1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * lustre/osd/osd_handler.c
5 * Top-level entry points into osd module
7 * Copyright (c) 2006 Cluster File Systems, Inc.
8 * Author: Nikita Danilov <nikita@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
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.
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.
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.
30 # define EXPORT_SYMTAB
32 #define DEBUG_SUBSYSTEM S_MDS
34 #include <linux/module.h>
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 */
42 /* XATTR_{REPLACE,CREATE} */
43 #include <linux/xattr.h>
45 * XXX temporary stuff: direct access to ldiskfs/jdb. Interface between osd
46 * and file system is not yet specified.
48 /* handle_t, journal_start(), journal_stop() */
49 #include <linux/jbd.h>
51 #include <linux/ldiskfs_fs.h>
52 #include <linux/ldiskfs_jbd.h>
57 * struct OBD_{ALLOC,FREE}*()
60 #include <obd_support.h>
61 /* struct ptlrpc_thread */
62 #include <lustre_net.h>
65 /* class_register_type(), class_unregister_type(), class_get_type() */
66 #include <obd_class.h>
67 #include <lustre_disk.h>
70 #include <lustre_fid.h>
71 #include <linux/lustre_iam.h>
73 #include "osd_internal.h"
76 struct osd_directory {
77 struct iam_container od_container;
78 struct iam_descr od_descr;
79 struct semaphore od_sem;
83 struct dt_object oo_dt;
85 * Inode for file system object represented by this osd_object. This
86 * inode is pinned for the whole duration of lu_object life.
88 * Not modified concurrently (either setup early during object
89 * creation, or assigned by osd_object_create() under write lock).
91 struct inode *oo_inode;
92 struct rw_semaphore oo_sem;
93 struct osd_directory *oo_dir;
94 /* protects inode attributes. */
97 const struct lu_env *oo_owner;
106 struct dt_device od_dt_dev;
107 /* information about underlying file system */
108 struct lustre_mount_info *od_mount;
112 * XXX temporary stuff for object index: directory where every object
113 * is named by its fid.
115 struct dentry *od_obj_area;
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.
123 struct lu_env od_env_for_commit;
128 unsigned int od_fl_capa:1;
129 unsigned long od_capa_timeout;
131 struct lustre_capa_key *od_capa_keys;
132 struct hlist_head *od_capa_hash;
135 * statfs optimization: we cache a bit.
137 cfs_time_t od_osfs_age;
138 struct kstatfs od_kstatfs;
139 spinlock_t od_osfs_lock;
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);
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 *,
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);
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,
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,
216 static int osd_object_is_root(const struct osd_object *obj);
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,
246 static struct thandle *osd_trans_start (const struct lu_env *env,
248 struct txn_param *p);
249 static journal_t *osd_journal (const struct osd_device *dev);
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;
265 struct thandle ot_super;
267 struct journal_callback ot_jcb;
271 * Invariants, assertions.
275 * XXX: do not enable this, until invariant checking code is made thread safe
276 * in the face of pdirops locking.
278 #define OSD_INVARIANT_CHECKS (0)
280 #if OSD_INVARIANT_CHECKS
281 static int osd_invariant(const struct osd_object *obj)
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);
293 #define osd_invariant(obj) (1)
296 static inline struct osd_thread_info *osd_oti_get(const struct lu_env *env)
298 return lu_context_key_get(&env->le_ctx, &osd_key);
303 * Concurrency: doesn't matter
305 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
307 return osd_oti_get(env)->oti_r_locks > 0;
311 * Concurrency: doesn't matter
313 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
315 struct osd_thread_info *oti = osd_oti_get(env);
316 return oti->oti_w_locks > 0 && o->oo_owner == env;
319 #define OSD_COUNTERS_DO(exp) exp
323 #define osd_read_locked(env, o) (1)
324 #define osd_write_locked(env, o) (1)
325 #define OSD_COUNTERS_DO(exp) ((void)0)
329 * Concurrency: doesn't access mutable data
331 static int osd_root_get(const struct lu_env *env,
332 struct dt_device *dev, struct lu_fid *f)
336 inode = osd_sb(osd_dt_dev(dev))->s_root->d_inode;
337 lu_igif_build(f, inode->i_ino, inode->i_generation);
342 * OSD object methods.
346 * Concurrency: no concurrent access is possible that early in object
349 static struct lu_object *osd_object_alloc(const struct lu_env *env,
350 const struct lu_object_header *hdr,
353 struct osd_object *mo;
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);
371 * Concurrency: shouldn't matter.
373 static void osd_object_init0(struct osd_object *obj)
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));
382 * Concurrency: no concurrent access is possible that early in object
385 static int osd_object_init(const struct lu_env *env, struct lu_object *l)
387 struct osd_object *obj = osd_obj(l);
390 LASSERT(osd_invariant(obj));
392 result = osd_fid_lookup(env, obj, lu_object_fid(l));
394 if (obj->oo_inode != NULL)
395 osd_object_init0(obj);
397 LASSERT(osd_invariant(obj));
402 * Concurrency: no concurrent access is possible that late in object
405 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
407 struct osd_object *obj = osd_obj(l);
409 LASSERT(osd_invariant(obj));
411 dt_object_fini(&obj->oo_dt);
415 static struct iam_path_descr *osd_ipd_get(const struct lu_env *env,
416 const struct iam_container *bag)
418 return bag->ic_descr->id_ops->id_ipd_alloc(bag,
419 osd_oti_get(env)->oti_ipd);
422 static void osd_ipd_put(const struct lu_env *env,
423 const struct iam_container *bag,
424 struct iam_path_descr *ipd)
426 bag->ic_descr->id_ops->id_ipd_free(ipd);
430 * Concurrency: no concurrent access is possible that late in object
433 static void osd_index_fini(struct osd_object *o)
435 struct iam_container *bag;
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);
443 OBD_FREE_PTR(o->oo_dir);
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.)
453 static int osd_inode_unlinked(const struct inode *inode)
455 return inode->i_nlink == 0;
459 OSD_TXN_OI_DELETE_CREDITS = 20,
460 OSD_TXN_INODE_DELETE_CREDITS = 20
464 * Concurrency: no concurrent access is possible that late in object
467 static int osd_inode_remove(const struct lu_env *env, struct osd_object *obj)
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;
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);
480 result = osd_oi_delete(oti, &osd->od_oi, fid, th);
481 osd_trans_stop(env, th);
483 result = PTR_ERR(th);
488 * Called just before object is freed. Releases all resources except for
489 * object itself (that is released by osd_object_free()).
491 * Concurrency: no concurrent access is possible that late in object
494 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
496 struct osd_object *obj = osd_obj(l);
497 struct inode *inode = obj->oo_inode;
499 LASSERT(osd_invariant(obj));
502 * If object is unlinked remove fid->ino mapping from object index.
504 * File body will be deleted by iput().
511 if (osd_inode_unlinked(inode)) {
512 result = osd_inode_remove(env, obj);
514 LU_OBJECT_DEBUG(D_ERROR, env, l,
515 "Failed to cleanup: %d\n",
519 obj->oo_inode = NULL;
524 * Concurrency: ->loo_object_release() is called under site spin-lock.
526 static void osd_object_release(const struct lu_env *env,
529 struct osd_object *o = osd_obj(l);
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);
537 * Concurrency: shouldn't matter.
539 static int osd_object_print(const struct lu_env *env, void *cookie,
540 lu_printer_t p, const struct lu_object *l)
542 struct osd_object *o = osd_obj(l);
545 if (o->oo_dir != NULL)
546 d = o->oo_dir->od_container.ic_descr;
549 return (*p)(env, cookie, LUSTRE_OSD_NAME"-object@%p(i:%p:%lu/%u)[%s]",
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");
557 * Concurrency: shouldn't matter.
