1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see [sun.com URL with a
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_handler.c
38 * Top-level entry points into osd module
40 * Author: Nikita Danilov <nikita@clusterfs.com>
44 # define EXPORT_SYMTAB
46 #define DEBUG_SUBSYSTEM S_MDS
48 #include <linux/module.h>
50 /* LUSTRE_VERSION_CODE */
51 #include <lustre_ver.h>
52 /* prerequisite for linux/xattr.h */
53 #include <linux/types.h>
54 /* prerequisite for linux/xattr.h */
56 /* XATTR_{REPLACE,CREATE} */
57 #include <linux/xattr.h>
59 * XXX temporary stuff: direct access to ldiskfs/jdb. Interface between osd
60 * and file system is not yet specified.
62 /* handle_t, journal_start(), journal_stop() */
63 #include <linux/jbd.h>
65 #include <linux/ldiskfs_fs.h>
66 #include <linux/ldiskfs_jbd.h>
71 * struct OBD_{ALLOC,FREE}*()
74 #include <obd_support.h>
75 /* struct ptlrpc_thread */
76 #include <lustre_net.h>
79 #include <lustre_fid.h>
80 #include <linux/lustre_iam.h>
82 #include "osd_internal.h"
85 struct osd_directory {
86 struct iam_container od_container;
87 struct iam_descr od_descr;
88 struct semaphore od_sem;
92 struct dt_object oo_dt;
94 * Inode for file system object represented by this osd_object. This
95 * inode is pinned for the whole duration of lu_object life.
97 * Not modified concurrently (either setup early during object
98 * creation, or assigned by osd_object_create() under write lock).
100 struct inode *oo_inode;
101 struct rw_semaphore oo_sem;
102 struct osd_directory *oo_dir;
103 /* protects inode attributes. */
106 const struct lu_env *oo_owner;
110 static int osd_root_get (const struct lu_env *env,
111 struct dt_device *dev, struct lu_fid *f);
113 static int lu_device_is_osd (const struct lu_device *d);
114 static void osd_mod_exit (void) __exit;
115 static int osd_mod_init (void) __init;
116 static int osd_type_init (struct lu_device_type *t);
117 static void osd_type_fini (struct lu_device_type *t);
118 static int osd_object_init (const struct lu_env *env,
119 struct lu_object *l);
120 static void osd_object_release(const struct lu_env *env,
121 struct lu_object *l);
122 static int osd_object_print (const struct lu_env *env, void *cookie,
123 lu_printer_t p, const struct lu_object *o);
124 static struct lu_device *osd_device_free (const struct lu_env *env,
125 struct lu_device *m);
126 static void *osd_key_init (const struct lu_context *ctx,
127 struct lu_context_key *key);
128 static void osd_key_fini (const struct lu_context *ctx,
129 struct lu_context_key *key, void *data);
130 static void osd_key_exit (const struct lu_context *ctx,
131 struct lu_context_key *key, void *data);
132 static int osd_has_index (const struct osd_object *obj);
133 static void osd_object_init0 (struct osd_object *obj);
134 static int osd_device_init (const struct lu_env *env,
135 struct lu_device *d, const char *,
137 static int osd_fid_lookup (const struct lu_env *env,
138 struct osd_object *obj,
139 const struct lu_fid *fid);
140 static void osd_inode_getattr (const struct lu_env *env,
141 struct inode *inode, struct lu_attr *attr);
142 static void osd_inode_setattr (const struct lu_env *env,
143 struct inode *inode, const struct lu_attr *attr);
144 static int osd_param_is_sane (const struct osd_device *dev,
145 const struct txn_param *param);
146 static int osd_index_lookup (const struct lu_env *env,
147 struct dt_object *dt,
148 struct dt_rec *rec, const struct dt_key *key,
149 struct lustre_capa *capa);
150 static int osd_index_insert (const struct lu_env *env,
151 struct dt_object *dt,
152 const struct dt_rec *rec,
153 const struct dt_key *key,
154 struct thandle *handle,
155 struct lustre_capa *capa);
156 static int osd_index_delete (const struct lu_env *env,
157 struct dt_object *dt, const struct dt_key *key,
158 struct thandle *handle,
159 struct lustre_capa *capa);
160 static int osd_index_probe (const struct lu_env *env,
161 struct osd_object *o,
162 const struct dt_index_features *feat);
163 static int osd_index_try (const struct lu_env *env,
164 struct dt_object *dt,
165 const struct dt_index_features *feat);
166 static void osd_index_fini (struct osd_object *o);
168 static void osd_it_fini (const struct lu_env *env, struct dt_it *di);
169 static int osd_it_get (const struct lu_env *env,
170 struct dt_it *di, const struct dt_key *key);
171 static void osd_it_put (const struct lu_env *env, struct dt_it *di);
172 static int osd_it_next (const struct lu_env *env, struct dt_it *di);
173 static int osd_it_del (const struct lu_env *env, struct dt_it *di,
175 static int osd_it_key_size (const struct lu_env *env,
176 const struct dt_it *di);
177 static void osd_conf_get (const struct lu_env *env,
178 const struct dt_device *dev,
179 struct dt_device_param *param);
180 static void osd_trans_stop (const struct lu_env *env,
182 static int osd_object_is_root(const struct osd_object *obj);
184 static struct osd_object *osd_obj (const struct lu_object *o);
185 static struct osd_device *osd_dev (const struct lu_device *d);
186 static struct osd_device *osd_dt_dev (const struct dt_device *d);
187 static struct osd_object *osd_dt_obj (const struct dt_object *d);
188 static struct osd_device *osd_obj2dev (const struct osd_object *o);
189 static struct lu_device *osd2lu_dev (struct osd_device *osd);
190 static struct lu_device *osd_device_fini (const struct lu_env *env,
191 struct lu_device *d);
192 static struct lu_device *osd_device_alloc (const struct lu_env *env,
193 struct lu_device_type *t,
194 struct lustre_cfg *cfg);
195 static struct lu_object *osd_object_alloc (const struct lu_env *env,
196 const struct lu_object_header *hdr,
197 struct lu_device *d);
198 static struct inode *osd_iget (struct osd_thread_info *info,
199 struct osd_device *dev,
200 const struct osd_inode_id *id);
201 static struct super_block *osd_sb (const struct osd_device *dev);
202 static struct dt_it *osd_it_init (const struct lu_env *env,
203 struct dt_object *dt, int wable,
204 struct lustre_capa *capa);
205 static struct dt_key *osd_it_key (const struct lu_env *env,
206 const struct dt_it *di);
207 static struct dt_rec *osd_it_rec (const struct lu_env *env,
208 const struct dt_it *di);
209 static struct timespec *osd_inode_time (const struct lu_env *env,
212 static struct thandle *osd_trans_start (const struct lu_env *env,
214 struct txn_param *p);
215 static journal_t *osd_journal (const struct osd_device *dev);
217 static struct lu_device_type_operations osd_device_type_ops;
218 static struct lu_device_type osd_device_type;
219 static struct lu_object_operations osd_lu_obj_ops;
220 static struct obd_ops osd_obd_device_ops;
221 static struct lu_device_operations osd_lu_ops;
222 static struct lu_context_key osd_key;
223 static struct dt_object_operations osd_obj_ops;
224 static struct dt_body_operations osd_body_ops;
225 static struct dt_index_operations osd_index_ops;
226 static struct dt_index_operations osd_index_compat_ops;
229 struct thandle ot_super;
231 struct journal_callback ot_jcb;
235 * Invariants, assertions.
239 * XXX: do not enable this, until invariant checking code is made thread safe
240 * in the face of pdirops locking.
