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 = sb->s_op->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 };
 816
 817 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
 818                           enum dt_txn_op op)
 819 {
 820         LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits));
 821         return osd_dto_credits[op];
 822 }
 823
 824 static struct dt_device_operations osd_dt_ops = {
 825         .dt_root_get       = osd_root_get,
 826         .dt_statfs         = osd_statfs,
 827         .dt_trans_start    = osd_trans_start,
 828         .dt_trans_stop     = osd_trans_stop,
 829         .dt_conf_get       = osd_conf_get,
 830         .dt_sync           = osd_sync,
 831         .dt_ro             = osd_ro,
 832         .dt_credit_get     = osd_credit_get,
 833         .dt_init_capa_ctxt = osd_init_capa_ctxt,
 834 };
 835
 836 static void osd_object_read_lock(const struct lu_env *env,
 837                                  struct dt_object *dt)
 838 {
 839         struct osd_object *obj = osd_dt_obj(dt);
 840
 841         LASSERT(osd_invariant(obj));
 842
 843         OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
 844         down_read(&obj->oo_sem);
 845 #if OSD_COUNTERS
 846         {
 847                 struct osd_thread_info *oti = osd_oti_get(env);
 848
 849                 LASSERT(obj->oo_owner == NULL);
 850                 oti->oti_r_locks++;
 851         }
 852 #endif
 853 }
 854
 855 static void osd_object_write_lock(const struct lu_env *env,
 856                                   struct dt_object *dt)
 857 {
 858         struct osd_object *obj = osd_dt_obj(dt);
 859
 860         LASSERT(osd_invariant(obj));
 861
 862         OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
 863         down_write(&obj->oo_sem);
 864 #if OSD_COUNTERS
 865         {
 866                 struct osd_thread_info *oti = osd_oti_get(env);
 867
 868                 LASSERT(obj->oo_owner == NULL);
 869                 obj->oo_owner = env;
 870                 oti->oti_w_locks++;
 871         }
 872 #endif
 873 }
 874
 875 static void osd_object_read_unlock(const struct lu_env *env,
 876                                    struct dt_object *dt)
 877 {
 878         struct osd_object *obj = osd_dt_obj(dt);
 879
 880         LASSERT(osd_invariant(obj));
 881 #if OSD_COUNTERS
 882         {
 883                 struct osd_thread_info *oti = osd_oti_get(env);
 884
 885                 LASSERT(oti->oti_r_locks > 0);
 886                 oti->oti_r_locks--;
 887         }
 888 #endif
 889         up_read(&obj->oo_sem);
 890 }
 891
 892 static void osd_object_write_unlock(const struct lu_env *env,
 893                                     struct dt_object *dt)
 894 {
 895         struct osd_object *obj = osd_dt_obj(dt);
 896
 897         LASSERT(osd_invariant(obj));
 898 #if OSD_COUNTERS
 899         {
 900                 struct osd_thread_info *oti = osd_oti_get(env);
 901
 902                 LASSERT(obj->oo_owner == env);
 903                 LASSERT(oti->oti_w_locks > 0);
 904                 oti->oti_w_locks--;
 905                 obj->oo_owner = NULL;
 906         }
 907 #endif
 908         up_write(&obj->oo_sem);
 909 }
 910
 911 static int capa_is_sane(const struct lu_env *env,
 912                         struct osd_device *dev,
 913                         struct lustre_capa *capa,
 914                         struct lustre_capa_key *keys)
 915 {
 916         struct osd_thread_info *oti = osd_oti_get(env);
 917         struct obd_capa *oc;
 918         int i, rc = 0;
 919         ENTRY;
 920
 921         oc = capa_lookup(dev->od_capa_hash, capa, 0);
 922         if (oc) {
 923                 if (capa_is_expired(oc)) {
 924                         DEBUG_CAPA(D_ERROR, capa, "expired");
 925                         rc = -ESTALE;
 926                 }
 927                 capa_put(oc);
 928                 RETURN(rc);
 929         }
 930
 931         spin_lock(&capa_lock);
 932         for (i = 0; i < 2; i++) {
 933                 if (keys[i].lk_keyid == capa->lc_keyid) {
 934                         oti->oti_capa_key = keys[i];
 935                         break;
 936                 }
 937         }
 938         spin_unlock(&capa_lock);
 939
 940         if (i == 2) {
 941                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
 942                 RETURN(-ESTALE);
 943         }
 944
 945         rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
 946         if (rc)
 947                 RETURN(rc);
 948         if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
 949         {
 950                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
 951                 RETURN(-EACCES);
 952         }
 953
 954         oc = capa_add(dev->od_capa_hash, capa);
 955         capa_put(oc);
 956
 957         RETURN(0);
 958 }
 959
 960 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
 961                            struct lustre_capa *capa, __u64 opc)
 962 {
 963         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
 964         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
 965         int rc;
 966
 967         if (!dev->od_fl_capa)
 968                 return 0;
 969
 970         if (capa == BYPASS_CAPA)
 971                 return 0;
 972
 973         if (!capa) {
 974                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
 975                 return -EACCES;
 976         }
 977
 978         if (!lu_fid_eq(fid, &capa->lc_fid)) {
 979                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
 980                            PFID(fid));
 981                 return -EACCES;
 982         }
 983
 984         if (!capa_opc_supported(capa, opc)) {
 985                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
 986                 return -EACCES;
 987         }
 988
 989         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
 990                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
 991                 return -EACCES;
 992         }
 993
 994         return 0;
 995 }
 996
 997 static int osd_attr_get(const struct lu_env *env,
 998                         struct dt_object *dt,
 999                         struct lu_attr *attr,
1000                         struct lustre_capa *capa)
1001 {
1002         struct osd_object *obj = osd_dt_obj(dt);
1003
1004         LASSERT(dt_object_exists(dt));
1005         LASSERT(osd_invariant(obj));
1006
1007         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1008                 return -EACCES;
1009
1010         spin_lock(&obj->oo_guard);
1011         osd_inode_getattr(env, obj->oo_inode, attr);
1012         spin_unlock(&obj->oo_guard);
1013         return 0;
1014 }
1015
1016 static int osd_attr_set(const struct lu_env *env,
1017                         struct dt_object *dt,
1018                         const struct lu_attr *attr,
1019                         struct thandle *handle,
1020                         struct lustre_capa *capa)
1021 {
1022         struct osd_object *obj = osd_dt_obj(dt);
1023
1024         LASSERT(handle != NULL);
1025         LASSERT(dt_object_exists(dt));
1026         LASSERT(osd_invariant(obj));
1027
1028         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1029                 return -EACCES;
1030
1031         spin_lock(&obj->oo_guard);
1032         osd_inode_setattr(env, obj->oo_inode, attr);
1033         spin_unlock(&obj->oo_guard);
1034
1035         mark_inode_dirty(obj->oo_inode);
1036         return 0;
1037 }
1038
1039 static struct timespec *osd_inode_time(const struct lu_env *env,
1040                                        struct inode *inode, __u64 seconds)
1041 {
1042         struct osd_thread_info *oti = osd_oti_get(env);
1043         struct timespec        *t   = &oti->oti_time;
1044
1045         t->tv_sec  = seconds;
1046         t->tv_nsec = 0;
1047         *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1048         return t;
1049 }
1050
1051 static void osd_inode_setattr(const struct lu_env *env,
1052                               struct inode *inode, const struct lu_attr *attr)
1053 {
1054         __u64 bits;
1055
1056         bits = attr->la_valid;
1057
1058         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1059
1060         if (bits & LA_ATIME)
1061                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1062         if (bits & LA_CTIME)
1063                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1064         if (bits & LA_MTIME)
1065                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1066         if (bits & LA_SIZE)
1067                 LDISKFS_I(inode)->i_disksize = inode->i_size = attr->la_size;
1068         if (bits & LA_BLOCKS)
1069                 inode->i_blocks = attr->la_blocks;
1070         if (bits & LA_MODE)
1071                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1072                         (attr->la_mode & ~S_IFMT);
1073         if (bits & LA_UID)
1074                 inode->i_uid    = attr->la_uid;
1075         if (bits & LA_GID)
1076                 inode->i_gid    = attr->la_gid;
1077         if (bits & LA_NLINK)
1078                 inode->i_nlink  = attr->la_nlink;
1079         if (bits & LA_RDEV)
1080                 inode->i_rdev   = attr->la_rdev;
1081         if (bits & LA_BLKSIZE)
1082                 inode->i_blksize = attr->la_blksize;
1083
1084         if (bits & LA_FLAGS) {
1085                 struct ldiskfs_inode_info *li = LDISKFS_I(inode);
1086
1087                 li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
1088                         (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
1089         }
1090 }
1091
1092 /*
1093  * Object creation.