559 static int osd_statfs(const struct lu_env *env,
560 struct dt_device *d, struct kstatfs *sfs)
562 struct osd_device *osd = osd_dt_dev(d);
563 struct super_block *sb = osd_sb(osd);
566 spin_lock(&osd->od_osfs_lock);
568 if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
569 result = ll_do_statfs(sb, &osd->od_kstatfs);
570 if (likely(result == 0)) /* N.B. statfs can't really fail */
571 osd->od_osfs_age = cfs_time_current_64();
574 if (likely(result == 0))
575 *sfs = osd->od_kstatfs;
576 spin_unlock(&osd->od_osfs_lock);
582 * Concurrency: doesn't access mutable data.
584 static void osd_conf_get(const struct lu_env *env,
585 const struct dt_device *dev,
586 struct dt_device_param *param)
589 * XXX should be taken from not-yet-existing fs abstraction layer.
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;
601 * Concurrency: doesn't access mutable data.
603 static int osd_param_is_sane(const struct osd_device *dev,
604 const struct txn_param *param)
606 return param->tp_credits <= osd_journal(dev)->j_max_transaction_buffers;
610 * Concurrency: shouldn't matter.
612 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
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;
618 LASSERT(dev != NULL);
619 LASSERT(oh->ot_handle == NULL);
622 CERROR("transaction @0x%p commit error: %d\n", th, error);
625 * This od_env_for_commit is only for commit usage. see
628 dt_txn_hook_commit(&osd_dt_dev(dev)->od_env_for_commit, th);
631 lu_device_put(&dev->dd_lu_dev);
634 lu_context_exit(&th->th_ctx);
635 lu_context_fini(&th->th_ctx);
640 * Concurrency: shouldn't matter.
642 static struct thandle *osd_trans_start(const struct lu_env *env,
646 struct osd_device *dev = osd_dt_dev(d);
648 struct osd_thandle *oh;
654 hook_res = dt_txn_hook_start(env, d, p);
656 RETURN(ERR_PTR(hook_res));
658 if (osd_param_is_sane(dev, p)) {
659 OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
662 * XXX temporary stuff. Some abstraction layer should
666 jh = journal_start(osd_journal(dev), p->tp_credits);
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);
681 struct osd_thread_info *oti =
684 LASSERT(oti->oti_txns == 0);
685 LASSERT(oti->oti_r_locks == 0);
686 LASSERT(oti->oti_w_locks == 0);
695 th = ERR_PTR(-ENOMEM);
697 CERROR("Invalid transaction parameters\n");
698 th = ERR_PTR(-EINVAL);
705 * Concurrency: shouldn't matter.
707 static void osd_trans_stop(const struct lu_env *env, struct thandle *th)
710 struct osd_thandle *oh;
714 oh = container_of0(th, struct osd_thandle, ot_super);
715 if (oh->ot_handle != NULL) {
716 handle_t *hdl = oh->ot_handle;
718 * XXX temporary stuff. Some abstraction layer should be used.
720 result = dt_txn_hook_stop(env, th);
722 CERROR("Failure in transaction hook: %d\n", result);
725 oh->ot_handle = NULL;
726 result = journal_stop(hdl);
728 CERROR("Failure to stop transaction: %d\n", result);
732 struct osd_thread_info *oti = osd_oti_get(env);
734 LASSERT(oti->oti_txns == 1);
735 LASSERT(oti->oti_r_locks == 0);
736 LASSERT(oti->oti_w_locks == 0);
745 * Concurrency: shouldn't matter.
747 static int osd_sync(const struct lu_env *env, struct dt_device *d)
749 CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_NAME);
750 return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
754 * Concurrency: shouldn't matter.
756 lvfs_sbdev_type fsfilt_ldiskfs_journal_sbdev(struct super_block *);
758 static void osd_ro(const struct lu_env *env, struct dt_device *d)
762 CERROR("*** setting device %s read-only ***\n", LUSTRE_OSD_NAME);
764 __lvfs_set_rdonly(lvfs_sbdev(osd_sb(osd_dt_dev(d))),
765 fsfilt_ldiskfs_journal_sbdev(osd_sb(osd_dt_dev(d))));
770 * Concurrency: serialization provided by callers.
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)
776 struct osd_device *dev = osd_dt_dev(d);
779 dev->od_fl_capa = mode;
780 dev->od_capa_timeout = timeout;
781 dev->od_capa_alg = alg;
782 dev->od_capa_keys = keys;
786 /* Note: we did not count into QUOTA here, If we mount with --data_journal
788 static const int osd_dto_credits[DTO_NR] = {
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
794 [DTO_INDEX_INSERT] = 16,
795 [DTO_INDEX_DELETE] = 16,
796 [DTO_IDNEX_UPDATE] = 16,
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)
802 [DTO_OBJECT_CREATE] = 23,
803 [DTO_OBJECT_DELETE] = 23,
805 * Attr set credits 3 inode, group, GDT
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?
813 [DTO_XATTR_SET] = 16,
815 /* creadits for inode change during write */
816 [DTO_WRITE_BASE] = 3,
817 /* credits for single block write */
818 [DTO_WRITE_BLOCK] = 12
821 static int osd_credit_get(const struct lu_env *env, struct dt_device *d,
824 LASSERT(0 <= op && op < ARRAY_SIZE(osd_dto_credits));
825 return osd_dto_credits[op];
828 static struct dt_device_operations osd_dt_ops = {
829 .dt_root_get = osd_root_get,
830 .dt_statfs = osd_statfs,
831 .dt_trans_start = osd_trans_start,
832 .dt_trans_stop = osd_trans_stop,
833 .dt_conf_get = osd_conf_get,
836 .dt_credit_get = osd_credit_get,
837 .dt_init_capa_ctxt = osd_init_capa_ctxt,
840 static void osd_object_read_lock(const struct lu_env *env,
841 struct dt_object *dt)
843 struct osd_object *obj = osd_dt_obj(dt);
845 LASSERT(osd_invariant(obj));
847 OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
848 down_read(&obj->oo_sem);
851 struct osd_thread_info *oti = osd_oti_get(env);
853 LASSERT(obj->oo_owner == NULL);
859 static void osd_object_write_lock(const struct lu_env *env,
860 struct dt_object *dt)
862 struct osd_object *obj = osd_dt_obj(dt);
864 LASSERT(osd_invariant(obj));
866 OSD_COUNTERS_DO(LASSERT(obj->oo_owner != env));
867 down_write(&obj->oo_sem);
870 struct osd_thread_info *oti = osd_oti_get(env);
872 LASSERT(obj->oo_owner == NULL);
879 static void osd_object_read_unlock(const struct lu_env *env,
880 struct dt_object *dt)
882 struct osd_object *obj = osd_dt_obj(dt);
884 LASSERT(osd_invariant(obj));
887 struct osd_thread_info *oti = osd_oti_get(env);
889 LASSERT(oti->oti_r_locks > 0);
893 up_read(&obj->oo_sem);
896 static void osd_object_write_unlock(const struct lu_env *env,
897 struct dt_object *dt)
899 struct osd_object *obj = osd_dt_obj(dt);
901 LASSERT(osd_invariant(obj));
904 struct osd_thread_info *oti = osd_oti_get(env);
906 LASSERT(obj->oo_owner == env);
907 LASSERT(oti->oti_w_locks > 0);
909 obj->oo_owner = NULL;
912 up_write(&obj->oo_sem);
915 static int capa_is_sane(const struct lu_env *env,
916 struct osd_device *dev,
917 struct lustre_capa *capa,
918 struct lustre_capa_key *keys)
920 struct osd_thread_info *oti = osd_oti_get(env);
925 oc = capa_lookup(dev->od_capa_hash, capa, 0);
927 if (capa_is_expired(oc)) {
928 DEBUG_CAPA(D_ERROR, capa, "expired");
935 spin_lock(&capa_lock);
936 for (i = 0; i < 2; i++) {
937 if (keys[i].lk_keyid == capa->lc_keyid) {
938 oti->oti_capa_key = keys[i];
942 spin_unlock(&capa_lock);
945 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
949 rc = capa_hmac(oti->oti_capa.lc_hmac, capa, oti->oti_capa_key.lk_key);
