lustre/osd/osd_handler.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/osd/osd_handler.c
   5  *  Top-level entry points into osd module
   6  *
   7  *  Copyright (c) 2006 Cluster File Systems, Inc.
   8  *   Author: Nikita Danilov <nikita@clusterfs.com>
   9  *
  10  *   This file is part of the Lustre file system, http://www.lustre.org
  11  *   Lustre is a trademark of Cluster File Systems, Inc.
  12  *
  13  *   You may have signed or agreed to another license before downloading
  14  *   this software.  If so, you are bound by the terms and conditions
  15  *   of that agreement, and the following does not apply to you.  See the
  16  *   LICENSE file included with this distribution for more information.
  17  *
  18  *   If you did not agree to a different license, then this copy of Lustre
  19  *   is open source software; you can redistribute it and/or modify it
  20  *   under the terms of version 2 of the GNU General Public License as
  21  *   published by the Free Software Foundation.
  22  *
  23  *   In either case, Lustre is distributed in the hope that it will be
  24  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  25  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  26  *   license text for more details.
  27  */
  28
  29 #ifndef EXPORT_SYMTAB
  30 # define EXPORT_SYMTAB
  31 #endif
  32 #define DEBUG_SUBSYSTEM S_MDS
  33
  34 #include <linux/module.h>
  35
  36 /* LUSTRE_VERSION_CODE */
  37 #include <lustre_ver.h>
  38 /* prerequisite for linux/xattr.h */
  39 #include <linux/types.h>
  40 /* prerequisite for linux/xattr.h */
  41 #include <linux/fs.h>
  42 /* XATTR_{REPLACE,CREATE} */
  43 #include <linux/xattr.h>
  44 /*
  45  * XXX temporary stuff: direct access to ldiskfs/jdb. Interface between osd
  46  * and file system is not yet specified.
  47  */
  48 /* handle_t, journal_start(), journal_stop() */
  49 #include <linux/jbd.h>
  50 /* LDISKFS_SB() */
  51 #include <linux/ldiskfs_fs.h>
  52 #include <linux/ldiskfs_jbd.h>
  53 /* simple_mkdir() */
  54 #include <lvfs.h>
  55
  56 /*
  57  * struct OBD_{ALLOC,FREE}*()
  58  * OBD_FAIL_CHECK
  59  */
  60 #include <obd_support.h>
  61 /* struct ptlrpc_thread */
  62 #include <lustre_net.h>
  63 /* LUSTRE_OSD_NAME */
  64 #include <obd.h>
  65 /* class_register_type(), class_unregister_type(), class_get_type() */
  66 #include <obd_class.h>
  67 #include <lustre_disk.h>
  68
  69 /* fid_is_local() */
  70 #include <lustre_fid.h>
  71 #include <linux/lustre_iam.h>
  72
  73 #include "osd_internal.h"
  74 #include "osd_igif.h"
  75
  76 struct osd_directory {
  77         struct iam_container od_container;
  78         struct iam_descr     od_descr;
  79         struct semaphore     od_sem;
  80 };
  81
  82 struct osd_object {
  83         struct dt_object       oo_dt;
  84         /*
  85          * Inode for file system object represented by this osd_object. This
  86          * inode is pinned for the whole duration of lu_object life.
  87          *
  88          * Not modified concurrently (either setup early during object
  89          * creation, or assigned by osd_object_create() under write lock).
  90          */
  91         struct inode          *oo_inode;
  92         struct rw_semaphore    oo_sem;
  93         struct osd_directory  *oo_dir;
  94         /* protects inode attributes. */
  95         spinlock_t             oo_guard;
  96 #if OSD_COUNTERS
  97         const struct lu_env   *oo_owner;
  98 #endif
  99 };
 100
 101 /*
 102  * osd device.
 103  */
 104 struct osd_device {
 105         /* super-class */
 106         struct dt_device          od_dt_dev;
 107         /* information about underlying file system */
 108         struct lustre_mount_info *od_mount;
 109         /* object index */
 110         struct osd_oi             od_oi;
 111         /*
 112          * XXX temporary stuff for object index: directory where every object
 113          * is named by its fid.
 114          */
 115         struct dentry            *od_obj_area;
 116
 117         /* Environment for transaction commit callback.
 118          * Currently, OSD is based on ext3/JBD. Transaction commit in ext3/JBD
 119          * is serialized, that is there is no more than one transaction commit
 120          * at a time (JBD journal_commit_transaction() is serialized).
 121          * This means that it's enough to have _one_ lu_context.
 122          */
 123         struct lu_env             od_env_for_commit;
 124
 125         /*
 126          * Fid Capability
 127          */
 128         unsigned int              od_fl_capa:1;
 129         unsigned long             od_capa_timeout;
 130         __u32                     od_capa_alg;
 131         struct lustre_capa_key   *od_capa_keys;
 132         struct hlist_head        *od_capa_hash;
 133
 134         /*
 135          * statfs optimization: we cache a bit.
 136          */
 137         cfs_time_t                od_osfs_age;
 138         struct kstatfs            od_kstatfs;
 139         spinlock_t                od_osfs_lock;
 140 };
 141
 142 static int   osd_root_get      (const struct lu_env *env,
 143                                 struct dt_device *dev, struct lu_fid *f);
 144 static int   osd_statfs        (const struct lu_env *env,
 145                                 struct dt_device *dev, struct kstatfs *sfs);
 146
 147 static int   lu_device_is_osd  (const struct lu_device *d);
 148 static void  osd_mod_exit      (void) __exit;
 149 static int   osd_mod_init      (void) __init;
 150 static int   osd_type_init     (struct lu_device_type *t);
 151 static void  osd_type_fini     (struct lu_device_type *t);
 152 static int   osd_object_init   (const struct lu_env *env,
 153                                 struct lu_object *l);
 154 static void  osd_object_release(const struct lu_env *env,
 155                                 struct lu_object *l);
 156 static int   osd_object_print  (const struct lu_env *env, void *cookie,
 157                                 lu_printer_t p, const struct lu_object *o);
 158 static void  osd_device_free   (const struct lu_env *env,
 159                                 struct lu_device *m);
 160 static void *osd_key_init      (const struct lu_context *ctx,
 161                                 struct lu_context_key *key);
 162 static void  osd_key_fini      (const struct lu_context *ctx,
 163                                 struct lu_context_key *key, void *data);
 164 static void  osd_key_exit      (const struct lu_context *ctx,
 165                                 struct lu_context_key *key, void *data);
 166 static int   osd_has_index     (const struct osd_object *obj);
 167 static void  osd_object_init0  (struct osd_object *obj);
 168 static int   osd_device_init   (const struct lu_env *env,
 169                                 struct lu_device *d, const char *,
 170                                 struct lu_device *);
 171 static int   osd_fid_lookup    (const struct lu_env *env,
 172                                 struct osd_object *obj,
 173                                 const struct lu_fid *fid);
 174 static void  osd_inode_getattr (const struct lu_env *env,
 175                                 struct inode *inode, struct lu_attr *attr);
 176 static void  osd_inode_setattr (const struct lu_env *env,
 177                                 struct inode *inode, const struct lu_attr *attr);
 178 static int   osd_param_is_sane (const struct osd_device *dev,
 179                                 const struct txn_param *param);
 180 static int   osd_index_lookup  (const struct lu_env *env,
 181                                 struct dt_object *dt,
 182                                 struct dt_rec *rec, const struct dt_key *key,
 183                                 struct lustre_capa *capa);
 184 static int   osd_index_insert  (const struct lu_env *env,
 185                                 struct dt_object *dt,
 186                                 const struct dt_rec *rec,
 187                                 const struct dt_key *key,
 188                                 struct thandle *handle,
 189                                 struct lustre_capa *capa);
 190 static int   osd_index_delete  (const struct lu_env *env,
 191                                 struct dt_object *dt, const struct dt_key *key,
 192                                 struct thandle *handle,
 193                                 struct lustre_capa *capa);
 194 static int   osd_index_probe   (const struct lu_env *env,
 195                                 struct osd_object *o,
 196                                 const struct dt_index_features *feat);
 197 static int   osd_index_try     (const struct lu_env *env,
 198                                 struct dt_object *dt,
 199                                 const struct dt_index_features *feat);
 200 static void  osd_index_fini    (struct osd_object *o);
 201
 202 static void  osd_it_fini       (const struct lu_env *env, struct dt_it *di);
 203 static int   osd_it_get        (const struct lu_env *env,
 204                                 struct dt_it *di, const struct dt_key *key);
 205 static void  osd_it_put        (const struct lu_env *env, struct dt_it *di);
 206 static int   osd_it_next       (const struct lu_env *env, struct dt_it *di);
 207 static int   osd_it_del        (const struct lu_env *env, struct dt_it *di,
 208                                 struct thandle *th);
 209 static int   osd_it_key_size   (const struct lu_env *env,
 210                                 const struct dt_it *di);
 211 static void  osd_conf_get      (const struct lu_env *env,
 212                                 const struct dt_device *dev,
 213                                 struct dt_device_param *param);
 214 static void  osd_trans_stop    (const struct lu_env *env,
 215                                 struct thandle *th);
 216 static int   osd_object_is_root(const struct osd_object *obj);
 217
 218 static struct osd_object  *osd_obj          (const struct lu_object *o);
 219 static struct osd_device  *osd_dev          (const struct lu_device *d);
 220 static struct osd_device  *osd_dt_dev       (const struct dt_device *d);
 221 static struct osd_object  *osd_dt_obj       (const struct dt_object *d);
 222 static struct osd_device  *osd_obj2dev      (const struct osd_object *o);
 223 static struct lu_device   *osd2lu_dev       (struct osd_device *osd);
 224 static struct lu_device   *osd_device_fini  (const struct lu_env *env,
 225                                              struct lu_device *d);
 226 static struct lu_device   *osd_device_alloc (const struct lu_env *env,
 227                                              struct lu_device_type *t,
 228                                              struct lustre_cfg *cfg);
 229 static struct lu_object   *osd_object_alloc (const struct lu_env *env,
 230                                              const struct lu_object_header *hdr,
 231                                              struct lu_device *d);
 232 static struct inode       *osd_iget         (struct osd_thread_info *info,
 233                                              struct osd_device *dev,
 234                                              const struct osd_inode_id *id);
 235 static struct super_block *osd_sb           (const struct osd_device *dev);
 236 static struct dt_it       *osd_it_init      (const struct lu_env *env,
 237                                              struct dt_object *dt, int wable,
 238                                              struct lustre_capa *capa);
 239 static struct dt_key      *osd_it_key       (const struct lu_env *env,
 240                                              const struct dt_it *di);
 241 static struct dt_rec      *osd_it_rec       (const struct lu_env *env,
 242                                              const struct dt_it *di);
 243 static struct timespec    *osd_inode_time   (const struct lu_env *env,
 244                                              struct inode *inode,
 245                                              __u64 seconds);
 246 static struct thandle     *osd_trans_start  (const struct lu_env *env,
 247                                              struct dt_device *d,
 248                                              struct txn_param *p);
 249 static journal_t          *osd_journal      (const struct osd_device *dev);
 250
 251 static struct lu_device_type_operations osd_device_type_ops;
 252 static struct lu_device_type            osd_device_type;
 253 static struct lu_object_operations      osd_lu_obj_ops;
 254 static struct obd_ops                   osd_obd_device_ops;
 255 static struct lprocfs_vars              lprocfs_osd_module_vars[];
 256 static struct lprocfs_vars              lprocfs_osd_obd_vars[];
 257 static struct lu_device_operations      osd_lu_ops;
 258 static struct lu_context_key            osd_key;
 259 static struct dt_object_operations      osd_obj_ops;
 260 static struct dt_body_operations        osd_body_ops;
 261 static struct dt_index_operations       osd_index_ops;
 262 static struct dt_index_operations       osd_index_compat_ops;
 263
 264 struct osd_thandle {
 265         struct thandle          ot_super;
 266         handle_t               *ot_handle;
 267         struct journal_callback ot_jcb;
 268 };
 269
 270 /*
 271  * Invariants, assertions.