1094  *
1095  * XXX temporary solution.
1096  */
1097
1098 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1099                           struct lu_attr *attr, struct thandle *th)
1100 {
1101         return 0;
1102 }
1103
1104 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1105                            struct lu_attr *attr, struct thandle *th)
1106 {
1107         LASSERT(obj->oo_inode != NULL);
1108
1109         osd_object_init0(obj);
1110         return 0;
1111 }
1112
1113 extern struct inode *ldiskfs_create_inode(handle_t *handle,
1114                                           struct inode * dir, int mode);
1115
1116 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1117                       umode_t mode,
1118                       struct dt_allocation_hint *hint,
1119                       struct thandle *th)
1120 {
1121         int result;
1122         struct osd_device  *osd = osd_obj2dev(obj);
1123         struct osd_thandle *oth;
1124         struct inode       *parent;
1125         struct inode       *inode;
1126
1127         LASSERT(osd_invariant(obj));
1128         LASSERT(obj->oo_inode == NULL);
1129         LASSERT(osd->od_obj_area != NULL);
1130
1131         oth = container_of(th, struct osd_thandle, ot_super);
1132         LASSERT(oth->ot_handle->h_transaction != NULL);
1133
1134         if (hint && hint->dah_parent)
1135                 parent = osd_dt_obj(hint->dah_parent)->oo_inode;
1136         else
1137                 parent = osd->od_obj_area->d_inode;
1138         LASSERT(parent->i_op != NULL);
1139
1140         inode = ldiskfs_create_inode(oth->ot_handle, parent, mode);
1141         if (!IS_ERR(inode)) {
1142                 obj->oo_inode = inode;
1143                 result = 0;
1144         } else
1145                 result = PTR_ERR(inode);
1146         LASSERT(osd_invariant(obj));
1147         return result;
1148 }
1149
1150
1151 extern int iam_lvar_create(struct inode *obj, int keysize, int ptrsize,
1152                            int recsize, handle_t *handle);
1153
1154 enum {
1155         OSD_NAME_LEN = 255
1156 };
1157
1158 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1159                      struct lu_attr *attr,
1160                      struct dt_allocation_hint *hint,
1161                      struct thandle *th)
1162 {
1163         int result;
1164         struct osd_thandle *oth;
1165
1166         LASSERT(S_ISDIR(attr->la_mode));
1167
1168         oth = container_of(th, struct osd_thandle, ot_super);
1169         LASSERT(oth->ot_handle->h_transaction != NULL);
1170         result = osd_mkfile(info, obj, (attr->la_mode &
1171                             (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1172         if (result == 0) {
1173                 LASSERT(obj->oo_inode != NULL);
1174                 /*
1175                  * XXX uh-oh... call low-level iam function directly.
1176                  */
1177                 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1178                                          sizeof (struct lu_fid_pack),
1179                                          oth->ot_handle);
1180         }
1181         return result;
1182 }
1183
1184 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1185                      struct lu_attr *attr,
1186                      struct dt_allocation_hint *hint,
1187                      struct thandle *th)
1188 {
1189         LASSERT(S_ISREG(attr->la_mode));
1190         return osd_mkfile(info, obj, (attr->la_mode &
1191                                (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1192 }
1193
1194 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1195                      struct lu_attr *attr,
1196                      struct dt_allocation_hint *hint,
1197                      struct thandle *th)
1198 {
1199         LASSERT(S_ISLNK(attr->la_mode));
1200         return osd_mkfile(info, obj, (attr->la_mode &
1201                               (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1202 }
1203
1204 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1205                      struct lu_attr *attr,
1206                      struct dt_allocation_hint *hint,
1207                      struct thandle *th)
1208 {
1209         int result;
1210         struct osd_device *osd = osd_obj2dev(obj);
1211         struct inode      *dir;
1212         umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1213
1214         LASSERT(osd_invariant(obj));
1215         LASSERT(obj->oo_inode == NULL);
1216         LASSERT(osd->od_obj_area != NULL);
1217         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1218                 S_ISFIFO(mode) || S_ISSOCK(mode));
1219
1220         dir = osd->od_obj_area->d_inode;
1221         LASSERT(dir->i_op != NULL);
1222
1223         result = osd_mkfile(info, obj, mode, hint, th);
1224         if (result == 0) {
1225                 LASSERT(obj->oo_inode != NULL);
1226                 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1227         }
1228         LASSERT(osd_invariant(obj));
1229         return result;
1230 }
1231
1232 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1233                               struct lu_attr *,
1234                               struct dt_allocation_hint *hint,
1235                               struct thandle *);
1236
1237 static osd_obj_type_f osd_create_type_f(__u32 mode)
1238 {
1239         osd_obj_type_f result;
1240
1241         switch (mode) {
1242         case S_IFDIR:
1243                 result = osd_mkdir;
1244                 break;
1245         case S_IFREG:
1246                 result = osd_mkreg;
1247                 break;
1248         case S_IFLNK:
1249                 result = osd_mksym;
1250                 break;
1251         case S_IFCHR:
1252         case S_IFBLK:
1253         case S_IFIFO:
1254         case S_IFSOCK:
1255                 result = osd_mknod;
1256                 break;
1257         default:
1258                 LBUG();
1259                 break;
1260         }
1261         return result;
1262 }
1263
1264
1265 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1266                         struct dt_object *parent, umode_t child_mode)
1267 {
1268         LASSERT(ah);
1269
1270         memset(ah, 0, sizeof(*ah));
1271         ah->dah_parent = parent;
1272         ah->dah_mode = child_mode;
1273 }
1274
1275
1276 /*
1277  * Concurrency: @dt is write locked.
1278  */
1279 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1280                              struct lu_attr *attr,
1281                              struct dt_allocation_hint *hint,
1282                              struct thandle *th)
1283 {
1284         const struct lu_fid    *fid  = lu_object_fid(&dt->do_lu);
1285         struct osd_object      *obj  = osd_dt_obj(dt);
1286         struct osd_device      *osd  = osd_obj2dev(obj);
1287         struct osd_thread_info *info = osd_oti_get(env);
1288         int result;
1289
1290         ENTRY;
1291
1292         LASSERT(osd_invariant(obj));
1293         LASSERT(!dt_object_exists(dt));
1294         LASSERT(osd_write_locked(env, obj));
1295         LASSERT(th != NULL);
1296
1297         /*
1298          * XXX missing: Quote handling.
