flags & (DT_IND_VARKEY |
2671                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2672                         ergo(feat->dif_flags & DT_IND_UPDATE,
2673                              1 /* XXX check that object (and file system) is
2674                                 * writable */);
2675         }
2676 }
2677
2678 static int osd_iam_container_init(const struct lu_env *env,
2679                                   struct osd_object *obj,
2680                                   struct osd_directory *dir)
2681 {
2682         struct iam_container *bag = &dir->od_container;
2683         int result;
2684
2685         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2686         if (result != 0)
2687                 return result;
2688
2689         result = iam_container_setup(bag);
2690         if (result == 0)
2691                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2692         else
2693                 iam_container_fini(bag);
2694
2695         return result;
2696 }
2697
2698
2699 /*
2700  * Concurrency: no external locking is necessary.
2701  */
2702 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2703                          const struct dt_index_features *feat)
2704 {
2705         int                      result;
2706         int                      skip_iam = 0;
2707         struct osd_object       *obj = osd_dt_obj(dt);
2708         struct osd_device       *osd = osd_obj2dev(obj);
2709
2710         LINVRNT(osd_invariant(obj));
2711         LASSERT(dt_object_exists(dt));
2712
2713         if (osd_object_is_root(obj)) {
2714                 dt->do_index_ops = &osd_index_ea_ops;
2715                 result = 0;
2716         } else if (feat == &dt_directory_features && osd->od_iop_mode) {
2717                 dt->do_index_ops = &osd_index_ea_ops;
2718                 if (S_ISDIR(obj->oo_inode->i_mode))
2719                         result = 0;
2720                 else
2721                         result = -ENOTDIR;
2722                 skip_iam = 1;
2723         } else if (unlikely(feat == &dt_otable_features)) {
2724                 dt->do_index_ops = &osd_otable_ops;
2725                 return 0;
2726         } else if (feat == &dt_acct_features) {
2727                 dt->do_index_ops = &osd_acct_index_ops;
2728                 result = 0;
2729                 skip_iam = 1;
2730         } else if (!osd_has_index(obj)) {
2731                 struct osd_directory *dir;
2732
2733                 OBD_ALLOC_PTR(dir);
2734                 if (dir != NULL) {
2735
2736                         cfs_spin_lock(&obj->oo_guard);
2737                         if (obj->oo_dir == NULL)
2738                                 obj->oo_dir = dir;
2739                         else
2740                                 /*
2741                                  * Concurrent thread allocated container data.
2742                                  */
2743                                 OBD_FREE_PTR(dir);
2744                         cfs_spin_unlock(&obj->oo_guard);
2745                         /*
2746                          * Now, that we have container data, serialize its
2747                          * initialization.
2748                          */
2749                         cfs_down_write(&obj->oo_ext_idx_sem);
2750                         /*
2751                          * recheck under lock.
2752                          */
2753                         if (!osd_has_index(obj))
2754                                 result = osd_iam_container_init(env, obj, dir);
2755                         else
2756                                 result = 0;
2757                         cfs_up_write(&obj->oo_ext_idx_sem);
2758                 } else {
2759                         result = -ENOMEM;
2760                 }
2761         } else {
2762                 result = 0;
2763         }
2764
2765         if (result == 0 && skip_iam == 0) {
2766                 if (!osd_iam_index_probe(env, obj, feat))
2767                         result = -ENOTDIR;
2768         }
2769         LINVRNT(osd_invariant(obj));
2770
2771         return result;
2772 }
2773
2774 static int osd_otable_it_attr_get(const struct lu_env *env,
2775                                  struct dt_object *dt,
2776                                  struct lu_attr *attr,
2777                                  struct lustre_capa *capa)
2778 {
2779         attr->la_valid = 0;
2780         return 0;
2781 }
2782
2783 static const struct dt_object_operations osd_obj_ops = {
2784         .do_read_lock         = osd_object_read_lock,
2785         .do_write_lock        = osd_object_write_lock,
2786         .do_read_unlock       = osd_object_read_unlock,
2787         .do_write_unlock      = osd_object_write_unlock,
2788         .do_write_locked      = osd_object_write_locked,
2789         .do_attr_get          = osd_attr_get,
2790         .do_declare_attr_set  = osd_declare_attr_set,
2791         .do_attr_set          = osd_attr_set,
2792         .do_ah_init           = osd_ah_init,
2793         .do_declare_create    = osd_declare_object_create,
2794         .do_create            = osd_object_create,
2795         .do_declare_destroy   = osd_declare_object_destroy,
2796         .do_destroy           = osd_object_destroy,
2797         .do_index_try         = osd_index_try,
2798         .do_declare_ref_add   = osd_declare_object_ref_add,
2799         .do_ref_add           = osd_object_ref_add,
2800         .do_declare_ref_del   = osd_declare_object_ref_del,
2801         .do_ref_del           = osd_object_ref_del,
2802         .do_xattr_get         = osd_xattr_get,
2803         .do_declare_xattr_set = osd_declare_xattr_set,
2804         .do_xattr_set         = osd_xattr_set,
2805         .do_declare_xattr_del = osd_declare_xattr_del,
2806         .do_xattr_del         = osd_xattr_del,
2807         .do_xattr_list        = osd_xattr_list,
2808         .do_capa_get          = osd_capa_get,
2809         .do_object_sync       = osd_object_sync,
2810         .do_data_get          = osd_data_get,
2811 };
2812
2813 /**
2814  * dt_object_operations for interoperability mode
2815  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2816  */
2817 static const struct dt_object_operations osd_obj_ea_ops = {
2818         .do_read_lock         = osd_object_read_lock,
2819         .do_write_lock        = osd_object_write_lock,
2820         .do_read_unlock       = osd_object_read_unlock,
2821         .do_write_unlock      = osd_object_write_unlock,
2822         .do_write_locked      = osd_object_write_locked,
2823         .do_attr_get          = osd_attr_get,
2824         .do_declare_attr_set  = osd_declare_attr_set,
2825         .do_attr_set          = osd_attr_set,
2826         .do_ah_init           = osd_ah_init,
2827         .do_declare_create    = osd_declare_object_create,
2828         .do_create            = osd_object_ea_create,
2829         .do_declare_destroy   = osd_declare_object_destroy,
2830         .do_destroy           = osd_object_destroy,
2831         .do_index_try         = osd_index_try,
2832         .do_declare_ref_add   = osd_declare_object_ref_add,
2833         .do_ref_add           = osd_object_ref_add,
2834         .do_declare_ref_del   = osd_declare_object_ref_del,
2835         .do_ref_del           = osd_object_ref_del,
2836         .do_xattr_get         = osd_xattr_get,
2837         .do_declare_xattr_set = osd_declare_xattr_set,
2838         .do_xattr_set         = osd_xattr_set,
2839         .do_declare_xattr_del = osd_declare_xattr_del,
2840         .do_xattr_del         = osd_xattr_del,
2841         .do_xattr_list        = osd_xattr_list,
2842         .do_capa_get          = osd_capa_get,
2843         .do_object_sync       = osd_object_sync,
2844         .do_data_get          = osd_data_get,
2845 };
2846
2847 static const struct dt_object_operations osd_obj_otable_it_ops = {
2848         .do_attr_get    = osd_otable_it_attr_get,
2849         .do_index_try   = osd_index_try,
2850 };
2851
2852 static int osd_index_declare_iam_delete(const struct lu_env *env,
2853                                         struct dt_object *dt,
2854                                         const struct dt_key *key,
2855                                         struct thandle *handle)
2856 {
2857         struct osd_thandle    *oh;
2858
2859         oh = container_of0(handle, struct osd_thandle, ot_super);
2860         LASSERT(oh->ot_handle == NULL);
2861
2862         OSD_DECLARE_OP(oh, delete);
2863         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2864
2865         return 0;
2866 }
2867
2868 /**
2869  *      delete a (key, value) pair from index \a dt specified by \a key
2870  *
2871  *      \param  dt      osd index object
2872  *      \param  key     key for index
2873  *      \param  rec     record reference
2874  *      \param  handle  transaction handler
2875  *
2876  *      \retval  0  success
2877  *      \retval -ve   failure
2878  */
2879
2880 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2881                                 const struct dt_key *key,
2882                                 struct thandle *handle,
2883                                 struct lustre_capa *capa)
2884 {
2885         struct osd_object     *obj = osd_dt_obj(dt);
2886         struct osd_thandle    *oh;
2887         struct iam_path_descr *ipd;
2888         struct iam_container  *bag = &obj->oo_dir->od_container;
2889         int                    rc;
2890
2891         ENTRY;
2892
2893         LINVRNT(osd_invariant(obj));
2894         LASSERT(dt_object_exists(dt));
2895         LASSERT(bag->ic_object == obj->oo_inode);
2896         LASSERT(handle != NULL);
2897
2898         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2899                 RETURN(-EACCES);
2900
2901         OSD_EXEC_OP(handle, delete);
2902
2903         ipd = osd_idx_ipd_get(env, bag);
2904         if (unlikely(ipd == NULL))
2905                 RETURN(-ENOMEM);
2906
2907         oh = container_of0(handle, struct osd_thandle, ot_super);
2908         LASSERT(oh->ot_handle != NULL);
2909         LASSERT(oh->ot_handle->h_transaction != NULL);
2910
2911         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
2912         osd_ipd_put(env, bag, ipd);
2913         LINVRNT(osd_invariant(obj));
2914         RETURN(rc);
2915 }
2916
2917 static int osd_index_declare_ea_delete(const struct lu_env *env,
2918                                        struct dt_object *dt,
2919                                        const struct dt_key *key,
2920                                        struct thandle *handle)
2921 {
2922         struct osd_thandle *oh;
2923
2924         LASSERT(dt_object_exists(dt));
2925         LASSERT(handle != NULL);
2926
2927         oh = container_of0(handle, struct osd_thandle, ot_super);
2928         LASSERT(oh->ot_handle == NULL);
2929
2930         OSD_DECLARE_OP(oh, delete);
2931         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_DELETE];
2932
2933         LASSERT(osd_dt_obj(dt)->oo_inode);
2934         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
2935                         osd_dt_obj(dt)->oo_inode);
2936         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
2937                         osd_dt_obj(dt)->oo_inode);
2938
2939         return 0;
2940 }
2941
2942 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
2943                                           struct dt_rec *fid)
2944 {
2945         struct osd_fid_pack *rec;
2946         int                  rc = -ENODATA;
2947
2948         if (de->file_type & LDISKFS_DIRENT_LUFID) {
2949                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
2950                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
2951         }
2952         RETURN(rc);
2953 }
2954
2955 /**
2956  * Index delete function for interoperability mode (b11826).