242 #define OSD_INVARIANT_CHECKS (0)
244 #if OSD_INVARIANT_CHECKS
245 static int osd_invariant(const struct osd_object *obj)
249 ergo(obj->oo_inode != NULL,
250 obj->oo_inode->i_sb == osd_sb(osd_obj2dev(obj)) &&
251 atomic_read(&obj->oo_inode->i_count) > 0) &&
252 ergo(obj->oo_dir != NULL &&
253 obj->oo_dir->od_conationer.ic_object != NULL,
254 obj->oo_dir->od_conationer.ic_object == obj->oo_inode);
257 #define osd_invariant(obj) (1)
260 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
262 return lu_context_key_get(&env->le_ctx, &osd_key);
267 * Concurrency: doesn't matter
269 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
271 return osd_oti_get(env)->oti_r_locks > 0;
275 * Concurrency: doesn't matter
277 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
279 struct osd_thread_info *oti = osd_oti_get(env);
280 return oti->oti_w_locks > 0 && o->oo_owner == env;
283 #define OSD_COUNTERS_DO(exp) exp
287 #define osd_read_locked(env, o) (1)
288 #define osd_write_locked(env, o) (1)
289 #define OSD_COUNTERS_DO(exp) ((void)0)
293 * Concurrency: doesn't access mutable data
295 static int osd_root_get(const struct lu_env *env,
296 struct dt_device *dev, struct lu_fid *f)
300 inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
301 lu_igif_build(f, inode->i_ino, inode->i_generation);
306 * OSD object methods.
310 * Concurrency: no concurrent access is possible that early in object
313 static struct lu_object *osd_object_alloc(const struct lu_env *env,
314 const struct lu_object_header *hdr,
317 struct osd_object *mo;
323 l = &mo->oo_dt.do_lu;
324 dt_object_init(&mo->oo_dt, NULL, d);
325 mo->oo_dt.do_ops = &osd_obj_ops;
326 l->lo_ops = &osd_lu_obj_ops;
327 init_rwsem(&mo->oo_sem);
328 spin_lock_init(&mo->oo_guard);
335 * Concurrency: shouldn't matter.
337 static void osd_object_init0(struct osd_object *obj)
339 LASSERT(obj->oo_inode != NULL);
340 obj->oo_dt.do_body_ops = &osd_body_ops;
341 obj->oo_dt.do_lu.lo_header->loh_attr |=
342 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
346 * Concurrency: no concurrent access is possible that early in object
349 static int osd_object_init(const struct lu_env *env, struct lu_object *l)
351 struct osd_object *obj = osd_obj(l);
354 LASSERT(osd_invariant(obj));
356 result = osd_fid_lookup(env, obj, lu_object_fid(l));
358 if (obj->oo_inode != NULL)
359 osd_object_init0(obj);
361 LASSERT(osd_invariant(obj));
366 * Concurrency: no concurrent access is possible that late in object
369 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
371 struct osd_object *obj = osd_obj(l);
373 LASSERT(osd_invariant(obj));
375 dt_object_fini(&obj->oo_dt);
379 static struct iam_path_descr *osd_ipd_get(const struct lu_env *env,
380 const struct iam_container *bag)
382 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
383 osd_oti_get(env)->oti_ipd);
386 static void osd_ipd_put(const struct lu_env *env,
387 const struct iam_container *bag,
388 struct iam_path_descr *ipd)
390 bag->ic_descr->id_ops->id_ipd_free(ipd);
394 * Concurrency: no concurrent access is possible that late in object
397 static void osd_index_fini(struct osd_object *o)
399 struct iam_container *bag;
401 if (o->oo_dir != NULL) {
402 bag = &o->oo_dir->od_container;
403 if (o->oo_inode != NULL) {
404 if (bag->ic_object == o->oo_inode)
405 iam_container_fini(bag);
407 OBD_FREE_PTR(o->oo_dir);
413 * Concurrency: no concurrent access is possible that late in object
414 * life-cycle (for all existing callers, that is. New callers have to provide
415 * their own locking.)
417 static int osd_inode_unlinked(const struct inode *inode)
419 return inode->i_nlink == 0;
423 OSD_TXN_OI_DELETE_CREDITS = 20,
424 OSD_TXN_INODE_DELETE_CREDITS = 20
428 * Concurrency: no concurrent access is possible that late in object
431 static int osd_inode_remove(const struct lu_env *env, struct osd_object *obj)
433 const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
434 struct osd_device *osd = osd_obj2dev(obj);
435 struct osd_thread_info *oti = osd_oti_get(env);
436 struct txn_param *prm = &oti->oti_txn;
440 txn_param_init(prm, OSD_TXN_OI_DELETE_CREDITS +
441 OSD_TXN_INODE_DELETE_CREDITS);
442 th = osd_trans_start(env, &osd->od_dt_dev, prm);
444 result = osd_oi_delete(oti, &osd->od_oi, fid, th);
445 osd_trans_stop(env, th);
447 result = PTR_ERR(th);
452 * Called just before object is freed. Releases all resources except for
453 * object itself (that is released by osd_object_free()).
455 * Concurrency: no concurrent access is possible that late in object
458 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
460 struct osd_object *obj = osd_obj(l);
461 struct inode *inode = obj->oo_inode;
463 LASSERT(osd_invariant(obj));
466 * If object is unlinked remove fid->ino mapping from object index.
468 * File body will be deleted by iput().
475 if (osd_inode_unlinked(inode)) {
476 result = osd_inode_remove(env, obj);
478 LU_OBJECT_DEBUG(D_ERROR, env, l,
479 "Failed to cleanup: %d\n",
483 obj->oo_inode = NULL;
488 * Concurrency: ->loo_object_release() is called under site spin-lock.
490 static void osd_object_release(const struct lu_env *env,
493 struct osd_object *o = osd_obj(l);
495 LASSERT(!lu_object_is_dying(l->lo_header));
496 if (o->oo_inode != NULL && osd_inode_unlinked(o->oo_inode))
497 set_bit(LU_OBJECT_HEARD_BANSHEE, &l->lo_header->loh_flags);
501 * Concurrency: shouldn't matter.
503 static int osd_object_print(const struct lu_env *env, void *cookie,
504 lu_printer_t p, const struct lu_object *l)
506 struct osd_object *o = osd_obj(l);
509 if (o->oo_dir != NULL)
510 d = o->oo_dir->od_container.ic_descr;
513 return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
515 o->oo_inode ? o->oo_inode->i_ino : 0UL,
516 o->oo_inode ? o->oo_inode->i_generation : 0,
517 d ? d->id_ops->id_name : "plain");
521 * Concurrency: shouldn't matter.
523 int osd_statfs(const struct lu_env *env, struct dt_device *d,
526 struct osd_device *osd = osd_dt_dev(d);
527 struct super_block *sb = osd_sb(osd);
530 spin_lock(&osd->od_osfs_lock);
532 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
533 result = ll_do_statfs(sb, &osd->od_kstatfs);
534 if (likely(result == 0)) /* N.B. statfs can't really fail */
535 osd->od_osfs_age = cfs_time_current_64();
538 if (likely(result == 0))
539 *sfs = osd->od_kstatfs;
540 spin_unlock(&osd->od_osfs_lock);
546 * Concurrency: doesn't access mutable data.
548 static void osd_conf_get(const struct lu_env *env,
549 const struct dt_device *dev,
550 struct dt_device_param *param)
553 * XXX should be taken from not-yet-existing fs abstraction layer.
555 param->ddp_max_name_len = LDISKFS_NAME_LEN;
556 param->ddp_max_nlink = LDISKFS_LINK_MAX;
557 param->ddp_block_shift = osd_sb(osd_dt_dev(dev))->s_blocksize_bits;
565 * Concurrency: doesn't access mutable data.
567 static int osd_param_is_sane(const struct osd_device *dev,
568 const struct txn_param *param)
570 return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
574 * Concurrency: shouldn't matter.
576 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
578 struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
579 struct thandle *th = &oh->ot_super;
580 struct dt_device *dev = th->th_dev;
582 LASSERT(dev != NULL);
583 LASSERT(oh->ot_handle == NULL);
586 CERROR("transaction @0x%p commit error: %d\n", th, error);
588 struct lu_env *env = &osd_dt_dev(dev)->od_env_for_commit;
590 * This od_env_for_commit is only for commit usage. see
593 lu_context_enter(&env->le_ctx);
594 dt_txn_hook_commit(env, th);
595 lu_context_exit(&env->le_ctx);
598 lu_device_put(&dev->dd_lu_dev);
601 lu_context_exit(&th->th_ctx);
602 lu_context_fini(&th->th_ctx);
607 * Concurrency: shouldn't matter.