1299          */
1300
1301         result = osd_create_pre(info, obj, attr, th);
1302         if (result == 0) {
1303                 result = osd_create_type_f(attr->la_mode & S_IFMT)(info, obj,
1304                                                                 attr, hint, th);
1305                 if (result == 0)
1306                         result = osd_create_post(info, obj, attr, th);
1307         }
1308         if (result == 0) {
1309                 struct osd_inode_id *id = &info->oti_id;
1310
1311                 LASSERT(obj->oo_inode != NULL);
1312
1313                 id->oii_ino = obj->oo_inode->i_ino;
1314                 id->oii_gen = obj->oo_inode->i_generation;
1315
1316                 result = osd_oi_insert(info, &osd->od_oi, fid, id, th);
1317         }
1318
1319         LASSERT(ergo(result == 0, dt_object_exists(dt)));
1320         LASSERT(osd_invariant(obj));
1321         RETURN(result);
1322 }
1323
1324 /*
1325  * Concurrency: @dt is write locked.
1326  */
1327 static void osd_object_ref_add(const struct lu_env *env,
1328                                struct dt_object *dt,
1329                                struct thandle *th)
1330 {
1331         struct osd_object *obj = osd_dt_obj(dt);
1332         struct inode *inode = obj->oo_inode;
1333
1334         LASSERT(osd_invariant(obj));
1335         LASSERT(dt_object_exists(dt));
1336         LASSERT(osd_write_locked(env, obj));
1337         LASSERT(th != NULL);
1338
1339         spin_lock(&obj->oo_guard);
1340         if (inode->i_nlink < LDISKFS_LINK_MAX) {
1341                 inode->i_nlink ++;
1342                 spin_unlock(&obj->oo_guard);
1343                 mark_inode_dirty(inode);
1344         } else {
1345                 spin_unlock(&obj->oo_guard);
1346                 LU_OBJECT_DEBUG(D_ERROR, env, &dt->do_lu,
1347                                 "Overflowed nlink\n");
1348         }
1349         LASSERT(osd_invariant(obj));
1350 }
1351
1352 /*
1353  * Concurrency: @dt is write locked.
1354  */
1355 static void osd_object_ref_del(const struct lu_env *env,
1356                                struct dt_object *dt,
1357                                struct thandle *th)
1358 {
1359         struct osd_object *obj = osd_dt_obj(dt);
1360         struct inode *inode = obj->oo_inode;
1361
1362         LASSERT(osd_invariant(obj));
1363         LASSERT(dt_object_exists(dt));
1364         LASSERT(osd_write_locked(env, obj));
1365         LASSERT(th != NULL);
1366
1367         spin_lock(&obj->oo_guard);
1368         if (inode->i_nlink > 0) {
1369                 inode->i_nlink --;
1370                 spin_unlock(&obj->oo_guard);
1371                 mark_inode_dirty(inode);
1372         } else {
1373                 spin_unlock(&obj->oo_guard);
1374                 LU_OBJECT_DEBUG(D_ERROR, env, &dt->do_lu,
1375                                 "Underflowed nlink\n");
1376         }
1377         LASSERT(osd_invariant(obj));
1378 }
1379
1380 /*
1381  * Concurrency: @dt is read locked.
1382  */
1383 static int osd_xattr_get(const struct lu_env *env,
1384                          struct dt_object *dt,
1385                          struct lu_buf *buf,
1386                          const char *name,
1387                          struct lustre_capa *capa)
1388 {
1389         struct osd_object      *obj    = osd_dt_obj(dt);
1390         struct inode           *inode  = obj->oo_inode;
1391         struct osd_thread_info *info   = osd_oti_get(env);
1392         struct dentry          *dentry = &info->oti_dentry;
1393
1394         LASSERT(dt_object_exists(dt));
1395         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1396         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1397
1398         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1399                 return -EACCES;
1400
1401         dentry->d_inode = inode;
1402         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
1403 }
1404
1405 /*
1406  * Concurrency: @dt is write locked.
1407  */
1408 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1409                          const struct lu_buf *buf, const char *name, int fl,
1410                          struct thandle *handle, struct lustre_capa *capa)
1411 {
1412         int fs_flags;
1413
1414         struct osd_object      *obj    = osd_dt_obj(dt);
1415         struct inode           *inode  = obj->oo_inode;
1416         struct osd_thread_info *info   = osd_oti_get(env);
1417         struct dentry          *dentry = &info->oti_dentry;
1418
1419         LASSERT(dt_object_exists(dt));
1420         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1421         LASSERT(osd_write_locked(env, obj));
1422         LASSERT(handle != NULL);
1423
1424         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1425                 return -EACCES;
1426
1427         dentry->d_inode = inode;
1428
1429         fs_flags = 0;
1430         if (fl & LU_XATTR_REPLACE)
1431                 fs_flags |= XATTR_REPLACE;
1432
1433         if (fl & LU_XATTR_CREATE)
1434                 fs_flags |= XATTR_CREATE;
1435
1436         return inode->i_op->setxattr(dentry, name,
1437                                      buf->lb_buf, buf->lb_len, fs_flags);
1438 }
1439
1440 /*
1441  * Concurrency: @dt is read locked.
1442  */
1443 static int osd_xattr_list(const struct lu_env *env,
1444                           struct dt_object *dt,
1445                           struct lu_buf *buf,
1446                           struct lustre_capa *capa)
1447 {
1448         struct osd_object      *obj    = osd_dt_obj(dt);
1449         struct inode           *inode  = obj->oo_inode;
1450         struct osd_thread_info *info   = osd_oti_get(env);
1451         struct dentry          *dentry = &info->oti_dentry;
1452
1453         LASSERT(dt_object_exists(dt));
1454         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
1455         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1456
1457         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1458                 return -EACCES;
1459
1460         dentry->d_inode = inode;
1461         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
1462 }
1463
1464 /*
1465  * Concurrency: @dt is write locked.
1466  */
1467 static int osd_xattr_del(const struct lu_env *env,
1468                          struct dt_object *dt,
1469                          const char *name,
1470                          struct thandle *handle,
1471                          struct lustre_capa *capa)
1472 {
1473         struct osd_object      *obj    = osd_dt_obj(dt);
1474         struct inode           *inode  = obj->oo_inode;
1475         struct osd_thread_info *info   = osd_oti_get(env);
1476         struct dentry          *dentry = &info->oti_dentry;
1477
1478         LASSERT(dt_object_exists(dt));
1479         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
1480         LASSERT(osd_write_locked(env, obj));
1481         LASSERT(handle != NULL);
1482
1483         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1484                 return -EACCES;
1485
1486         dentry->d_inode = inode;
1487         return inode->i_op->removexattr(dentry, name);
1488 }
1489
1490 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1491                                      struct dt_object *dt,
1492                                      struct lustre_capa *old,
1493                                      __u32 uid, __u64 opc)
1494 {
1495         struct osd_thread_info *info = osd_oti_get(env);
1496         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1497         struct osd_object *obj = osd_dt_obj(dt);
1498         struct osd_device *dev = osd_obj2dev(obj);
1499         struct lustre_capa_key *key = &info->oti_capa_key;
1500         struct lustre_capa *capa = &info->oti_capa;
1501         struct obd_capa *oc;
1502         int rc;
1503         ENTRY;
1504
1505         if (!dev->od_fl_capa)
1506                 RETURN(ERR_PTR(-ENOENT));
1507
1508         LASSERT(dt_object_exists(dt));
1509         LASSERT(osd_invariant(obj));
1510
1511         /* renewal sanity check */
1512         if (old && osd_object_auth(env, dt, old, opc))
1513                 RETURN(ERR_PTR(-EACCES));
1514
1515         capa->lc_fid = *fid;
1516         capa->lc_opc = opc;
1517         capa->lc_uid = uid;
1518         capa->lc_flags = dev->od_capa_alg << 24;
1519         capa->lc_timeout = dev->od_capa_timeout;
1520         capa->lc_expiry = 0;
1521
1522         oc = capa_lookup(dev->od_capa_hash, capa, 1);
1523         if (oc) {
1524                 LASSERT(!capa_is_expired(oc));
1525                 RETURN(oc);
1526         }
1527
1528         spin_lock(&capa_lock);
1529         *key = dev->od_capa_keys[1];
1530         spin_unlock(&capa_lock);
1531
1532         capa->lc_keyid = key->lk_keyid;
1533         capa->lc_expiry = CURRENT_SECONDS + dev->od_capa_timeout;
1534
1535         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1536         if (rc) {
1537                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1538                 RETURN(ERR_PTR(rc));
1539         }
1540
1541         oc = capa_add(dev->od_capa_hash, capa);
1542         RETURN(oc);
1543 }
1544
1545 static struct dt_object_operations osd_obj_ops = {
1546         .do_read_lock    = osd_object_read_lock,
1547         .do_write_lock   = osd_object_write_lock,
1548         .do_read_unlock  = osd_object_read_unlock,
1549         .do_write_unlock = osd_object_write_unlock,
1550         .do_attr_get     = osd_attr_get,
1551         .do_attr_set     = osd_attr_set,
1552         .do_ah_init      = osd_ah_init,
1553         .do_create       = osd_object_create,
1554         .do_index_try    = osd_index_try,
1555         .do_ref_add      = osd_object_ref_add,
1556         .do_ref_del      = osd_object_ref_del,
1557         .do_xattr_get    = osd_xattr_get,
1558         .do_xattr_set    = osd_xattr_set,
1559         .do_xattr_del    = osd_xattr_del,
1560         .do_xattr_list   = osd_xattr_list,
1561         .do_capa_get     = osd_capa_get,
1562 };
1563
1564 /*
1565  * Body operations.
