4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osp/osp_internal.h
38 * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com>
41 #ifndef _OSP_INTERNAL_H
42 #define _OSP_INTERNAL_H
45 #include <obd_class.h>
46 #include <dt_object.h>
47 #include <md_object.h>
48 #include <lustre_fid.h>
49 #include <lustre_update.h>
50 #include <lu_target.h>
51 #include <lustre_mdc.h>
54 * Infrastructure to support tracking of last committed llog record
56 struct osp_id_tracker {
59 __u32 otr_committed_id;
60 /* callback is register once per diskfs -- that's the whole point */
61 struct dt_txn_callback otr_tx_cb;
62 /* single node can run many clusters */
63 struct list_head otr_wakeup_list;
64 struct list_head otr_list;
65 /* underlying shared device */
66 struct dt_device *otr_dev;
67 /* how many users of this tracker */
68 atomic_t otr_refcount;
71 struct osp_precreate {
75 spinlock_t osp_pre_lock;
77 /* last fid to assign in creation */
78 struct lu_fid osp_pre_used_fid;
79 /* last created id OST reported, next-created - available id's */
80 struct lu_fid osp_pre_last_created_fid;
81 /* how many ids are reserved in declare, we shouldn't block in create */
82 __u64 osp_pre_reserved;
83 /* consumers (who needs new ids) wait here */
84 wait_queue_head_t osp_pre_user_waitq;
85 /* current precreation status: working, failed, stopping? */
87 /* how many to precreate next time */
88 int osp_pre_grow_count;
89 int osp_pre_min_grow_count;
90 int osp_pre_max_grow_count;
91 /* whether to grow precreation window next time or not */
92 int osp_pre_grow_slow;
93 /* cleaning up orphans or recreating missing objects */
94 int osp_pre_recovering;
98 struct dt_device opd_dt_dev;
99 /* corresponded OST index */
102 /* corrsponded MDT index, which will be used when connecting to OST
103 * for validating the connection (see ofd_parse_connect_data) */
105 /* device used to store persistent state (llogs, last ids) */
106 struct obd_export *opd_storage_exp;
107 struct dt_device *opd_storage;
108 struct dt_object *opd_last_used_oid_file;
109 struct dt_object *opd_last_used_seq_file;
111 /* stored persistently in LE format, updated directly to/from disk
112 * and required le64_to_cpu() conversion before use.
113 * Protected by opd_pre_lock */
114 struct lu_fid opd_last_used_fid;
115 struct lu_fid opd_gap_start_fid;
117 /* connection to OST */
118 struct obd_device *opd_obd;
119 struct obd_export *opd_exp;
120 struct obd_uuid opd_cluuid;
121 struct obd_connect_data *opd_connect_data;
123 struct proc_dir_entry *opd_proc_entry;
124 struct lprocfs_stats *opd_stats;
125 /* connection status. */
126 unsigned int opd_new_connection:1,
127 opd_got_disconnected:1,
130 opd_imp_seen_connected:1,
133 /* whether local recovery is completed:
134 * reported via ->ldo_recovery_complete() */
135 int opd_recovery_completed;
137 /* precreate structure for OSP */
138 struct osp_precreate *opd_pre;
139 /* dedicate precreate thread */
140 struct ptlrpc_thread opd_pre_thread;
141 /* thread waits for signals about pool going empty */
142 wait_queue_head_t opd_pre_waitq;
145 * OST synchronization
147 spinlock_t opd_syn_lock;
148 /* unique generation, to recognize start of new records in the llog */
149 struct llog_gen opd_syn_generation;
150 /* number of changes to sync, used to wake up sync thread */
151 unsigned long opd_syn_changes;
152 /* processing of changes from previous mount is done? */
153 int opd_syn_prev_done;
155 struct ptlrpc_thread opd_syn_thread;
156 wait_queue_head_t opd_syn_waitq;
157 /* list of remotely committed rpc */
158 struct list_head opd_syn_committed_there;
159 /* number of changes being under sync */
160 int opd_syn_sync_in_progress;
161 /* number of RPCs in flight - flow control */
162 int opd_syn_rpc_in_flight;
163 int opd_syn_max_rpc_in_flight;
164 /* number of RPC in processing (including non-committed by OST) */
165 int opd_syn_rpc_in_progress;
166 int opd_syn_max_rpc_in_progress;
167 /* osd api's commit cb control structure */
168 struct dt_txn_callback opd_syn_txn_cb;
169 /* last used change number -- semantically similar to transno */
170 unsigned long opd_syn_last_used_id;
171 /* last committed change number -- semantically similar to
173 unsigned long opd_syn_last_committed_id;
174 /* last processed (taken from llog) id */