609 static struct thandle *osd_trans_start(const struct lu_env *env,
613 struct osd_device *dev = osd_dt_dev(d);
615 struct osd_thandle *oh;
621 hook_res = dt_txn_hook_start(env, d, p);
623 RETURN(ERR_PTR(hook_res));
625 if (osd_param_is_sane(dev, p)) {
626 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
629 * XXX temporary stuff. Some abstraction layer should
633 jh = journal_start(osd_journal(dev), p->tp_credits);
639 jh->h_sync = p->tp_sync;
640 lu_device_get(&d->dd_lu_dev);
641 /* add commit callback */
642 lu_context_init(&th->th_ctx, LCT_TX_HANDLE);
643 lu_context_enter(&th->th_ctx);
644 journal_callback_set(jh, osd_trans_commit_cb,
645 (struct journal_callback *)&oh->ot_jcb);
648 struct osd_thread_info *oti =
651 LASSERT(oti->oti_txns == 0);
652 LASSERT(oti->oti_r_locks == 0);
653 LASSERT(oti->oti_w_locks == 0);
662 th = ERR_PTR(-ENOMEM);
664 CERROR("Invalid transaction parameters\n");
665 th = ERR_PTR(-EINVAL);
672 * Concurrency: shouldn't matter.
674 static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
677 struct osd_thandle *oh;
681 oh = container_of0(th, struct osd_thandle, ot_super);
682 if (oh->ot_handle != NULL) {
683 handle_t *hdl = oh->ot_handle;
685 * XXX temporary stuff. Some abstraction layer should be used.
687 result = dt_txn_hook_stop(env, th);
689 CERROR("Failure in transaction hook: %d\n", result);
692 oh->ot_handle = NULL;
693 result = journal_stop(hdl);
695 CERROR("Failure to stop transaction: %d\n", result);
699 struct osd_thread_info *oti = osd_oti_get(env);
701 LASSERT(oti->oti_txns == 1);
702 LASSERT(oti->oti_r_locks == 0);
703 LASSERT(oti->oti_w_locks == 0);
712 * Concurrency: shouldn't matter.
714 static int osd_sync(const struct lu_env *env, struct dt_device *d)
716 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
717 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
721 * Concurrency: shouldn't matter.
723 lvfs_sbdev_type fsfilt_ldiskfs_journal_sbdev(struct super_block *);
725 static void osd_ro(const struct lu_env *env, struct dt_device *d)
729 CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
731 __lvfs_set_rdonly(lvfs_sbdev(osd_sb(osd_dt_dev(d))),
732 fsfilt_ldiskfs_journal_sbdev(osd_sb(osd_dt_dev(d))));
737 * Concurrency: serialization provided by callers.
739 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
740 int mode, unsigned long timeout, __u32 alg,
741 struct lustre_capa_key *keys)
743 struct osd_device *dev = osd_dt_dev(d);
746 dev->od_fl_capa = mode;
747 dev->od_capa_timeout = timeout;
748 dev->od_capa_alg = alg;
749 dev->od_capa_keys = keys;
753 /* Note: we did not count into QUOTA here, If we mount with --data_journal
755 static const int osd_dto_credits[DTO_NR] = {
757 * Insert/Delete. IAM EXT3_INDEX_EXTRA_TRANS_BLOCKS(8) +
758 * EXT3_SINGLEDATA_TRANS_BLOCKS 8 XXX Note: maybe iam need more,since
759 * iam have more level than Ext3 htree
761 [DTO_INDEX_INSERT] = 16,
762 [DTO_INDEX_DELETE] = 16,
763 [DTO_IDNEX_UPDATE] = 16,
765 * Create a object. Same as create object in Ext3 filesystem, but did
766 * not count QUOTA i EXT3_DATA_TRANS_BLOCKS(12) +
767 * INDEX_EXTRA_BLOCKS(8) + 3(inode bits,groups, GDT)
769 [DTO_OBJECT_CREATE] = 23,
770 [DTO_OBJECT_DELETE] = 23,
772 * Attr set credits 3 inode, group, GDT
776 * XATTR_SET. SAME AS XATTR of EXT3 EXT3_DATA_TRANS_BLOCKS XXX Note:
777 * in original MDS implmentation EXT3_INDEX_EXTRA_TRANS_BLOCKS are
778 * also counted in. Do not know why?
780 [DTO_XATTR_SET] = 16,
782 /* creadits for inode change during write */
783 [DTO_WRITE_BASE] = 3,
784 /* credits for single block write */
785 [DTO_WRITE_BLOCK] = 12
788 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
791 LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits));
792 return osd_dto_credits[op];
795 static struct dt_device_operations osd_dt_ops = {
796 .dt_root_get = osd_root_get,
797 .dt_statfs = osd_statfs,
798 .dt_trans_start = osd_trans_start,
799 .dt_trans_stop = osd_trans_stop,
800 .dt_conf_get = osd_conf_get,
803 .dt_credit_get = osd_credit_get,
804 .dt_init_capa_ctxt = osd_init_capa_ctxt,
807 static void osd_object_read_lock(const struct lu_env *env,
808 struct dt_object *dt)
810 struct osd_object *obj = osd_dt_obj(dt);
812 LASSERT(osd_invariant(obj));
814 OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
815 down_read(&obj->oo_sem);
818 struct osd_thread_info *oti = osd_oti_get(env);
820 LASSERT(obj->oo_owner == NULL);
826 static void osd_object_write_lock(const struct lu_env *env,
827 struct dt_object *dt)
829 struct osd_object *obj = osd_dt_obj(dt);
831 LASSERT(osd_invariant(obj));
833 OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
834 down_write(&obj->oo_sem);
837 struct osd_thread_info *oti = osd_oti_get(env);
839 LASSERT(obj->oo_owner == NULL);
846 static void osd_object_read_unlock(const struct lu_env *env,
847 struct dt_object *dt)
849 struct osd_object *obj = osd_dt_obj(dt);
851 LASSERT(osd_invariant(obj));
854 struct osd_thread_info *oti = osd_oti_get(env);
856 LASSERT(oti->oti_r_locks > 0);
860 up_read(&obj->oo_sem);
863 static void osd_object_write_unlock(const struct lu_env *env,
864 struct dt_object *dt)
866 struct osd_object *obj = osd_dt_obj(dt);
868 LASSERT(osd_invariant(obj));
871 struct osd_thread_info *oti = osd_oti_get(env);
873 LASSERT(obj->oo_owner == env);
874 LASSERT(oti->oti_w_locks > 0);
876 obj->oo_owner = NULL;
879 up_write(&obj->oo_sem);
882 static int capa_is_sane(const struct lu_env *env,
883 struct osd_device *dev,
884 struct lustre_capa *capa,
885 struct lustre_capa_key *keys)
887 struct osd_thread_info *oti = osd_oti_get(env);
892 oc = capa_lookup(dev->od_capa_hash, capa, 0);
894 if (capa_is_expired(oc)) {
895 DEBUG_CAPA(D_ERROR, capa, "expired");
902 spin_lock(&capa_lock);
903 for (i = 0; i < 2; i++) {
904 if (keys[i].lk_keyid == capa->lc_keyid) {
905 oti->oti_capa_key = keys[i];
909 spin_unlock(&capa_lock);
912 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
916 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
919 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
921 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
925 oc = capa_add(dev->od_capa_hash, capa);
931 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
932 struct lustre_capa *capa, __u64 opc)
934 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
935 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
938 if (!dev->od_fl_capa)
941 if (capa == BYPASS_CAPA)
945 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
949 if (!lu_fid_eq(fid, &capa->lc_fid)) {
950 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
955 if (!capa_opc_supported(capa, opc)) {
956 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
960 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
961 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
968 static int osd_attr_get(const struct lu_env *env,
969 struct dt_object *dt,
970 struct lu_attr *attr,
971 struct lustre_capa *capa)
973 struct osd_object *obj = osd_dt_obj(dt);
975 LASSERT(dt_object_exists(dt));
976 LASSERT(osd_invariant(obj));
978 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
981 spin_lock(&obj->oo_guard);
982 osd_inode_getattr(env, obj->oo_inode, attr);
983 spin_unlock(&obj->oo_guard);
987 static int osd_attr_set(const struct lu_env *env,
988 struct dt_object *dt,
989 const struct lu_attr *attr,
990 struct thandle *handle,
991 struct lustre_capa *capa)
993 struct osd_object *obj = osd_dt_obj(dt);
995 LASSERT(handle != NULL);
996 LASSERT(dt_object_exists(dt));
997 LASSERT(osd_invariant(obj));
999 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1002 spin_lock(&obj->oo_guard);
1003 osd_inode_setattr(env, obj->oo_inode, attr);
1004 spin_unlock(&obj->oo_guard);
1006 mark_inode_dirty(obj->oo_inode);
1010 static struct timespec *osd_inode_time(const struct lu_env *env,
1011 struct inode *inode, __u64 seconds)
1013 struct osd_thread_info *oti = osd_oti_get(env);
1014 struct timespec *t = &oti->oti_time;
1016 t->tv_sec = seconds;
1018 *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1022 static void osd_inode_setattr(const struct lu_env *env,
1023 struct inode *inode, const struct lu_attr *attr)
1027 bits = attr->la_valid;
1029 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1031 if (bits & LA_ATIME)
1032 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1033 if (bits & LA_CTIME)
1034 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1035 if (bits & LA_MTIME)
1036 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1037 if (bits & LA_SIZE) {
1038 LDISKFS_I(inode)->i_disksize = attr->la_size;
1039 i_size_write(inode, attr->la_size);
1041 if (bits & LA_BLOCKS)
1042 inode->i_blocks = attr->la_blocks;
1044 inode->i_mode = (inode->i_mode & S_IFMT) |
1045 (attr->la_mode & ~S_IFMT);
1047 inode->i_uid = attr->la_uid;
1049 inode->i_gid = attr->la_gid;
1050 if (bits & LA_NLINK)
1051 inode->i_nlink = attr->la_nlink;
1053 inode->i_rdev = attr->la_rdev;
1055 if (bits & LA_FLAGS) {
1056 struct ldiskfs_inode_info *li = LDISKFS_I(inode);
1058 li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
1059 (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
1066 * XXX temporary solution.