1566  */
1567
1568 /*
1569  * XXX: Another layering violation for now.
1570  *
1571  * We don't want to use ->f_op->read methods, because generic file write
1572  *
1573  *         - serializes on ->i_sem, and
1574  *
1575  *         - does a lot of extra work like balance_dirty_pages(),
1576  *
1577  * which doesn't work for globally shared files like /last-received.
1578  */
1579 int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1580 int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
1581                                 loff_t *offs, handle_t *handle);
1582
1583 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
1584                         struct lu_buf *buf, loff_t *pos,
1585                         struct lustre_capa *capa)
1586 {
1587         struct inode *inode = osd_dt_obj(dt)->oo_inode;
1588
1589         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
1590                 RETURN(-EACCES);
1591
1592         return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
1593 }
1594
1595 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
1596                          const struct lu_buf *buf, loff_t *pos,
1597                          struct thandle *handle, struct lustre_capa *capa)
1598 {
1599         struct inode       *inode = osd_dt_obj(dt)->oo_inode;
1600         struct osd_thandle *oh;
1601         ssize_t             result;
1602
1603         LASSERT(handle != NULL);
1604
1605         if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
1606                 RETURN(-EACCES);
1607
1608         oh = container_of(handle, struct osd_thandle, ot_super);
1609         LASSERT(oh->ot_handle->h_transaction != NULL);
1610         result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
1611                                              pos, oh->ot_handle);
1612         if (result == 0)
1613                 result = buf->lb_len;
1614         return result;
1615 }
1616
1617 static struct dt_body_operations osd_body_ops = {
1618         .dbo_read  = osd_read,
1619         .dbo_write = osd_write
1620 };
1621
1622 /*
1623  * Index operations.
1624  */
1625
1626 static int osd_object_is_root(const struct osd_object *obj)
1627 {
1628         return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
1629 }
1630
1631 static int osd_index_probe(const struct lu_env *env, struct osd_object *o,
1632                            const struct dt_index_features *feat)
1633 {
1634         struct iam_descr *descr;
1635
1636         if (osd_object_is_root(o))
1637                 return feat == &dt_directory_features;
1638
1639         LASSERT(o->oo_dir != NULL);
1640
1641         descr = o->oo_dir->od_container.ic_descr;
1642         if (feat == &dt_directory_features)
1643                 return descr == &iam_htree_compat_param ||
1644                         (descr->id_rec_size == sizeof(struct lu_fid_pack) &&
1645                          1 /*
1646                             * XXX check that index looks like directory.
1647                             */
1648                                 );
1649         else
1650                 return
1651                         feat->dif_keysize_min <= descr->id_key_size &&
1652                         descr->id_key_size <= feat->dif_keysize_max &&
1653                         feat->dif_recsize_min <= descr->id_rec_size &&
1654                         descr->id_rec_size <= feat->dif_recsize_max &&
1655                         !(feat->dif_flags & (DT_IND_VARKEY |
1656                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
1657                         ergo(feat->dif_flags & DT_IND_UPDATE,
1658                              1 /* XXX check that object (and file system) is
1659                                 * writable */);
1660 }
1661
1662 static int osd_container_init(const struct lu_env *env,
1663                               struct osd_object *obj,
1664                               struct osd_directory *dir)
1665 {
1666         int result;
1667         struct iam_container *bag;
1668
1669         bag    = &dir->od_container;
1670         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
1671         if (result == 0) {
1672                 result = iam_container_setup(bag);
1673                 if (result == 0)
1674                         obj->oo_dt.do_index_ops = &osd_index_ops;
1675                 else
1676                         iam_container_fini(bag);
1677         }
1678         return result;
1679 }
1680
1681 /*
1682  * Concurrency: no external locking is necessary.
1683  */
1684 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1685                          const struct dt_index_features *feat)
1686 {
1687         int result;
1688         struct osd_object *obj = osd_dt_obj(dt);
1689
1690         LASSERT(osd_invariant(obj));
1691         LASSERT(dt_object_exists(dt));
1692
1693         if (osd_object_is_root(obj)) {
1694                 dt->do_index_ops = &osd_index_compat_ops;
1695                 result = 0;
1696         } else if (!osd_has_index(obj)) {
1697                 struct osd_directory *dir;
1698
1699                 OBD_ALLOC_PTR(dir);
1700                 if (dir != NULL) {
1701                         sema_init(&dir->od_sem, 1);
1702
1703                         spin_lock(&obj->oo_guard);
1704                         if (obj->oo_dir == NULL)
1705                                 obj->oo_dir = dir;
1706                         else
1707                                 /*
1708                                  * Concurrent thread allocated container data.
1709                                  */
1710                                 OBD_FREE_PTR(dir);
1711                         spin_unlock(&obj->oo_guard);
1712                         /*
1713                          * Now, that we have container data, serialize its
1714                          * initialization.
1715                          */
1716                         down(&obj->oo_dir->od_sem);
1717                         /*
1718                          * recheck under lock.