2957  * It will remove the directory entry added by osd_index_ea_insert().
2958  * This entry is needed to maintain name->fid mapping.
2959  *
2960  * \param key,  key i.e. file entry to be deleted
2961  *
2962  * \retval   0, on success
2963  * \retval -ve, on error
2964  */
2965 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
2966                                const struct dt_key *key,
2967                                struct thandle *handle,
2968                                struct lustre_capa *capa)
2969 {
2970         struct osd_object          *obj    = osd_dt_obj(dt);
2971         struct inode               *dir    = obj->oo_inode;
2972         struct dentry              *dentry;
2973         struct osd_thandle         *oh;
2974         struct ldiskfs_dir_entry_2 *de;
2975         struct buffer_head         *bh;
2976         struct htree_lock          *hlock = NULL;
2977         int                         rc;
2978
2979         ENTRY;
2980
2981         LINVRNT(osd_invariant(obj));
2982         LASSERT(dt_object_exists(dt));
2983         LASSERT(handle != NULL);
2984
2985         OSD_EXEC_OP(handle, delete);
2986
2987         oh = container_of(handle, struct osd_thandle, ot_super);
2988         LASSERT(oh->ot_handle != NULL);
2989         LASSERT(oh->ot_handle->h_transaction != NULL);
2990
2991         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2992                 RETURN(-EACCES);
2993
2994         dentry = osd_child_dentry_get(env, obj,
2995                                       (char *)key, strlen((char *)key));
2996
2997         if (obj->oo_hl_head != NULL) {
2998                 hlock = osd_oti_get(env)->oti_hlock;
2999                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3000                                    dir, LDISKFS_HLOCK_DEL);
3001         } else {
3002                 cfs_down_write(&obj->oo_ext_idx_sem);
3003         }
3004
3005         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3006         if (bh) {
3007                 rc = ldiskfs_delete_entry(oh->ot_handle,
3008                                           dir, de, bh);
3009                 brelse(bh);
3010         } else {
3011                 rc = -ENOENT;
3012         }
3013         if (hlock != NULL)
3014                 ldiskfs_htree_unlock(hlock);
3015         else
3016                 cfs_up_write(&obj->oo_ext_idx_sem);
3017
3018         LASSERT(osd_invariant(obj));
3019         RETURN(rc);
3020 }
3021
3022 /**
3023  *      Lookup index for \a key and copy record to \a rec.
3024  *
3025  *      \param  dt      osd index object
3026  *      \param  key     key for index
3027  *      \param  rec     record reference
3028  *
3029  *      \retval  +ve  success : exact mach
3030  *      \retval  0    return record with key not greater than \a key
3031  *      \retval -ve   failure
3032  */
3033 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3034                                 struct dt_rec *rec, const struct dt_key *key,
3035                                 struct lustre_capa *capa)
3036 {
3037         struct osd_object      *obj = osd_dt_obj(dt);
3038         struct iam_path_descr  *ipd;
3039         struct iam_container   *bag = &obj->oo_dir->od_container;
3040         struct osd_thread_info *oti = osd_oti_get(env);
3041         struct iam_iterator    *it = &oti->oti_idx_it;
3042         struct iam_rec         *iam_rec;
3043         int                     rc;
3044
3045         ENTRY;
3046
3047         LASSERT(osd_invariant(obj));
3048         LASSERT(dt_object_exists(dt));
3049         LASSERT(bag->ic_object == obj->oo_inode);
3050
3051         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3052                 RETURN(-EACCES);
3053
3054         ipd = osd_idx_ipd_get(env, bag);
3055         if (IS_ERR(ipd))
3056                 RETURN(-ENOMEM);
3057
3058         /* got ipd now we can start iterator. */
3059         iam_it_init(it, bag, 0, ipd);
3060
3061         rc = iam_it_get(it, (struct iam_key *)key);
3062         if (rc >= 0) {
3063                 if (S_ISDIR(obj->oo_inode->i_mode))
3064                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3065                 else
3066                         iam_rec = (struct iam_rec *) rec;
3067
3068                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3069                 if (S_ISDIR(obj->oo_inode->i_mode))
3070                         osd_fid_unpack((struct lu_fid *) rec,
3071                                        (struct osd_fid_pack *)iam_rec);
3072         }
3073         iam_it_put(it);
3074         iam_it_fini(it);
3075         osd_ipd_put(env, bag, ipd);
3076
3077         LINVRNT(osd_invariant(obj));
3078
3079         RETURN(rc);
3080 }
3081
3082 static int osd_index_declare_iam_insert(const struct lu_env *env,
3083                                         struct dt_object *dt,
3084                                         const struct dt_rec *rec,
3085                                         const struct dt_key *key,
3086                                         struct thandle *handle)
3087 {
3088         struct osd_thandle *oh;
3089
3090         LASSERT(dt_object_exists(dt));
3091         LASSERT(handle != NULL);
3092
3093         oh = container_of0(handle, struct osd_thandle, ot_super);
3094         LASSERT(oh->ot_handle == NULL);
3095
3096         OSD_DECLARE_OP(oh, insert);
3097         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3098
3099         return 0;
3100 }
3101
3102 /**
3103  *      Inserts (key, value) pair in \a dt index object.
3104  *
3105  *      \param  dt      osd index object
3106  *      \param  key     key for index
3107  *      \param  rec     record reference
3108  *      \param  th      transaction handler
3109  *
3110  *      \retval  0  success
3111  *      \retval -ve failure
3112  */
3113 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3114                                 const struct dt_rec *rec,
3115                                 const struct dt_key *key, struct thandle *th,
3116                                 struct lustre_capa *capa, int ignore_quota)
3117 {
3118         struct osd_object     *obj = osd_dt_obj(dt);
3119         struct iam_path_descr *ipd;
3120         struct osd_thandle    *oh;
3121         struct iam_container  *bag = &obj->oo_dir->od_container;
3122 #ifdef HAVE_QUOTA_SUPPORT
3123         cfs_cap_t              save = cfs_curproc_cap_pack();
3124 #endif
3125         struct osd_thread_info *oti = osd_oti_get(env);
3126         struct iam_rec         *iam_rec = (struct iam_rec *)oti->oti_ldp;
3127         int                     rc;
3128
3129         ENTRY;
3130
3131         LINVRNT(osd_invariant(obj));
3132         LASSERT(dt_object_exists(dt));
3133         LASSERT(bag->ic_object == obj->oo_inode);
3134         LASSERT(th != NULL);
3135
3136         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3137                 RETURN(-EACCES);
3138
3139         OSD_EXEC_OP(th, insert);
3140
3141         ipd = osd_idx_ipd_get(env, bag);
3142         if (unlikely(ipd == NULL))
3143                 RETURN(-ENOMEM);
3144
3145         oh = container_of0(th, struct osd_thandle, ot_super);
3146         LASSERT(oh->ot_handle != NULL);
3147         LASSERT(oh->ot_handle->h_transaction != NULL);
3148 #ifdef HAVE_QUOTA_SUPPORT
3149         if (ignore_quota)
3150                 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3151         else
3152                 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3153 #endif
3154         if (S_ISDIR(obj->oo_inode->i_mode))
3155                 osd_fid_pack((struct osd_fid_pack *)iam_rec, rec, &oti->oti_fid);
3156         else
3157                 iam_rec = (struct iam_rec *) rec;
3158         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
3159                         iam_rec, ipd);
3160 #ifdef HAVE_QUOTA_SUPPORT
3161         cfs_curproc_cap_unpack(save);
3162 #endif
3163         osd_ipd_put(env, bag, ipd);
3164         LINVRNT(osd_invariant(obj));
3165         RETURN(rc);
3166 }
3167
3168 /**
3169  * Calls ldiskfs_add_entry() to add directory entry
3170  * into the directory. This is required for
3171  * interoperability mode (b11826)
3172  *
3173  * \retval   0, on success
3174  * \retval -ve, on error
3175  */
3176 static int __osd_ea_add_rec(struct osd_thread_info *info,
3177                             struct osd_object *pobj, struct inode  *cinode,
3178                             const char *name, const struct dt_rec *fid,
3179                             struct htree_lock *hlock, struct thandle *th)
3180 {
3181         struct ldiskfs_dentry_param *ldp;
3182         struct dentry               *child;
3183         struct osd_thandle          *oth;
3184         int                          rc;
3185
3186         oth = container_of(th, struct osd_thandle, ot_super);
3187         LASSERT(oth->ot_handle != NULL);
3188         LASSERT(oth->ot_handle->h_transaction != NULL);
3189
3190         child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
3191
3192         /* XXX: remove fid_is_igif() check here.