1069 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1070 struct lu_attr *attr, struct thandle *th)
1075 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1076 struct lu_attr *attr, struct thandle *th)
1078 LASSERT(obj->oo_inode != NULL);
1080 osd_object_init0(obj);
1084 extern struct inode *ldiskfs_create_inode(handle_t *handle,
1085 struct inode * dir, int mode);
1087 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1089 struct dt_allocation_hint *hint,
1093 struct osd_device *osd = osd_obj2dev(obj);
1094 struct osd_thandle *oth;
1095 struct inode *parent;
1096 struct inode *inode;
1098 LASSERT(osd_invariant(obj));
1099 LASSERT(obj->oo_inode == NULL);
1100 LASSERT(osd->od_obj_area != NULL);
1102 oth = container_of(th, struct osd_thandle, ot_super);
1103 LASSERT(oth->ot_handle->h_transaction != NULL);
1105 if (hint && hint->dah_parent)
1106 parent = osd_dt_obj(hint->dah_parent)->oo_inode;
1108 parent = osd->od_obj_area->d_inode;
1109 LASSERT(parent->i_op != NULL);
1111 inode = ldiskfs_create_inode(oth->ot_handle, parent, mode);
1112 if (!IS_ERR(inode)) {
1113 obj->oo_inode = inode;
1116 result = PTR_ERR(inode);
1117 LASSERT(osd_invariant(obj));
1122 extern int iam_lvar_create(struct inode *obj, int keysize, int ptrsize,
1123 int recsize, handle_t *handle);
1129 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1130 struct lu_attr *attr,
1131 struct dt_allocation_hint *hint,
1135 struct osd_thandle *oth;
1137 LASSERT(S_ISDIR(attr->la_mode));
1139 oth = container_of(th, struct osd_thandle, ot_super);
1140 LASSERT(oth->ot_handle->h_transaction != NULL);
1141 result = osd_mkfile(info, obj, (attr->la_mode &
1142 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1144 LASSERT(obj->oo_inode != NULL);
1146 * XXX uh-oh... call low-level iam function directly.
1148 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1149 sizeof (struct lu_fid_pack),
1155 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1156 struct lu_attr *attr,
1157 struct dt_allocation_hint *hint,
1160 LASSERT(S_ISREG(attr->la_mode));
1161 return osd_mkfile(info, obj, (attr->la_mode &
1162 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1165 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1166 struct lu_attr *attr,
1167 struct dt_allocation_hint *hint,
1170 LASSERT(S_ISLNK(attr->la_mode));
1171 return osd_mkfile(info, obj, (attr->la_mode &
1172 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1175 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1176 struct lu_attr *attr,
1177 struct dt_allocation_hint *hint,
1181 struct osd_device *osd = osd_obj2dev(obj);
1183 umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1185 LASSERT(osd_invariant(obj));
1186 LASSERT(obj->oo_inode == NULL);
1187 LASSERT(osd->od_obj_area != NULL);
1188 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1189 S_ISFIFO(mode) || S_ISSOCK(mode));
1191 dir = osd->od_obj_area->d_inode;
1192 LASSERT(dir->i_op != NULL);
1194 result = osd_mkfile(info, obj, mode, hint, th);
1196 LASSERT(obj->oo_inode != NULL);
1197 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1199 LASSERT(osd_invariant(obj));
1203 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1205 struct dt_allocation_hint *hint,
1208 static osd_obj_type_f osd_create_type_f(__u32 mode)
1210 osd_obj_type_f result;
1236 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1237 struct dt_object *parent, umode_t child_mode)
1241 memset(ah, 0, sizeof(*ah));
1242 ah->dah_parent = parent;
1243 ah->dah_mode = child_mode;
1248 * Concurrency: @dt is write locked.
1250 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1251 struct lu_attr *attr,
1252 struct dt_allocation_hint *hint,
1255 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1256 struct osd_object *obj = osd_dt_obj(dt);
1257 struct osd_device *osd = osd_obj2dev(obj);
1258 struct osd_thread_info *info = osd_oti_get(env);
1263 LASSERT(osd_invariant(obj));
1264 LASSERT(!dt_object_exists(dt));
1265 LASSERT(osd_write_locked(env, obj));
1266 LASSERT(th != NULL);
1269 * XXX missing: Quote handling.
1272 result = osd_create_pre(info, obj, attr, th);
1274 result = osd_create_type_f(attr->la_mode & S_IFMT)(info, obj,
1277 result = osd_create_post(info, obj, attr, th);
1280 struct osd_inode_id *id = &info->oti_id;
1282 LASSERT(obj->oo_inode != NULL);
1284 id->oii_ino = obj->oo_inode->i_ino;
1285 id->oii_gen = obj->oo_inode->i_generation;
1287 result = osd_oi_insert(info, &osd->od_oi, fid, id, th);
1290 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1291 LASSERT(osd_invariant(obj));
1296 * Concurrency: @dt is write locked.
1298 static void osd_object_ref_add(const struct lu_env *env,
1299 struct dt_object *dt,
1302 struct osd_object *obj = osd_dt_obj(dt);
1303 struct inode *inode = obj->oo_inode;
1305 LASSERT(osd_invariant(obj));
1306 LASSERT(dt_object_exists(dt));
1307 LASSERT(osd_write_locked(env, obj));
1308 LASSERT(th != NULL);
1310 spin_lock(&obj->oo_guard);
1311 LASSERT(inode->i_nlink < LDISKFS_LINK_MAX);
1313 spin_unlock(&obj->oo_guard);
1314 mark_inode_dirty(inode);
1315 LASSERT(osd_invariant(obj));
1319 * Concurrency: @dt is write locked.
1321 static void osd_object_ref_del(const struct lu_env *env,
1322 struct dt_object *dt,
1325 struct osd_object *obj = osd_dt_obj(dt);
1326 struct inode *inode = obj->oo_inode;
1328 LASSERT(osd_invariant(obj));
1329 LASSERT(dt_object_exists(dt));
1330 LASSERT(osd_write_locked(env, obj));
1331 LASSERT(th != NULL);
1333 spin_lock(&obj->oo_guard);
1334 LASSERT(inode->i_nlink > 0);
1336 spin_unlock(&obj->oo_guard);
1337 mark_inode_dirty(inode);
1338 LASSERT(osd_invariant(obj));
1342 * Concurrency: @dt is read locked.