1719                          */
1720                         if (!osd_has_index(obj))
1721                                 result = osd_container_init(env, obj, dir);
1722                         else
1723                                 result = 0;
1724                         up(&obj->oo_dir->od_sem);
1725                 } else
1726                         result = -ENOMEM;
1727         } else
1728                 result = 0;
1729
1730         if (result == 0) {
1731                 if (!osd_index_probe(env, obj, feat))
1732                         result = -ENOTDIR;
1733         }
1734         LASSERT(osd_invariant(obj));
1735
1736         return result;
1737 }
1738
1739 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1740                             const struct dt_key *key, struct thandle *handle,
1741                             struct lustre_capa *capa)
1742 {
1743         struct osd_object     *obj = osd_dt_obj(dt);
1744         struct osd_thandle    *oh;
1745         struct iam_path_descr *ipd;
1746         struct iam_container  *bag = &obj->oo_dir->od_container;
1747         int rc;
1748
1749         ENTRY;
1750
1751         LASSERT(osd_invariant(obj));
1752         LASSERT(dt_object_exists(dt));
1753         LASSERT(bag->ic_object == obj->oo_inode);
1754         LASSERT(handle != NULL);
1755
1756         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
1757                 RETURN(-EACCES);
1758
1759         ipd = osd_ipd_get(env, bag);
1760         if (unlikely(ipd == NULL))
1761                 RETURN(-ENOMEM);
1762
1763         oh = container_of0(handle, struct osd_thandle, ot_super);
1764         LASSERT(oh->ot_handle != NULL);
1765         LASSERT(oh->ot_handle->h_transaction != NULL);
1766
1767         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
1768         osd_ipd_put(env, bag, ipd);
1769         LASSERT(osd_invariant(obj));
1770         RETURN(rc);
1771 }
1772
1773 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1774                             struct dt_rec *rec, const struct dt_key *key,
1775                             struct lustre_capa *capa)
1776 {
1777         struct osd_object     *obj = osd_dt_obj(dt);
1778         struct iam_path_descr *ipd;
1779         struct iam_container  *bag = &obj->oo_dir->od_container;
1780         int rc;
1781
1782         ENTRY;
1783
1784         LASSERT(osd_invariant(obj));
1785         LASSERT(dt_object_exists(dt));
1786         LASSERT(bag->ic_object == obj->oo_inode);
1787
1788         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
1789                 return -EACCES;
1790
1791         ipd = osd_ipd_get(env, bag);
1792         if (unlikely(ipd == NULL))
1793                 RETURN(-ENOMEM);
1794
1795         rc = iam_lookup(bag, (const struct iam_key *)key,
1796                         (struct iam_rec *)rec, ipd);
1797         osd_ipd_put(env, bag, ipd);
1798         LASSERT(osd_invariant(obj));
1799
1800         RETURN(rc);
1801 }
1802
1803 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1804                             const struct dt_rec *rec, const struct dt_key *key,
1805                             struct thandle *th, struct lustre_capa *capa)
1806 {
1807         struct osd_object     *obj = osd_dt_obj(dt);
1808         struct iam_path_descr *ipd;
1809         struct osd_thandle    *oh;
1810         struct iam_container  *bag = &obj->oo_dir->od_container;
1811         int rc;
1812
1813         ENTRY;
1814
1815         LASSERT(osd_invariant(obj));
1816         LASSERT(dt_object_exists(dt));
1817         LASSERT(bag->ic_object == obj->oo_inode);
1818         LASSERT(th != NULL);
1819
1820         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
1821                 return -EACCES;
1822
1823         ipd = osd_ipd_get(env, bag);
1824         if (unlikely(ipd == NULL))
1825                 RETURN(-ENOMEM);
1826
1827         oh = container_of0(th, struct osd_thandle, ot_super);
1828         LASSERT(oh->ot_handle != NULL);
1829         LASSERT(oh->ot_handle->h_transaction != NULL);
1830         rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
1831                         (struct iam_rec *)rec, ipd);
1832         osd_ipd_put(env, bag, ipd);
1833         LASSERT(osd_invariant(obj));
1834         RETURN(rc);
1835 }
1836
1837 /*
1838  * Iterator operations.
1839  */
1840 struct osd_it {
1841         struct osd_object     *oi_obj;
1842         struct iam_path_descr *oi_ipd;
1843         struct iam_iterator    oi_it;
1844 };
1845
1846 static struct dt_it *osd_it_init(const struct lu_env *env,
1847                                  struct dt_object *dt, int writable,
1848                                  struct lustre_capa *capa)
1849 {
1850         struct osd_it         *it;
1851         struct osd_object     *obj = osd_dt_obj(dt);
1852         struct lu_object      *lo  = &dt->do_lu;
1853         struct iam_path_descr *ipd;
1854         struct iam_container  *bag = &obj->oo_dir->od_container;
1855         __u32                  flags;
1856
1857         LASSERT(lu_object_exists(lo));
1858
1859         if (osd_object_auth(env, dt, capa, writable ? CAPA_OPC_BODY_WRITE :
1860                             CAPA_OPC_BODY_READ))
1861                 return ERR_PTR(-EACCES);
1862
1863         flags = writable ? IAM_IT_MOVE|IAM_IT_WRITE : IAM_IT_MOVE;
1864         OBD_ALLOC_PTR(it);
1865         if (it != NULL) {
1866                 /*
1867                  * XXX: as ipd is allocated within osd_thread_info, assignment
1868                  * below implies that iterator usage is confined within single
1869                  * environment.
1870                  */
1871                 ipd = osd_ipd_get(env, bag);
1872                 if (likely(ipd != NULL)) {
1873                         it->oi_obj = obj;
1874                         it->oi_ipd = ipd;
1875                         lu_object_get(lo);
1876                         iam_it_init(&it->oi_it, bag, flags, ipd);
1877                         return (struct dt_it *)it;
1878                 } else
1879                         OBD_FREE_PTR(it);
1880         }
1881         return ERR_PTR(-ENOMEM);
1882 }
1883
1884 static void osd_it_fini(const struct lu_env *env, struct dt_it *di)
1885 {
1886         struct osd_it     *it = (struct osd_it *)di;
1887         struct osd_object *obj = it->oi_obj;
1888
1889         iam_it_fini(&it->oi_it);
1890         osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
1891         lu_object_put(env, &obj->oo_dt.do_lu);
1892         OBD_FREE_PTR(it);
1893 }
1894
1895 static int osd_it_get(const struct lu_env *env,
1896                       struct dt_it *di, const struct dt_key *key)
1897 {
1898         struct osd_it *it = (struct osd_it *)di;
1899
1900         return iam_it_get(&it->oi_it, (const struct iam_key *)key);
1901 }
1902
1903 static void osd_it_put(const struct lu_env *env, struct dt_it *di)
1904 {
1905         struct osd_it *it = (struct osd_it *)di;
1906
1907         iam_it_put(&it->oi_it);
1908 }
1909
1910 static int osd_it_next(const struct lu_env *env, struct dt_it *di)
1911 {
1912         struct osd_it *it = (struct osd_it *)di;
1913
1914         return iam_it_next(&it->oi_it);
1915 }
1916
1917 static int osd_it_del(const struct lu_env *env, struct dt_it *di,
1918                       struct thandle *th)
1919 {
1920         struct osd_it      *it = (struct osd_it *)di;
1921         struct osd_thandle *oh;
1922
1923         LASSERT(th != NULL);
1924
1925         oh = container_of0(th, struct osd_thandle, ot_super);
1926         LASSERT(oh->ot_handle != NULL);
1927         LASSERT(oh->ot_handle->h_transaction != NULL);
1928
1929         return iam_it_rec_delete(oh->ot_handle, &it->oi_it);
1930 }
1931
1932 static struct dt_key *osd_it_key(const struct lu_env *env,
1933                                  const struct dt_it *di)
1934 {
1935         struct osd_it *it = (struct osd_it *)di;
1936
1937         return (struct dt_key *)iam_it_key_get(&it->oi_it);
1938 }
1939
1940 static int osd_it_key_size(const struct lu_env *env, const struct dt_it *di)
1941 {
1942         struct osd_it *it = (struct osd_it *)di;
1943
1944         return iam_it_key_size(&it->oi_it);
1945 }
1946
1947 static struct dt_rec *osd_it_rec(const struct lu_env *env,
1948                                  const struct dt_it *di)
1949 {
1950         struct osd_it *it = (struct osd_it *)di;
1951
1952         return (struct dt_rec *)iam_it_rec_get(&it->oi_it);
1953 }
1954
1955 static __u32 osd_it_store(const struct lu_env *env, const struct dt_it *di)
1956 {
1957         struct osd_it *it = (struct osd_it *)di;
1958
1959         return iam_it_store(&it->oi_it);
1960 }
1961
1962 static int osd_it_load(const struct lu_env *env,
1963                        const struct dt_it *di, __u32 hash)
1964 {
1965         struct osd_it *it = (struct osd_it *)di;
1966
1967         return iam_it_load(&it->oi_it, hash);
1968 }
1969
1970 static struct dt_index_operations osd_index_ops = {
1971         .dio_lookup = osd_index_lookup,
1972         .dio_insert = osd_index_insert,
1973         .dio_delete = osd_index_delete,
1974         .dio_it     = {
1975                 .init     = osd_it_init,
1976                 .fini     = osd_it_fini,
1977                 .get      = osd_it_get,
1978                 .put      = osd_it_put,
1979                 .del      = osd_it_del,
1980                 .next     = osd_it_next,
1981                 .key      = osd_it_key,
1982                 .key_size = osd_it_key_size,
1983                 .rec      = osd_it_rec,
1984                 .store    = osd_it_store,
1985                 .load     = osd_it_load
1986         }
1987 };
1988
1989 static int osd_index_compat_delete(const struct lu_env *env,
1990                                    struct dt_object *dt,
1991                                    const struct dt_key *key,
1992                                    struct thandle *handle,
1993                                    struct lustre_capa *capa)
1994 {
1995         struct osd_object *obj = osd_dt_obj(dt);
1996
1997         LASSERT(handle != NULL);
1998         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
1999         ENTRY;
2000
2001 #if 0
2002         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2003                 RETURN(-EACCES);
2004 #endif
2005
2006         RETURN(-EOPNOTSUPP);
2007 }
2008
2009 /*
2010  * Compatibility index operations.