3193          * IGIF check is just to handle insertion of .. when it is 'ROOT',
3194          * it is IGIF now but needs FID in dir entry as well for readdir
3195          * to work.
3196          * LU-838 should fix that and remove fid_is_igif() check */
3197         if (fid_is_igif((struct lu_fid *)fid) ||
3198             fid_is_norm((struct lu_fid *)fid)) {
3199                 ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3200                 osd_get_ldiskfs_dirent_param(ldp, fid);
3201                 child->d_fsdata = (void *)ldp;
3202         } else {
3203                 child->d_fsdata = NULL;
3204         }
3205         rc = osd_ldiskfs_add_entry(oth->ot_handle, child, cinode, hlock);
3206
3207         RETURN(rc);
3208 }
3209
3210 /**
3211  * Calls ldiskfs_add_dot_dotdot() to add dot and dotdot entries
3212  * into the directory.Also sets flags into osd object to
3213  * indicate dot and dotdot are created. This is required for
3214  * interoperability mode (b11826)
3215  *
3216  * \param dir   directory for dot and dotdot fixup.
3217  * \param obj   child object for linking
3218  *
3219  * \retval   0, on success
3220  * \retval -ve, on error
3221  */
3222 static int osd_add_dot_dotdot(struct osd_thread_info *info,
3223                               struct osd_object *dir,
3224                               struct inode  *parent_dir, const char *name,
3225                               const struct dt_rec *dot_fid,
3226                               const struct dt_rec *dot_dot_fid,
3227                               struct thandle *th)
3228 {
3229         struct inode                *inode = dir->oo_inode;
3230         struct ldiskfs_dentry_param *dot_ldp;
3231         struct ldiskfs_dentry_param *dot_dot_ldp;
3232         struct osd_thandle          *oth;
3233         int result = 0;
3234
3235         oth = container_of(th, struct osd_thandle, ot_super);
3236         LASSERT(oth->ot_handle->h_transaction != NULL);
3237         LASSERT(S_ISDIR(dir->oo_inode->i_mode));
3238
3239         if (strcmp(name, dot) == 0) {
3240                 if (dir->oo_compat_dot_created) {
3241                         result = -EEXIST;
3242                 } else {
3243                         LASSERT(inode == parent_dir);
3244                         dir->oo_compat_dot_created = 1;
3245                         result = 0;
3246                 }
3247         } else if(strcmp(name, dotdot) == 0) {
3248                 dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
3249                 dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
3250
3251                 if (!dir->oo_compat_dot_created)
3252                         return -EINVAL;
3253                 if (!fid_is_igif((struct lu_fid *)dot_fid)) {
3254                         osd_get_ldiskfs_dirent_param(dot_ldp, dot_fid);
3255                         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
3256                 } else {
3257                         dot_ldp = NULL;
3258                         dot_dot_ldp = NULL;
3259                 }
3260                 /* in case of rename, dotdot is already created */
3261                 if (dir->oo_compat_dotdot_created) {
3262                         return __osd_ea_add_rec(info, dir, parent_dir, name,
3263                                                 dot_dot_fid, NULL, th);
3264                 }
3265
3266                 result = ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
3267                                                 inode, dot_ldp, dot_dot_ldp);
3268                 if (result == 0)
3269                        dir->oo_compat_dotdot_created = 1;
3270         }
3271
3272         return result;
3273 }
3274
3275
3276 /**
3277  * It will call the appropriate osd_add* function and return the
3278  * value, return by respective functions.
3279  */
3280 static int osd_ea_add_rec(const struct lu_env *env, struct osd_object *pobj,
3281                           struct inode *cinode, const char *name,
3282                           const struct dt_rec *fid, struct thandle *th)
3283 {
3284         struct osd_thread_info *info   = osd_oti_get(env);
3285         struct htree_lock      *hlock;
3286         int                     rc;
3287
3288         hlock = pobj->oo_hl_head != NULL ? info->oti_hlock : NULL;
3289
3290         if (name[0] == '.' && (name[1] == '\0' || (name[1] == '.' &&
3291                                                    name[2] =='\0'))) {
3292                 if (hlock != NULL) {
3293                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3294                                            pobj->oo_inode, 0);
3295                 } else {
3296                         cfs_down_write(&pobj->oo_ext_idx_sem);
3297                 }
3298                 rc = osd_add_dot_dotdot(info, pobj, cinode, name,
3299                      (struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
3300                                         fid, th);
3301         } else {
3302                 if (hlock != NULL) {
3303                         ldiskfs_htree_lock(hlock, pobj->oo_hl_head,
3304                                            pobj->oo_inode, LDISKFS_HLOCK_ADD);
3305                 } else {
3306                         cfs_down_write(&pobj->oo_ext_idx_sem);
3307                 }
3308
3309                 rc = __osd_ea_add_rec(info, pobj, cinode, name, fid,
3310                                       hlock, th);
3311         }
3312         if (hlock != NULL)
3313                 ldiskfs_htree_unlock(hlock);
3314         else
3315                 cfs_up_write(&pobj->oo_ext_idx_sem);
3316
3317         return rc;
3318 }
3319
3320 static int
3321 osd_consistency_check(struct osd_thread_info *oti, struct osd_device *dev,
3322                       struct osd_idmap_cache *oic)
3323 {
3324         struct osd_scrub    *scrub = &dev->od_scrub;
3325         struct lu_fid       *fid   = &oic->oic_fid;
3326         struct osd_inode_id *id    = &oti->oti_id;
3327         int                  once  = 0;
3328         int                  rc;
3329         ENTRY;
3330
3331 again:
3332         rc = osd_oi_lookup(oti, dev, fid, id);
3333         if (rc != 0 && rc != -ENOENT)
3334                 RETURN(rc);
3335
3336         if (rc == 0 && osd_id_eq(id, &oic->oic_lid))
3337                 RETURN(0);
3338
3339         if (thread_is_running(&scrub->os_thread)) {
3340                 rc = osd_oii_insert(dev, oic, rc == -ENOENT);
3341                 /* There is race condition between osd_oi_lookup and OI scrub.
3342                  * The OI scrub finished just after osd_oi_lookup() failure.