952 if (memcmp(oti->oti_capa.lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac)))
954 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
958 oc = capa_add(dev->od_capa_hash, capa);
964 static int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
965 struct lustre_capa *capa, __u64 opc)
967 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
968 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
971 if (!dev->od_fl_capa)
974 if (capa == BYPASS_CAPA)
978 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
982 if (!lu_fid_eq(fid, &capa->lc_fid)) {
983 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
988 if (!capa_opc_supported(capa, opc)) {
989 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
993 if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
994 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1001 static int osd_attr_get(const struct lu_env *env,
1002 struct dt_object *dt,
1003 struct lu_attr *attr,
1004 struct lustre_capa *capa)
1006 struct osd_object *obj = osd_dt_obj(dt);
1008 LASSERT(dt_object_exists(dt));
1009 LASSERT(osd_invariant(obj));
1011 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1014 spin_lock(&obj->oo_guard);
1015 osd_inode_getattr(env, obj->oo_inode, attr);
1016 spin_unlock(&obj->oo_guard);
1020 static int osd_attr_set(const struct lu_env *env,
1021 struct dt_object *dt,
1022 const struct lu_attr *attr,
1023 struct thandle *handle,
1024 struct lustre_capa *capa)
1026 struct osd_object *obj = osd_dt_obj(dt);
1028 LASSERT(handle != NULL);
1029 LASSERT(dt_object_exists(dt));
1030 LASSERT(osd_invariant(obj));
1032 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1035 spin_lock(&obj->oo_guard);
1036 osd_inode_setattr(env, obj->oo_inode, attr);
1037 spin_unlock(&obj->oo_guard);
1039 mark_inode_dirty(obj->oo_inode);
1043 static struct timespec *osd_inode_time(const struct lu_env *env,
1044 struct inode *inode, __u64 seconds)
1046 struct osd_thread_info *oti = osd_oti_get(env);
1047 struct timespec *t = &oti->oti_time;
1049 t->tv_sec = seconds;
1051 *t = timespec_trunc(*t, get_sb_time_gran(inode->i_sb));
1055 static void osd_inode_setattr(const struct lu_env *env,
1056 struct inode *inode, const struct lu_attr *attr)
1060 bits = attr->la_valid;
1062 LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1064 if (bits & LA_ATIME)
1065 inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
1066 if (bits & LA_CTIME)
1067 inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
1068 if (bits & LA_MTIME)
1069 inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
1070 if (bits & LA_SIZE) {
1071 LDISKFS_I(inode)->i_disksize = attr->la_size;
1072 i_size_write(inode, attr->la_size);
1074 if (bits & LA_BLOCKS)
1075 inode->i_blocks = attr->la_blocks;
1077 inode->i_mode = (inode->i_mode & S_IFMT) |
1078 (attr->la_mode & ~S_IFMT);
1080 inode->i_uid = attr->la_uid;
1082 inode->i_gid = attr->la_gid;
1083 if (bits & LA_NLINK)
1084 inode->i_nlink = attr->la_nlink;
1086 inode->i_rdev = attr->la_rdev;
1087 if (bits & LA_BLKSIZE)
1088 inode->i_blksize = attr->la_blksize;
1090 if (bits & LA_FLAGS) {
1091 struct ldiskfs_inode_info *li = LDISKFS_I(inode);
1093 li->i_flags = (li->i_flags & ~LDISKFS_FL_USER_MODIFIABLE) |
1094 (attr->la_flags & LDISKFS_FL_USER_MODIFIABLE);
1101 * XXX temporary solution.
1104 static int osd_create_pre(struct osd_thread_info *info, struct osd_object *obj,
1105 struct lu_attr *attr, struct thandle *th)
1110 static int osd_create_post(struct osd_thread_info *info, struct osd_object *obj,
1111 struct lu_attr *attr, struct thandle *th)
1113 LASSERT(obj->oo_inode != NULL);
1115 osd_object_init0(obj);
1119 extern struct inode *ldiskfs_create_inode(handle_t *handle,
1120 struct inode * dir, int mode);
1122 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1124 struct dt_allocation_hint *hint,
1128 struct osd_device *osd = osd_obj2dev(obj);
1129 struct osd_thandle *oth;
1130 struct inode *parent;
1131 struct inode *inode;
1133 LASSERT(osd_invariant(obj));
1134 LASSERT(obj->oo_inode == NULL);
1135 LASSERT(osd->od_obj_area != NULL);
1137 oth = container_of(th, struct osd_thandle, ot_super);
1138 LASSERT(oth->ot_handle->h_transaction != NULL);
1140 if (hint && hint->dah_parent)
1141 parent = osd_dt_obj(hint->dah_parent)->oo_inode;
1143 parent = osd->od_obj_area->d_inode;
1144 LASSERT(parent->i_op != NULL);
1146 inode = ldiskfs_create_inode(oth->ot_handle, parent, mode);
1147 if (!IS_ERR(inode)) {
1148 obj->oo_inode = inode;
1151 result = PTR_ERR(inode);
1152 LASSERT(osd_invariant(obj));
1157 extern int iam_lvar_create(struct inode *obj, int keysize, int ptrsize,
1158 int recsize, handle_t *handle);
1164 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1165 struct lu_attr *attr,
1166 struct dt_allocation_hint *hint,
1170 struct osd_thandle *oth;
1172 LASSERT(S_ISDIR(attr->la_mode));
1174 oth = container_of(th, struct osd_thandle, ot_super);
1175 LASSERT(oth->ot_handle->h_transaction != NULL);
1176 result = osd_mkfile(info, obj, (attr->la_mode &
1177 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1179 LASSERT(obj->oo_inode != NULL);
1181 * XXX uh-oh... call low-level iam function directly.
1183 result = iam_lvar_create(obj->oo_inode, OSD_NAME_LEN, 4,
1184 sizeof (struct lu_fid_pack),
1190 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1191 struct lu_attr *attr,
1192 struct dt_allocation_hint *hint,
1195 LASSERT(S_ISREG(attr->la_mode));
1196 return osd_mkfile(info, obj, (attr->la_mode &
1197 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1200 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1201 struct lu_attr *attr,
1202 struct dt_allocation_hint *hint,
1205 LASSERT(S_ISLNK(attr->la_mode));
1206 return osd_mkfile(info, obj, (attr->la_mode &
1207 (S_IFMT | S_IRWXUGO | S_ISVTX)), hint, th);
1210 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1211 struct lu_attr *attr,
1212 struct dt_allocation_hint *hint,
1216 struct osd_device *osd = osd_obj2dev(obj);
1218 umode_t mode = attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX);
1220 LASSERT(osd_invariant(obj));
1221 LASSERT(obj->oo_inode == NULL);
1222 LASSERT(osd->od_obj_area != NULL);
1223 LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1224 S_ISFIFO(mode) || S_ISSOCK(mode));
1226 dir = osd->od_obj_area->d_inode;
1227 LASSERT(dir->i_op != NULL);
1229 result = osd_mkfile(info, obj, mode, hint, th);
1231 LASSERT(obj->oo_inode != NULL);
1232 init_special_inode(obj->oo_inode, mode, attr->la_rdev);
1234 LASSERT(osd_invariant(obj));
1238 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1240 struct dt_allocation_hint *hint,
1243 static osd_obj_type_f osd_create_type_f(__u32 mode)
1245 osd_obj_type_f result;
1271 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1272 struct dt_object *parent, umode_t child_mode)
1276 memset(ah, 0, sizeof(*ah));
1277 ah->dah_parent = parent;
1278 ah->dah_mode = child_mode;
1283 * Concurrency: @dt is write locked.