1344 static int osd_xattr_get(const struct lu_env *env,
1345 struct dt_object *dt,
1348 struct lustre_capa *capa)
1350 struct osd_object *obj = osd_dt_obj(dt);
1351 struct inode *inode = obj->oo_inode;
1352 struct osd_thread_info *info = osd_oti_get(env);
1353 struct dentry *dentry = &info->oti_dentry;
1355 LASSERT(dt_object_exists(dt));
1356 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1357 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1359 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1362 dentry->d_inode = inode;
1363 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
1367 * Concurrency: @dt is write locked.
1369 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1370 const struct lu_buf *buf, const char *name, int fl,
1371 struct thandle *handle, struct lustre_capa *capa)
1373 struct osd_object *obj = osd_dt_obj(dt);
1374 struct inode *inode = obj->oo_inode;
1375 struct osd_thread_info *info = osd_oti_get(env);
1376 struct dentry *dentry = &info->oti_dentry;
1377 struct timespec *t = &info->oti_time;
1378 int fs_flags = 0, rc;
1380 LASSERT(dt_object_exists(dt));
1381 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1382 LASSERT(osd_write_locked(env, obj));
1383 LASSERT(handle != NULL);
1385 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1388 if (fl & LU_XATTR_REPLACE)
1389 fs_flags |= XATTR_REPLACE;
1391 if (fl & LU_XATTR_CREATE)
1392 fs_flags |= XATTR_CREATE;
1394 dentry->d_inode = inode;
1395 *t = inode->i_ctime;
1396 rc = inode->i_op->setxattr(dentry, name,
1397 buf->lb_buf, buf->lb_len, fs_flags);
1398 if (likely(rc == 0)) {
1399 /* ctime should not be updated with server-side time. */
1400 spin_lock(&obj->oo_guard);
1401 inode->i_ctime = *t;
1402 spin_unlock(&obj->oo_guard);
1403 mark_inode_dirty(inode);
1409 * Concurrency: @dt is read locked.
1411 static int osd_xattr_list(const struct lu_env *env,
1412 struct dt_object *dt,
1414 struct lustre_capa *capa)
1416 struct osd_object *obj = osd_dt_obj(dt);
1417 struct inode *inode = obj->oo_inode;
1418 struct osd_thread_info *info = osd_oti_get(env);
1419 struct dentry *dentry = &info->oti_dentry;
1421 LASSERT(dt_object_exists(dt));
1422 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
1423 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1425 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1428 dentry->d_inode = inode;
1429 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
1433 * Concurrency: @dt is write locked.
1435 static int osd_xattr_del(const struct lu_env *env,
1436 struct dt_object *dt,
1438 struct thandle *handle,
1439 struct lustre_capa *capa)
1441 struct osd_object *obj = osd_dt_obj(dt);
1442 struct inode *inode = obj->oo_inode;
1443 struct osd_thread_info *info = osd_oti_get(env);
1444 struct dentry *dentry = &info->oti_dentry;
1445 struct timespec *t = &info->oti_time;
1448 LASSERT(dt_object_exists(dt));
1449 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
1450 LASSERT(osd_write_locked(env, obj));
1451 LASSERT(handle != NULL);
1453 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1456 dentry->d_inode = inode;
1457 *t = inode->i_ctime;
1458 rc = inode->i_op->removexattr(dentry, name);
1459 if (likely(rc == 0)) {
1460 /* ctime should not be updated with server-side time. */
1461 spin_lock(&obj->oo_guard);
1462 inode->i_ctime = *t;
1463 spin_unlock(&obj->oo_guard);
1464 mark_inode_dirty(inode);
1469 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1470 struct dt_object *dt,
1471 struct lustre_capa *old,
1474 struct osd_thread_info *info = osd_oti_get(env);
1475 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1476 struct osd_object *obj = osd_dt_obj(dt);
1477 struct osd_device *dev = osd_obj2dev(obj);
1478 struct lustre_capa_key *key = &info->oti_capa_key;
1479 struct lustre_capa *capa = &info->oti_capa;
1480 struct obd_capa *oc;
1484 if (!dev->od_fl_capa)
1485 RETURN(ERR_PTR(-ENOENT));
1487 LASSERT(dt_object_exists(dt));
1488 LASSERT(osd_invariant(obj));
1490 /* renewal sanity check */
1491 if (old && osd_object_auth(env, dt, old, opc))
1492 RETURN(ERR_PTR(-EACCES));
1494 capa->lc_fid = *fid;
1497 capa->lc_flags = dev->od_capa_alg << 24;
1498 capa->lc_timeout = dev->od_capa_timeout;
1499 capa->lc_expiry = 0;
1501 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1503 LASSERT(!capa_is_expired(oc));
1507 spin_lock(&capa_lock);
1508 *key = dev->od_capa_keys[1];
1509 spin_unlock(&capa_lock);
1511 capa->lc_keyid = key->lk_keyid;
1512 capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
1514 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1516 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1517 RETURN(ERR_PTR(rc));
1520 oc = capa_add(dev->od_capa_hash, capa);
1524 static struct dt_object_operations osd_obj_ops = {
1525 .do_read_lock = osd_object_read_lock,
1526 .do_write_lock = osd_object_write_lock,
1527 .do_read_unlock = osd_object_read_unlock,
1528 .do_write_unlock = osd_object_write_unlock,
1529 .do_attr_get = osd_attr_get,
1530 .do_attr_set = osd_attr_set,
1531 .do_ah_init = osd_ah_init,
1532 .do_create = osd_object_create,
1533 .do_index_try = osd_index_try,
1534 .do_ref_add = osd_object_ref_add,
1535 .do_ref_del = osd_object_ref_del,
1536 .do_xattr_get = osd_xattr_get,
1537 .do_xattr_set = osd_xattr_set,
1538 .do_xattr_del = osd_xattr_del,
1539 .do_xattr_list = osd_xattr_list,
1540 .do_capa_get = osd_capa_get,
1548 * XXX: Another layering violation for now.
1550 * We don't want to use ->f_op->read methods, because generic file write
1552 * - serializes on ->i_sem, and
1554 * - does a lot of extra work like balance_dirty_pages(),
1556 * which doesn't work for globally shared files like /last-received.
1558 int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1559 int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
1560 loff_t *offs, handle_t *handle);
1562 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
1563 struct lu_buf *buf, loff_t *pos,
1564 struct lustre_capa *capa)
1566 struct inode *inode = osd_dt_obj(dt)->oo_inode;
1568 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
1571 return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
1574 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
1575 const struct lu_buf *buf, loff_t *pos,
1576 struct thandle *handle, struct lustre_capa *capa)
1578 struct inode *inode = osd_dt_obj(dt)->oo_inode;
1579 struct osd_thandle *oh;
1582 LASSERT(handle != NULL);
1584 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
1587 oh = container_of(handle, struct osd_thandle, ot_super);
1588 LASSERT(oh->ot_handle->h_transaction != NULL);
1589 result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
1590 pos, oh->ot_handle);
1592 result = buf->lb_len;
1596 static struct dt_body_operations osd_body_ops = {
1597 .dbo_read = osd_read,
1598 .dbo_write = osd_write
1605 static int osd_object_is_root(const struct osd_object *obj)
1607 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
1610 static int osd_index_probe(const struct lu_env *env, struct osd_object *o,
1611 const struct dt_index_features *feat)
1613 struct iam_descr *descr;
1615 if (osd_object_is_root(o))
1616 return feat == &dt_directory_features;
1618 LASSERT(o->oo_dir != NULL);
1620 descr = o->oo_dir->od_container.ic_descr;
1621 if (feat == &dt_directory_features)
1622 return descr == &iam_htree_compat_param ||
1623 (descr->id_rec_size == sizeof(struct lu_fid_pack) &&
1625 * XXX check that index looks like directory.
1630 feat->dif_keysize_min <= descr->id_key_size &&
1631 descr->id_key_size <= feat->dif_keysize_max &&
1632 feat->dif_recsize_min <= descr->id_rec_size &&
1633 descr->id_rec_size <= feat->dif_recsize_max &&
1634 !(feat->dif_flags & (DT_IND_VARKEY |
1635 DT_IND_VARREC | DT_IND_NONUNQ)) &&
1636 ergo(feat->dif_flags & DT_IND_UPDATE,
1637 1 /* XXX check that object (and file system) is
1641 static int osd_container_init(const struct lu_env *env,
1642 struct osd_object *obj,
1643 struct osd_directory *dir)
1646 struct iam_container *bag;
1648 bag = &dir->od_container;
1649 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
1651 result = iam_container_setup(bag);
1653 obj->oo_dt.do_index_ops = &osd_index_ops;
1655 iam_container_fini(bag);
1661 * Concurrency: no external locking is necessary.