3343                  * Under such case, it is unnecessary to trigger OI scrub again,
3344                  * but try to call osd_oi_lookup() again. */
3345                 if (unlikely(rc == -EAGAIN))
3346                         goto again;
3347
3348                 RETURN(rc);
3349         }
3350
3351         if (!scrub->os_no_scrub && ++once == 1) {
3352                 CDEBUG(D_LFSCK, "Trigger OI scrub by RPC for "DFID"\n",
3353                        PFID(fid));
3354                 rc = osd_scrub_start(dev);
3355                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC for "DFID
3356                                ", rc = %d [2]\n",
3357                                LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
3358                                PFID(fid), rc);
3359                 if (rc == 0)
3360                         goto again;
3361         }
3362
3363         RETURN(rc = -EREMCHG);
3364 }
3365
3366 /**
3367  * Calls ->lookup() to find dentry. From dentry get inode and
3368  * read inode's ea to get fid. This is required for  interoperability
3369  * mode (b11826)
3370  *
3371  * \retval   0, on success
3372  * \retval -ve, on error
3373  */
3374 static int osd_ea_lookup_rec(const struct lu_env *env, struct osd_object *obj,
3375                              struct dt_rec *rec, const struct dt_key *key)
3376 {
3377         struct inode               *dir    = obj->oo_inode;
3378         struct dentry              *dentry;
3379         struct ldiskfs_dir_entry_2 *de;
3380         struct buffer_head         *bh;
3381         struct lu_fid              *fid = (struct lu_fid *) rec;
3382         struct htree_lock          *hlock = NULL;
3383         int                         ino;
3384         int                         rc;
3385
3386         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
3387
3388         dentry = osd_child_dentry_get(env, obj,
3389                                       (char *)key, strlen((char *)key));
3390
3391         if (obj->oo_hl_head != NULL) {
3392                 hlock = osd_oti_get(env)->oti_hlock;
3393                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3394                                    dir, LDISKFS_HLOCK_LOOKUP);
3395         } else {
3396                 cfs_down_read(&obj->oo_ext_idx_sem);
3397         }
3398
3399         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3400         if (bh) {
3401                 struct osd_thread_info *oti = osd_oti_get(env);
3402                 struct osd_idmap_cache *oic = &oti->oti_cache;
3403                 struct osd_device *dev = osd_obj2dev(obj);
3404                 struct osd_scrub *scrub = &dev->od_scrub;
3405                 struct scrub_file *sf = &scrub->os_file;
3406
3407                 ino = le32_to_cpu(de->inode);
3408                 rc = osd_get_fid_from_dentry(de, rec);
3409
3410                 /* done with de, release bh */
3411                 brelse(bh);
3412                 if (rc != 0)
3413                         rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
3414                 else
3415                         osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
3416                 if (rc != 0 || !fid_is_norm(fid)) {
3417                         fid_zero(&oic->oic_fid);
3418                         GOTO(out, rc);
3419                 }
3420
3421                 oic->oic_fid = *fid;
3422                 if ((scrub->os_pos_current <= ino) &&
3423                     (sf->sf_flags & SF_INCONSISTENT ||
3424                      ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
3425                                       sf->sf_oi_bitmap)))
3426                         rc = osd_consistency_check(oti, dev, oic);
3427         } else {
3428                 rc = -ENOENT;
3429         }
3430
3431         GOTO(out, rc);
3432
3433 out:
3434         if (hlock != NULL)
3435                 ldiskfs_htree_unlock(hlock);
3436         else
3437                 cfs_up_read(&obj->oo_ext_idx_sem);
3438         return rc;
3439 }
3440
3441 /**
3442  * Find the osd object for given fid.
3443  *
3444  * \param fid need to find the osd object having this fid
3445  *
3446  * \retval osd_object on success
3447  * \retval        -ve on error
3448  */
3449 struct osd_object *osd_object_find(const struct lu_env *env,
3450                                    struct dt_object *dt,
3451                                    const struct lu_fid *fid)
3452 {
3453         struct lu_device  *ludev = dt->do_lu.lo_dev;
3454         struct osd_object *child = NULL;
3455         struct lu_object  *luch;
3456         struct lu_object  *lo;
3457
3458         luch = lu_object_find(env, ludev, fid, NULL);
3459         if (!IS_ERR(luch)) {
3460                 if (lu_object_exists(luch)) {
3461                         lo = lu_object_locate(luch->lo_header, ludev->ld_type);
3462                         if (lo != NULL)
3463                                 child = osd_obj(lo);
3464                         else
3465                                 LU_OBJECT_DEBUG(D_ERROR, env, luch,
3466                                                 "lu_object can't be located"
3467                                                 DFID"\n", PFID(fid));
3468
3469                         if (child == NULL) {
3470                                 lu_object_put(env, luch);
3471                                 CERROR("Unable to get osd_object\n");
3472                                 child = ERR_PTR(-ENOENT);
3473                         }
3474                 } else {
3475                         LU_OBJECT_DEBUG(D_ERROR, env, luch,
3476                                         "lu_object does not exists "DFID"\n",
3477                                         PFID(fid));
3478                         lu_object_put(env, luch);
3479                         child = ERR_PTR(-ENOENT);
3480                 }
3481         } else
3482                 child = (void *)luch;
3483
3484         return child;
3485 }
3486
3487 /**
3488  * Put the osd object once done with it.
3489  *
3490  * \param obj osd object that needs to be put
3491  */
3492 static inline void osd_object_put(const struct lu_env *env,
3493                                   struct osd_object *obj)
3494 {
3495         lu_object_put(env, &obj->oo_dt.do_lu);
3496 }
3497
3498 static int osd_index_declare_ea_insert(const struct lu_env *env,
3499                                        struct dt_object *dt,
3500                                        const struct dt_rec *rec,
3501                                        const struct dt_key *key,
3502                                        struct thandle *handle)
3503 {
3504         struct osd_thandle *oh;
3505
3506         LASSERT(dt_object_exists(dt));
3507         LASSERT(handle != NULL);
3508
3509         oh = container_of0(handle, struct osd_thandle, ot_super);
3510         LASSERT(oh->ot_handle == NULL);
3511
3512         OSD_DECLARE_OP(oh, insert);
3513         oh->ot_credits += osd_dto_credits_noquota[DTO_INDEX_INSERT];
3514
3515         LASSERT(osd_dt_obj(dt)->oo_inode);
3516         osd_declare_qid(dt, oh, USRQUOTA, osd_dt_obj(dt)->oo_inode->i_uid,
3517                         osd_dt_obj(dt)->oo_inode);
3518         osd_declare_qid(dt, oh, GRPQUOTA, osd_dt_obj(dt)->oo_inode->i_gid,
3519                         osd_dt_obj(dt)->oo_inode);
3520
3521         return 0;
3522 }
3523
3524 /**
3525  * Index add function for interoperability mode (b11826).
3526  * It will add the directory entry.This entry is needed to
3527  * maintain name->fid mapping.
3528  *
3529  * \param key it is key i.e. file entry to be inserted
3530  * \param rec it is value of given key i.e. fid
3531  *
3532  * \retval   0, on success
3533  * \retval -ve, on error
3534  */
3535 static int osd_index_ea_insert(const struct lu_env *env, struct dt_object *dt,
3536                                const struct dt_rec *rec,
3537                                const struct dt_key *key, struct thandle *th,
3538                                struct lustre_capa *capa, int ignore_quota)
3539 {
3540         struct osd_object *obj   = osd_dt_obj(dt);
3541         struct lu_fid     *fid   = (struct lu_fid *) rec;
3542         const char        *name  = (const char *)key;
3543         struct osd_object *child;
3544 #ifdef HAVE_QUOTA_SUPPORT
3545         cfs_cap_t          save  = cfs_curproc_cap_pack();
3546 #endif
3547         int                rc;
3548
3549         ENTRY;
3550
3551         LASSERT(osd_invariant(obj));
3552         LASSERT(dt_object_exists(dt));
3553         LASSERT(th != NULL);
3554
3555         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
3556                 RETURN(-EACCES);
3557
3558         child = osd_object_find(env, dt, fid);
3559         if (!IS_ERR(child)) {
3560 #ifdef HAVE_QUOTA_SUPPORT
3561                 if (ignore_quota)
3562                         cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
3563                 else
3564                         cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
3565 #endif
3566                 rc = osd_ea_add_rec(env, obj, child->oo_inode, name, rec, th);
3567 #ifdef HAVE_QUOTA_SUPPORT
3568                 cfs_curproc_cap_unpack(save);
3569 #endif
3570                 osd_object_put(env, child);
3571         } else {
3572                 rc = PTR_ERR(child);
3573         }
3574
3575         LASSERT(osd_invariant(obj));
3576         RETURN(rc);
3577 }
3578
3579 /**
3580  *  Initialize osd Iterator for given osd index object.
3581  *
3582  *  \param  dt      osd index object
3583  */
3584
3585 static struct dt_it *osd_it_iam_init(const struct lu_env *env,
3586                                      struct dt_object *dt,
3587                                      __u32 unused,
3588                                      struct lustre_capa *capa)
3589 {
3590         struct osd_it_iam      *it;
3591         struct osd_thread_info *oti = osd_oti_get(env);
3592         struct osd_object      *obj = osd_dt_obj(dt);
3593         struct lu_object       *lo  = &dt->do_lu;
3594         struct iam_path_descr  *ipd;
3595         struct iam_container   *bag = &obj->oo_dir->od_container;
3596
3597         LASSERT(lu_object_exists(lo));
3598
3599         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
3600                 return ERR_PTR(-EACCES);
3601
3602         it = &oti->oti_it;
3603         ipd = osd_it_ipd_get(env, bag);
3604         if (likely(ipd != NULL)) {
3605                 it->oi_obj = obj;
3606                 it->oi_ipd = ipd;
3607                 lu_object_get(lo);
3608                 iam_it_init(&it->oi_it, bag, IAM_IT_MOVE, ipd);
3609                 return (struct dt_it *)it;
3610         }
3611         return ERR_PTR(-ENOMEM);
3612 }
3613
3614 /**
3615  * free given Iterator.