 272  */
 273
 274 /*
 275  * XXX: do not enable this, until invariant checking code is made thread safe
 276  * in the face of pdirops locking.
 277  */
 278 #define OSD_INVARIANT_CHECKS (0)
 279
 280 #if OSD_INVARIANT_CHECKS
 281 static int osd_invariant(const struct osd_object *obj)
 282 {
 283         return
 284                 obj != NULL &&
 285                 ergo(obj->oo_inode != NULL,
 286                      obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
 287                      atomic_read(&obj->oo_inode->i_count) > 0) &&
 288                 ergo(obj->oo_dir != NULL &&
 289                      obj->oo_dir->od_conationer.ic_object != NULL,
 290                      obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
 291 }
 292 #else
 293 #define osd_invariant(obj) (1)
 294 #endif
 295
 296 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
 297 {
 298         return lu_context_key_get(&env->le_ctx, &osd_key);
 299 }
 300
 301 #if OSD_COUNTERS
 302 /*
 303  * Concurrency: doesn't matter
 304  */
 305 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
 306 {
 307         return osd_oti_get(env)->oti_r_locks > 0;
 308 }
 309
 310 /*
 311  * Concurrency: doesn't matter
 312  */
 313 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
 314 {
 315         struct osd_thread_info *oti = osd_oti_get(env);
 316         return oti->oti_w_locks > 0 && o->oo_owner == env;
 317 }
 318
 319 #define OSD_COUNTERS_DO(exp) exp
 320 #else
 321
 322
 323 #define osd_read_locked(env, o) (1)
 324 #define osd_write_locked(env, o) (1)
 325 #define OSD_COUNTERS_DO(exp) ((void)0)
 326 #endif
 327
 328 /*
 329  * Concurrency: doesn't access mutable data
 330  */
 331 static int osd_root_get(const struct lu_env *env,
 332                         struct dt_device *dev, struct lu_fid *f)
 333 {
 334         struct inode *inode;
 335
 336         inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
 337         lu_igif_build(f, inode->i_ino, inode->i_generation);
 338         return 0;
 339 }
 340
 341 /*
 342  * OSD object methods.
 343  */
 344
 345 /*
 346  * Concurrency: no concurrent access is possible that early in object
 347  * life-cycle.
 348  */
 349 static struct lu_object *osd_object_alloc(const struct lu_env *env,
 350                                           const struct lu_object_header *hdr,
 351                                           struct lu_device *d)
 352 {
 353         struct osd_object *mo;
 354
 355         OBD_ALLOC_PTR(mo);
 356         if (mo != NULL) {
 357                 struct lu_object *l;
 358
 359                 l = &mo->oo_dt.do_lu;
 360                 dt_object_init(&mo->oo_dt, NULL, d);
 361                 mo->oo_dt.do_ops = &osd_obj_ops;
 362                 l->lo_ops = &osd_lu_obj_ops;
 363                 init_rwsem(&mo->oo_sem);
 364                 spin_lock_init(&mo->oo_guard);
 365                 return l;
 366         } else
 367                 return NULL;
 368 }
 369
 370 /*
 371  * Concurrency: shouldn't matter.
 372  */
 373 static void osd_object_init0(struct osd_object *obj)
 374 {
 375         LASSERT(obj->oo_inode != NULL);
 376         obj->oo_dt.do_body_ops = &osd_body_ops;
 377         obj->oo_dt.do_lu.lo_header->loh_attr |=
 378                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
 379 }
 380
 381 /*
 382  * Concurrency: no concurrent access is possible that early in object
 383  * life-cycle.
 384  */
 385 static int osd_object_init(const struct lu_env *env, struct lu_object *l)
 386 {
 387         struct osd_object *obj = osd_obj(l);
 388         int result;
 389
 390         LASSERT(osd_invariant(obj));
 391
 392         result = osd_fid_lookup(env, obj, lu_object_fid(l));
 393         if (result == 0) {
 394                 if (obj->oo_inode != NULL)
 395                         osd_object_init0(obj);
 396         }
 397         LASSERT(osd_invariant(obj));
 398         return result;
 399 }
 400
 401 /*
 402  * Concurrency: no concurrent access is possible that late in object
 403  * life-cycle.
 404  */
 405 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
 406 {
 407         struct osd_object *obj = osd_obj(l);
 408
 409         LASSERT(osd_invariant(obj));
 410
 411         dt_object_fini(&obj->oo_dt);
 412         OBD_FREE_PTR(obj);
 413 }
 414
 415 static struct iam_path_descr *osd_ipd_get(const struct lu_env *env,
 416                                           const struct iam_container *bag)
 417 {
 418         return bag->ic_descr->id_ops->id_ipd_alloc(bag,
 419                                                    osd_oti_get(env)->oti_ipd);
 420 }
 421
 422 static void osd_ipd_put(const struct lu_env *env,
 423                         const struct iam_container *bag,
 424                         struct iam_path_descr *ipd)
 425 {
 426         bag->ic_descr->id_ops->id_ipd_free(ipd);
 427 }
 428
 429 /*
 430  * Concurrency: no concurrent access is possible that late in object
 431  * life-cycle.
 432  */
 433 static void osd_index_fini(struct osd_object *o)
 434 {
 435         struct iam_container *bag;
 436
 437         if (o->oo_dir != NULL) {
 438                 bag = &o->oo_dir->od_container;
 439                 if (o->oo_inode != NULL) {
 440                         if (bag->ic_object == o->oo_inode)
 441                                 iam_container_fini(bag);
 442                 }
 443                 OBD_FREE_PTR(o->oo_dir);
 444                 o->oo_dir = NULL;
 445         }
 446 }
 447
 448 /*
 449  * Concurrency: no concurrent access is possible that late in object
 450  * life-cycle (for all existing callers, that is. New callers have to provide
 451  * their own locking.)
 452  */
 453 static int osd_inode_unlinked(const struct inode *inode)
 454 {
 455         return inode->i_nlink == 0;
 456 }
 457
 458 enum {
 459         OSD_TXN_OI_DELETE_CREDITS    = 20,
 460         OSD_TXN_INODE_DELETE_CREDITS = 20
 461 };
 462
 463 /*
 464  * Concurrency: no concurrent access is possible that late in object
 465  * life-cycle.
 466  */
 467 static int osd_inode_remove(const struct lu_env *env, struct osd_object *obj)
 468 {
 469         const struct lu_fid    *fid = lu_object_fid(&obj->oo_dt.do_lu);
 470         struct osd_device      *osd = osd_obj2dev(obj);
 471         struct osd_thread_info *oti = osd_oti_get(env);
 472         struct txn_param       *prm = &oti->oti_txn;
 473         struct thandle         *th;
 474         int result;
 475
 476         txn_param_init(prm, OSD_TXN_OI_DELETE_CREDITS +
 477                             OSD_TXN_INODE_DELETE_CREDITS);
 478         th = osd_trans_start(env, &osd->od_dt_dev, prm);
 479         if (!IS_ERR(th)) {
 480                 result = osd_oi_delete(oti, &osd->od_oi, fid, th);
 481                 osd_trans_stop(env, th);
 482         } else
 483                 result = PTR_ERR(th);
 484         return result;
 485 }
 486
 487 /*
 488  * Called just before object is freed. Releases all resources except for
 489  * object itself (that is released by osd_object_free()).
 490  *
 491  * Concurrency: no concurrent access is possible that late in object
 492  * life-cycle.
 493  */
 494 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
 495 {
 496         struct osd_object *obj   = osd_obj(l);
 497         struct inode      *inode = obj->oo_inode;
 498
 499         LASSERT(osd_invariant(obj));
 500
 501         /*
 502          * If object is unlinked remove fid->ino mapping from object index.
 503          *
 504          * File body will be deleted by iput().
 505          */
 506
 507         osd_index_fini(obj);
 508         if (inode != NULL) {
 509                 int result;
 510
 511                 if (osd_inode_unlinked(inode)) {
 512                         result = osd_inode_remove(env, obj);
 513                         if (result != 0)
 514                                 LU_OBJECT_DEBUG(D_ERROR, env, l,
 515                                                 "Failed to cleanup: %d\n",
 516                                                 result);
 517                 }
 518                 iput(inode);
 519                 obj->oo_inode = NULL;
 520         }
 521 }
 522
 523 /*
 524  * Concurrency: ->loo_object_release() is called under site spin-lock.
 525  */
 526 static void osd_object_release(const struct lu_env *env,
 527                                struct lu_object *l)
 528 {
 529         struct osd_object *o = osd_obj(l);
 530
 531         LASSERT(!lu_object_is_dying(l->lo_header));
 532         if (o->oo_inode != NULL && osd_inode_unlinked(o->oo_inode))
 533                 set_bit(LU_OBJECT_HEARD_BANSHEE, &l->lo_header->loh_flags);
 534 }
 535
 536 /*
 537  * Concurrency: shouldn't matter.
 538  */
 539 static int osd_object_print(const struct lu_env *env, void *cookie,
 540                             lu_printer_t p, const struct lu_object *l)
 541 {
 542         struct osd_object *o = osd_obj(l);
 543         struct iam_descr  *d;
 544
 545         if (o->oo_dir != NULL)
 546                 d = o->oo_dir->od_container.ic_descr;
 547         else
 548                 d = NULL;
 549         return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
 550                     o, o->oo_inode,
 551                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
 552                     o->oo_inode ? o->oo_inode->i_generation : 0,
 553                     d ? d->id_ops->id_name : "plain");
 554 }
 555
 556 /*
 557  * Concurrency: shouldn't matter.
 558  */
 559 static int osd_statfs(const struct lu_env *env,
 560                       struct dt_device *d, struct kstatfs *sfs)
 561 {
 562         struct osd_device *osd = osd_dt_dev(d);
 563         struct super_block *sb = osd_sb(osd);
 564         int result = 0;
 565
 566         spin_lock(&osd->od_osfs_lock);
 567         /* cache 1 second */
 568         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
 569                 result = ll_do_statfs(sb, &osd->od_kstatfs);
 570                 if (likely(result == 0)) /* N.B. statfs can't really fail */
 571                         osd->od_osfs_age = cfs_time_current_64();
 572         }
 573
 574         if (likely(result == 0))
 575                 *sfs = osd->od_kstatfs;
 576         spin_unlock(&osd->od_osfs_lock);
 577
 578         return result;
 579 }
 580
 581 /*
 582  * Concurrency: doesn't access mutable data.
 583  */
 584 static void osd_conf_get(const struct lu_env *env,
 585                          const struct dt_device *dev,
 586                          struct dt_device_param *param)
 587 {
 588         /*
 589          * XXX should be taken from not-yet-existing fs abstraction layer.
 590          */
 591         param->ddp_max_name_len  = LDISKFS_NAME_LEN;
 592         param->ddp_max_nlink     = LDISKFS_LINK_MAX;
 593         param->ddp_block_shift   = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
 594 }
 595
 596 /*
 597  * Journal
 598  */
 599
 600 /*
 601  * Concurrency: doesn't access mutable data.