1285 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
1286 struct lu_attr *attr,
1287 struct dt_allocation_hint *hint,
1290 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1291 struct osd_object *obj = osd_dt_obj(dt);
1292 struct osd_device *osd = osd_obj2dev(obj);
1293 struct osd_thread_info *info = osd_oti_get(env);
1298 LASSERT(osd_invariant(obj));
1299 LASSERT(!dt_object_exists(dt));
1300 LASSERT(osd_write_locked(env, obj));
1301 LASSERT(th != NULL);
1304 * XXX missing: Quote handling.
1307 result = osd_create_pre(info, obj, attr, th);
1309 result = osd_create_type_f(attr->la_mode & S_IFMT)(info, obj,
1312 result = osd_create_post(info, obj, attr, th);
1315 struct osd_inode_id *id = &info->oti_id;
1317 LASSERT(obj->oo_inode != NULL);
1319 id->oii_ino = obj->oo_inode->i_ino;
1320 id->oii_gen = obj->oo_inode->i_generation;
1322 result = osd_oi_insert(info, &osd->od_oi, fid, id, th);
1325 LASSERT(ergo(result == 0, dt_object_exists(dt)));
1326 LASSERT(osd_invariant(obj));
1331 * Concurrency: @dt is write locked.
1333 static void osd_object_ref_add(const struct lu_env *env,
1334 struct dt_object *dt,
1337 struct osd_object *obj = osd_dt_obj(dt);
1338 struct inode *inode = obj->oo_inode;
1340 LASSERT(osd_invariant(obj));
1341 LASSERT(dt_object_exists(dt));
1342 LASSERT(osd_write_locked(env, obj));
1343 LASSERT(th != NULL);
1345 spin_lock(&obj->oo_guard);
1346 if (inode->i_nlink < LDISKFS_LINK_MAX) {
1348 spin_unlock(&obj->oo_guard);
1349 mark_inode_dirty(inode);
1351 spin_unlock(&obj->oo_guard);
1352 LU_OBJECT_DEBUG(D_ERROR, env, &dt->do_lu,
1353 "Overflowed nlink\n");
1355 LASSERT(osd_invariant(obj));
1359 * Concurrency: @dt is write locked.
1361 static void osd_object_ref_del(const struct lu_env *env,
1362 struct dt_object *dt,
1365 struct osd_object *obj = osd_dt_obj(dt);
1366 struct inode *inode = obj->oo_inode;
1368 LASSERT(osd_invariant(obj));
1369 LASSERT(dt_object_exists(dt));
1370 LASSERT(osd_write_locked(env, obj));
1371 LASSERT(th != NULL);
1373 spin_lock(&obj->oo_guard);
1374 if (inode->i_nlink > 0) {
1376 spin_unlock(&obj->oo_guard);
1377 mark_inode_dirty(inode);
1379 spin_unlock(&obj->oo_guard);
1380 LU_OBJECT_DEBUG(D_ERROR, env, &dt->do_lu,
1381 "Underflowed nlink\n");
1383 LASSERT(osd_invariant(obj));
1387 * Concurrency: @dt is read locked.
1389 static int osd_xattr_get(const struct lu_env *env,
1390 struct dt_object *dt,
1393 struct lustre_capa *capa)
1395 struct osd_object *obj = osd_dt_obj(dt);
1396 struct inode *inode = obj->oo_inode;
1397 struct osd_thread_info *info = osd_oti_get(env);
1398 struct dentry *dentry = &info->oti_dentry;
1400 LASSERT(dt_object_exists(dt));
1401 LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
1402 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1404 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1407 dentry->d_inode = inode;
1408 return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
1412 * Concurrency: @dt is write locked.
1414 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
1415 const struct lu_buf *buf, const char *name, int fl,
1416 struct thandle *handle, struct lustre_capa *capa)
1420 struct osd_object *obj = osd_dt_obj(dt);
1421 struct inode *inode = obj->oo_inode;
1422 struct osd_thread_info *info = osd_oti_get(env);
1423 struct dentry *dentry = &info->oti_dentry;
1425 LASSERT(dt_object_exists(dt));
1426 LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
1427 LASSERT(osd_write_locked(env, obj));
1428 LASSERT(handle != NULL);
1430 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1433 dentry->d_inode = inode;
1436 if (fl & LU_XATTR_REPLACE)
1437 fs_flags |= XATTR_REPLACE;
1439 if (fl & LU_XATTR_CREATE)
1440 fs_flags |= XATTR_CREATE;
1442 return inode->i_op->setxattr(dentry, name,
1443 buf->lb_buf, buf->lb_len, fs_flags);
1447 * Concurrency: @dt is read locked.
1449 static int osd_xattr_list(const struct lu_env *env,
1450 struct dt_object *dt,
1452 struct lustre_capa *capa)
1454 struct osd_object *obj = osd_dt_obj(dt);
1455 struct inode *inode = obj->oo_inode;
1456 struct osd_thread_info *info = osd_oti_get(env);
1457 struct dentry *dentry = &info->oti_dentry;
1459 LASSERT(dt_object_exists(dt));
1460 LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
1461 LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
1463 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1466 dentry->d_inode = inode;
1467 return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
1471 * Concurrency: @dt is write locked.
1473 static int osd_xattr_del(const struct lu_env *env,
1474 struct dt_object *dt,
1476 struct thandle *handle,
1477 struct lustre_capa *capa)
1479 struct osd_object *obj = osd_dt_obj(dt);
1480 struct inode *inode = obj->oo_inode;
1481 struct osd_thread_info *info = osd_oti_get(env);
1482 struct dentry *dentry = &info->oti_dentry;
1484 LASSERT(dt_object_exists(dt));
1485 LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
1486 LASSERT(osd_write_locked(env, obj));
1487 LASSERT(handle != NULL);
1489 if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1492 dentry->d_inode = inode;
1493 return inode->i_op->removexattr(dentry, name);
1496 static struct obd_capa *osd_capa_get(const struct lu_env *env,
1497 struct dt_object *dt,
1498 struct lustre_capa *old,
1501 struct osd_thread_info *info = osd_oti_get(env);
1502 const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1503 struct osd_object *obj = osd_dt_obj(dt);
1504 struct osd_device *dev = osd_obj2dev(obj);
1505 struct lustre_capa_key *key = &info->oti_capa_key;
1506 struct lustre_capa *capa = &info->oti_capa;
1507 struct obd_capa *oc;
1511 if (!dev->od_fl_capa)
1512 RETURN(ERR_PTR(-ENOENT));
1514 LASSERT(dt_object_exists(dt));
1515 LASSERT(osd_invariant(obj));
1517 /* renewal sanity check */
1518 if (old && osd_object_auth(env, dt, old, opc))
1519 RETURN(ERR_PTR(-EACCES));
1521 capa->lc_fid = *fid;
1524 capa->lc_flags = dev->od_capa_alg << 24;
1525 capa->lc_timeout = dev->od_capa_timeout;
1526 capa->lc_expiry = 0;
1528 oc = capa_lookup(dev->od_capa_hash, capa, 1);
1530 LASSERT(!capa_is_expired(oc));
1534 spin_lock(&capa_lock);
1535 *key = dev->od_capa_keys[1];
1536 spin_unlock(&capa_lock);
1538 capa->lc_keyid = key->lk_keyid;
1539 capa->lc_expiry = CURRENT_SECONDS + dev->od_capa_timeout;
1541 rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
1543 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
1544 RETURN(ERR_PTR(rc));
1547 oc = capa_add(dev->od_capa_hash, capa);
1551 static struct dt_object_operations osd_obj_ops = {
1552 .do_read_lock = osd_object_read_lock,
1553 .do_write_lock = osd_object_write_lock,
1554 .do_read_unlock = osd_object_read_unlock,
1555 .do_write_unlock = osd_object_write_unlock,
1556 .do_attr_get = osd_attr_get,
1557 .do_attr_set = osd_attr_set,
1558 .do_ah_init = osd_ah_init,
1559 .do_create = osd_object_create,
1560 .do_index_try = osd_index_try,
1561 .do_ref_add = osd_object_ref_add,
1562 .do_ref_del = osd_object_ref_del,
1563 .do_xattr_get = osd_xattr_get,
1564 .do_xattr_set = osd_xattr_set,
1565 .do_xattr_del = osd_xattr_del,
1566 .do_xattr_list = osd_xattr_list,
1567 .do_capa_get = osd_capa_get,
1575 * XXX: Another layering violation for now.