3616  */
3617
3618 static void osd_it_iam_fini(const struct lu_env *env, struct dt_it *di)
3619 {
3620         struct osd_it_iam *it = (struct osd_it_iam *)di;
3621         struct osd_object *obj = it->oi_obj;
3622
3623         iam_it_fini(&it->oi_it);
3624         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
3625         lu_object_put(env, &obj->oo_dt.do_lu);
3626 }
3627
3628 /**
3629  *  Move Iterator to record specified by \a key
3630  *
3631  *  \param  di      osd iterator
3632  *  \param  key     key for index
3633  *
3634  *  \retval +ve  di points to record with least key not larger than key
3635  *  \retval  0   di points to exact matched key
3636  *  \retval -ve  failure
3637  */
3638
3639 static int osd_it_iam_get(const struct lu_env *env,
3640                           struct dt_it *di, const struct dt_key *key)
3641 {
3642         struct osd_it_iam *it = (struct osd_it_iam *)di;
3643
3644         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
3645 }
3646
3647 /**
3648  *  Release Iterator
3649  *
3650  *  \param  di      osd iterator
3651  */
3652
3653 static void osd_it_iam_put(const struct lu_env *env, struct dt_it *di)
3654 {
3655         struct osd_it_iam *it = (struct osd_it_iam *)di;
3656
3657         iam_it_put(&it->oi_it);
3658 }
3659
3660 /**
3661  *  Move iterator by one record
3662  *
3663  *  \param  di      osd iterator
3664  *
3665  *  \retval +1   end of container reached
3666  *  \retval  0   success
3667  *  \retval -ve  failure
3668  */
3669
3670 static int osd_it_iam_next(const struct lu_env *env, struct dt_it *di)
3671 {
3672         struct osd_it_iam *it = (struct osd_it_iam *)di;
3673
3674         return iam_it_next(&it->oi_it);
3675 }
3676
3677 /**
3678  * Return pointer to the key under iterator.
3679  */
3680
3681 static struct dt_key *osd_it_iam_key(const struct lu_env *env,
3682                                  const struct dt_it *di)
3683 {
3684         struct osd_it_iam *it = (struct osd_it_iam *)di;
3685
3686         return (struct dt_key *)iam_it_key_get(&it->oi_it);
3687 }
3688
3689 /**
3690  * Return size of key under iterator (in bytes)
3691  */
3692
3693 static int osd_it_iam_key_size(const struct lu_env *env, const struct dt_it *di)
3694 {
3695         struct osd_it_iam *it = (struct osd_it_iam *)di;
3696
3697         return iam_it_key_size(&it->oi_it);
3698 }
3699
3700 static inline void osd_it_append_attrs(struct lu_dirent *ent, __u32 attr,
3701                                        int len, __u16 type)
3702 {
3703         struct luda_type *lt;
3704         const unsigned    align = sizeof(struct luda_type) - 1;
3705
3706         /* check if file type is required */
3707         if (attr & LUDA_TYPE) {
3708                         len = (len + align) & ~align;
3709
3710                         lt = (void *) ent->lde_name + len;
3711                         lt->lt_type = cpu_to_le16(CFS_DTTOIF(type));
3712                         ent->lde_attrs |= LUDA_TYPE;
3713         }
3714
3715         ent->lde_attrs = cpu_to_le32(ent->lde_attrs);
3716 }
3717
3718 /**
3719  * build lu direct from backend fs dirent.
3720  */
3721
3722 static inline void osd_it_pack_dirent(struct lu_dirent *ent,
3723                                       struct lu_fid *fid, __u64 offset,
3724                                       char *name, __u16 namelen,
3725                                       __u16 type, __u32 attr)
3726 {
3727         fid_cpu_to_le(&ent->lde_fid, fid);
3728         ent->lde_attrs = LUDA_FID;
3729
3730         ent->lde_hash = cpu_to_le64(offset);
3731         ent->lde_reclen = cpu_to_le16(lu_dirent_calc_size(namelen, attr));
3732
3733         strncpy(ent->lde_name, name, namelen);
3734         ent->lde_namelen = cpu_to_le16(namelen);
3735
3736         /* append lustre attributes */
3737         osd_it_append_attrs(ent, attr, namelen, type);
3738 }
3739
3740 /**
3741  * Return pointer to the record under iterator.
3742  */
3743 static int osd_it_iam_rec(const struct lu_env *env,
3744                           const struct dt_it *di,
3745                           struct dt_rec *dtrec, __u32 attr)
3746 {
3747         struct osd_it_iam      *it   = (struct osd_it_iam *)di;
3748         struct osd_thread_info *info = osd_oti_get(env);
3749         ENTRY;
3750
3751         if (S_ISDIR(it->oi_obj->oo_inode->i_mode)) {
3752                 const struct osd_fid_pack *rec;
3753                 struct lu_fid             *fid = &info->oti_fid;
3754                 struct lu_dirent          *lde = (struct lu_dirent *)dtrec;
3755                 char                      *name;
3756                 int                        namelen;
3757                 __u64                      hash;
3758                 int                        rc;
3759
3760                 name = (char *)iam_it_key_get(&it->oi_it);
3761                 if (IS_ERR(name))
3762                         RETURN(PTR_ERR(name));
3763
3764                 namelen = iam_it_key_size(&it->oi_it);
3765
3766                 rec = (const struct osd_fid_pack *)iam_it_rec_get(&it->oi_it);
3767                 if (IS_ERR(rec))
3768                         RETURN(PTR_ERR(rec));
3769
3770                 rc = osd_fid_unpack(fid, rec);
3771                 if (rc)
3772                         RETURN(rc);
3773
3774                 hash = iam_it_store(&it->oi_it);
3775
3776                 /* IAM does not store object type in IAM index (dir) */
3777                 osd_it_pack_dirent(lde, fid, hash, name, namelen,
3778                                    0, LUDA_FID);
3779         } else {
3780                 iam_reccpy(&it->oi_it.ii_path.ip_leaf,
3781                            (struct iam_rec *)dtrec);
3782         }
3783
3784         RETURN(0);
3785 }
3786
3787 /**
3788  * Returns cookie for current Iterator position.
3789  */
3790 static __u64 osd_it_iam_store(const struct lu_env *env, const struct dt_it *di)
3791 {
3792         struct osd_it_iam *it = (struct osd_it_iam *)di;
3793
3794         return iam_it_store(&it->oi_it);
3795 }
3796
3797 /**
3798  * Restore iterator from cookie.
3799  *
3800  * \param  di      osd iterator
3801  * \param  hash    Iterator location cookie
3802  *
3803  * \retval +ve  di points to record with least key not larger than key.
3804  * \retval  0   di points to exact matched key
3805  * \retval -ve  failure
3806  */
3807
3808 static int osd_it_iam_load(const struct lu_env *env,
3809                            const struct dt_it *di, __u64 hash)
3810 {
3811         struct osd_it_iam *it = (struct osd_it_iam *)di;
3812
3813         return iam_it_load(&it->oi_it, hash);
3814 }
3815
3816 static const struct dt_index_operations osd_index_iam_ops = {
3817         .dio_lookup         = osd_index_iam_lookup,
3818         .dio_declare_insert = osd_index_declare_iam_insert,
3819         .dio_insert         = osd_index_iam_insert,
3820         .dio_declare_delete = osd_index_declare_iam_delete,
3821         .dio_delete         = osd_index_iam_delete,
3822         .dio_it     = {
3823                 .init     = osd_it_iam_init,
3824                 .fini     = osd_it_iam_fini,
3825                 .get      = osd_it_iam_get,
3826                 .put      = osd_it_iam_put,
3827                 .next     = osd_it_iam_next,
3828                 .key      = osd_it_iam_key,
3829                 .key_size = osd_it_iam_key_size,
3830                 .rec      = osd_it_iam_rec,
3831                 .store    = osd_it_iam_store,
3832                 .load     = osd_it_iam_load
3833         }
3834 };
3835
3836 /**
3837  * Creates or initializes iterator context.
3838  *
3839  * \retval struct osd_it_ea, iterator structure on success
3840  *
3841  */
3842 static struct dt_it *osd_it_ea_init(const struct lu_env *env,
3843                                     struct dt_object *dt,
3844                                     __u32 attr,
3845                                     struct lustre_capa *capa)
3846 {
3847         struct osd_object       *obj  = osd_dt_obj(dt);
3848         struct osd_thread_info  *info = osd_oti_get(env);
3849         struct osd_it_ea        *it   = &info->oti_it_ea;
3850         struct lu_object        *lo   = &dt->do_lu;
3851         struct dentry           *obj_dentry = &info->oti_it_dentry;
3852         ENTRY;
3853         LASSERT(lu_object_exists(lo));
3854
3855         obj_dentry->d_inode = obj->oo_inode;
3856         obj_dentry->d_sb = osd_sb(osd_obj2dev(obj));
3857         obj_dentry->d_name.hash = 0;
3858
3859         it->oie_rd_dirent       = 0;
3860         it->oie_it_dirent       = 0;
3861         it->oie_dirent          = NULL;
3862         it->oie_buf             = info->oti_it_ea_buf;
3863         it->oie_obj             = obj;
3864         it->oie_file.f_pos      = 0;
3865         it->oie_file.f_dentry   = obj_dentry;
3866         if (attr & LUDA_64BITHASH)
3867                 it->oie_file.f_mode |= FMODE_64BITHASH;
3868         else
3869                 it->oie_file.f_mode |= FMODE_32BITHASH;
3870         it->oie_file.f_mapping    = obj->oo_inode->i_mapping;
3871         it->oie_file.f_op         = obj->oo_inode->i_fop;
3872         it->oie_file.private_data = NULL;
3873         lu_object_get(lo);
3874         RETURN((struct dt_it *) it);
3875 }
3876
3877 /**
3878  * Destroy or finishes iterator context.