 602  */
 603 static int osd_param_is_sane(const struct osd_device *dev,
 604                              const struct txn_param *param)
 605 {
 606         return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
 607 }
 608
 609 /*
 610  * Concurrency: shouldn't matter.
 611  */
 612 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
 613 {
 614         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
 615         struct thandle     *th = &oh->ot_super;
 616         struct dt_device   *dev = th->th_dev;
 617
 618         LASSERT(dev != NULL);
 619         LASSERT(oh->ot_handle == NULL);
 620
 621         if (error) {
 622                 CERROR("transaction @0x%p commit error: %d\n", th, error);
 623         } else {
 624                 /*
 625                  * This od_env_for_commit is only for commit usage.  see
 626                  * "struct dt_device"
 627                  */
 628                 dt_txn_hook_commit(&osd_dt_dev(dev)->od_env_for_commit, th);
 629         }
 630
 631         lu_device_put(&dev->dd_lu_dev);
 632         th->th_dev = NULL;
 633
 634         lu_context_exit(&th->th_ctx);
 635         lu_context_fini(&th->th_ctx);
 636         OBD_FREE_PTR(oh);
 637 }
 638
 639 /*
 640  * Concurrency: shouldn't matter.
 641  */
 642 static struct thandle *osd_trans_start(const struct lu_env *env,
 643                                        struct dt_device *d,
 644                                        struct txn_param *p)
 645 {
 646         struct osd_device  *dev = osd_dt_dev(d);
 647         handle_t           *jh;
 648         struct osd_thandle *oh;
 649         struct thandle     *th;
 650         int hook_res;
 651
 652         ENTRY;
 653
 654         hook_res = dt_txn_hook_start(env, d, p);
 655         if (hook_res != 0)
 656                 RETURN(ERR_PTR(hook_res));
 657
 658         if (osd_param_is_sane(dev, p)) {
 659                 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
 660                 if (oh != NULL) {
 661                         /*
 662                          * XXX temporary stuff. Some abstraction layer should
 663                          * be used.
 664                          */
 665
 666                         jh = journal_start(osd_journal(dev), p->tp_credits);
 667                         if (!IS_ERR(jh)) {
 668                                 oh->ot_handle = jh;
 669                                 th = &oh->ot_super;
 670                                 th->th_dev = d;
 671                                 th->th_result = 0;
 672                                 jh->h_sync = p->tp_sync;
 673                                 lu_device_get(&d->dd_lu_dev);
 674                                 /* add commit callback */
 675                                 lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
 676                                 lu_context_enter(&th->th_ctx);
 677                                 journal_callback_set(jh, osd_trans_commit_cb,
 678                                                      (struct journal_callback *)&oh->ot_jcb);
 679 #if OSD_COUNTERS
 680                                 {
 681                                         struct osd_thread_info *oti =
 682                                                 osd_oti_get(env);
 683
 684                                         LASSERT(oti->oti_txns == 0);
 685                                         LASSERT(oti->oti_r_locks == 0);
 686                                         LASSERT(oti->oti_w_locks == 0);
 687                                         oti->oti_txns++;
 688                                 }
 689 #endif
 690                         } else {
 691                                 OBD_FREE_PTR(oh);
 692                                 th = (void *)jh;
 693                         }
 694                 } else
 695                         th = ERR_PTR(-ENOMEM);
 696         } else {
 697                 CERROR("Invalid transaction parameters\n");
 698                 th = ERR_PTR(-EINVAL);
 699         }
 700
 701         RETURN(th);
 702 }
 703
 704 /*
 705  * Concurrency: shouldn't matter.
 706  */
 707 static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
 708 {
 709         int result;
 710         struct osd_thandle *oh;
 711
 712         ENTRY;
 713
 714         oh = container_of0(th, struct osd_thandle, ot_super);
 715         if (oh->ot_handle != NULL) {
 716                 handle_t *hdl = oh->ot_handle;
 717                 /*
 718                  * XXX temporary stuff. Some abstraction layer should be used.
 719                  */
 720                 result = dt_txn_hook_stop(env, th);
 721                 if (result != 0)
 722                         CERROR("Failure in transaction hook: %d\n", result);
 723
 724                 /**/
 725                 oh->ot_handle = NULL;
 726                 result = journal_stop(hdl);
 727                 if (result != 0)
 728                         CERROR("Failure to stop transaction: %d\n", result);
 729
 730 #if OSD_COUNTERS
 731                 {
 732                         struct osd_thread_info *oti = osd_oti_get(env);
 733
 734                         LASSERT(oti->oti_txns == 1);
 735                         LASSERT(oti->oti_r_locks == 0);
 736                         LASSERT(oti->oti_w_locks == 0);
 737                         oti->oti_txns--;
 738                 }
 739 #endif
 740         }
 741         EXIT;
 742 }
 743
 744 /*
 745  * Concurrency: shouldn't matter.
 746  */
 747 static int osd_sync(const struct lu_env *env, struct dt_device *d)
 748 {
 749         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
 750         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
 751 }
 752
 753 /*
 754  * Concurrency: shouldn't matter.
 755  */
 756 lvfs_sbdev_type fsfilt_ldiskfs_journal_sbdev(struct super_block *);
 757
 758 static void osd_ro(const struct lu_env *env, struct dt_device *d)
 759 {
 760         ENTRY;
 761
 762         CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
 763
 764         __lvfs_set_rdonly(lvfs_sbdev(osd_sb(osd_dt_dev(d))),
 765                           fsfilt_ldiskfs_journal_sbdev(osd_sb(osd_dt_dev(d))));
 766         EXIT;
 767 }
 768
 769 /*
 770  * Concurrency: serialization provided by callers.
 771  */
 772 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
 773                               int mode, unsigned long timeout, __u32 alg,
 774                               struct lustre_capa_key *keys)
 775 {
 776         struct osd_device *dev = osd_dt_dev(d);
 777         ENTRY;
 778
 779         dev->od_fl_capa = mode;
 780         dev->od_capa_timeout = timeout;
 781         dev->od_capa_alg = alg;
 782         dev->od_capa_keys = keys;
 783         RETURN(0);
 784 }
 785
 786 /* Note: we did not count into QUOTA here, If we mount with --data_journal
 787  * we may need more*/
 788 static const int osd_dto_credits[DTO_NR] = {
 789         /*
 790          * Insert/Delete. IAM EXT3_INDEX_EXTRA_TRANS_BLOCKS(8) +
 791          * EXT3_SINGLEDATA_TRANS_BLOCKS 8 XXX Note: maybe iam need more,since
 792          * iam have more level than Ext3 htree
 793          */
 794         [DTO_INDEX_INSERT]  = 16,
 795         [DTO_INDEX_DELETE]  = 16,
 796         [DTO_IDNEX_UPDATE]  = 16,
 797         /*
 798          * Create a object. Same as create object in Ext3 filesystem, but did
 799          * not count QUOTA i EXT3_DATA_TRANS_BLOCKS(12) +
 800          * INDEX_EXTRA_BLOCKS(8) + 3(inode bits,groups, GDT)
 801          */
 802         [DTO_OBJECT_CREATE] = 23,
 803         [DTO_OBJECT_DELETE] = 23,
 804         /*
 805          * Attr set credits 3 inode, group, GDT
 806          */
 807         [DTO_ATTR_SET]      = 3,
 808         /*
 809          * XATTR_SET. SAME AS XATTR of EXT3 EXT3_DATA_TRANS_BLOCKS XXX Note:
 810          * in original MDS implmentation EXT3_INDEX_EXTRA_TRANS_BLOCKS are
 811          * also counted in. Do not know why?
 812          */
 813         [DTO_XATTR_SET]     = 16,
 814         [DTO_LOG_REC]       = 16,
 815         /* creadits for inode change during write */
 816         [DTO_WRITE_BASE]    = 3,
 817         /* credits for single block write */
 818         [DTO_WRITE_BLOCK]   = 12
 819 };
 820
 821 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
 822                           enum dt_txn_op op)
 823 {
 824         LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits));
 825         return osd_dto_credits[op];
 826 }
 827
 828 static struct dt_device_operations osd_dt_ops = {
 829         .dt_root_get       = osd_root_get,
 830         .dt_statfs         = osd_statfs,
 831         .dt_trans_start    = osd_trans_start,
 832         .dt_trans_stop     = osd_trans_stop,
 833         .dt_conf_get       = osd_conf_get,
 834         .dt_sync           = osd_sync,
 835         .dt_ro             = osd_ro,
 836         .dt_credit_get     = osd_credit_get,
 837         .dt_init_capa_ctxt = osd_init_capa_ctxt,
 838 };
 839
 840 static void osd_object_read_lock(const struct lu_env *env,
 841                                  struct dt_object *dt)
 842 {
 843         struct osd_object *obj = osd_dt_obj(dt);
 844
 845         LASSERT(osd_invariant(obj));
 846
 847         OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
 848         down_read(&obj->oo_sem);
 849 #if OSD_COUNTERS
 850         {
 851                 struct osd_thread_info *oti = osd_oti_get(env);
 852
 853                 LASSERT(obj->oo_owner == NULL);
 854                 oti->oti_r_locks++;
 855         }
 856 #endif
 857 }
 858
 859 static void osd_object_write_lock(const struct lu_env *env,
 860                                   struct dt_object *dt)
 861 {
 862         struct osd_object *obj = osd_dt_obj(dt);
 863
 864         LASSERT(osd_invariant(obj));
 865
 866         OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
 867         down_write(&obj->oo_sem);
 868 #if OSD_COUNTERS
 869         {
 870                 struct osd_thread_info *oti = osd_oti_get(env);
 871
 872                 LASSERT(obj->oo_owner == NULL);
 873                 obj->oo_owner = env;
 874                 oti->oti_w_locks++;
 875         }
 876 #endif
 877 }
 878
 879 static void osd_object_read_unlock(const struct lu_env *env,
 880                                    struct dt_object *dt)
 881 {
 882         struct osd_object *obj = osd_dt_obj(dt);
 883
 884         LASSERT(osd_invariant(obj));
 885 #if OSD_COUNTERS
 886         {
 887                 struct osd_thread_info *oti = osd_oti_get(env);
 888
 889                 LASSERT(oti->oti_r_locks > 0);
 890                 oti->oti_r_locks--;
 891         }
 892 #endif
 893         up_read(&obj->oo_sem);
 894 }
 895
 896 static void osd_object_write_unlock(const struct lu_env *env,
 897                                     struct dt_object *dt)
 898 {
 899         struct osd_object *obj = osd_dt_obj(dt);
 900
 901         LASSERT(osd_invariant(obj));
 902 #if OSD_COUNTERS
 903         {
 904                 struct osd_thread_info *oti = osd_oti_get(env);
 905
 906                 LASSERT(obj->oo_owner == env);
 907                 LASSERT(oti->oti_w_locks > 0);
 908                 oti->oti_w_locks--;
 909                 obj->oo_owner = NULL;
 910         }
 911 #endif
 912         up_write(&obj->oo_sem);
 913 }
 914
 915 static int capa_is_sane(const struct lu_env *env,
 916                         struct osd_device *dev,
 917                         struct lustre_capa *capa,
 918                         struct lustre_capa_key *keys)
 919 {
 920         struct osd_thread_info *oti = osd_oti_get(env);
 921         struct obd_capa *oc;
 922         int i, rc = 0;
 923         ENTRY;
 924
 925         oc = capa_lookup(dev->od_capa_hash, capa, 0);
 926         if (oc) {
 927                 if (capa_is_expired(oc)) {
 928                         DEBUG_CAPA(D_ERROR, capa, "expired");
 929                         rc = -ESTALE;
 930                 }
 931                 capa_put(oc);
 932                 RETURN(rc);
 933         }
 934
 935         spin_lock(&capa_lock);
 936         for (i = 0; i < 2; i++) {
 937                 if (keys[i].lk_keyid == capa->lc_keyid) {
 938                         oti->oti_capa_key = keys[i];
 939                         break;
 940                 }
 941         }
 942         spin_unlock(&capa_lock);
 943
 944         if (i == 2) {
 945                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
 946                 RETURN(-ESTALE);
 947         }