1577 * We don't want to use ->f_op->read methods, because generic file write
1579 * - serializes on ->i_sem, and
1581 * - does a lot of extra work like balance_dirty_pages(),
1583 * which doesn't work for globally shared files like /last-received.
1585 int fsfilt_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs);
1586 int fsfilt_ldiskfs_write_handle(struct inode *inode, void *buf, int bufsize,
1587 loff_t *offs, handle_t *handle);
1589 static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt,
1590 struct lu_buf *buf, loff_t *pos,
1591 struct lustre_capa *capa)
1593 struct inode *inode = osd_dt_obj(dt)->oo_inode;
1595 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ))
1598 return fsfilt_ldiskfs_read(inode, buf->lb_buf, buf->lb_len, pos);
1601 static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
1602 const struct lu_buf *buf, loff_t *pos,
1603 struct thandle *handle, struct lustre_capa *capa)
1605 struct inode *inode = osd_dt_obj(dt)->oo_inode;
1606 struct osd_thandle *oh;
1609 LASSERT(handle != NULL);
1611 if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE))
1614 oh = container_of(handle, struct osd_thandle, ot_super);
1615 LASSERT(oh->ot_handle->h_transaction != NULL);
1616 result = fsfilt_ldiskfs_write_handle(inode, buf->lb_buf, buf->lb_len,
1617 pos, oh->ot_handle);
1619 result = buf->lb_len;
1623 static struct dt_body_operations osd_body_ops = {
1624 .dbo_read = osd_read,
1625 .dbo_write = osd_write
1632 static int osd_object_is_root(const struct osd_object *obj)
1634 return osd_sb(osd_obj2dev(obj))->s_root->d_inode == obj->oo_inode;
1637 static int osd_index_probe(const struct lu_env *env, struct osd_object *o,
1638 const struct dt_index_features *feat)
1640 struct iam_descr *descr;
1642 if (osd_object_is_root(o))
1643 return feat == &dt_directory_features;
1645 LASSERT(o->oo_dir != NULL);
1647 descr = o->oo_dir->od_container.ic_descr;
1648 if (feat == &dt_directory_features)
1649 return descr == &iam_htree_compat_param ||
1650 (descr->id_rec_size == sizeof(struct lu_fid_pack) &&
1652 * XXX check that index looks like directory.
1657 feat->dif_keysize_min <= descr->id_key_size &&
1658 descr->id_key_size <= feat->dif_keysize_max &&
1659 feat->dif_recsize_min <= descr->id_rec_size &&
1660 descr->id_rec_size <= feat->dif_recsize_max &&
1661 !(feat->dif_flags & (DT_IND_VARKEY |
1662 DT_IND_VARREC | DT_IND_NONUNQ)) &&
1663 ergo(feat->dif_flags & DT_IND_UPDATE,
1664 1 /* XXX check that object (and file system) is
1668 static int osd_container_init(const struct lu_env *env,
1669 struct osd_object *obj,
1670 struct osd_directory *dir)
1673 struct iam_container *bag;
1675 bag = &dir->od_container;
1676 result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
1678 result = iam_container_setup(bag);
1680 obj->oo_dt.do_index_ops = &osd_index_ops;
1682 iam_container_fini(bag);
1688 * Concurrency: no external locking is necessary.
1690 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
1691 const struct dt_index_features *feat)
1694 struct osd_object *obj = osd_dt_obj(dt);
1696 LASSERT(osd_invariant(obj));
1697 LASSERT(dt_object_exists(dt));
1699 if (osd_object_is_root(obj)) {
1700 dt->do_index_ops = &osd_index_compat_ops;
1702 } else if (!osd_has_index(obj)) {
1703 struct osd_directory *dir;
1707 sema_init(&dir->od_sem, 1);
1709 spin_lock(&obj->oo_guard);
1710 if (obj->oo_dir == NULL)
1714 * Concurrent thread allocated container data.
1717 spin_unlock(&obj->oo_guard);
1719 * Now, that we have container data, serialize its
1722 down(&obj->oo_dir->od_sem);
1724 * recheck under lock.
1726 if (!osd_has_index(obj))
1727 result = osd_container_init(env, obj, dir);
1730 up(&obj->oo_dir->od_sem);
1737 if (!osd_index_probe(env, obj, feat))
1740 LASSERT(osd_invariant(obj));
1745 static int osd_index_delete(const struct lu_env *env, struct dt_object *dt,
1746 const struct dt_key *key, struct thandle *handle,
1747 struct lustre_capa *capa)
1749 struct osd_object *obj = osd_dt_obj(dt);
1750 struct osd_thandle *oh;
1751 struct iam_path_descr *ipd;
1752 struct iam_container *bag = &obj->oo_dir->od_container;
1757 LASSERT(osd_invariant(obj));
1758 LASSERT(dt_object_exists(dt));
1759 LASSERT(bag->ic_object == obj->oo_inode);
1760 LASSERT(handle != NULL);
1762 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
1765 ipd = osd_ipd_get(env, bag);
1766 if (unlikely(ipd == NULL))
1769 oh = container_of0(handle, struct osd_thandle, ot_super);
1770 LASSERT(oh->ot_handle != NULL);
1771 LASSERT(oh->ot_handle->h_transaction != NULL);
1773 rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
1774 osd_ipd_put(env, bag, ipd);
1775 LASSERT(osd_invariant(obj));
1779 static int osd_index_lookup(const struct lu_env *env, struct dt_object *dt,
1780 struct dt_rec *rec, const struct dt_key *key,
1781 struct lustre_capa *capa)
1783 struct osd_object *obj = osd_dt_obj(dt);
1784 struct iam_path_descr *ipd;
1785 struct iam_container *bag = &obj->oo_dir->od_container;
1790 LASSERT(osd_invariant(obj));
1791 LASSERT(dt_object_exists(dt));
1792 LASSERT(bag->ic_object == obj->oo_inode);
1794 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
1797 ipd = osd_ipd_get(env, bag);
1798 if (unlikely(ipd == NULL))
1801 rc = iam_lookup(bag, (const struct iam_key *)key,
1802 (struct iam_rec *)rec, ipd);
1803 osd_ipd_put(env, bag, ipd);
1804 LASSERT(osd_invariant(obj));
1809 static int osd_index_insert(const struct lu_env *env, struct dt_object *dt,
1810 const struct dt_rec *rec, const struct dt_key *key,
1811 struct thandle *th, struct lustre_capa *capa)
1813 struct osd_object *obj = osd_dt_obj(dt);
1814 struct iam_path_descr *ipd;
1815 struct osd_thandle *oh;
1816 struct iam_container *bag = &obj->oo_dir->od_container;
1821 LASSERT(osd_invariant(obj));
1822 LASSERT(dt_object_exists(dt));
1823 LASSERT(bag->ic_object == obj->oo_inode);
1824 LASSERT(th != NULL);
1826 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
1829 ipd = osd_ipd_get(env, bag);
1830 if (unlikely(ipd == NULL))
1833 oh = container_of0(th, struct osd_thandle, ot_super);
1834 LASSERT(oh->ot_handle != NULL);
1835 LASSERT(oh->ot_handle->h_transaction != NULL);
1836 rc = iam_insert(oh->ot_handle, bag, (const struct iam_key *)key,
1837 (struct iam_rec *)rec, ipd);
1838 osd_ipd_put(env, bag, ipd);
1839 LASSERT(osd_invariant(obj));
1844 * Iterator operations.