3879  *
3880  * \param di iterator structure to be destroyed
3881  */
3882 static void osd_it_ea_fini(const struct lu_env *env, struct dt_it *di)
3883 {
3884         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3885         struct osd_object    *obj  = it->oie_obj;
3886         struct inode       *inode  = obj->oo_inode;
3887
3888         ENTRY;
3889         it->oie_file.f_op->release(inode, &it->oie_file);
3890         lu_object_put(env, &obj->oo_dt.do_lu);
3891         EXIT;
3892 }
3893
3894 /**
3895  * It position the iterator at given key, so that next lookup continues from
3896  * that key Or it is similar to dio_it->load() but based on a key,
3897  * rather than file position.
3898  *
3899  * As a special convention, osd_it_ea_get(env, di, "") has to rewind iterator
3900  * to the beginning.
3901  *
3902  * TODO: Presently return +1 considering it is only used by mdd_dir_is_empty().
3903  */
3904 static int osd_it_ea_get(const struct lu_env *env,
3905                          struct dt_it *di, const struct dt_key *key)
3906 {
3907         struct osd_it_ea     *it   = (struct osd_it_ea *)di;
3908
3909         ENTRY;
3910         LASSERT(((const char *)key)[0] == '\0');
3911         it->oie_file.f_pos      = 0;
3912         it->oie_rd_dirent       = 0;
3913         it->oie_it_dirent       = 0;
3914         it->oie_dirent          = NULL;
3915
3916         RETURN(+1);
3917 }
3918
3919 /**
3920  * Does nothing
3921  */
3922 static void osd_it_ea_put(const struct lu_env *env, struct dt_it *di)
3923 {
3924 }
3925
3926 /**
3927  * It is called internally by ->readdir(). It fills the
3928  * iterator's in-memory data structure with required
3929  * information i.e. name, namelen, rec_size etc.
3930  *
3931  * \param buf in which information to be filled in.
3932  * \param name name of the file in given dir
3933  *
3934  * \retval 0 on success
3935  * \retval 1 on buffer full
3936  */
3937 static int osd_ldiskfs_filldir(char *buf, const char *name, int namelen,
3938                                loff_t offset, __u64 ino,
3939                                unsigned d_type)
3940 {
3941         struct osd_it_ea        *it   = (struct osd_it_ea *)buf;
3942         struct osd_it_ea_dirent *ent  = it->oie_dirent;
3943         struct lu_fid           *fid  = &ent->oied_fid;
3944         struct osd_fid_pack     *rec;
3945         ENTRY;
3946
3947         /* this should never happen */
3948         if (unlikely(namelen == 0 || namelen > LDISKFS_NAME_LEN)) {
3949                 CERROR("ldiskfs return invalid namelen %d\n", namelen);
3950                 RETURN(-EIO);
3951         }
3952
3953         if ((void *) ent - it->oie_buf + sizeof(*ent) + namelen >
3954             OSD_IT_EA_BUFSIZE)
3955                 RETURN(1);
3956
3957         if (d_type & LDISKFS_DIRENT_LUFID) {
3958                 rec = (struct osd_fid_pack*) (name + namelen + 1);
3959
3960                 if (osd_fid_unpack(fid, rec) != 0)
3961                         fid_zero(fid);
3962
3963                 d_type &= ~LDISKFS_DIRENT_LUFID;
3964         } else {
3965                 fid_zero(fid);
3966         }
3967
3968         ent->oied_ino     = ino;
3969         ent->oied_off     = offset;
3970         ent->oied_namelen = namelen;
3971         ent->oied_type    = d_type;
3972
3973         memcpy(ent->oied_name, name, namelen);
3974
3975         it->oie_rd_dirent++;
3976         it->oie_dirent = (void *) ent + cfs_size_round(sizeof(*ent) + namelen);
3977         RETURN(0);
3978 }
3979
3980 /**
3981  * Calls ->readdir() to load a directory entry at a time
3982  * and stored it in iterator's in-memory data structure.
3983  *
3984  * \param di iterator's in memory structure
3985  *
3986  * \retval   0 on success
3987  * \retval -ve on error
3988  */
3989 static int osd_ldiskfs_it_fill(const struct lu_env *env,
3990                                const struct dt_it *di)
3991 {
3992         struct osd_it_ea   *it    = (struct osd_it_ea *)di;
3993         struct osd_object  *obj   = it->oie_obj;
3994         struct inode       *inode = obj->oo_inode;
3995         struct htree_lock  *hlock = NULL;
3996         int                 result = 0;
3997
3998         ENTRY;
3999         it->oie_dirent = it->oie_buf;
4000         it->oie_rd_dirent = 0;
4001
4002         if (obj->oo_hl_head != NULL) {
4003                 hlock = osd_oti_get(env)->oti_hlock;
4004                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
4005                                    inode, LDISKFS_HLOCK_READDIR);
4006         } else {
4007                 cfs_down_read(&obj->oo_ext_idx_sem);
4008         }
4009
4010         result = inode->i_fop->readdir(&it->oie_file, it,
4011                                        (filldir_t) osd_ldiskfs_filldir);
4012
4013         if (hlock != NULL)
4014                 ldiskfs_htree_unlock(hlock);
4015         else
4016                 cfs_up_read(&obj->oo_ext_idx_sem);
4017
4018         if (it->oie_rd_dirent == 0) {
4019                 result = -EIO;
4020         } else {
4021                 it->oie_dirent = it->oie_buf;
4022                 it->oie_it_dirent = 1;
4023         }
4024
4025         RETURN(result);
4026 }
4027
4028 /**
4029  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4030  * to load a directory entry at a time and stored it in
4031  * iterator's in-memory data structure.
4032  *
4033  * \param di iterator's in memory structure
4034  *
4035  * \retval +ve iterator reached to end
4036  * \retval   0 iterator not reached to end
4037  * \retval -ve on error
4038  */
4039 static int osd_it_ea_next(const struct lu_env *env, struct dt_it *di)
4040 {
4041         struct osd_it_ea *it = (struct osd_it_ea *)di;
4042         int rc;
4043
4044         ENTRY;
4045
4046         if (it->oie_it_dirent < it->oie_rd_dirent) {
4047                 it->oie_dirent =
4048                         (void *) it->oie_dirent +
4049                         cfs_size_round(sizeof(struct osd_it_ea_dirent) +
4050                                        it->oie_dirent->oied_namelen);
4051                 it->oie_it_dirent++;
4052                 RETURN(0);
4053         } else {
4054                 if (it->oie_file.f_pos == ldiskfs_get_htree_eof(&it->oie_file))
4055                         rc = +1;
4056                 else
4057                         rc = osd_ldiskfs_it_fill(env, di);
4058         }
4059
4060         RETURN(rc);
4061 }
4062
4063 /**
4064  * Returns the key at current position from iterator's in memory structure.
4065  *
4066  * \param di iterator's in memory structure
4067  *
4068  * \retval key i.e. struct dt_key on success
4069  */
4070 static struct dt_key *osd_it_ea_key(const struct lu_env *env,
4071                                     const struct dt_it *di)
4072 {
4073         struct osd_it_ea *it = (struct osd_it_ea *)di;
4074
4075         return (struct dt_key *)it->oie_dirent->oied_name;
4076 }
4077
4078 /**
4079  * Returns the key's size at current position from iterator's in memory structure.
4080  *
4081  * \param di iterator's in memory structure
4082  *
4083  * \retval key_size i.e. struct dt_key on success
4084  */
4085 static int osd_it_ea_key_size(const struct lu_env *env, const struct dt_it *di)
4086 {
4087         struct osd_it_ea *it = (struct osd_it_ea *)di;
4088
4089         return it->oie_dirent->oied_namelen;
4090 }
4091
4092
4093 /**
4094  * Returns the value (i.e. fid/igif) at current position from iterator's
4095  * in memory structure.
4096  *
4097  * \param di struct osd_it_ea, iterator's in memory structure
4098  * \param attr attr requested for dirent.
4099  * \param lde lustre dirent
4100  *
4101  * \retval   0 no error and \param lde has correct lustre dirent.