2011  */
2012
2013
2014 static void osd_build_pack(const struct lu_env *env, struct osd_device *osd,
2015                            struct dentry *dentry, struct lu_fid_pack *pack)
2016 {
2017         struct inode  *inode = dentry->d_inode;
2018         struct lu_fid *fid   = &osd_oti_get(env)->oti_fid;
2019
2020         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2021         fid_cpu_to_be(fid, fid);
2022         pack->fp_len = sizeof *fid + 1;
2023         memcpy(pack->fp_area, fid, sizeof *fid);
2024 }
2025
2026 static int osd_index_compat_lookup(const struct lu_env *env,
2027                                    struct dt_object *dt,
2028                                    struct dt_rec *rec, const struct dt_key *key,
2029                                    struct lustre_capa *capa)
2030 {
2031         struct osd_object *obj = osd_dt_obj(dt);
2032
2033         struct osd_device      *osd  = osd_obj2dev(obj);
2034         struct osd_thread_info *info = osd_oti_get(env);
2035         struct inode           *dir;
2036
2037         int result;
2038
2039         /*
2040          * XXX temporary solution.
2041          */
2042         struct dentry *dentry;
2043         struct dentry *parent;
2044
2045         LASSERT(osd_invariant(obj));
2046         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2047         LASSERT(osd_has_index(obj));
2048
2049         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2050                 return -EACCES;
2051
2052         info->oti_str.name = (const char *)key;
2053         info->oti_str.len  = strlen((const char *)key);
2054
2055         dir = obj->oo_inode;
2056         LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2057
2058         parent = d_alloc_root(dir);
2059         if (parent == NULL)
2060                 return -ENOMEM;
2061         igrab(dir);
2062         dentry = d_alloc(parent, &info->oti_str);
2063         if (dentry != NULL) {
2064                 struct dentry *d;
2065
2066                 /*
2067                  * XXX passing NULL for nameidata should work for
2068                  * ext3/ldiskfs.
2069                  */
2070                 d = dir->i_op->lookup(dir, dentry, NULL);
2071                 if (d == NULL) {
2072                         /*
2073                          * normal case, result is in @dentry.
2074                          */
2075                         if (dentry->d_inode != NULL) {
2076                                 osd_build_pack(env, osd, dentry,
2077                                                (struct lu_fid_pack *)rec);
2078                                 result = 0;
2079                         } else
2080                                 result = -ENOENT;
2081                  } else {
2082                         /* What? Disconnected alias? Ppheeeww... */
2083                         CERROR("Aliasing where not expected\n");
2084                         result = -EIO;
2085                         dput(d);
2086                 }
2087                 dput(dentry);
2088         } else
2089                 result = -ENOMEM;
2090         dput(parent);
2091         LASSERT(osd_invariant(obj));
2092         return result;
2093 }
2094
2095 static int osd_add_rec(struct osd_thread_info *info, struct osd_device *dev,
2096                        struct inode *dir, struct inode *inode, const char *name)
2097 {
2098         struct dentry *old;
2099         struct dentry *new;
2100         struct dentry *parent;
2101
2102         int result;
2103
2104         info->oti_str.name = name;
2105         info->oti_str.len  = strlen(name);
2106
2107         LASSERT(atomic_read(&dir->i_count) > 0);
2108         result = -ENOMEM;
2109         old = d_alloc(dev->od_obj_area, &info->oti_str);
2110         if (old != NULL) {
2111                 d_instantiate(old, inode);
2112                 igrab(inode);
2113                 LASSERT(atomic_read(&dir->i_count) > 0);
2114                 parent = d_alloc_root(dir);
2115                 if (parent != NULL) {
2116                         igrab(dir);
2117                         LASSERT(atomic_read(&dir->i_count) > 1);
2118                         new = d_alloc(parent, &info->oti_str);
2119                         LASSERT(atomic_read(&dir->i_count) > 1);
2120                         if (new != NULL) {
2121                                 LASSERT(atomic_read(&dir->i_count) > 1);
2122                                 result = dir->i_op->link(old, dir, new);
2123                                 LASSERT(atomic_read(&dir->i_count) > 1);
2124                                 dput(new);
2125                                 LASSERT(atomic_read(&dir->i_count) > 1);
2126                         }
2127                         LASSERT(atomic_read(&dir->i_count) > 1);
2128                         dput(parent);
2129                         LASSERT(atomic_read(&dir->i_count) > 0);
2130                 }
2131                 dput(old);
2132         }
2133         LASSERT(atomic_read(&dir->i_count) > 0);
2134         return result;
2135 }
2136
2137
2138 /*
2139  * XXX Temporary stuff.