1847 struct osd_object *oi_obj;
1848 struct iam_path_descr *oi_ipd;
1849 struct iam_iterator oi_it;
1852 static struct dt_it *osd_it_init(const struct lu_env *env,
1853 struct dt_object *dt, int writable,
1854 struct lustre_capa *capa)
1857 struct osd_object *obj = osd_dt_obj(dt);
1858 struct lu_object *lo = &dt->do_lu;
1859 struct iam_path_descr *ipd;
1860 struct iam_container *bag = &obj->oo_dir->od_container;
1863 LASSERT(lu_object_exists(lo));
1865 if (osd_object_auth(env, dt, capa, writable ? CAPA_OPC_BODY_WRITE :
1866 CAPA_OPC_BODY_READ))
1867 return ERR_PTR(-EACCES);
1869 flags = writable ? IAM_IT_MOVE|IAM_IT_WRITE : IAM_IT_MOVE;
1873 * XXX: as ipd is allocated within osd_thread_info, assignment
1874 * below implies that iterator usage is confined within single
1877 ipd = osd_ipd_get(env, bag);
1878 if (likely(ipd != NULL)) {
1882 iam_it_init(&it->oi_it, bag, flags, ipd);
1883 return (struct dt_it *)it;
1887 return ERR_PTR(-ENOMEM);
1890 static void osd_it_fini(const struct lu_env *env, struct dt_it *di)
1892 struct osd_it *it = (struct osd_it *)di;
1893 struct osd_object *obj = it->oi_obj;
1895 iam_it_fini(&it->oi_it);
1896 osd_ipd_put(env, &obj->oo_dir->od_container, it->oi_ipd);
1897 lu_object_put(env, &obj->oo_dt.do_lu);
1901 static int osd_it_get(const struct lu_env *env,
1902 struct dt_it *di, const struct dt_key *key)
1904 struct osd_it *it = (struct osd_it *)di;
1906 return iam_it_get(&it->oi_it, (const struct iam_key *)key);
1909 static void osd_it_put(const struct lu_env *env, struct dt_it *di)
1911 struct osd_it *it = (struct osd_it *)di;
1913 iam_it_put(&it->oi_it);
1916 static int osd_it_next(const struct lu_env *env, struct dt_it *di)
1918 struct osd_it *it = (struct osd_it *)di;
1920 return iam_it_next(&it->oi_it);
1923 static int osd_it_del(const struct lu_env *env, struct dt_it *di,
1926 struct osd_it *it = (struct osd_it *)di;
1927 struct osd_thandle *oh;
1929 LASSERT(th != NULL);
1931 oh = container_of0(th, struct osd_thandle, ot_super);
1932 LASSERT(oh->ot_handle != NULL);
1933 LASSERT(oh->ot_handle->h_transaction != NULL);
1935 return iam_it_rec_delete(oh->ot_handle, &it->oi_it);
1938 static struct dt_key *osd_it_key(const struct lu_env *env,
1939 const struct dt_it *di)
1941 struct osd_it *it = (struct osd_it *)di;
1943 return (struct dt_key *)iam_it_key_get(&it->oi_it);
1946 static int osd_it_key_size(const struct lu_env *env, const struct dt_it *di)
1948 struct osd_it *it = (struct osd_it *)di;
1950 return iam_it_key_size(&it->oi_it);
1953 static struct dt_rec *osd_it_rec(const struct lu_env *env,
1954 const struct dt_it *di)
1956 struct osd_it *it = (struct osd_it *)di;
1958 return (struct dt_rec *)iam_it_rec_get(&it->oi_it);
1961 static __u32 osd_it_store(const struct lu_env *env, const struct dt_it *di)
1963 struct osd_it *it = (struct osd_it *)di;
1965 return iam_it_store(&it->oi_it);
1968 static int osd_it_load(const struct lu_env *env,
1969 const struct dt_it *di, __u32 hash)
1971 struct osd_it *it = (struct osd_it *)di;
1973 return iam_it_load(&it->oi_it, hash);
1976 static struct dt_index_operations osd_index_ops = {
1977 .dio_lookup = osd_index_lookup,
1978 .dio_insert = osd_index_insert,
1979 .dio_delete = osd_index_delete,
1981 .init = osd_it_init,
1982 .fini = osd_it_fini,
1986 .next = osd_it_next,
1988 .key_size = osd_it_key_size,
1990 .store = osd_it_store,
1995 static int osd_index_compat_delete(const struct lu_env *env,
1996 struct dt_object *dt,
1997 const struct dt_key *key,
1998 struct thandle *handle,
1999 struct lustre_capa *capa)
2001 struct osd_object *obj = osd_dt_obj(dt);
2003 LASSERT(handle != NULL);
2004 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2008 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2012 RETURN(-EOPNOTSUPP);
2016 * Compatibility index operations.
2020 static void osd_build_pack(const struct lu_env *env, struct osd_device *osd,
2021 struct dentry *dentry, struct lu_fid_pack *pack)
2023 struct inode *inode = dentry->d_inode;
2024 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
2026 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2027 fid_cpu_to_be(fid, fid);
2028 pack->fp_len = sizeof *fid + 1;
2029 memcpy(pack->fp_area, fid, sizeof *fid);
2032 static int osd_index_compat_lookup(const struct lu_env *env,
2033 struct dt_object *dt,
2034 struct dt_rec *rec, const struct dt_key *key,
2035 struct lustre_capa *capa)
2037 struct osd_object *obj = osd_dt_obj(dt);
2039 struct osd_device *osd = osd_obj2dev(obj);
2040 struct osd_thread_info *info = osd_oti_get(env);
2046 * XXX temporary solution.
2048 struct dentry *dentry;
2049 struct dentry *parent;
2051 LASSERT(osd_invariant(obj));
2052 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2053 LASSERT(osd_has_index(obj));
2055 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
2058 info->oti_str.name = (const char *)key;
2059 info->oti_str.len = strlen((const char *)key);
2061 dir = obj->oo_inode;
2062 LASSERT(dir->i_op != NULL && dir->i_op->lookup != NULL);
2064 parent = d_alloc_root(dir);
2068 dentry = d_alloc(parent, &info->oti_str);
2069 if (dentry != NULL) {
2073 * XXX passing NULL for nameidata should work for
2076 d = dir->i_op->lookup(dir, dentry, NULL);
2079 * normal case, result is in @dentry.