4102  * \retval -ve on error
4103  */
4104 static inline int osd_it_ea_rec(const struct lu_env *env,
4105                                 const struct dt_it *di,
4106                                 struct dt_rec *dtrec, __u32 attr)
4107 {
4108         struct osd_it_ea       *it    = (struct osd_it_ea *)di;
4109         struct osd_object      *obj   = it->oie_obj;
4110         struct osd_device      *dev   = osd_obj2dev(obj);
4111         struct osd_scrub       *scrub = &dev->od_scrub;
4112         struct scrub_file      *sf    = &scrub->os_file;
4113         struct osd_thread_info *oti   = osd_oti_get(env);
4114         struct osd_idmap_cache *oic   = &oti->oti_cache;
4115         struct lu_fid          *fid   = &it->oie_dirent->oied_fid;
4116         struct lu_dirent       *lde   = (struct lu_dirent *)dtrec;
4117         __u32                   ino   = it->oie_dirent->oied_ino;
4118         int                     rc    = 0;
4119         ENTRY;
4120
4121         if (!fid_is_sane(fid)) {
4122                 rc = osd_ea_fid_get(env, obj, ino, fid, &oic->oic_lid);
4123                 if (rc != 0)
4124                         RETURN(rc);
4125         } else {
4126                 osd_id_gen(&oic->oic_lid, ino, OSD_OII_NOGEN);
4127         }
4128
4129         osd_it_pack_dirent(lde, fid, it->oie_dirent->oied_off,
4130                            it->oie_dirent->oied_name,
4131                            it->oie_dirent->oied_namelen,
4132                            it->oie_dirent->oied_type, attr);
4133         if (!fid_is_norm(fid)) {
4134                 fid_zero(&oic->oic_fid);
4135                 RETURN(0);
4136         }
4137
4138         oic->oic_fid = *fid;
4139         if ((scrub->os_pos_current <= ino) &&
4140             (sf->sf_flags & SF_INCONSISTENT ||
4141              ldiskfs_test_bit(osd_oi_fid2idx(dev, fid), sf->sf_oi_bitmap)))
4142                 rc = osd_consistency_check(oti, dev, oic);
4143
4144         RETURN(rc);
4145 }
4146
4147 /**
4148  * Returns a cookie for current position of the iterator head, so that
4149  * user can use this cookie to load/start the iterator next time.
4150  *
4151  * \param di iterator's in memory structure
4152  *
4153  * \retval cookie for current position, on success
4154  */
4155 static __u64 osd_it_ea_store(const struct lu_env *env, const struct dt_it *di)
4156 {
4157         struct osd_it_ea *it = (struct osd_it_ea *)di;
4158
4159         return it->oie_dirent->oied_off;
4160 }
4161
4162 /**
4163  * It calls osd_ldiskfs_it_fill() which will use ->readdir()
4164  * to load a directory entry at a time and stored it i inn,
4165  * in iterator's in-memory data structure.
4166  *
4167  * \param di struct osd_it_ea, iterator's in memory structure
4168  *
4169  * \retval +ve on success
4170  * \retval -ve on error
4171  */
4172 static int osd_it_ea_load(const struct lu_env *env,
4173                           const struct dt_it *di, __u64 hash)
4174 {
4175         struct osd_it_ea *it = (struct osd_it_ea *)di;
4176         int rc;
4177
4178         ENTRY;
4179         it->oie_file.f_pos = hash;
4180
4181         rc =  osd_ldiskfs_it_fill(env, di);
4182         if (rc == 0)
4183                 rc = +1;
4184
4185         RETURN(rc);
4186 }
4187
4188 /**
4189  * Index lookup function for interoperability mode (b11826).
4190  *
4191  * \param key,  key i.e. file name to be searched
4192  *
4193  * \retval +ve, on success
4194  * \retval -ve, on error
4195  */
4196 static int osd_index_ea_lookup(const struct lu_env *env, struct dt_object *dt,
4197                                struct dt_rec *rec, const struct dt_key *key,
4198                                struct lustre_capa *capa)
4199 {
4200         struct osd_object *obj = osd_dt_obj(dt);
4201         int rc = 0;
4202
4203         ENTRY;
4204
4205         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
4206         LINVRNT(osd_invariant(obj));
4207
4208         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
4209                 return -EACCES;
4210
4211         rc = osd_ea_lookup_rec(env, obj, rec, key);
4212
4213         if (rc == 0)
4214                 rc = +1;
4215         RETURN(rc);
4216 }
4217
4218 /**
4219  * Index and Iterator operations for interoperability
4220  * mode (i.e. to run 2.0 mds on 1.8 disk) (b11826)
4221  */
4222 static const struct dt_index_operations osd_index_ea_ops = {
4223         .dio_lookup         = osd_index_ea_lookup,
4224         .dio_declare_insert = osd_index_declare_ea_insert,
4225         .dio_insert         = osd_index_ea_insert,
4226         .dio_declare_delete = osd_index_declare_ea_delete,
4227         .dio_delete         = osd_index_ea_delete,
4228         .dio_it     = {
4229                 .init     = osd_it_ea_init,
4230                 .fini     = osd_it_ea_fini,
4231                 .get      = osd_it_ea_get,
4232                 .put      = osd_it_ea_put,
4233                 .next     = osd_it_ea_next,
4234                 .key      = osd_it_ea_key,
4235                 .key_size = osd_it_ea_key_size,
4236                 .rec      = osd_it_ea_rec,
4237                 .store    = osd_it_ea_store,
4238                 .load     = osd_it_ea_load
4239         }
4240 };
4241
4242 static void *osd_key_init(const struct lu_context *ctx,
4243                           struct lu_context_key *key)
4244 {
4245         struct osd_thread_info *info;
4246
4247         OBD_ALLOC_PTR(info);
4248         if (info == NULL)
4249                 return ERR_PTR(-ENOMEM);
4250
4251         OBD_ALLOC(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4252         if (info->oti_it_ea_buf == NULL)
4253                 goto out_free_info;
4254
4255         info->oti_env = container_of(ctx, struct lu_env, le_ctx);
4256
4257         info->oti_hlock = ldiskfs_htree_lock_alloc();
4258         if (info->oti_hlock == NULL)
4259                 goto out_free_ea;
4260
4261         return info;
4262
4263  out_free_ea:
4264         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4265  out_free_info:
4266         OBD_FREE_PTR(info);
4267         return ERR_PTR(-ENOMEM);
4268 }
4269
4270 static void osd_key_fini(const struct lu_context *ctx,
4271                          struct lu_context_key *key, void* data)
4272 {
4273         struct osd_thread_info *info = data;
4274
4275         if (info->oti_hlock != NULL)
4276                 ldiskfs_htree_lock_free(info->oti_hlock);
4277         OBD_FREE(info->oti_it_ea_buf, OSD_IT_EA_BUFSIZE);
4278         OBD_FREE_PTR(info);
4279 }
4280
4281 static void osd_key_exit(const struct lu_context *ctx,
4282                          struct lu_context_key *key, void *data)
4283 {
4284         struct osd_thread_info *info = data;
4285
4286         LASSERT(info->oti_r_locks == 0);
4287         LASSERT(info->oti_w_locks == 0);
4288         LASSERT(info->oti_txns    == 0);
4289 }
4290
4291 /* type constructor/destructor: osd_type_init, osd_type_fini */
4292 LU_TYPE_INIT_FINI(osd, &osd_key);
4293
4294 struct lu_context_key osd_key = {
4295         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD | LCT_MG_THREAD | LCT_LOCAL,
4296         .lct_init = osd_key_init,
4297         .lct_fini = osd_key_fini,
4298         .lct_exit = osd_key_exit
4299 };
4300
4301
4302 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
4303                            const char *name, struct lu_device *next)
4304 {
4305         struct osd_device *osd = osd_dev(d);
4306
4307         strncpy(osd->od_svname, name, MAX_OBD_NAME);
4308         return osd_procfs_init(osd, name);
4309 }
4310
4311 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
4312 {
4313         ENTRY;
4314
4315         osd_scrub_cleanup(env, o);
4316
4317         if (o->od_fsops) {
4318                 fsfilt_put_ops(o->od_fsops);
4319                 o->od_fsops = NULL;
4320         }
4321
4322         /* shutdown quota slave instance associated with the device */
4323         if (o->od_quota_slave != NULL) {
4324                 qsd_fini(env, o->od_quota_slave);
4325                 o->od_quota_slave = NULL;
4326         }
4327
4328         RETURN(0);
4329 }
4330
4331 static int osd_mount(const struct lu_env *env,
4332                      struct osd_device *o, struct lustre_cfg *cfg)
4333 {
4334         struct lustre_mount_info *lmi;
4335         const char               *dev  = lustre_cfg_string(cfg, 0);
4336         struct lustre_sb_info    *lsi;
4337         int                       rc = 0;
4338
4339         ENTRY;
4340
4341         o->od_fsops = fsfilt_get_ops(mt_str(LDD_MT_LDISKFS));
4342         if (o->od_fsops == NULL) {
4343                 CERROR("Can't find fsfilt_ldiskfs\n");
4344                 RETURN(-ENOTSUPP);
4345         }
4346
4347         if (o->od_mount != NULL) {