2140  */
2141 static int osd_index_compat_insert(const struct lu_env *env,
2142                                    struct dt_object *dt,
2143                                    const struct dt_rec *rec,
2144                                    const struct dt_key *key, struct thandle *th,
2145                                    struct lustre_capa *capa)
2146 {
2147         struct osd_object     *obj = osd_dt_obj(dt);
2148
2149         const char          *name = (const char *)key;
2150
2151         struct lu_device    *ludev = dt->do_lu.lo_dev;
2152         struct lu_object    *luch;
2153
2154         struct osd_thread_info   *info = osd_oti_get(env);
2155         const struct lu_fid_pack *pack  = (const struct lu_fid_pack *)rec;
2156         struct lu_fid            *fid   = &osd_oti_get(env)->oti_fid;
2157
2158         int result;
2159
2160         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2161         LASSERT(osd_invariant(obj));
2162         LASSERT(th != NULL);
2163
2164         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2165                 return -EACCES;
2166
2167         fid_unpack(pack, fid);
2168         luch = lu_object_find(env, ludev->ld_site, fid);
2169         if (!IS_ERR(luch)) {
2170                 if (lu_object_exists(luch)) {
2171                         struct osd_object *child;
2172
2173                         child = osd_obj(lu_object_locate(luch->lo_header,
2174                                                          ludev->ld_type));
2175                         if (child != NULL)
2176                                 result = osd_add_rec(info, osd_obj2dev(obj),
2177                                                      obj->oo_inode,
2178                                                      child->oo_inode, name);
2179                         else {
2180                                 CERROR("No osd slice.\n");
2181                                 result = -ENOENT;
2182                         }
2183                         LASSERT(osd_invariant(obj));
2184                         LASSERT(osd_invariant(child));
2185                 } else {
2186                         CERROR("Sorry.\n");
2187                         result = -ENOENT;
2188                 }
2189                 lu_object_put(env, luch);
2190         } else
2191                 result = PTR_ERR(luch);
2192         LASSERT(osd_invariant(obj));
2193         return result;
2194 }
2195
2196 static struct dt_index_operations osd_index_compat_ops = {
2197         .dio_lookup = osd_index_compat_lookup,
2198         .dio_insert = osd_index_compat_insert,
2199         .dio_delete = osd_index_compat_delete
2200 };
2201
2202 /*
2203  * OSD device type methods
2204  */
2205 static int osd_type_init(struct lu_device_type *t)
2206 {
2207         LU_CONTEXT_KEY_INIT(&osd_key);
2208         return lu_context_key_register(&osd_key);
2209 }
2210
2211 static void osd_type_fini(struct lu_device_type *t)
2212 {
2213         lu_context_key_degister(&osd_key);
2214 }
2215
2216 static struct lu_context_key osd_key = {
2217         .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
2218         .lct_init = osd_key_init,
2219         .lct_fini = osd_key_fini,
2220         .lct_exit = osd_key_exit
2221 };
2222
2223 static void *osd_key_init(const struct lu_context *ctx,
2224                           struct lu_context_key *key)
2225 {
2226         struct osd_thread_info *info;
2227
2228         OBD_ALLOC_PTR(info);
2229         if (info != NULL)
2230                 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
2231         else
2232                 info = ERR_PTR(-ENOMEM);
2233         return info;
2234 }
2235
2236 static void osd_key_fini(const struct lu_context *ctx,
2237                          struct lu_context_key *key, void *data)
2238 {
2239         struct osd_thread_info *info = data;
2240         OBD_FREE_PTR(info);
2241 }
2242
2243 static void osd_key_exit(const struct lu_context *ctx,
2244                          struct lu_context_key *key, void *data)
2245 {
2246 #if OSD_COUNTERS
2247         struct osd_thread_info *info = data;
2248
2249         LASSERT(info->oti_r_locks == 0);
2250         LASSERT(info->oti_w_locks == 0);
2251         LASSERT(info->oti_txns    == 0);
2252 #endif
2253 }
2254
2255 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
2256                            const char *name, struct lu_device *next)
2257 {
2258         return lu_env_init(&osd_dev(d)->od_env_for_commit, NULL, LCT_MD_THREAD);
2259 }
2260
2261 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
2262 {
2263         struct osd_thread_info *info = osd_oti_get(env);
2264         ENTRY;
2265         if (o->od_obj_area != NULL) {
2266                 dput(o->od_obj_area);
2267                 o->od_obj_area = NULL;
2268         }
2269         osd_oi_fini(info, &o->od_oi);
2270
2271         RETURN(0);
2272 }
2273
2274 static int osd_mount(const struct lu_env *env,
2275                      struct osd_device *o, struct lustre_cfg *cfg)
2276 {
2277         struct lustre_mount_info *lmi;
2278         const char               *dev  = lustre_cfg_string(cfg, 0);
2279         struct osd_thread_info   *info = osd_oti_get(env);
2280         int result;
2281
2282         ENTRY;
2283
2284         if (o->od_mount != NULL) {
2285                 CERROR("Already mounted (%s)\n", dev);
2286                 RETURN(-EEXIST);
2287         }
2288
2289         /* get mount */
2290         lmi = server_get_mount(dev);
2291         if (lmi == NULL) {
2292                 CERROR("Cannot get mount info for %s!\n", dev);
2293                 RETURN(-EFAULT);
2294         }
2295
2296         LASSERT(lmi != NULL);
2297         /* save lustre_mount_info in dt_device */
2298         o->od_mount = lmi;
2299
2300         result = osd_oi_init(info, &o->od_oi, &o->od_dt_dev);
2301         if (result == 0) {
2302                 struct dentry *d;
2303
2304                 d = simple_mkdir(osd_sb(o)->s_root, "*OBJ-TEMP*", 0777, 1);
2305                 if (!IS_ERR(d)) {
2306                         o->od_obj_area = d;
2307                 } else
2308                         result = PTR_ERR(d);
2309         }
2310         if (result != 0)
2311                 osd_shutdown(env, o);
2312         RETURN(result);
2313 }
2314
2315 static struct lu_device *osd_device_fini(const struct lu_env *env,
2316                                          struct lu_device *d)
2317 {
2318         ENTRY;
2319
2320         shrink_dcache_sb(osd_sb(osd_dev(d)));
2321         osd_sync(env, lu2dt_dev(d));
2322
2323         if (osd_dev(d)->od_mount)
2324                 server_put_mount(osd_dev(d)->od_mount->lmi_name,
2325                                  osd_dev(d)->od_mount->lmi_mnt);
2326         osd_dev(d)->od_mount = NULL;
2327
2328         lu_env_fini(&osd_dev(d)->od_env_for_commit);
2329         RETURN(NULL);
2330 }
2331
2332 static struct lu_device *osd_device_alloc(const struct lu_env *env,
2333                                           struct lu_device_type *t,
2334                                           struct lustre_cfg *cfg)
2335 {
2336         struct lu_device  *l;
2337         struct osd_device *o;
2338
2339         OBD_ALLOC_PTR(o);
2340         if (o != NULL) {
2341                 int result;
2342
2343                 result = dt_device_init(&o->od_dt_dev, t);
2344                 if (result == 0) {
2345                         l = osd2lu_dev(o);
2346                         l->ld_ops = &osd_lu_ops;
2347                         o->od_dt_dev.dd_ops = &osd_dt_ops;
2348                         spin_lock_init(&o->od_osfs_lock);
2349                         o->od_osfs_age = cfs_time_shift_64(-1000);
2350                         o->od_capa_hash = init_capa_hash();
2351                         if (o->od_capa_hash == NULL)
2352                                 l = ERR_PTR(-ENOMEM);
2353                 } else
2354                         l = ERR_PTR(result);
2355         } else
2356                 l = ERR_PTR(-ENOMEM);
2357         return l;
2358 }
2359
2360 static void osd_device_free(const struct lu_env *env, struct lu_device *d)
2361 {
2362         struct osd_device *o = osd_dev(d);
2363
2364         cleanup_capa_hash(o->od_capa_hash);
2365         dt_device_fini(&o->od_dt_dev);
2366         OBD_FREE_PTR(o);
2367 }
2368
2369 static int osd_process_config(const struct lu_env *env,
2370                               struct lu_device *d, struct lustre_cfg *cfg)
2371 {
2372         struct osd_device *o = osd_dev(d);
2373         int err;
2374         ENTRY;
2375
2376         switch(cfg->lcfg_command) {
2377         case LCFG_SETUP:
2378                 err = osd_mount(env, o, cfg);
2379                 break;
2380         case LCFG_CLEANUP:
2381                 err = osd_shutdown(env, o);
2382                 break;
2383         default:
2384                 err = -ENOTTY;
2385         }
2386
2387         RETURN(err);
2388 }
2389 extern void ldiskfs_orphan_cleanup (struct super_block * sb,
2390                                     struct ldiskfs_super_block * es);
2391
2392 static int osd_recovery_complete(const struct lu_env *env,
2393                                  struct lu_device *d)
2394 {
2395         struct osd_device *o = osd_dev(d);
2396         ENTRY;
2397         /* TODO: orphans handling */
2398         ldiskfs_orphan_cleanup(osd_sb(o), LDISKFS_SB(osd_sb(o))->s_es);
2399         RETURN(0);
2400 }
2401
2402 static struct inode *osd_iget(struct osd_thread_info *info,
2403                               struct osd_device *dev,
2404                               const struct osd_inode_id *id)
2405 {
2406         struct inode *inode;
2407
2408         inode = iget(osd_sb(dev), id->oii_ino);
2409         if (inode == NULL) {
2410                 CERROR("no inode\n");
2411                 inode = ERR_PTR(-EACCES);
2412         } else if (is_bad_inode(inode)) {
2413                 CERROR("bad inode\n");
2414                 iput(inode);
2415                 inode = ERR_PTR(-ENOENT);
2416         } else if (inode->i_generation != id->oii_gen) {
2417                 CERROR("stale inode\n");
2418                 iput(inode);
2419                 inode = ERR_PTR(-ESTALE);
2420         }
2421
2422         return inode;
2423
2424 }
2425
2426 static int osd_fid_lookup(const struct lu_env *env,
2427                           struct osd_object *obj, const struct lu_fid *fid)
2428 {
2429         struct osd_thread_info *info;
2430         struct lu_device       *ldev = obj->oo_dt.do_lu.lo_dev;
2431         struct osd_device      *dev;
2432         struct osd_inode_id    *id;
2433         struct osd_oi          *oi;
2434         struct inode           *inode;
2435         int                     result;
2436
2437         LASSERT(osd_invariant(obj));
2438         LASSERT(obj->oo_inode == NULL);
2439         LASSERT(fid_is_sane(fid));
2440         /*
2441          * This assertion checks that osd layer sees only local
2442          * fids. Unfortunately it is somewhat expensive (does a
2443          * cache-lookup). Disabling it for production/acceptance-testing.