2081 if (dentry->d_inode != NULL) {
2082 osd_build_pack(env, osd, dentry,
2083 (struct lu_fid_pack *)rec);
2088 /* What? Disconnected alias? Ppheeeww... */
2089 CERROR("Aliasing where not expected\n");
2097 LASSERT(osd_invariant(obj));
2101 static int osd_add_rec(struct osd_thread_info *info, struct osd_device *dev,
2102 struct inode *dir, struct inode *inode, const char *name)
2106 struct dentry *parent;
2110 info->oti_str.name = name;
2111 info->oti_str.len = strlen(name);
2113 LASSERT(atomic_read(&dir->i_count) > 0);
2115 old = d_alloc(dev->od_obj_area, &info->oti_str);
2117 d_instantiate(old, inode);
2119 LASSERT(atomic_read(&dir->i_count) > 0);
2120 parent = d_alloc_root(dir);
2121 if (parent != NULL) {
2123 LASSERT(atomic_read(&dir->i_count) > 1);
2124 new = d_alloc(parent, &info->oti_str);
2125 LASSERT(atomic_read(&dir->i_count) > 1);
2127 LASSERT(atomic_read(&dir->i_count) > 1);
2128 result = dir->i_op->link(old, dir, new);
2129 LASSERT(atomic_read(&dir->i_count) > 1);
2131 LASSERT(atomic_read(&dir->i_count) > 1);
2133 LASSERT(atomic_read(&dir->i_count) > 1);
2135 LASSERT(atomic_read(&dir->i_count) > 0);
2139 LASSERT(atomic_read(&dir->i_count) > 0);
2145 * XXX Temporary stuff.
2147 static int osd_index_compat_insert(const struct lu_env *env,
2148 struct dt_object *dt,
2149 const struct dt_rec *rec,
2150 const struct dt_key *key, struct thandle *th,
2151 struct lustre_capa *capa)
2153 struct osd_object *obj = osd_dt_obj(dt);
2155 const char *name = (const char *)key;
2157 struct lu_device *ludev = dt->do_lu.lo_dev;
2158 struct lu_object *luch;
2160 struct osd_thread_info *info = osd_oti_get(env);
2161 const struct lu_fid_pack *pack = (const struct lu_fid_pack *)rec;
2162 struct lu_fid *fid = &osd_oti_get(env)->oti_fid;
2166 LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2167 LASSERT(osd_invariant(obj));
2168 LASSERT(th != NULL);
2170 if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_INSERT))
2173 result = fid_unpack(pack, fid);
2177 luch = lu_object_find(env, ludev->ld_site, fid);
2178 if (!IS_ERR(luch)) {
2179 if (lu_object_exists(luch)) {
2180 struct osd_object *child;
2182 child = osd_obj(lu_object_locate(luch->lo_header,
2185 result = osd_add_rec(info, osd_obj2dev(obj),
2187 child->oo_inode, name);
2189 CERROR("No osd slice.\n");
2192 LASSERT(osd_invariant(obj));
2193 LASSERT(osd_invariant(child));
2198 lu_object_put(env, luch);
2200 result = PTR_ERR(luch);
2201 LASSERT(osd_invariant(obj));
2205 static struct dt_index_operations osd_index_compat_ops = {
2206 .dio_lookup = osd_index_compat_lookup,
2207 .dio_insert = osd_index_compat_insert,
2208 .dio_delete = osd_index_compat_delete
2211 /* type constructor/destructor: osd_type_init, osd_type_fini */
2212 LU_TYPE_INIT_FINI(osd, &osd_key);
2214 static struct lu_context_key osd_key = {
2215 .lct_tags = LCT_DT_THREAD | LCT_MD_THREAD,
2216 .lct_init = osd_key_init,
2217 .lct_fini = osd_key_fini,
2218 .lct_exit = osd_key_exit
2221 static void *osd_key_init(const struct lu_context *ctx,
2222 struct lu_context_key *key)
2224 struct osd_thread_info *info;
2226 OBD_ALLOC_PTR(info);
2228 info->oti_env = container_of(ctx, struct lu_env, le_ctx);
2230 info = ERR_PTR(-ENOMEM);
2234 /* context key destructor: osd_key_fini */
2235 LU_KEY_FINI(osd, struct osd_thread_info);
2237 static void osd_key_exit(const struct lu_context *ctx,
2238 struct lu_context_key *key, void *data)
2241 struct osd_thread_info *info = data;
2243 LASSERT(info->oti_r_locks == 0);
2244 LASSERT(info->oti_w_locks == 0);
2245 LASSERT(info->oti_txns == 0);
2249 static int osd_device_init(const struct lu_env *env, struct lu_device *d,
2250 const char *name, struct lu_device *next)
2252 return lu_env_init(&osd_dev(d)->od_env_for_commit, NULL, LCT_MD_THREAD);
2255 static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
2257 struct osd_thread_info *info = osd_oti_get(env);
2259 if (o->od_obj_area != NULL) {
2260 dput(o->od_obj_area);
2261 o->od_obj_area = NULL;
2263 osd_oi_fini(info, &o->od_oi);
2268 static int osd_mount(const struct lu_env *env,
2269 struct osd_device *o, struct lustre_cfg *cfg)
2271 struct lustre_mount_info *lmi;
2272 const char *dev = lustre_cfg_string(cfg, 0);
2273 struct osd_thread_info *info = osd_oti_get(env);
2278 if (o->od_mount != NULL) {
2279 CERROR("Already mounted (%s)\n", dev);
2284 lmi = server_get_mount(dev);
2286 CERROR("Cannot get mount info for %s!\n", dev);
2290 LASSERT(lmi != NULL);
2291 /* save lustre_mount_info in dt_device */
2294 result = osd_oi_init(info, &o->od_oi, &o->od_dt_dev);
2298 d = simple_mkdir(osd_sb(o)->s_root, "*OBJ-TEMP*", 0777, 1);
2302 result = PTR_ERR(d);
2305 osd_shutdown(env, o);
2309 static struct lu_device *osd_device_fini(const struct lu_env *env,
2310 struct lu_device *d)
2314 shrink_dcache_sb(osd_sb(osd_dev(d)));
2315 osd_sync(env, lu2dt_dev(d));
2317 if (osd_dev(d)->od_mount)
2318 server_put_mount(osd_dev(d)->od_mount->lmi_name,
2319 osd_dev(d)->od_mount->lmi_mnt);
2320 osd_dev(d)->od_mount = NULL;
2322 lu_env_fini(&osd_dev(d)->od_env_for_commit);
2326 static struct lu_device *osd_device_alloc(const struct lu_env *env,
2327 struct lu_device_type *t,
2328 struct lustre_cfg *cfg)
2330 struct lu_device *l;
2331 struct osd_device *o;
2337 result = dt_device_init(&o->od_dt_dev, t);
2340 l->ld_ops = &osd_lu_ops;
2341 o->od_dt_dev.dd_ops = &osd_dt_ops;
2342 spin_lock_init(&o->od_osfs_lock);
2343 o->od_osfs_age = cfs_time_shift_64(-1000);
2344 o->od_capa_hash = init_capa_hash();
2345 if (o->od_capa_hash == NULL)
2346 l = ERR_PTR(-ENOMEM);
2348 l = ERR_PTR(result);
2350 l = ERR_PTR(-ENOMEM);
2354 static void osd_device_free(const struct lu_env *env, struct lu_device *d)
2356 struct osd_device *o = osd_dev(d);
2358 cleanup_capa_hash(o->od_capa_hash);
2359 dt_device_fini(&o->od_dt_dev);
2363 static int osd_process_config(const struct lu_env *env,
2364 struct lu_device *d, struct lustre_cfg *cfg)
2366 struct osd_device *o = osd_dev(d);
2370 switch(cfg->lcfg_command) {
2372 err = osd_mount(env, o, cfg);
2375 err = osd_shutdown(env, o);
2383 extern void ldiskfs_orphan_cleanup (struct super_block * sb,
2384 struct ldiskfs_super_block * es);
2386 static int osd_recovery_complete(const struct lu_env *env,
2387 struct lu_device *d)
2389 struct osd_device *o = osd_dev(d);
2391 /* TODO: orphans handling */
2392 ldiskfs_orphan_cleanup(osd_sb(o), LDISKFS_SB(osd_sb(o))->s_es);
2396 static struct inode *osd_iget(struct osd_thread_info *info,
2397 struct osd_device *dev,
2398 const struct osd_inode_id *id)
2400 struct inode *inode;
2402 inode = iget(osd_sb(dev), id->oii_ino);
2403 if (inode == NULL) {
2404 CERROR("no inode\n");
2405 inode = ERR_PTR(-EACCES);
2406 } else if (is_bad_inode(inode)) {
2407 CERROR("bad inode\n");
2409 inode = ERR_PTR(-ENOENT);
2410 } else if (inode->i_generation != id->oii_gen) {
2411 CERROR("stale inode\n");
2413 inode = ERR_PTR(-ESTALE);
2420 static int osd_fid_lookup(const struct lu_env *env,
2421 struct osd_object *obj, const struct lu_fid *fid)
2423 struct osd_thread_info *info;
2424 struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
2425 struct osd_device *dev;
2426 struct osd_inode_id *id;
2428 struct inode *inode;
2431 LASSERT(osd_invariant(obj));
2432 LASSERT(obj->oo_inode == NULL);
2433 LASSERT(fid_is_sane(fid));
2435 * This assertion checks that osd layer sees only local
2436 * fids. Unfortunately it is somewhat expensive (does a
2437 * cache-lookup). Disabling it for production/acceptance-testing.