 948
 949         rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
 950         if (rc)
 951                 RETURN(rc);
 952         if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
 953         {
 954                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
 955                 RETURN(-EACCES);
 956         }
 957
 958         oc = capa_add(dev->od_capa_hash, capa);
 959         capa_put(oc);
 960
 961         RETURN(0);
 962 }
 963
 964 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
 965                            struct lustre_capa *capa, __u64 opc)
 966 {
 967         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
 968         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
 969         int rc;
 970
 971         if (!dev->od_fl_capa)
 972                 return 0;
 973
 974         if (capa == BYPASS_CAPA)
 975                 return 0;
 976
 977         if (!capa) {
 978                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
 979                 return -EACCES;
 980         }
 981
 982         if (!lu_fid_eq(fid, &capa->lc_fid)) {
 983                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
 984                            PFID(fid));
 985                 return -EACCES;
 986         }
 987
 988         if (!capa_opc_supported(capa, opc)) {
 989                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
 990                 return -EACCES;
 991         }
 992
 993         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
 994                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
 995                 return -EACCES;
 996         }
 997
 998         return 0;
 999 }
1000
1001 static int osd_attr_get(const struct lu_env *env,
1002                         struct dt_object *dt,
1003                         struct lu_attr *attr,
1004                         struct lustre_capa *capa)
1005 {
1006         struct osd_object *obj = osd_dt_obj(dt);
1007
1008         LASSERT(dt_object_exists(dt));
1009         LASSERT(osd_invariant(obj));
1010
1011         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1012                 return -EACCES;
1013
1014         spin_lock(&obj->oo_guard);
1015         osd_inode_getattr(env, obj->oo_inode, attr);
1016         spin_unlock(&obj->oo_guard);
1017         return 0;
1018 }
1019
1020 static int osd_attr_set(const struct lu_env *env,
1021                         struct dt_object *dt,
1022                         const struct lu_attr *attr,
1023                         struct thandle *handle,
1024                         struct lustre_capa *capa)
1025 {
1026         struct osd_object *obj = osd_dt_obj(dt);
1027
1028         LASSERT(handle != NULL);
1029         LASSERT(dt_object_exists(dt));
1030         LASSERT(osd_invariant(obj));
1031
1032         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1033                 return -EACCES;
1034
1035         spin_lock(&obj->oo_guard);
1036         osd_inode_setattr(env, obj->oo_inode, attr);
1037         spin_unlock(&obj->oo_guard);
1038
1039         mark_inode_dirty(obj->oo_inode);
1040         return 0;
1041 }
1042
1043 static struct timespec *osd_inode_time(const struct lu_env *env,
1044                                        struct inode *inode, __u64 seconds)
1045 {
1046         struct osd_thread_info *oti = osd_oti_get(env);
1047         struct timespec        *t   = &oti->oti_time;
1048
1049         t->tv_sec  = seconds;
1050         t->tv_nsec = 0;
1051         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1052         return t;
1053 }
1054
1055 static void osd_inode_setattr(const struct lu_env *env,
1056                               struct inode *inode, const struct lu_attr *attr)
1057 {
1058         __u64 bits;
1059
1060         bits = attr->la_valid;
1061
1062         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1063
1064         if (bits & LA_ATIME)
1065                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1066         if (bits & LA_CTIME)
1067                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1068         if (bits & LA_MTIME)
1069                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1070         if (bits & LA_SIZE) {
1071                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1072                 i_size_write(inode, attr->la_size);
1073         }
1074         if (bits & LA_BLOCKS)
1075                 inode->i_blocks = attr->la_blocks;
1076         if (bits & LA_MODE)
1077                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1078                         (attr->la_mode & ~S_IFMT);
1079         if (bits & LA_UID)
1080                 inode->i_uid    = attr->la_uid;
1081         if (bits & LA_GID)
1082                 inode->i_gid    = attr->la_gid;
1083         if (bits & LA_NLINK)
1084                 inode->i_nlink  = attr->la_nlink;
1085         if (bits & LA_RDEV)
1086                 inode->i_rdev   = attr->la_rdev;
1087
1088         if (bits & LA_FLAGS) {
1089                 struct ldiskfs_inode_info *li = LDISKFS_I(inode);
1090
1091                 li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
1092                         (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
1093         }
1094 }
1095
1096 /*
1097  * Object creation.
1098  *
1099  * XXX temporary solution.
1100  */
1101
1102 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1103                           struct lu_attr *attr, struct thandle *th)
1104 {
1105         return 0;
1106 }
1107
1108 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1109                            struct lu_attr *attr, struct thandle *th)
1110 {
1111         LASSERT(obj->oo_inode != NULL);
1112
1113         osd_object_init0(obj);
1114         return 0;
1115 }
1116
1117 extern struct inode *ldiskfs_create_inode(handle_t *handle,
1118                                           struct inode * dir, int mode);
1119
1120 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1121                       umode_t mode,
1122                       struct dt_allocation_hint *hint,
1123                       struct thandle *th)
1124 {
1125         int result;
1126         struct osd_device  *osd = osd_obj2dev(obj);
1127         struct osd_thandle *oth;
1128         struct inode       *parent;
1129         struct inode       *inode;
1130
1131         LASSERT(osd_invariant(obj));
1132         LASSERT(obj->oo_inode == NULL);
1133         LASSERT(osd->od_obj_area != NULL);
1134
1135         oth = container_of(th, struct osd_thandle, ot_super);
1136         LASSERT(oth->ot_handle->h_transaction != NULL);
1137
1138         if (hint && hint->dah_parent)
1139                 parent = osd_dt_obj(hint->dah_parent)->oo_inode;
1140         else
1141                 parent = osd->od_obj_area->d_inode;
1142         LASSERT(parent->i_op != NULL);
1143
1144         inode = ldiskfs_create_inode(oth->ot_handle, parent, mode);
1145         if (!IS_ERR(inode)) {
1146                 obj->oo_inode = inode;
1147                 result = 0;
1148         } else
1149                 result = PTR_ERR(inode);
1150         LASSERT(osd_invariant(obj));
1151         return result;
1152 }
1153
1154
1155 extern int iam_lvar_create(struct inode *obj, int keysize, int ptrsize,
1156                            int recsize, handle_t *handle);
1157
1158 enum {
1159         OSD_NAME_LEN = 255
1160 };
1161
1162 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1163                      struct lu_attr *attr,
1164                      struct dt_allocation_hint *hint,
1165                      struct thandle *th)
1166 {
1167         int result;
1168         struct osd_thandle *oth;
1169
1170         LASSERT(S_ISDIR(attr->la_mode));
1171
1172         oth = container_of(th, struct osd_thandle, ot_super);
1173         LASSERT(oth->ot_handle->h_transaction != NULL);
1174         result = osd_mkfile(info, obj, (attr->la_mode &
1175                             (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1176         if (result == 0) {
1177                 LASSERT(obj->oo_inode != NULL);
1178                 /*
1179                  * XXX uh-oh... call low-level iam function directly.
1180                  */
1181                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1182                                          sizeof (struct lu_fid_pack),
1183                                          oth->ot_handle);
1184         }
1185         return result;
1186 }
1187
1188 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1189                      struct lu_attr *attr,
1190                      struct dt_allocation_hint *hint,
1191                      struct thandle *th)
1192 {
1193         LASSERT(S_ISREG(attr->la_mode));
1194         return osd_mkfile(info, obj, (attr->la_mode &
1195                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1196 }
1197
1198 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1199                      struct lu_attr *attr,
1200                      struct dt_allocation_hint *hint,
1201                      struct thandle *th)
1202 {
1203         LASSERT(S_ISLNK(attr->la_mode));
1204         return osd_mkfile(info, obj, (attr->la_mode &
1205                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1206 }
1207
1208 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1209                      struct lu_attr *attr,
1210                      struct dt_allocation_hint *hint,
1211                      struct thandle *th)
1212 {
1213         int result;
1214         struct osd_device *osd = osd_obj2dev(obj);
1215         struct inode      *dir;
1216         umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1217
1218         LASSERT(osd_invariant(obj));
1219         LASSERT(obj->oo_inode == NULL);
1220         LASSERT(osd->od_obj_area != NULL);
1221         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1222                 S_ISFIFO(mode) || S_ISSOCK(mode));
1223
1224         dir = osd->od_obj_area->d_inode;
1225         LASSERT(dir->i_op != NULL);
1226
1227         result = osd_mkfile(info, obj, mode, hint, th);
1228         if (result == 0) {
1229                 LASSERT(obj->oo_inode != NULL);
1230                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1231         }
1232         LASSERT(osd_invariant(obj));
1233         return result;
1234 }
1235
1236 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1237                               struct lu_attr *,
1238                               struct dt_allocation_hint *hint,
1239                               struct thandle *);
1240
1241 static osd_obj_type_f osd_create_type_f(__u32 mode)
1242 {
1243         osd_obj_type_f result;
1244
1245         switch (mode) {
1246         case S_IFDIR:
1247                 result = osd_mkdir;
1248                 break;
1249         case S_IFREG:
1250                 result = osd_mkreg;
1251                 break;
1252         case S_IFLNK:
1253                 result = osd_mksym;
1254                 break;
1255         case S_IFCHR:
1256         case S_IFBLK:
1257         case S_IFIFO:
1258         case S_IFSOCK:
1259                 result = osd_mknod;
1260                 break;
1261         default:
1262                 LBUG();
1263                 break;
1264         }
1265         return result;
1266 }
1267
1268
1269 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1270                         struct dt_object *parent, umode_t child_mode)
1271 {
1272         LASSERT(ah);
1273
1274         memset(ah, 0, sizeof(*ah));
1275         ah->dah_parent = parent;
1276         ah->dah_mode = child_mode;
1277 }
1278
1279
1280 /*
1281  * Concurrency: @dt is write locked.
1282  */
1283 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1284                              struct lu_attr *attr,
1285                              struct dt_allocation_hint *hint,
1286                              struct thandle *th)
1287 {
1288         const struct lu_fid    *fid  = lu_object_fid(&dt->do_lu);
1289         struct osd_object      *obj  = osd_dt_obj(dt);
1290         struct osd_device      *osd  = osd_obj2dev(obj);
1291         struct osd_thread_info *info = osd_oti_get(env);
1292         int result;
1293
1294         ENTRY;
1295
1296         LASSERT(osd_invariant(obj));
1297         LASSERT(!dt_object_exists(dt));
1298         LASSERT(osd_write_locked(env, obj));
1299         LASSERT(th != NULL);
1300
1301         /*
1302          * XXX missing: Quote handling.