1663 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1664 const struct dt_index_features *feat)
1667 struct osd_object *obj = osd_dt_obj(dt);
1669 LASSERT(osd_invariant(obj));
1670 LASSERT(dt_object_exists(dt));
1672 if (osd_object_is_root(obj)) {
1673 dt->do_index_ops = &osd_index_compat_ops;
1675 } else if (!osd_has_index(obj)) {
1676 struct osd_directory *dir;
1680 sema_init(&dir->od_sem, 1);
1682 spin_lock(&obj->oo_guard);
1683 if (obj->oo_dir == NULL)
1687 * Concurrent thread allocated container data.
1690 spin_unlock(&obj->oo_guard);
1692 * Now, that we have container data, serialize its
1695 down(&obj->oo_dir->od_sem);
1697 * recheck under lock.
1699 if (!osd_has_index(obj))
1700 result = osd_container_init(env, obj, dir);
1703 up(&obj->oo_dir->od_sem);
1710 if (!osd_index_probe(env, obj, feat))
1713 LASSERT(osd_invariant(obj));
1718 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1719 const struct dt_key *key, struct thandle *handle,
1720 struct lustre_capa *capa)
1722 struct osd_object *obj = osd_dt_obj(dt);
1723 struct osd_thandle *oh;
1724 struct iam_path_descr *ipd;
1725 struct iam_container *bag = &obj->oo_dir->od_container;
1730 LASSERT(osd_invariant(obj));
1731 LASSERT(dt_object_exists(dt));
1732 LASSERT(bag->ic_object == obj->oo_inode);
1733 LASSERT(handle != NULL);
1735 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
1738 ipd = osd_ipd_get(env, bag);
1739 if (unlikely(ipd == NULL))
1742 oh = container_of0(handle, struct osd_thandle, ot_super);
1743 LASSERT(oh->ot_handle != NULL);
1744 LASSERT(oh->ot_handle->h_transaction != NULL);
1746 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
1747 osd_ipd_put(env, bag, ipd);
1748 LASSERT(osd_invariant(obj));
1752 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1753 struct dt_rec *rec, const struct dt_key *key,
1754 struct lustre_capa *capa)
1756 struct osd_object *obj = osd_dt_obj(dt);
1757 struct iam_path_descr *ipd;
1758 struct iam_container *bag = &obj->oo_dir->od_container;
1763 LASSERT(osd_invariant(obj));
1764 LASSERT(dt_object_exists(dt));
1765 LASSERT(bag->ic_object == obj->oo_inode);
1767 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
1770 ipd = osd_ipd_get(env, bag);
1771 if (unlikely(ipd == NULL))
1774 rc = iam_lookup(bag, (const struct iam_key *)key,
1775 (struct iam_rec *)rec, ipd);
1776 osd_ipd_put(env, bag, ipd);
1777 LASSERT(osd_invariant(obj));
1782 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1783 const struct dt_rec *rec, const struct dt_key *key,
1784 struct thandle *th, struct lustre_capa *capa)
1786 struct osd_object *obj = osd_dt_obj(dt);
1787 struct iam_path_descr *ipd;
1788 struct osd_thandle *oh;
1789 struct iam_container *bag = &obj->oo_dir->od_container;
1794 LASSERT(osd_invariant(obj));
1795 LASSERT(dt_object_exists(dt));
1796 LASSERT(bag->ic_object == obj->oo_inode);
1797 LASSERT(th != NULL);
1799 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
1802 ipd = osd_ipd_get(env, bag);
1803 if (unlikely(ipd == NULL))
1806 oh = container_of0(th, struct osd_thandle, ot_super);
1807 LASSERT(oh->ot_handle != NULL);
1808 LASSERT(oh->ot_handle->h_transaction != NULL);
1809 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
1810 (struct iam_rec *)rec, ipd);
1811 osd_ipd_put(env, bag, ipd);
1812 LASSERT(osd_invariant(obj));
1817 * Iterator operations.
1820 struct osd_object *oi_obj;
1821 struct iam_path_descr *oi_ipd;
1822 struct iam_iterator oi_it;
1825 static struct dt_it *osd_it_init(const struct lu_env *env,
1826 struct dt_object *dt, int writable,
1827 struct lustre_capa *capa)
1830 struct osd_object *obj = osd_dt_obj(dt);
1831 struct lu_object *lo = &dt->do_lu;
1832 struct iam_path_descr *ipd;
1833 struct iam_container *bag = &obj->oo_dir->od_container;
1836 LASSERT(lu_object_exists(lo));
1838 if (osd_object_auth(env, dt, capa, writable ? CAPA_OPC_BODY_WRITE :
1839 CAPA_OPC_BODY_READ))
1840 return ERR_PTR(-EACCES);
1842 flags = writable ? IAM_IT_MOVE|IAM_IT_WRITE : IAM_IT_MOVE;
1846 * XXX: as ipd is allocated within osd_thread_info, assignment
1847 * below implies that iterator usage is confined within single
1850 ipd = osd_ipd_get(env, bag);
1851 if (likely(ipd != NULL)) {
1855 iam_it_init(&it->oi_it, bag, flags, ipd);
1856 return (struct dt_it *)it;
1860 return ERR_PTR(-ENOMEM);
1863 static void osd_it_fini(const struct lu_env *env, struct dt_it *di)
1865 struct osd_it *it = (struct osd_it *)di;
1866 struct osd_object *obj = it->oi_obj;
1868 iam_it_fini(&it->oi_it);
1869 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
1870 lu_object_put(env, &obj->oo_dt.do_lu);
1874 static int osd_it_get(const struct lu_env *env,
1875 struct dt_it *di, const struct dt_key *key)
1877 struct osd_it *it = (struct osd_it *)di;
1879 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
1882 static void osd_it_put(const struct lu_env *env, struct dt_it *di)
1884 struct osd_it *it = (struct osd_it *)di;
1886 iam_it_put(&it->oi_it);
1889 static int osd_it_next(const struct lu_env *env, struct dt_it *di)
1891 struct osd_it *it = (struct osd_it *)di;
1893 return iam_it_next(&it->oi_it);
1896 static int osd_it_del(const struct lu_env *env, struct dt_it *di,
1899 struct osd_it *it = (struct osd_it *)di;
1900 struct osd_thandle *oh;
1902 LASSERT(th != NULL);
1904 oh = container_of0(th, struct osd_thandle, ot_super);
1905 LASSERT(oh->ot_handle != NULL);
1906 LASSERT(oh->ot_handle->h_transaction != NULL);
1908 return iam_it_rec_delete(oh->ot_handle, &it->oi_it);
1911 static struct dt_key *osd_it_key(const struct lu_env *env,
1912 const struct dt_it *di)
1914 struct osd_it *it = (struct osd_it *)di;
1916 return (struct dt_key *)iam_it_key_get(&it->oi_it);
1919 static int osd_it_key_size(const struct lu_env *env, const struct dt_it *di)
1921 struct osd_it *it = (struct osd_it *)di;
1923 return iam_it_key_size(&it->oi_it);
1926 static struct dt_rec *osd_it_rec(const struct lu_env *env,
1927 const struct dt_it *di)
1929 struct osd_it *it = (struct osd_it *)di;
1931 return (struct dt_rec *)iam_it_rec_get(&it->oi_it);
1934 static __u64 osd_it_store(const struct lu_env *env, const struct dt_it *di)
1936 struct osd_it *it = (struct osd_it *)di;
1938 return iam_it_store(&it->oi_it);
1941 static int osd_it_load(const struct lu_env *env,
1942 const struct dt_it *di, __u64 hash)
1944 struct osd_it *it = (struct osd_it *)di;
1946 return iam_it_load(&it->oi_it, hash);
1949 static struct dt_index_operations osd_index_ops = {
1950 .dio_lookup = osd_index_lookup,
1951 .dio_insert = osd_index_insert,
1952 .dio_delete = osd_index_delete,
1954 .init = osd_it_init,
1955 .fini = osd_it_fini,
1959 .next = osd_it_next,
1961 .key_size = osd_it_key_size,
1963 .store = osd_it_store,
1968 static int osd_index_compat_delete(const struct lu_env *env,
1969 struct dt_object *dt,
1970 const struct dt_key *key,
1971 struct thandle *handle,
1972 struct lustre_capa *capa)
1974 struct osd_object *obj = osd_dt_obj(dt);
1976 LASSERT(handle != NULL);
1977 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
1981 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
1985 RETURN(-EOPNOTSUPP);
1989 * Compatibility index operations.