2444          */
2445         LASSERT(1 || fid_is_local(ldev->ld_site, fid));
2446
2447         ENTRY;
2448
2449         info = osd_oti_get(env);
2450         dev  = osd_dev(ldev);
2451         id   = &info->oti_id;
2452         oi   = &dev->od_oi;
2453
2454         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
2455                 RETURN(-ENOENT);
2456
2457         result = osd_oi_lookup(info, oi, fid, id);
2458         if (result == 0) {
2459                 inode = osd_iget(info, dev, id);
2460                 if (!IS_ERR(inode)) {
2461                         obj->oo_inode = inode;
2462                         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
2463                         result = 0;
2464                 } else
2465                         /*
2466                          * If fid wasn't found in oi, inode-less object is
2467                          * created, for which lu_object_exists() returns
2468                          * false. This is used in a (frequent) case when
2469                          * objects are created as locking anchors or
2470                          * place holders for objects yet to be created.
2471                          */
2472                         result = PTR_ERR(inode);
2473         } else if (result == -ENOENT)
2474                 result = 0;
2475         LASSERT(osd_invariant(obj));
2476         RETURN(result);
2477 }
2478
2479 static void osd_inode_getattr(const struct lu_env *env,
2480                               struct inode *inode, struct lu_attr *attr)
2481 {
2482         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2483                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2484                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
2485
2486         attr->la_atime      = LTIME_S(inode->i_atime);
2487         attr->la_mtime      = LTIME_S(inode->i_mtime);
2488         attr->la_ctime      = LTIME_S(inode->i_ctime);
2489         attr->la_mode       = inode->i_mode;
2490         attr->la_size       = inode->i_size;
2491         attr->la_blocks     = inode->i_blocks;
2492         attr->la_uid        = inode->i_uid;
2493         attr->la_gid        = inode->i_gid;
2494         attr->la_flags      = LDISKFS_I(inode)->i_flags;
2495         attr->la_nlink      = inode->i_nlink;
2496         attr->la_rdev       = inode->i_rdev;
2497         attr->la_blksize    = inode->i_blksize;
2498 }
2499
2500 /*
2501  * Helpers.
2502  */
2503
2504 static int lu_device_is_osd(const struct lu_device *d)
2505 {
2506         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
2507 }
2508
2509 static struct osd_object *osd_obj(const struct lu_object *o)
2510 {
2511         LASSERT(lu_device_is_osd(o->lo_dev));
2512         return container_of0(o, struct osd_object, oo_dt.do_lu);
2513 }
2514
2515 static struct osd_device *osd_dt_dev(const struct dt_device *d)
2516 {
2517         LASSERT(lu_device_is_osd(&d->dd_lu_dev));
2518         return container_of0(d, struct osd_device, od_dt_dev);
2519 }
2520
2521 static struct osd_device *osd_dev(const struct lu_device *d)
2522 {
2523         LASSERT(lu_device_is_osd(d));
2524         return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
2525 }
2526
2527 static struct osd_object *osd_dt_obj(const struct dt_object *d)
2528 {
2529         return osd_obj(&d->do_lu);
2530 }
2531
2532 static struct osd_device *osd_obj2dev(const struct osd_object *o)
2533 {
2534         return osd_dev(o->oo_dt.do_lu.lo_dev);
2535 }
2536
2537 static struct lu_device *osd2lu_dev(struct osd_device *osd)
2538 {
2539         return &osd->od_dt_dev.dd_lu_dev;
2540 }
2541
2542 static struct super_block *osd_sb(const struct osd_device *dev)
2543 {
2544         return dev->od_mount->lmi_mnt->mnt_sb;
2545 }
2546
2547 static journal_t *osd_journal(const struct osd_device *dev)
2548 {
2549         return LDISKFS_SB(osd_sb(dev))->s_journal;
2550 }
2551
2552 static int osd_has_index(const struct osd_object *obj)
2553 {
2554         return obj->oo_dt.do_index_ops != NULL;
2555 }
2556
2557 static int osd_object_invariant(const struct lu_object *l)
2558 {
2559         return osd_invariant(osd_obj(l));
2560 }
2561
2562 static struct lu_object_operations osd_lu_obj_ops = {
2563         .loo_object_init      = osd_object_init,
2564         .loo_object_delete    = osd_object_delete,
2565         .loo_object_release   = osd_object_release,
2566         .loo_object_free      = osd_object_free,
2567         .loo_object_print     = osd_object_print,
2568         .loo_object_invariant = osd_object_invariant
2569 };
2570
2571 static struct lu_device_operations osd_lu_ops = {
2572         .ldo_object_alloc      = osd_object_alloc,
2573         .ldo_process_config    = osd_process_config,
2574         .ldo_recovery_complete = osd_recovery_complete
2575 };
2576
2577 static struct lu_device_type_operations osd_device_type_ops = {
2578         .ldto_init = osd_type_init,
2579         .ldto_fini = osd_type_fini,
2580
2581         .ldto_device_alloc = osd_device_alloc,
2582         .ldto_device_free  = osd_device_free,
2583
2584         .ldto_device_init    = osd_device_init,
2585         .ldto_device_fini    = osd_device_fini
2586 };
2587
2588 static struct lu_device_type osd_device_type = {
2589         .ldt_tags     = LU_DEVICE_DT,
2590         .ldt_name     = LUSTRE_OSD_NAME,
2591         .ldt_ops      = &osd_device_type_ops,
2592         .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
2593 };
2594
2595 /*
2596  * lprocfs legacy support.
2597  */
2598 static struct lprocfs_vars lprocfs_osd_obd_vars[] = {
2599         { 0 }
2600 };
2601
2602 static struct lprocfs_vars lprocfs_osd_module_vars[] = {
2603         { 0 }
2604 };
2605
2606 static struct obd_ops osd_obd_device_ops = {
2607         .o_owner = THIS_MODULE
2608 };
2609
2610 LPROCFS_INIT_VARS(osd, lprocfs_osd_module_vars, lprocfs_osd_obd_vars);
2611
2612 static int __init osd_mod_init(void)
2613 {
2614         struct lprocfs_static_vars lvars;
2615
2616         lprocfs_init_vars(osd, &lvars);
2617         return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
2618                                    LUSTRE_OSD_NAME, &osd_device_type);
2619 }
2620
2621 static void __exit osd_mod_exit(void)
2622 {
2623         class_unregister_type(LUSTRE_OSD_NAME);
2624 }
2625
2626 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2627 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
2628 MODULE_LICENSE("GPL");
2629
2630 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);