1303          */
1304
1305         result = osd_create_pre(info, obj, attr, th);
1306         if (result == 0) {
1307                 result = osd_create_type_f(attr->la_mode & S_IFMT)(info, obj,
1308                                                                 attr, hint, th);
1309                 if (result == 0)
1310                         result = osd_create_post(info, obj, attr, th);
1311         }
1312         if (result == 0) {
1313                 struct osd_inode_id *id = &info->oti_id;
1314
1315                 LASSERT(obj->oo_inode != NULL);
1316
1317                 id->oii_ino = obj->oo_inode->i_ino;
1318                 id->oii_gen = obj->oo_inode->i_generation;
1319
1320                 result = osd_oi_insert(info, &osd->od_oi, fid, id, th);
1321         }
1322
1323         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1324         LASSERT(osd_invariant(obj));
1325         RETURN(result);
1326 }
1327
1328 /*
1329  * Concurrency: @dt is write locked.
1330  */
1331 static void osd_object_ref_add(const struct lu_env *env,
1332                                struct dt_object *dt,
1333                                struct thandle *th)
1334 {
1335         struct osd_object *obj = osd_dt_obj(dt);
1336         struct inode *inode = obj->oo_inode;
1337
1338         LASSERT(osd_invariant(obj));
1339         LASSERT(dt_object_exists(dt));
1340         LASSERT(osd_write_locked(env, obj));
1341         LASSERT(th != NULL);
1342
1343         spin_lock(&obj->oo_guard);
1344         if (inode->i_nlink < LDISKFS_LINK_MAX) {
1345                 inode->i_nlink ++;
1346                 spin_unlock(&obj->oo_guard);
1347                 mark_inode_dirty(inode);
1348         } else {
1349                 spin_unlock(&obj->oo_guard);
1350                 LU_OBJECT_DEBUG(D_ERROR, env, &dt->do_lu,
1351                                 "Overflowed nlink\n");
1352         }
1353         LASSERT(osd_invariant(obj));
1354 }
1355
1356 /*
1357  * Concurrency: @dt is write locked.
1358  */
1359 static void osd_object_ref_del(const struct lu_env *env,
1360                                struct dt_object *dt,
1361                                struct thandle *th)
1362 {
1363         struct osd_object *obj = osd_dt_obj(dt);
1364         struct inode *inode = obj->oo_inode;
1365
1366         LASSERT(osd_invariant(obj));
1367         LASSERT(dt_object_exists(dt));
1368         LASSERT(osd_write_locked(env, obj));
1369         LASSERT(th != NULL);
1370
1371         spin_lock(&obj->oo_guard);
1372         if (inode->i_nlink > 0) {
1373                 inode->i_nlink --;
1374                 spin_unlock(&obj->oo_guard);
1375                 mark_inode_dirty(inode);
1376         } else {
1377                 spin_unlock(&obj->oo_guard);
1378                 LU_OBJECT_DEBUG(D_ERROR, env, &dt->do_lu,
1379                                 "Underflowed nlink\n");
1380         }
1381         LASSERT(osd_invariant(obj));
1382 }
1383
1384 /*
1385  * Concurrency: @dt is read locked.
1386  */
1387 static int osd_xattr_get(const struct lu_env *env,
1388                          struct dt_object *dt,
1389                          struct lu_buf *buf,
1390                          const char *name,
1391                          struct lustre_capa *capa)
1392 {
1393         struct osd_object      *obj    = osd_dt_obj(dt);
1394         struct inode           *inode  = obj->oo_inode;
1395         struct osd_thread_info *info   = osd_oti_get(env);
1396         struct dentry          *dentry = &info->oti_dentry;
1397
1398         LASSERT(dt_object_exists(dt));
1399         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1400         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1401
1402         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1403                 return -EACCES;
1404
1405         dentry->d_inode = inode;
1406         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
1407 }
1408
1409 /*
1410  * Concurrency: @dt is write locked.
1411  */
1412 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1413                          const struct lu_buf *buf, const char *name, int fl,
1414                          struct thandle *handle, struct lustre_capa *capa)
1415 {
1416         int fs_flags;
1417
1418         struct osd_object      *obj    = osd_dt_obj(dt);
1419         struct inode           *inode  = obj->oo_inode;
1420         struct osd_thread_info *info   = osd_oti_get(env);
1421         struct dentry          *dentry = &info->oti_dentry;
1422
1423         LASSERT(dt_object_exists(dt));
1424         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1425         LASSERT(osd_write_locked(env, obj));
1426         LASSERT(handle != NULL);
1427
1428         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1429                 return -EACCES;
1430
1431         dentry->d_inode = inode;
1432
1433         fs_flags = 0;
1434         if (fl & LU_XATTR_REPLACE)
1435                 fs_flags |= XATTR_REPLACE;
1436
1437         if (fl & LU_XATTR_CREATE)
1438                 fs_flags |= XATTR_CREATE;
1439
1440         return inode->i_op->setxattr(dentry, name,
1441                                      buf->lb_buf, buf->lb_len, fs_flags);
1442 }
1443
1444 /*
1445  * Concurrency: @dt is read locked.
1446  */
1447 static int osd_xattr_list(const struct lu_env *env,
1448                           struct dt_object *dt,
1449                           struct lu_buf *buf,
1450                           struct lustre_capa *capa)
1451 {
1452         struct osd_object      *obj    = osd_dt_obj(dt);
1453         struct inode           *inode  = obj->oo_inode;
1454         struct osd_thread_info *info   = osd_oti_get(env);
1455         struct dentry          *dentry = &info->oti_dentry;
1456
1457         LASSERT(dt_object_exists(dt));
1458         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
1459         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1460
1461         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1462                 return -EACCES;
1463
1464         dentry->d_inode = inode;
1465         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
1466 }
1467
1468 /*
1469  * Concurrency: @dt is write locked.
1470  */
1471 static int osd_xattr_del(const struct lu_env *env,
1472                          struct dt_object *dt,
1473                          const char *name,
1474                          struct thandle *handle,
1475                          struct lustre_capa *capa)
1476 {
1477         struct osd_object      *obj    = osd_dt_obj(dt);
1478         struct inode           *inode  = obj->oo_inode;
1479         struct osd_thread_info *info   = osd_oti_get(env);
1480         struct dentry          *dentry = &info->oti_dentry;
1481
1482         LASSERT(dt_object_exists(dt));
1483         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
1484         LASSERT(osd_write_locked(env, obj));
1485         LASSERT(handle != NULL);
1486
1487         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1488                 return -EACCES;
1489
1490         dentry->d_inode = inode;
1491         return inode->i_op->removexattr(dentry, name);
1492 }
1493
1494 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1495                                      struct dt_object *dt,
1496                                      struct lustre_capa *old,
1497                                      __u64 opc)
1498 {
1499         struct osd_thread_info *info = osd_oti_get(env);
1500         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1501         struct osd_object *obj = osd_dt_obj(dt);
1502         struct osd_device *dev = osd_obj2dev(obj);
1503         struct lustre_capa_key *key = &info->oti_capa_key;
1504         struct lustre_capa *capa = &info->oti_capa;
1505         struct obd_capa *oc;
1506         int rc;
1507         ENTRY;
1508
1509         if (!dev->od_fl_capa)
1510                 RETURN(ERR_PTR(-ENOENT));
1511
1512         LASSERT(dt_object_exists(dt));
1513         LASSERT(osd_invariant(obj));
1514
1515         /* renewal sanity check */
1516         if (old && osd_object_auth(env, dt, old, opc))
1517                 RETURN(ERR_PTR(-EACCES));
1518
1519         capa->lc_fid = *fid;
1520         capa->lc_opc = opc;
1521         capa->lc_uid = 0;
1522         capa->lc_flags = dev->od_capa_alg << 24;
1523         capa->lc_timeout = dev->od_capa_timeout;
1524         capa->lc_expiry = 0;
1525
1526         oc = capa_lookup(dev->od_capa_hash, capa, 1);
1527         if (oc) {
1528                 LASSERT(!capa_is_expired(oc));
1529                 RETURN(oc);
1530         }
1531
1532         spin_lock(&capa_lock);
1533         *key = dev->od_capa_keys[1];
1534         spin_unlock(&capa_lock);
1535
1536         capa->lc_keyid = key->lk_keyid;
1537         capa->lc_expiry = CURRENT_SECONDS + dev->od_capa_timeout;
1538
1539         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1540         if (rc) {
1541                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1542                 RETURN(ERR_PTR(rc));
1543         }
1544
1545         oc = capa_add(dev->od_capa_hash, capa);
1546         RETURN(oc);
1547 }
1548
1549 static struct dt_object_operations osd_obj_ops = {
1550         .do_read_lock    = osd_object_read_lock,
1551         .do_write_lock   = osd_object_write_lock,
1552         .do_read_unlock  = osd_object_read_unlock,
1553         .do_write_unlock = osd_object_write_unlock,
1554         .do_attr_get     = osd_attr_get,
1555         .do_attr_set     = osd_attr_set,
1556         .do_ah_init      = osd_ah_init,
1557         .do_create       = osd_object_create,
1558         .do_index_try    = osd_index_try,
1559         .do_ref_add      = osd_object_ref_add,
1560         .do_ref_del      = osd_object_ref_del,
1561         .do_xattr_get    = osd_xattr_get,
1562         .do_xattr_set    = osd_xattr_set,
1563         .do_xattr_del    = osd_xattr_del,
1564         .do_xattr_list   = osd_xattr_list,
1565         .do_capa_get     = osd_capa_get,
1566 };
1567
1568 /*
1569  * Body operations.
1570  */
1571
1572 /*
1573  * XXX: Another layering violation for now.
1574  *
1575  * We don't want to use ->f_op->read methods, because generic file write
1576  *
1577  *         - serializes on ->i_sem, and
1578  *
1579  *         - does a lot of extra work like balance_dirty_pages(),
1580  *
1581  * which doesn't work for globally shared files like /last-received.
1582  */
1583 int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1584 int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
1585                                 loff_t *offs, handle_t *handle);
1586
1587 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
1588                         struct lu_buf *buf, loff_t *pos,
1589                         struct lustre_capa *capa)
1590 {
1591         struct inode *inode = osd_dt_obj(dt)->oo_inode;
1592
1593         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
1594                 RETURN(-EACCES);
1595
1596         return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
1597 }
1598
1599 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
1600                          const struct lu_buf *buf, loff_t *pos,
1601                          struct thandle *handle, struct lustre_capa *capa)
1602 {
1603         struct inode       *inode = osd_dt_obj(dt)->oo_inode;
1604         struct osd_thandle *oh;
1605         ssize_t             result;
1606
1607         LASSERT(handle != NULL);
1608
1609         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
1610                 RETURN(-EACCES);
1611
1612         oh = container_of(handle, struct osd_thandle, ot_super);
1613         LASSERT(oh->ot_handle->h_transaction != NULL);
1614         result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
1615                                              pos, oh->ot_handle);
1616         if (result == 0)
1617                 result = buf->lb_len;
1618         return result;
1619 }
1620
1621 static struct dt_body_operations osd_body_ops = {
1622         .dbo_read  = osd_read,
1623         .dbo_write = osd_write
1624 };
1625
1626 /*
1627  * Index operations.
1628  */
1629
1630 static int osd_object_is_root(const struct osd_object *obj)
1631 {
1632         return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
1633 }
1634
1635 static int osd_index_probe(const struct lu_env *env, struct osd_object *o,
1636                            const struct dt_index_features *feat)
1637 {
1638         struct iam_descr *descr;
1639
1640         if (osd_object_is_root(o))
1641                 return feat == &dt_directory_features;
1642
1643         LASSERT(o->oo_dir != NULL);
1644
1645         descr = o->oo_dir->od_container.ic_descr;
1646         if (feat == &dt_directory_features)
1647                 return descr == &iam_htree_compat_param ||
1648                         (descr->id_rec_size == sizeof(struct lu_fid_pack) &&
1649                          1 /*
1650                             * XXX check that index looks like directory.