4348                 CERROR("Already mounted (%s)\n", dev);
4349                 RETURN(-EEXIST);
4350         }
4351
4352         /* get mount */
4353         lmi = server_get_mount(dev);
4354         if (lmi == NULL) {
4355                 CERROR("Cannot get mount info for %s!\n", dev);
4356                 RETURN(-EFAULT);
4357         }
4358
4359         LASSERT(lmi != NULL);
4360         /* save lustre_mount_info in dt_device */
4361         o->od_mount = lmi;
4362         o->od_mnt = lmi->lmi_mnt;
4363
4364         lsi = s2lsi(lmi->lmi_sb);
4365
4366         if (lsi->lsi_flags & LDD_F_IAM_DIR) {
4367                 o->od_iop_mode = 0;
4368                 LCONSOLE_WARN("%s: OSD: IAM mode enabled\n", dev);
4369         } else
4370                 o->od_iop_mode = 1;
4371
4372         if (lsi->lsi_flags & LDD_F_SV_TYPE_OST) {
4373                 rc = osd_compat_init(o);
4374                 if (rc)
4375                         CERROR("%s: can't initialize compats: %d\n", dev, rc);
4376         }
4377
4378         RETURN(rc);
4379 }
4380
4381 static struct lu_device *osd_device_fini(const struct lu_env *env,
4382                                          struct lu_device *d)
4383 {
4384         int rc;
4385         ENTRY;
4386
4387         osd_compat_fini(osd_dev(d));
4388
4389         shrink_dcache_sb(osd_sb(osd_dev(d)));
4390         osd_sync(env, lu2dt_dev(d));
4391
4392         rc = osd_procfs_fini(osd_dev(d));
4393         if (rc) {
4394                 CERROR("proc fini error %d \n", rc);
4395                 RETURN (ERR_PTR(rc));
4396         }
4397
4398         if (osd_dev(d)->od_mount)
4399                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
4400                                  osd_dev(d)->od_mount->lmi_mnt);
4401         osd_dev(d)->od_mount = NULL;
4402
4403         RETURN(NULL);
4404 }
4405
4406 static struct lu_device *osd_device_alloc(const struct lu_env *env,
4407                                           struct lu_device_type *t,
4408                                           struct lustre_cfg *cfg)
4409 {
4410         struct lu_device  *l;
4411         struct osd_device *o;
4412
4413         OBD_ALLOC_PTR(o);
4414         if (o != NULL) {
4415                 int result;
4416
4417                 result = dt_device_init(&o->od_dt_dev, t);
4418                 if (result == 0) {
4419                         l = osd2lu_dev(o);
4420                         l->ld_ops = &osd_lu_ops;
4421                         o->od_dt_dev.dd_ops = &osd_dt_ops;
4422                         cfs_spin_lock_init(&o->od_osfs_lock);
4423                         cfs_mutex_init(&o->od_otable_mutex);
4424                         o->od_osfs_age = cfs_time_shift_64(-1000);
4425                         o->od_capa_hash = init_capa_hash();
4426                         if (o->od_capa_hash == NULL) {
4427                                 dt_device_fini(&o->od_dt_dev);
4428                                 l = ERR_PTR(-ENOMEM);
4429                         }
4430                 } else
4431                         l = ERR_PTR(result);
4432
4433                 if (IS_ERR(l))
4434                         OBD_FREE_PTR(o);
4435         } else
4436                 l = ERR_PTR(-ENOMEM);
4437         return l;
4438 }
4439
4440 static struct lu_device *osd_device_free(const struct lu_env *env,
4441                                          struct lu_device *d)
4442 {
4443         struct osd_device *o = osd_dev(d);
4444         ENTRY;
4445
4446         cleanup_capa_hash(o->od_capa_hash);
4447         dt_device_fini(&o->od_dt_dev);
4448         OBD_FREE_PTR(o);
4449         RETURN(NULL);
4450 }
4451
4452 static int osd_process_config(const struct lu_env *env,
4453                               struct lu_device *d, struct lustre_cfg *cfg)
4454 {
4455         struct osd_device *o = osd_dev(d);
4456         int err;
4457         ENTRY;
4458
4459         switch(cfg->lcfg_command) {
4460         case LCFG_SETUP:
4461                 err = osd_mount(env, o, cfg);
4462                 break;
4463         case LCFG_CLEANUP:
4464                 lu_dev_del_linkage(d->ld_site, d);
4465                 err = osd_shutdown(env, o);
4466                 break;
4467         default:
4468                 err = -ENOSYS;
4469         }
4470
4471         RETURN(err);
4472 }
4473
4474 static int osd_recovery_complete(const struct lu_env *env,
4475                                  struct lu_device *d)
4476 {
4477         RETURN(0);
4478 }
4479
4480 static int osd_prepare(const struct lu_env *env, struct lu_device *pdev,
4481                        struct lu_device *dev)
4482 {
4483         struct osd_device *osd = osd_dev(dev);
4484         int                result;
4485         ENTRY;
4486
4487         /* 1. setup scrub, including OI files initialization */
4488         result = osd_scrub_setup(env, osd);
4489         if (result < 0)
4490                 RETURN(result);
4491
4492         /* 2. setup quota slave instance */
4493         osd->od_quota_slave = qsd_init(env, osd->od_svname, &osd->od_dt_dev,
4494                                        osd->od_proc_entry);
4495         if (IS_ERR(osd->od_quota_slave)) {
4496                 result = PTR_ERR(osd->od_quota_slave);
4497                 osd->od_quota_slave = NULL;
4498                 RETURN(result);
4499         }
4500
4501 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 3, 55, 0)
4502         /* Unfortunately, the current MDD implementation relies on some specific
4503          * code to be executed in the OSD layer. Since OFD now also uses the OSD
4504          * module, we need a way to skip the metadata-specific code when running
4505          * with OFD.
4506          * The hack here is to check the type of the parent device which is
4507          * either MD (i.e. MDD device) with the current MDT stack or DT (i.e.
4508          * OFD device) on an OST. As a reminder, obdfilter does not use the OSD
4509          * layer and still relies on lvfs. This hack won't work any more when
4510          * LOD is landed since LOD is of DT type.
4511          * This code should be removed once the orion MDT changes (LOD/OSP, ...)
4512          * have been landed */
4513         osd->od_is_md = lu_device_is_md(pdev);
4514 #else
4515 #warning "all is_md checks must be removed from osd-ldiskfs"
4516 #endif
4517
4518         if (!osd->od_is_md)
4519                 RETURN(0);
4520
4521         /* 3. setup local objects */
4522         result = llo_local_objects_setup(env, lu2md_dev(pdev), lu2dt_dev(dev));
4523         RETURN(result);
4524 }
4525
4526 static const struct lu_object_operations osd_lu_obj_ops = {
4527         .loo_object_init      = osd_object_init,
4528         .loo_object_delete    = osd_object_delete,
4529         .loo_object_release   = osd_object_release,
4530         .loo_object_free      = osd_object_free,
4531         .loo_object_print     = osd_object_print,
4532         .loo_object_invariant = osd_object_invariant
4533 };
4534
4535 const struct lu_device_operations osd_lu_ops = {
4536         .ldo_object_alloc      = osd_object_alloc,
4537         .ldo_process_config    = osd_process_config,
4538         .ldo_recovery_complete = osd_recovery_complete,
4539         .ldo_prepare           = osd_prepare,
4540 };
4541
4542 static const struct lu_device_type_operations osd_device_type_ops = {
4543         .ldto_init = osd_type_init,
4544         .ldto_fini = osd_type_fini,
4545
4546         .ldto_start = osd_type_start,
4547         .ldto_stop  = osd_type_stop,
4548
4549         .ldto_device_alloc = osd_device_alloc,
4550         .ldto_device_free  = osd_device_free,
4551
4552         .ldto_device_init    = osd_device_init,
4553         .ldto_device_fini    = osd_device_fini
4554 };
4555
4556 static struct lu_device_type osd_device_type = {
4557         .ldt_tags     = LU_DEVICE_DT,
4558         .ldt_name     = LUSTRE_OSD_NAME,
4559         .ldt_ops      = &osd_device_type_ops,
4560         .ldt_ctx_tags = LCT_LOCAL,
4561 };
4562
4563 /*
4564  * lprocfs legacy support.
4565  */
4566 static struct obd_ops osd_obd_device_ops = {
4567         .o_owner = THIS_MODULE
4568 };
4569
4570 static int __init osd_mod_init(void)
4571 {
4572         struct lprocfs_static_vars lvars;
4573
4574         osd_oi_mod_init();
4575         lprocfs_osd_init_vars(&lvars);
4576         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
4577                                    LUSTRE_OSD_NAME, &osd_device_type);
4578 }
4579
4580 static void __exit osd_mod_exit(void)
4581 {
4582         class_unregister_type(LUSTRE_OSD_NAME);
4583 }
4584
4585 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
4586 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
4587 MODULE_LICENSE("GPL");
4588
4589 cfs_module(osd, "0.1.0", osd_mod_init, osd_mod_exit);