1993 static void osd_build_pack(const struct lu_env *env, struct osd_device *osd,
1994 struct dentry *dentry, struct lu_fid_pack *pack)
1996 struct inode *inode = dentry->d_inode;
1997 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
1999 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2000 fid_cpu_to_be(fid, fid);
2001 pack->fp_len = sizeof *fid + 1;
2002 memcpy(pack->fp_area, fid, sizeof *fid);
2005 static int osd_index_compat_lookup(const struct lu_env *env,
2006 struct dt_object *dt,
2007 struct dt_rec *rec, const struct dt_key *key,
2008 struct lustre_capa *capa)
2010 struct osd_object *obj = osd_dt_obj(dt);
2012 struct osd_device *osd = osd_obj2dev(obj);
2013 struct osd_thread_info *info = osd_oti_get(env);
2019 * XXX temporary solution.
2021 struct dentry *dentry;
2022 struct dentry *parent;
2024 LASSERT(osd_invariant(obj));
2025 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2026 LASSERT(osd_has_index(obj));
2028 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2031 info->oti_str.name = (const char *)key;
2032 info->oti_str.len = strlen((const char *)key);
2034 dir = obj->oo_inode;
2035 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2037 parent = d_alloc_root(dir);
2041 dentry = d_alloc(parent, &info->oti_str);
2042 if (dentry != NULL) {
2046 * XXX passing NULL for nameidata should work for
2049 d = dir->i_op->lookup(dir, dentry, NULL);
2052 * normal case, result is in @dentry.
2054 if (dentry->d_inode != NULL) {
2055 osd_build_pack(env, osd, dentry,
2056 (struct lu_fid_pack *)rec);
2061 /* What? Disconnected alias? Ppheeeww... */
2062 CERROR("Aliasing where not expected\n");
2070 LASSERT(osd_invariant(obj));
2074 static int osd_add_rec(struct osd_thread_info *info, struct osd_device *dev,
2075 struct inode *dir, struct inode *inode, const char *name)
2079 struct dentry *parent;
2083 info->oti_str.name = name;
2084 info->oti_str.len = strlen(name);
2086 LASSERT(atomic_read(&dir->i_count) > 0);
2088 old = d_alloc(dev->od_obj_area, &info->oti_str);
2090 d_instantiate(old, inode);
2092 LASSERT(atomic_read(&dir->i_count) > 0);
2093 parent = d_alloc_root(dir);
2094 if (parent != NULL) {
2096 LASSERT(atomic_read(&dir->i_count) > 1);
2097 new = d_alloc(parent, &info->oti_str);
2098 LASSERT(atomic_read(&dir->i_count) > 1);
2100 LASSERT(atomic_read(&dir->i_count) > 1);
2101 result = dir->i_op->link(old, dir, new);
2102 LASSERT(atomic_read(&dir->i_count) > 1);
2104 LASSERT(atomic_read(&dir->i_count) > 1);
2106 LASSERT(atomic_read(&dir->i_count) > 1);
2108 LASSERT(atomic_read(&dir->i_count) > 0);
2112 LASSERT(atomic_read(&dir->i_count) > 0);
2118 * XXX Temporary stuff.
2120 static int osd_index_compat_insert(const struct lu_env *env,
2121 struct dt_object *dt,
2122 const struct dt_rec *rec,
2123 const struct dt_key *key, struct thandle *th,
2124 struct lustre_capa *capa)
2126 struct osd_object *obj = osd_dt_obj(dt);
2128 const char *name = (const char *)key;
2130 struct lu_device *ludev = dt->do_lu.lo_dev;
2131 struct lu_object *luch;
2133 struct osd_thread_info *info = osd_oti_get(env);
2134 const struct lu_fid_pack *pack = (const struct lu_fid_pack *)rec;
2135 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
2139 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2140 LASSERT(osd_invariant(obj));
2141 LASSERT(th != NULL);
2143 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2146 result = fid_unpack(pack, fid);
2150 luch = lu_object_find(env, ludev->ld_site, fid);
2151 if (!IS_ERR(luch)) {
2152 if (lu_object_exists(luch)) {
2153 struct osd_object *child;
2155 child = osd_obj(lu_object_locate(luch->lo_header,
2158 result = osd_add_rec(info, osd_obj2dev(obj),
2160 child->oo_inode, name);
2162 CERROR("No osd slice.\n");
2165 LASSERT(osd_invariant(obj));
2166 LASSERT(osd_invariant(child));
2171 lu_object_put(env, luch);
2173 result = PTR_ERR(luch);
2174 LASSERT(osd_invariant(obj));
2178 static struct dt_index_operations osd_index_compat_ops = {
2179 .dio_lookup = osd_index_compat_lookup,
2180 .dio_insert = osd_index_compat_insert,
2181 .dio_delete = osd_index_compat_delete
2184 /* type constructor/destructor: osd_type_init, osd_type_fini */
2185 LU_TYPE_INIT_FINI(osd, &osd_key);
2187 static struct lu_context_key osd_key = {
2188 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
2189 .lct_init = osd_key_init,
2190 .lct_fini = osd_key_fini,
2191 .lct_exit = osd_key_exit
2194 static void *osd_key_init(const struct lu_context *ctx,
2195 struct lu_context_key *key)
2197 struct osd_thread_info *info;
2199 OBD_ALLOC_PTR(info);
2201 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
2203 info = ERR_PTR(-ENOMEM);
2207 /* context key destructor: osd_key_fini */
2208 LU_KEY_FINI(osd, struct osd_thread_info);
2210 static void osd_key_exit(const struct lu_context *ctx,
2211 struct lu_context_key *key, void *data)
2214 struct osd_thread_info *info = data;
2216 LASSERT(info->oti_r_locks == 0);
2217 LASSERT(info->oti_w_locks == 0);
2218 LASSERT(info->oti_txns == 0);
2222 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
2223 const char *name, struct lu_device *next)
2226 /* context for commit hooks */
2227 rc = lu_context_init(&osd_dev(d)->od_env_for_commit.le_ctx,
2230 rc = osd_procfs_init(osd_dev(d), name);
2234 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
2236 struct osd_thread_info *info = osd_oti_get(env);
2238 if (o->od_obj_area != NULL) {
2239 dput(o->od_obj_area);
2240 o->od_obj_area = NULL;
2242 osd_oi_fini(info, &o->od_oi);
2247 static int osd_mount(const struct lu_env *env,
2248 struct osd_device *o, struct lustre_cfg *cfg)
2250 struct lustre_mount_info *lmi;
2251 const char *dev = lustre_cfg_string(cfg, 0);
2252 struct osd_thread_info *info = osd_oti_get(env);
2257 if (o->od_mount != NULL) {
2258 CERROR("Already mounted (%s)\n", dev);
2263 lmi = server_get_mount(dev);
2265 CERROR("Cannot get mount info for %s!\n", dev);
2269 LASSERT(lmi != NULL);
2270 /* save lustre_mount_info in dt_device */
2273 result = osd_oi_init(info, &o->od_oi, &o->od_dt_dev);
2277 d = simple_mkdir(osd_sb(o)->s_root, lmi->lmi_mnt, "*OBJ-TEMP*",
2282 result = PTR_ERR(d);
2285 osd_shutdown(env, o);
2289 static struct lu_device *osd_device_fini(const struct lu_env *env,
2290 struct lu_device *d)
2295 shrink_dcache_sb(osd_sb(osd_dev(d)));
2296 osd_sync(env, lu2dt_dev(d));
2298 rc = osd_procfs_fini(osd_dev(d));
2300 CERROR("proc fini error %d \n", rc);
2301 RETURN (ERR_PTR(rc));
2304 if (osd_dev(d)->od_mount)
2305 server_put_mount(osd_dev(d)->od_mount->lmi_name,
2306 osd_dev(d)->od_mount->lmi_mnt);
2307 osd_dev(d)->od_mount = NULL;
2309 lu_context_fini(&osd_dev(d)->od_env_for_commit.le_ctx);
2313 static struct lu_device *osd_device_alloc(const struct lu_env *env,
2314 struct lu_device_type *t,
2315 struct lustre_cfg *cfg)
2317 struct lu_device *l;
2318 struct osd_device *o;
2324 result = dt_device_init(&o->od_dt_dev, t);