1651                             */
1652                                 );
1653         else
1654                 return
1655                         feat->dif_keysize_min <= descr->id_key_size &&
1656                         descr->id_key_size <= feat->dif_keysize_max &&
1657                         feat->dif_recsize_min <= descr->id_rec_size &&
1658                         descr->id_rec_size <= feat->dif_recsize_max &&
1659                         !(feat->dif_flags & (DT_IND_VARKEY |
1660                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
1661                         ergo(feat->dif_flags & DT_IND_UPDATE,
1662                              1 /* XXX check that object (and file system) is
1663                                 * writable */);
1664 }
1665
1666 static int osd_container_init(const struct lu_env *env,
1667                               struct osd_object *obj,
1668                               struct osd_directory *dir)
1669 {
1670         int result;
1671         struct iam_container *bag;
1672
1673         bag    = &dir->od_container;
1674         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
1675         if (result == 0) {
1676                 result = iam_container_setup(bag);
1677                 if (result == 0)
1678                         obj->oo_dt.do_index_ops = &osd_index_ops;
1679                 else
1680                         iam_container_fini(bag);
1681         }
1682         return result;
1683 }
1684
1685 /*
1686  * Concurrency: no external locking is necessary.
1687  */
1688 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1689                          const struct dt_index_features *feat)
1690 {
1691         int result;
1692         struct osd_object *obj = osd_dt_obj(dt);
1693
1694         LASSERT(osd_invariant(obj));
1695         LASSERT(dt_object_exists(dt));
1696
1697         if (osd_object_is_root(obj)) {
1698                 dt->do_index_ops = &osd_index_compat_ops;
1699                 result = 0;
1700         } else if (!osd_has_index(obj)) {
1701                 struct osd_directory *dir;
1702
1703                 OBD_ALLOC_PTR(dir);
1704                 if (dir != NULL) {
1705                         sema_init(&dir->od_sem, 1);
1706
1707                         spin_lock(&obj->oo_guard);
1708                         if (obj->oo_dir == NULL)
1709                                 obj->oo_dir = dir;
1710                         else
1711                                 /*
1712                                  * Concurrent thread allocated container data.
1713                                  */
1714                                 OBD_FREE_PTR(dir);
1715                         spin_unlock(&obj->oo_guard);
1716                         /*
1717                          * Now, that we have container data, serialize its
1718                          * initialization.
1719                          */
1720                         down(&obj->oo_dir->od_sem);
1721                         /*
1722                          * recheck under lock.
1723                          */
1724                         if (!osd_has_index(obj))
1725                                 result = osd_container_init(env, obj, dir);
1726                         else
1727                                 result = 0;
1728                         up(&obj->oo_dir->od_sem);
1729                 } else
1730                         result = -ENOMEM;
1731         } else
1732                 result = 0;
1733
1734         if (result == 0) {
1735                 if (!osd_index_probe(env, obj, feat))
1736                         result = -ENOTDIR;
1737         }
1738         LASSERT(osd_invariant(obj));
1739
1740         return result;
1741 }
1742
1743 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1744                             const struct dt_key *key, struct thandle *handle,
1745                             struct lustre_capa *capa)
1746 {
1747         struct osd_object     *obj = osd_dt_obj(dt);
1748         struct osd_thandle    *oh;
1749         struct iam_path_descr *ipd;
1750         struct iam_container  *bag = &obj->oo_dir->od_container;
1751         int rc;
1752
1753         ENTRY;
1754
1755         LASSERT(osd_invariant(obj));
1756         LASSERT(dt_object_exists(dt));
1757         LASSERT(bag->ic_object == obj->oo_inode);
1758         LASSERT(handle != NULL);
1759
1760         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
1761                 RETURN(-EACCES);
1762
1763         ipd = osd_ipd_get(env, bag);
1764         if (unlikely(ipd == NULL))
1765                 RETURN(-ENOMEM);
1766
1767         oh = container_of0(handle, struct osd_thandle, ot_super);
1768         LASSERT(oh->ot_handle != NULL);
1769         LASSERT(oh->ot_handle->h_transaction != NULL);
1770
1771         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
1772         osd_ipd_put(env, bag, ipd);
1773         LASSERT(osd_invariant(obj));
1774         RETURN(rc);
1775 }
1776
1777 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1778                             struct dt_rec *rec, const struct dt_key *key,
1779                             struct lustre_capa *capa)
1780 {
1781         struct osd_object     *obj = osd_dt_obj(dt);
1782         struct iam_path_descr *ipd;
1783         struct iam_container  *bag = &obj->oo_dir->od_container;
1784         int rc;
1785
1786         ENTRY;
1787
1788         LASSERT(osd_invariant(obj));
1789         LASSERT(dt_object_exists(dt));
1790         LASSERT(bag->ic_object == obj->oo_inode);
1791
1792         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
1793                 return -EACCES;
1794
1795         ipd = osd_ipd_get(env, bag);
1796         if (unlikely(ipd == NULL))
1797                 RETURN(-ENOMEM);
1798
1799         rc = iam_lookup(bag, (const struct iam_key *)key,
1800                         (struct iam_rec *)rec, ipd);
1801         osd_ipd_put(env, bag, ipd);
1802         LASSERT(osd_invariant(obj));
1803
1804         RETURN(rc);
1805 }
1806
1807 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1808                             const struct dt_rec *rec, const struct dt_key *key,
1809                             struct thandle *th, struct lustre_capa *capa)
1810 {
1811         struct osd_object     *obj = osd_dt_obj(dt);
1812         struct iam_path_descr *ipd;
1813         struct osd_thandle    *oh;
1814         struct iam_container  *bag = &obj->oo_dir->od_container;
1815         int rc;
1816
1817         ENTRY;
1818
1819         LASSERT(osd_invariant(obj));
1820         LASSERT(dt_object_exists(dt));
1821         LASSERT(bag->ic_object == obj->oo_inode);
1822         LASSERT(th != NULL);
1823
1824         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
1825                 return -EACCES;
1826
1827         ipd = osd_ipd_get(env, bag);
1828         if (unlikely(ipd == NULL))
1829                 RETURN(-ENOMEM);
1830
1831         oh = container_of0(th, struct osd_thandle, ot_super);
1832         LASSERT(oh->ot_handle != NULL);
1833         LASSERT(oh->ot_handle->h_transaction != NULL);
1834         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
1835                         (struct iam_rec *)rec, ipd);
1836         osd_ipd_put(env, bag, ipd);
1837         LASSERT(osd_invariant(obj));
1838         RETURN(rc);
1839 }
1840
1841 /*
1842  * Iterator operations.
1843  */
1844 struct osd_it {
1845         struct osd_object     *oi_obj;
1846         struct iam_path_descr *oi_ipd;
1847         struct iam_iterator    oi_it;
1848 };
1849
1850 static struct dt_it *osd_it_init(const struct lu_env *env,
1851                                  struct dt_object *dt, int writable,
1852                                  struct lustre_capa *capa)
1853 {
1854         struct osd_it         *it;
1855         struct osd_object     *obj = osd_dt_obj(dt);
1856         struct lu_object      *lo  = &dt->do_lu;
1857         struct iam_path_descr *ipd;
1858         struct iam_container  *bag = &obj->oo_dir->od_container;
1859         __u32                  flags;
1860
1861         LASSERT(lu_object_exists(lo));
1862
1863         if (osd_object_auth(env, dt, capa, writable ? CAPA_OPC_BODY_WRITE :
1864                             CAPA_OPC_BODY_READ))
1865                 return ERR_PTR(-EACCES);
1866
1867         flags = writable ? IAM_IT_MOVE|IAM_IT_WRITE : IAM_IT_MOVE;
1868         OBD_ALLOC_PTR(it);
1869         if (it != NULL) {
1870                 /*
1871                  * XXX: as ipd is allocated within osd_thread_info, assignment
1872                  * below implies that iterator usage is confined within single
1873                  * environment.
1874                  */
1875                 ipd = osd_ipd_get(env, bag);
1876                 if (likely(ipd != NULL)) {
1877                         it->oi_obj = obj;
1878                         it->oi_ipd = ipd;
1879                         lu_object_get(lo);
1880                         iam_it_init(&it->oi_it, bag, flags, ipd);
1881                         return (struct dt_it *)it;
1882                 } else
1883                         OBD_FREE_PTR(it);
1884         }
1885         return ERR_PTR(-ENOMEM);
1886 }
1887
1888 static void osd_it_fini(const struct lu_env *env, struct dt_it *di)
1889 {
1890         struct osd_it     *it = (struct osd_it *)di;
1891         struct osd_object *obj = it->oi_obj;
1892
1893         iam_it_fini(&it->oi_it);
1894         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
1895         lu_object_put(env, &obj->oo_dt.do_lu);
1896         OBD_FREE_PTR(it);
1897 }
1898
1899 static int osd_it_get(const struct lu_env *env,
1900                       struct dt_it *di, const struct dt_key *key)
1901 {
1902         struct osd_it *it = (struct osd_it *)di;
1903
1904         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
1905 }
1906
1907 static void osd_it_put(const struct lu_env *env, struct dt_it *di)
1908 {
1909         struct osd_it *it = (struct osd_it *)di;
1910
1911         iam_it_put(&it->oi_it);
1912 }
1913
1914 static int osd_it_next(const struct lu_env *env, struct dt_it *di)
1915 {
1916         struct osd_it *it = (struct osd_it *)di;
1917
1918         return iam_it_next(&it->oi_it);
1919 }
1920
1921 static int osd_it_del(const struct lu_env *env, struct dt_it *di,
1922                       struct thandle *th)
1923 {
1924         struct osd_it      *it = (struct osd_it *)di;
1925         struct osd_thandle *oh;
1926
1927         LASSERT(th != NULL);
1928
1929         oh = container_of0(th, struct osd_thandle, ot_super);
1930         LASSERT(oh->ot_handle != NULL);
1931         LASSERT(oh->ot_handle->h_transaction != NULL);
1932
1933         return iam_it_rec_delete(oh->ot_handle, &it->oi_it);
1934 }
1935
1936 static struct dt_key *osd_it_key(const struct lu_env *env,
1937                                  const struct dt_it *di)
1938 {
1939         struct osd_it *it = (struct osd_it *)di;
1940
1941         return (struct dt_key *)iam_it_key_get(&it->oi_it);
1942 }
1943
1944 static int osd_it_key_size(const struct lu_env *env, const struct dt_it *di)
1945 {
1946         struct osd_it *it = (struct osd_it *)di;
1947
1948         return iam_it_key_size(&it->oi_it);
1949 }
1950
1951 static struct dt_rec *osd_it_rec(const struct lu_env *env,
1952                                  const struct dt_it *di)
1953 {
1954         struct osd_it *it = (struct osd_it *)di;
1955
1956         return (struct dt_rec *)iam_it_rec_get(&it->oi_it);
1957 }
1958
1959 static __u32 osd_it_store(const struct lu_env *env, const struct dt_it *di)
1960 {
1961         struct osd_it *it = (struct osd_it *)di;
1962
1963         return iam_it_store(&it->oi_it);
1964 }
1965
1966 static int osd_it_load(const struct lu_env *env,
1967                        const struct dt_it *di, __u32 hash)
1968 {
1969         struct osd_it *it = (struct osd_it *)di;
1970
1971         return iam_it_load(&it->oi_it, hash);
1972 }
1973
1974 static struct dt_index_operations osd_index_ops = {
1975         .dio_lookup = osd_index_lookup,
1976         .dio_insert = osd_index_insert,
1977         .dio_delete = osd_index_delete,
1978         .dio_it     = {
1979                 .init     = osd_it_init,
1980                 .fini     = osd_it_fini,
1981                 .get      = osd_it_get,
1982                 .put      = osd_it_put,
1983                 .del      = osd_it_del,
1984                 .next     = osd_it_next,
1985                 .key      = osd_it_key,
1986                 .key_size = osd_it_key_size,
1987                 .rec      = osd_it_rec,
1988                 .store    = osd_it_store,
1989                 .load     = osd_it_load
1990         }
1991 };
1992
1993 static int osd_index_compat_delete(const struct lu_env *env,
1994                                    struct dt_object *dt,
1995                                    const struct dt_key *key,
1996                                    struct thandle *handle,
1997                                    struct lustre_capa *capa)
1998 {
1999         struct osd_object *obj = osd_dt_obj(dt);
2000
2001         LASSERT(handle != NULL);
2002         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2003         ENTRY;
2004
2005 #if 0
2006         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2007                 RETURN(-EACCES);
2008 #endif
2009
2010         RETURN(-EOPNOTSUPP);
2011 }
2012
2013 /*
2014  * Compatibility index operations.