2439 LASSERT(1 || fid_is_local(ldev->ld_site, fid));
2443 info = osd_oti_get(env);
2444 dev = osd_dev(ldev);
2448 if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
2451 result = osd_oi_lookup(info, oi, fid, id);
2453 inode = osd_iget(info, dev, id);
2454 if (!IS_ERR(inode)) {
2455 obj->oo_inode = inode;
2456 LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
2460 * If fid wasn't found in oi, inode-less object is
2461 * created, for which lu_object_exists() returns
2462 * false. This is used in a (frequent) case when
2463 * objects are created as locking anchors or
2464 * place holders for objects yet to be created.
2466 result = PTR_ERR(inode);
2467 } else if (result == -ENOENT)
2469 LASSERT(osd_invariant(obj));
2473 static void osd_inode_getattr(const struct lu_env *env,
2474 struct inode *inode, struct lu_attr *attr)
2476 attr->la_valid |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2477 LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2478 LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
2480 attr->la_atime = LTIME_S(inode->i_atime);
2481 attr->la_mtime = LTIME_S(inode->i_mtime);
2482 attr->la_ctime = LTIME_S(inode->i_ctime);
2483 attr->la_mode = inode->i_mode;
2484 attr->la_size = i_size_read(inode);
2485 attr->la_blocks = inode->i_blocks;
2486 attr->la_uid = inode->i_uid;
2487 attr->la_gid = inode->i_gid;
2488 attr->la_flags = LDISKFS_I(inode)->i_flags;
2489 attr->la_nlink = inode->i_nlink;
2490 attr->la_rdev = inode->i_rdev;
2491 attr->la_blksize = inode->i_blksize;
2498 static int lu_device_is_osd(const struct lu_device *d)
2500 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osd_lu_ops);
2503 static struct osd_object *osd_obj(const struct lu_object *o)
2505 LASSERT(lu_device_is_osd(o->lo_dev));
2506 return container_of0(o, struct osd_object, oo_dt.do_lu);
2509 static struct osd_device *osd_dt_dev(const struct dt_device *d)
2511 LASSERT(lu_device_is_osd(&d->dd_lu_dev));
2512 return container_of0(d, struct osd_device, od_dt_dev);
2515 static struct osd_device *osd_dev(const struct lu_device *d)
2517 LASSERT(lu_device_is_osd(d));
2518 return osd_dt_dev(container_of0(d, struct dt_device, dd_lu_dev));
2521 static struct osd_object *osd_dt_obj(const struct dt_object *d)
2523 return osd_obj(&d->do_lu);
2526 static struct osd_device *osd_obj2dev(const struct osd_object *o)
2528 return osd_dev(o->oo_dt.do_lu.lo_dev);
2531 static struct lu_device *osd2lu_dev(struct osd_device *osd)
2533 return &osd->od_dt_dev.dd_lu_dev;
2536 static struct super_block *osd_sb(const struct osd_device *dev)
2538 return dev->od_mount->lmi_mnt->mnt_sb;
2541 static journal_t *osd_journal(const struct osd_device *dev)
2543 return LDISKFS_SB(osd_sb(dev))->s_journal;
2546 static int osd_has_index(const struct osd_object *obj)
2548 return obj->oo_dt.do_index_ops != NULL;
2551 static int osd_object_invariant(const struct lu_object *l)
2553 return osd_invariant(osd_obj(l));
2556 static struct lu_object_operations osd_lu_obj_ops = {
2557 .loo_object_init = osd_object_init,
2558 .loo_object_delete = osd_object_delete,
2559 .loo_object_release = osd_object_release,
2560 .loo_object_free = osd_object_free,
2561 .loo_object_print = osd_object_print,
2562 .loo_object_invariant = osd_object_invariant
2565 static struct lu_device_operations osd_lu_ops = {
2566 .ldo_object_alloc = osd_object_alloc,
2567 .ldo_process_config = osd_process_config,
2568 .ldo_recovery_complete = osd_recovery_complete
2571 static struct lu_device_type_operations osd_device_type_ops = {
2572 .ldto_init = osd_type_init,
2573 .ldto_fini = osd_type_fini,
2575 .ldto_device_alloc = osd_device_alloc,
2576 .ldto_device_free = osd_device_free,
2578 .ldto_device_init = osd_device_init,
2579 .ldto_device_fini = osd_device_fini
2582 static struct lu_device_type osd_device_type = {
2583 .ldt_tags = LU_DEVICE_DT,
2584 .ldt_name = LUSTRE_OSD_NAME,
2585 .ldt_ops = &osd_device_type_ops,
2586 .ldt_ctx_tags = LCT_MD_THREAD|LCT_DT_THREAD
2590 * lprocfs legacy support.
2592 static struct lprocfs_vars lprocfs_osd_obd_vars[] = {
2596 static struct lprocfs_vars lprocfs_osd_module_vars[] = {
2600 static struct obd_ops osd_obd_device_ops = {
2601 .o_owner = THIS_MODULE
2604 static void lprocfs_osd_init_vars(struct lprocfs_static_vars *lvars)
2606 lvars->module_vars = lprocfs_osd_module_vars;
2607 lvars->obd_vars = lprocfs_osd_obd_vars;
2611 static int __init osd_mod_init(void)
2613 struct lprocfs_static_vars lvars;
2615 lprocfs_osd_init_vars(&lvars);
2616 return class_register_type(&osd_obd_device_ops, NULL, lvars.module_vars,
2617 LUSTRE_OSD_NAME, &osd_device_type);
2620 static void __exit osd_mod_exit(void)
2622 class_unregister_type(LUSTRE_OSD_NAME);
2625 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
2626 MODULE_DESCRIPTION("Lustre Object Storage Device ("LUSTRE_OSD_NAME")");
2627 MODULE_LICENSE("GPL");
2629 cfs_module(osd, "0.0.2", osd_mod_init, osd_mod_exit);