2327 l->ld_ops = &osd_lu_ops;
2328 o->od_dt_dev.dd_ops = &osd_dt_ops;
2329 spin_lock_init(&o->od_osfs_lock);
2330 o->od_osfs_age = cfs_time_shift_64(-1000);
2331 o->od_capa_hash = init_capa_hash();
2332 if (o->od_capa_hash == NULL) {
2333 dt_device_fini(&o->od_dt_dev);
2334 l = ERR_PTR(-ENOMEM);
2337 l = ERR_PTR(result);
2342 l = ERR_PTR(-ENOMEM);
2346 static struct lu_device *osd_device_free(const struct lu_env *env,
2347 struct lu_device *d)
2349 struct osd_device *o = osd_dev(d);
2352 cleanup_capa_hash(o->od_capa_hash);
2353 dt_device_fini(&o->od_dt_dev);
2358 static int osd_process_config(const struct lu_env *env,
2359 struct lu_device *d, struct lustre_cfg *cfg)
2361 struct osd_device *o = osd_dev(d);
2365 switch(cfg->lcfg_command) {
2367 err = osd_mount(env, o, cfg);
2370 err = osd_shutdown(env, o);
2378 extern void ldiskfs_orphan_cleanup (struct super_block * sb,
2379 struct ldiskfs_super_block * es);
2381 static int osd_recovery_complete(const struct lu_env *env,
2382 struct lu_device *d)
2384 struct osd_device *o = osd_dev(d);
2386 /* TODO: orphans handling */
2387 ldiskfs_orphan_cleanup(osd_sb(o), LDISKFS_SB(osd_sb(o))->s_es);
2391 static struct inode *osd_iget(struct osd_thread_info *info,
2392 struct osd_device *dev,
2393 const struct osd_inode_id *id)
2395 struct inode *inode;
2397 inode = iget(osd_sb(dev), id->oii_ino);
2398 if (inode == NULL) {
2399 CERROR("no inode\n");
2400 inode = ERR_PTR(-EACCES);
2401 } else if (is_bad_inode(inode)) {
2402 CERROR("bad inode\n");
2404 inode = ERR_PTR(-ENOENT);
2405 } else if (inode->i_generation != id->oii_gen) {
2406 CERROR("stale inode\n");
2408 inode = ERR_PTR(-ESTALE);
2415 static int osd_fid_lookup(const struct lu_env *env,
2416 struct osd_object *obj, const struct lu_fid *fid)
2418 struct osd_thread_info *info;
2419 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
2420 struct osd_device *dev;
2421 struct osd_inode_id *id;
2423 struct inode *inode;
2426 LASSERT(osd_invariant(obj));
2427 LASSERT(obj->oo_inode == NULL);
2428 LASSERT(fid_is_sane(fid));
2430 * This assertion checks that osd layer sees only local
2431 * fids. Unfortunately it is somewhat expensive (does a
2432 * cache-lookup). Disabling it for production/acceptance-testing.
2434 LASSERT(1 || fid_is_local(ldev->ld_site, fid));
2438 info = osd_oti_get(env);
2439 dev = osd_dev(ldev);
2443 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
2446 result = osd_oi_lookup(info, oi, fid, id);
2448 inode = osd_iget(info, dev, id);
2449 if (!IS_ERR(inode)) {
2450 obj->oo_inode = inode;
2451 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
2455 * If fid wasn't found in oi, inode-less object is
2456 * created, for which lu_object_exists() returns
2457 * false. This is used in a (frequent) case when
2458 * objects are created as locking anchors or
2459 * place holders for objects yet to be created.
2461 result = PTR_ERR(inode);
2462 } else if (result == -ENOENT)
2464 LASSERT(osd_invariant(obj));
2468 static void osd_inode_getattr(const struct lu_env *env,
2469 struct inode *inode, struct lu_attr *attr)
2471 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2472 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2473 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
2475 attr->la_atime = LTIME_S(inode->i_atime);
2476 attr->la_mtime = LTIME_S(inode->i_mtime);
2477 attr->la_ctime = LTIME_S(inode->i_ctime);
2478 attr->la_mode = inode->i_mode;
2479 attr->la_size = i_size_read(inode);
2480 attr->la_blocks = inode->i_blocks;
2481 attr->la_uid = inode->i_uid;
2482 attr->la_gid = inode->i_gid;
2483 attr->la_flags = LDISKFS_I(inode)->i_flags;
2484 attr->la_nlink = inode->i_nlink;
2485 attr->la_rdev = inode->i_rdev;
2486 attr->la_blksize = ll_inode_blksize(inode);
2487 attr->la_blkbits = inode->i_blkbits;
2494 static int lu_device_is_osd(const struct lu_device *d)
2496 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
2499 static struct osd_object *osd_obj(const struct lu_object *o)
2501 LASSERT(lu_device_is_osd(o->lo_dev));
2502 return container_of0(o, struct osd_object, oo_dt.do_lu);
2505 static struct osd_device *osd_dt_dev(const struct dt_device *d)
2507 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
2508 return container_of0(d, struct osd_device, od_dt_dev);
2511 static struct osd_device *osd_dev(const struct lu_device *d)
2513 LASSERT(lu_device_is_osd(d));
2514 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
2517 static struct osd_object *osd_dt_obj(const struct dt_object *d)
2519 return osd_obj(&d->do_lu);
2522 static struct osd_device *osd_obj2dev(const struct osd_object *o)
2524 return osd_dev(o->oo_dt.do_lu.lo_dev);
2527 static struct lu_device *osd2lu_dev(struct osd_device *osd)
2529 return &osd->od_dt_dev.dd_lu_dev;
2532 static struct super_block *osd_sb(const struct osd_device *dev)
2534 return dev->od_mount->lmi_mnt->mnt_sb;
2537 static journal_t *osd_journal(const struct osd_device *dev)
2539 return LDISKFS_SB(osd_sb(dev))->s_journal;
2542 static int osd_has_index(const struct osd_object *obj)
2544 return obj->oo_dt.do_index_ops != NULL;
2547 static int osd_object_invariant(const struct lu_object *l)
2549 return osd_invariant(osd_obj(l));
2552 static struct lu_object_operations osd_lu_obj_ops = {
2553 .loo_object_init = osd_object_init,
2554 .loo_object_delete = osd_object_delete,
2555 .loo_object_release = osd_object_release,
2556 .loo_object_free = osd_object_free,
2557 .loo_object_print = osd_object_print,
2558 .loo_object_invariant = osd_object_invariant
2561 static struct lu_device_operations osd_lu_ops = {
2562 .ldo_object_alloc = osd_object_alloc,
2563 .ldo_process_config = osd_process_config,
2564 .ldo_recovery_complete = osd_recovery_complete
2567 static struct lu_device_type_operations osd_device_type_ops = {
2568 .ldto_init = osd_type_init,
2569 .ldto_fini = osd_type_fini,
2571 .ldto_device_alloc = osd_device_alloc,
2572 .ldto_device_free = osd_device_free,
2574 .ldto_device_init = osd_device_init,
2575 .ldto_device_fini = osd_device_fini
2578 static struct lu_device_type osd_device_type = {
2579 .ldt_tags = LU_DEVICE_DT,
2580 .ldt_name = LUSTRE_OSD_NAME,
2581 .ldt_ops = &osd_device_type_ops,
2582 .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
2586 * lprocfs legacy support.
2588 static struct obd_ops osd_obd_device_ops = {
2589 .o_owner = THIS_MODULE
2592 static int __init osd_mod_init(void)
2594 struct lprocfs_static_vars lvars;
2596 lprocfs_osd_init_vars(&lvars);
2597 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
2598 LUSTRE_OSD_NAME, &osd_device_type);
2601 static void __exit osd_mod_exit(void)
2603 class_unregister_type(LUSTRE_OSD_NAME);
2606 MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
2607 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
2608 MODULE_LICENSE("GPL");
2610 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);