2015  */
2016
2017
2018 static void osd_build_pack(const struct lu_env *env, struct osd_device *osd,
2019                            struct dentry *dentry, struct lu_fid_pack *pack)
2020 {
2021         struct inode  *inode = dentry->d_inode;
2022         struct lu_fid *fid   = &osd_oti_get(env)->oti_fid;
2023
2024         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2025         fid_cpu_to_be(fid, fid);
2026         pack->fp_len = sizeof *fid + 1;
2027         memcpy(pack->fp_area, fid, sizeof *fid);
2028 }
2029
2030 static int osd_index_compat_lookup(const struct lu_env *env,
2031                                    struct dt_object *dt,
2032                                    struct dt_rec *rec, const struct dt_key *key,
2033                                    struct lustre_capa *capa)
2034 {
2035         struct osd_object *obj = osd_dt_obj(dt);
2036
2037         struct osd_device      *osd  = osd_obj2dev(obj);
2038         struct osd_thread_info *info = osd_oti_get(env);
2039         struct inode           *dir;
2040
2041         int result;
2042
2043         /*
2044          * XXX temporary solution.
2045          */
2046         struct dentry *dentry;
2047         struct dentry *parent;
2048
2049         LASSERT(osd_invariant(obj));
2050         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2051         LASSERT(osd_has_index(obj));
2052
2053         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2054                 return -EACCES;
2055
2056         info->oti_str.name = (const char *)key;
2057         info->oti_str.len  = strlen((const char *)key);
2058
2059         dir = obj->oo_inode;
2060         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2061
2062         parent = d_alloc_root(dir);
2063         if (parent == NULL)
2064                 return -ENOMEM;
2065         igrab(dir);
2066         dentry = d_alloc(parent, &info->oti_str);
2067         if (dentry != NULL) {
2068                 struct dentry *d;
2069
2070                 /*
2071                  * XXX passing NULL for nameidata should work for
2072                  * ext3/ldiskfs.
2073                  */
2074                 d = dir->i_op->lookup(dir, dentry, NULL);
2075                 if (d == NULL) {
2076                         /*
2077                          * normal case, result is in @dentry.
2078                          */
2079                         if (dentry->d_inode != NULL) {
2080                                 osd_build_pack(env, osd, dentry,
2081                                                (struct lu_fid_pack *)rec);
2082                                 result = 0;
2083                         } else
2084                                 result = -ENOENT;
2085                  } else {
2086                         /* What? Disconnected alias? Ppheeeww... */
2087                         CERROR("Aliasing where not expected\n");
2088                         result = -EIO;
2089                         dput(d);
2090                 }
2091                 dput(dentry);
2092         } else
2093                 result = -ENOMEM;
2094         dput(parent);
2095         LASSERT(osd_invariant(obj));
2096         return result;
2097 }
2098
2099 static int osd_add_rec(struct osd_thread_info *info, struct osd_device *dev,
2100                        struct inode *dir, struct inode *inode, const char *name)
2101 {
2102         struct dentry *old;
2103         struct dentry *new;
2104         struct dentry *parent;
2105
2106         int result;
2107
2108         info->oti_str.name = name;
2109         info->oti_str.len  = strlen(name);
2110
2111         LASSERT(atomic_read(&dir->i_count) > 0);
2112         result = -ENOMEM;
2113         old = d_alloc(dev->od_obj_area, &info->oti_str);
2114         if (old != NULL) {
2115                 d_instantiate(old, inode);
2116                 igrab(inode);
2117                 LASSERT(atomic_read(&dir->i_count) > 0);
2118                 parent = d_alloc_root(dir);
2119                 if (parent != NULL) {
2120                         igrab(dir);
2121                         LASSERT(atomic_read(&dir->i_count) > 1);
2122                         new = d_alloc(parent, &info->oti_str);
2123                         LASSERT(atomic_read(&dir->i_count) > 1);
2124                         if (new != NULL) {
2125                                 LASSERT(atomic_read(&dir->i_count) > 1);
2126                                 result = dir->i_op->link(old, dir, new);
2127                                 LASSERT(atomic_read(&dir->i_count) > 1);
2128                                 dput(new);
2129                                 LASSERT(atomic_read(&dir->i_count) > 1);
2130                         }
2131                         LASSERT(atomic_read(&dir->i_count) > 1);
2132                         dput(parent);
2133                         LASSERT(atomic_read(&dir->i_count) > 0);
2134                 }
2135                 dput(old);
2136         }
2137         LASSERT(atomic_read(&dir->i_count) > 0);
2138         return result;
2139 }
2140
2141
2142 /*
2143  * XXX Temporary stuff.
2144  */
2145 static int osd_index_compat_insert(const struct lu_env *env,
2146                                    struct dt_object *dt,
2147                                    const struct dt_rec *rec,
2148                                    const struct dt_key *key, struct thandle *th,
2149                                    struct lustre_capa *capa)
2150 {
2151         struct osd_object     *obj = osd_dt_obj(dt);
2152
2153         const char          *name = (const char *)key;
2154
2155         struct lu_device    *ludev = dt->do_lu.lo_dev;
2156         struct lu_object    *luch;
2157
2158         struct osd_thread_info   *info = osd_oti_get(env);
2159         const struct lu_fid_pack *pack  = (const struct lu_fid_pack *)rec;
2160         struct lu_fid            *fid   = &osd_oti_get(env)->oti_fid;
2161
2162         int result;
2163
2164         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2165         LASSERT(osd_invariant(obj));
2166         LASSERT(th != NULL);
2167
2168         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2169                 return -EACCES;
2170
2171         result = fid_unpack(pack, fid);
2172         if (result != 0)
2173                 return result;
2174
2175         luch = lu_object_find(env, ludev->ld_site, fid);
2176         if (!IS_ERR(luch)) {
2177                 if (lu_object_exists(luch)) {
2178                         struct osd_object *child;
2179
2180                         child = osd_obj(lu_object_locate(luch->lo_header,
2181                                                          ludev->ld_type));
2182                         if (child != NULL)
2183                                 result = osd_add_rec(info, osd_obj2dev(obj),
2184                                                      obj->oo_inode,
2185                                                      child->oo_inode, name);
2186                         else {
2187                                 CERROR("No osd slice.\n");
2188                                 result = -ENOENT;
2189                         }
2190                         LASSERT(osd_invariant(obj));
2191                         LASSERT(osd_invariant(child));
2192                 } else {
2193                         CERROR("Sorry.\n");
2194                         result = -ENOENT;
2195                 }
2196                 lu_object_put(env, luch);
2197         } else
2198                 result = PTR_ERR(luch);
2199         LASSERT(osd_invariant(obj));
2200         return result;
2201 }
2202
2203 static struct dt_index_operations osd_index_compat_ops = {
2204         .dio_lookup = osd_index_compat_lookup,
2205         .dio_insert = osd_index_compat_insert,
2206         .dio_delete = osd_index_compat_delete
2207 };
2208
2209 /* type constructor/destructor: osd_type_init, osd_type_fini */
2210 LU_TYPE_INIT_FINI(osd, &osd_key);
2211
2212 static struct lu_context_key osd_key = {
2213         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
2214         .lct_init = osd_key_init,
2215         .lct_fini = osd_key_fini,
2216         .lct_exit = osd_key_exit
2217 };
2218
2219 static void *osd_key_init(const struct lu_context *ctx,
2220                           struct lu_context_key *key)
2221 {
2222         struct osd_thread_info *info;
2223
2224         OBD_ALLOC_PTR(info);
2225         if (info != NULL)
2226                 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
2227         else
2228                 info = ERR_PTR(-ENOMEM);
2229         return info;
2230 }
2231
2232 /* context key destructor: osd_key_fini */
2233 LU_KEY_FINI(osd, struct osd_thread_info);
2234
2235 static void osd_key_exit(const struct lu_context *ctx,
2236                          struct lu_context_key *key, void *data)
2237 {
2238 #if OSD_COUNTERS
2239         struct osd_thread_info *info = data;
2240
2241         LASSERT(info->oti_r_locks == 0);
2242         LASSERT(info->oti_w_locks == 0);
2243         LASSERT(info->oti_txns    == 0);
2244 #endif
2245 }
2246
2247 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
2248                            const char *name, struct lu_device *next)
2249 {
2250         return lu_env_init(&osd_dev(d)->od_env_for_commit, NULL, LCT_MD_THREAD);
2251 }
2252
2253 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
2254 {
2255         struct osd_thread_info *info = osd_oti_get(env);
2256         ENTRY;
2257         if (o->od_obj_area != NULL) {
2258                 dput(o->od_obj_area);
2259                 o->od_obj_area = NULL;
2260         }
2261         osd_oi_fini(info, &o->od_oi);
2262
2263         RETURN(0);
2264 }
2265
2266 static int osd_mount(const struct lu_env *env,
2267                      struct osd_device *o, struct lustre_cfg *cfg)
2268 {
2269         struct lustre_mount_info *lmi;
2270         const char               *dev  = lustre_cfg_string(cfg, 0);
2271         struct osd_thread_info   *info = osd_oti_get(env);
2272         int result;
2273
2274         ENTRY;
2275
2276         if (o->od_mount != NULL) {
2277                 CERROR("Already mounted (%s)\n", dev);
2278                 RETURN(-EEXIST);
2279         }
2280
2281         /* get mount */
2282         lmi = server_get_mount(dev);
2283         if (lmi == NULL) {
2284                 CERROR("Cannot get mount info for %s!\n", dev);
2285                 RETURN(-EFAULT);
2286         }
2287
2288         LASSERT(lmi != NULL);
2289         /* save lustre_mount_info in dt_device */
2290         o->od_mount = lmi;
2291
2292         result = osd_oi_init(info, &o->od_oi, &o->od_dt_dev);
2293         if (result == 0) {
2294                 struct dentry *d;
2295
2296                 d = simple_mkdir(osd_sb(o)->s_root, "*OBJ-TEMP*", 0777, 1);
2297                 if (!IS_ERR(d)) {
2298                         o->od_obj_area = d;
2299                 } else
2300                         result = PTR_ERR(d);
2301         }
2302         if (result != 0)
2303                 osd_shutdown(env, o);
2304         RETURN(result);
2305 }
2306
2307 static struct lu_device *osd_device_fini(const struct lu_env *env,
2308                                          struct lu_device *d)
2309 {
2310         ENTRY;
2311
2312         shrink_dcache_sb(osd_sb(osd_dev(d)));
2313         osd_sync(env, lu2dt_dev(d));
2314
2315         if (osd_dev(d)->od_mount)
2316                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
2317                                  osd_dev(d)->od_mount->lmi_mnt);
2318         osd_dev(d)->od_mount = NULL;
2319
2320         lu_env_fini(&osd_dev(d)->od_env_for_commit);
2321         RETURN(NULL);
2322 }
2323
2324 static struct lu_device *osd_device_alloc(const struct lu_env *env,
2325                                           struct lu_device_type *t,
2326                                           struct lustre_cfg *cfg)
2327 {
2328         struct lu_device  *l;
2329         struct osd_device *o;
2330
2331         OBD_ALLOC_PTR(o);
2332         if (o != NULL) {
2333                 int result;
2334
2335                 result = dt_device_init(&o->od_dt_dev, t);
2336                 if (result == 0) {
2337                         l = osd2lu_dev(o);
2338                         l->ld_ops = &osd_lu_ops;
2339                         o->od_dt_dev.dd_ops = &osd_dt_ops;
2340                         spin_lock_init(&o->od_osfs_lock);
2341                         o->od_osfs_age = cfs_time_shift_64(-1000);
2342                         o->od_capa_hash = init_capa_hash();
2343                         if (o->od_capa_hash == NULL)
2344                                 l = ERR_PTR(-ENOMEM);
2345                 } else
2346                         l = ERR_PTR(result);
2347         } else
2348                 l = ERR_PTR(-ENOMEM);
2349         return l;
2350 }
2351
2352 static void osd_device_free(const struct lu_env *env, struct lu_device *d)
2353 {
2354         struct osd_device *o = osd_dev(d);
2355
2356         cleanup_capa_hash(o->od_capa_hash);
2357         dt_device_fini(&o->od_dt_dev);
2358         OBD_FREE_PTR(o);
2359 }
2360
2361 static int osd_process_config(const struct lu_env *env,
2362                               struct lu_device *d, struct lustre_cfg *cfg)
2363 {
2364         struct osd_device *o = osd_dev(d);
2365         int err;
2366         ENTRY;
2367
2368         switch(cfg->lcfg_command) {
2369         case LCFG_SETUP:
2370                 err = osd_mount(env, o, cfg);
2371                 break;
2372         case LCFG_CLEANUP:
2373                 err = osd_shutdown(env, o);
2374                 break;
2375         default:
2376                 err = -ENOTTY;
2377         }
2378
2379         RETURN(err);
2380 }
2381 extern void ldiskfs_orphan_cleanup (struct super_block * sb,
2382                                     struct ldiskfs_super_block * es);
2383
2384 static int osd_recovery_complete(const struct lu_env *env,
2385                                  struct lu_device *d)
2386 {
2387         struct osd_device *o = osd_dev(d);
2388         ENTRY;
2389         /* TODO: orphans handling */
2390         ldiskfs_orphan_cleanup(osd_sb(o), LDISKFS_SB(osd_sb(o))->s_es);
2391         RETURN(0);
2392 }
2393
2394 static struct inode *osd_iget(struct osd_thread_info *info,
2395                               struct osd_device *dev,
2396                               const struct osd_inode_id *id)
2397 {
2398         struct inode *inode;
2399
2400         inode = iget(osd_sb(dev), id->oii_ino);
2401         if (inode == NULL) {
2402                 CERROR("no inode\n");
2403                 inode = ERR_PTR(-EACCES);
2404         } else if (is_bad_inode(inode)) {
2405                 CERROR("bad inode\n");
2406                 iput(inode);
2407                 inode = ERR_PTR(-ENOENT);
2408         } else if (inode->i_generation != id->oii_gen) {
2409                 CERROR("stale inode\n");
2410                 iput(inode);
2411                 inode = ERR_PTR(-ESTALE);
2412         }
2413
2414         return inode;
2415
2416 }
2417
2418 static int osd_fid_lookup(const struct lu_env *env,
2419                           struct osd_object *obj, const struct lu_fid *fid)
2420 {
2421         struct osd_thread_info *info;
2422         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
2423         struct osd_device      *dev;
2424         struct osd_inode_id    *id;
2425         struct osd_oi          *oi;
2426         struct inode           *inode;
2427         int                     result;
2428
2429         LASSERT(osd_invariant(obj));
2430         LASSERT(obj->oo_inode == NULL);
2431         LASSERT(fid_is_sane(fid));
2432         /*
2433          * This assertion checks that osd layer sees only local
2434          * fids. Unfortunately it is somewhat expensive (does a
2435          * cache-lookup). Disabling it for production/acceptance-testing.
2436          */
2437         LASSERT(1 || fid_is_local(ldev->ld_site, fid));
2438
2439         ENTRY;
2440
2441         info = osd_oti_get(env);
2442         dev  = osd_dev(ldev);
2443         id   = &info->oti_id;
2444         oi   = &dev->od_oi;
2445
2446         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
2447                 RETURN(-ENOENT);
2448
2449         result = osd_oi_lookup(info, oi, fid, id);
2450         if (result == 0) {
2451                 inode = osd_iget(info, dev, id);
2452                 if (!IS_ERR(inode)) {
2453                         obj->oo_inode = inode;
2454                         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
2455                         result = 0;
2456                 } else
2457                         /*
2458                          * If fid wasn't found in oi, inode-less object is
2459                          * created, for which lu_object_exists() returns
2460                          * false. This is used in a (frequent) case when
2461                          * objects are created as locking anchors or
2462                          * place holders for objects yet to be created.
2463                          */
2464                         result = PTR_ERR(inode);
2465         } else if (result == -ENOENT)
2466                 result = 0;
2467         LASSERT(osd_invariant(obj));
2468         RETURN(result);
2469 }
2470
2471 static void osd_inode_getattr(const struct lu_env *env,
2472                               struct inode *inode, struct lu_attr *attr)
2473 {
2474         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2475                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2476                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
2477
2478         attr->la_atime      = LTIME_S(inode->i_atime);
2479         attr->la_mtime      = LTIME_S(inode->i_mtime);
2480         attr->la_ctime      = LTIME_S(inode->i_ctime);
2481         attr->la_mode       = inode->i_mode;
2482         attr->la_size       = i_size_read(inode);
2483         attr->la_blocks     = inode->i_blocks;
2484         attr->la_uid        = inode->i_uid;
2485         attr->la_gid        = inode->i_gid;
2486         attr->la_flags      = LDISKFS_I(inode)->i_flags;
2487         attr->la_nlink      = inode->i_nlink;
2488         attr->la_rdev       = inode->i_rdev;
2489         attr->la_blksize    = ll_inode_blksize(inode);
2490         attr->la_blkbits    = inode->i_blkbits;
2491 }
2492
2493 /*
2494  * Helpers.
2495  */
2496
2497 static int lu_device_is_osd(const struct lu_device *d)
2498 {
2499         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
2500 }
2501
2502 static struct osd_object *osd_obj(const struct lu_object *o)
2503 {
2504         LASSERT(lu_device_is_osd(o->lo_dev));
2505         return container_of0(o, struct osd_object, oo_dt.do_lu);
2506 }
2507
2508 static struct osd_device *osd_dt_dev(const struct dt_device *d)
2509 {
2510         LASSERT(lu_device_is_osd(&d->dd_lu_dev));
2511         return container_of0(d, struct osd_device, od_dt_dev);
2512 }
2513
2514 static struct osd_device *osd_dev(const struct lu_device *d)
2515 {
2516         LASSERT(lu_device_is_osd(d));
2517         return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
2518 }
2519
2520 static struct osd_object *osd_dt_obj(const struct dt_object *d)
2521 {
2522         return osd_obj(&d->do_lu);
2523 }
2524
2525 static struct osd_device *osd_obj2dev(const struct osd_object *o)
2526 {
2527         return osd_dev(o->oo_dt.do_lu.lo_dev);
2528 }
2529
2530 static struct lu_device *osd2lu_dev(struct osd_device *osd)
2531 {
2532         return &osd->od_dt_dev.dd_lu_dev;
2533 }
2534
2535 static struct super_block *osd_sb(const struct osd_device *dev)
2536 {
2537         return dev->od_mount->lmi_mnt->mnt_sb;
2538 }
2539
2540 static journal_t *osd_journal(const struct osd_device *dev)
2541 {
2542         return LDISKFS_SB(osd_sb(dev))->s_journal;
2543 }
2544
2545 static int osd_has_index(const struct osd_object *obj)
2546 {
2547         return obj->oo_dt.do_index_ops != NULL;
2548 }
2549
2550 static int osd_object_invariant(const struct lu_object *l)
2551 {
2552         return osd_invariant(osd_obj(l));
2553 }
2554
2555 static struct lu_object_operations osd_lu_obj_ops = {
2556         .loo_object_init      = osd_object_init,
2557         .loo_object_delete    = osd_object_delete,
2558         .loo_object_release   = osd_object_release,
2559         .loo_object_free      = osd_object_free,
2560         .loo_object_print     = osd_object_print,
2561         .loo_object_invariant = osd_object_invariant
2562 };
2563
2564 static struct lu_device_operations osd_lu_ops = {
2565         .ldo_object_alloc      = osd_object_alloc,
2566         .ldo_process_config    = osd_process_config,
2567         .ldo_recovery_complete = osd_recovery_complete
2568 };
2569
2570 static struct lu_device_type_operations osd_device_type_ops = {
2571         .ldto_init = osd_type_init,
2572         .ldto_fini = osd_type_fini,
2573
2574         .ldto_device_alloc = osd_device_alloc,
2575         .ldto_device_free  = osd_device_free,
2576
2577         .ldto_device_init    = osd_device_init,
2578         .ldto_device_fini    = osd_device_fini
2579 };
2580
2581 static struct lu_device_type osd_device_type = {
2582         .ldt_tags     = LU_DEVICE_DT,
2583         .ldt_name     = LUSTRE_OSD_NAME,
2584         .ldt_ops      = &osd_device_type_ops,
2585         .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
2586 };
2587
2588 /*
2589  * lprocfs legacy support.
2590  */
2591 static struct lprocfs_vars lprocfs_osd_obd_vars[] = {
2592         { 0 }
2593 };
2594
2595 static struct lprocfs_vars lprocfs_osd_module_vars[] = {
2596         { 0 }
2597 };
2598
2599 static struct obd_ops osd_obd_device_ops = {
2600         .o_owner = THIS_MODULE
2601 };
2602
2603 static void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars)
2604 {
2605     lvars->module_vars  = lprocfs_osd_module_vars;
2606     lvars->obd_vars     = lprocfs_osd_obd_vars;
2607 }
2608
2609
2610 static int __init osd_mod_init(void)
2611 {
2612         struct lprocfs_static_vars lvars;
2613
2614         lprocfs_osd_init_vars(&lvars);
2615         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
2616                                    LUSTRE_OSD_NAME, &osd_device_type);
2617 }
2618
2619 static void __exit osd_mod_exit(void)
2620 {
2621         class_unregister_type(LUSTRE_OSD_NAME);
2622 }
2623
2624 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2625 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
2626 MODULE_LICENSE("GPL");
2627
2628 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);