175 unsigned long opd_syn_last_processed_id;
176 struct osp_id_tracker *opd_syn_tracker;
177 struct list_head opd_syn_ontrack;
178 /* stop processing new requests until barrier=0 */
179 atomic_t opd_syn_barrier;
180 wait_queue_head_t opd_syn_barrier_waitq;
183 * statfs related fields: OSP maintains it on its own
185 struct obd_statfs opd_statfs;
186 cfs_time_t opd_statfs_fresh_till;
187 struct timer_list opd_statfs_timer;
188 int opd_statfs_update_in_progress;
189 /* how often to update statfs data */
190 int opd_statfs_maxage;
192 struct proc_dir_entry *opd_symlink;
194 /* If the caller wants to do some idempotent async operations on
195 * remote server, it can append the async remote requests on the
196 * osp_device::opd_async_requests via declare() functions, these
197 * requests can be packed together and sent to the remote server
198 * via single OUT RPC later. */
199 struct dt_update_request *opd_async_requests;
200 /* Protect current operations on opd_async_requests. */
201 struct mutex opd_async_requests_mutex;
202 struct list_head opd_async_updates;
203 struct rw_semaphore opd_async_updates_rwsem;
204 atomic_t opd_async_updates_count;
207 #define opd_pre_lock opd_pre->osp_pre_lock
208 #define opd_pre_used_fid opd_pre->osp_pre_used_fid
209 #define opd_pre_last_created_fid opd_pre->osp_pre_last_created_fid
210 #define opd_pre_reserved opd_pre->osp_pre_reserved
211 #define opd_pre_user_waitq opd_pre->osp_pre_user_waitq
212 #define opd_pre_status opd_pre->osp_pre_status
213 #define opd_pre_grow_count opd_pre->osp_pre_grow_count
214 #define opd_pre_min_grow_count opd_pre->osp_pre_min_grow_count
215 #define opd_pre_max_grow_count opd_pre->osp_pre_max_grow_count
216 #define opd_pre_grow_slow opd_pre->osp_pre_grow_slow
217 #define opd_pre_recovering opd_pre->osp_pre_recovering
219 extern struct kmem_cache *osp_object_kmem;
221 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
222 * The left part is for value, binary mode. */
223 struct osp_xattr_entry {
224 struct list_head oxe_list;
230 unsigned int oxe_exist:1,
235 struct osp_object_attr {
236 struct lu_attr ooa_attr;
237 struct list_head ooa_xattr_list;
240 /* this is a top object */
242 struct lu_object_header opo_header;
243 struct dt_object opo_obj;
244 unsigned int opo_reserved:1,
247 /* read/write lock for md osp object */
248 struct rw_semaphore opo_sem;
249 const struct lu_env *opo_owner;
250 struct osp_object_attr *opo_ooa;
251 /* Protect opo_ooa. */
255 extern struct lu_object_operations osp_lu_obj_ops;
256 extern const struct dt_device_operations osp_dt_ops;
257 extern struct dt_object_operations osp_md_obj_ops;
258 extern struct dt_body_operations osp_md_body_ops;
260 struct osp_thread_info {
261 struct lu_buf osi_lb;
262 struct lu_buf osi_lb2;
263 struct lu_fid osi_fid;
264 struct lu_attr osi_attr;
265 struct ost_id osi_oi;
266 struct ost_id osi_oi2;
270 struct llog_rec_hdr osi_hdr;
271 struct llog_unlink64_rec osi_unlink;
272 struct llog_setattr64_rec osi_setattr;
273 struct llog_gen_rec osi_gen;
275 struct llog_cookie osi_cookie;
276 struct llog_catid osi_cid;
277 struct lu_seq_range osi_seq;
278 struct ldlm_res_id osi_resid;
279 struct obdo osi_obdo;
282 /* Iterator for OSP */
285 __u32 ooi_pos_lu_page;
288 int ooi_total_npages;
289 int ooi_valid_npages;
290 unsigned int ooi_swab:1;
292 struct dt_object *ooi_obj;
294 struct page *ooi_cur_page;
295 struct lu_idxpage *ooi_cur_idxpage;
296 struct page **ooi_pages;
300 struct thandle ot_super;
301 struct dt_update_request *ot_dur;
303 /* OSP will use this thandle to update last oid*/
304 struct thandle *ot_storage_th;
305 struct list_head ot_dcb_list;
306 atomic_t ot_refcount;
309 static inline struct osp_thandle *
310 thandle_to_osp_thandle(struct thandle *th)
312 return container_of(th, struct osp_thandle, ot_super);
315 static inline struct dt_update_request *
316 thandle_to_dt_update_request(struct thandle *th)
318 struct osp_thandle *oth;
320 oth = thandle_to_osp_thandle(th);
324 /* The transaction only include the updates on the remote node, and
325 * no local updates at all */
326 static inline bool is_only_remote_trans(struct thandle *th)
328 return th->th_top == NULL;
331 static inline void osp_objid_buf_prep(struct lu_buf *buf, loff_t *off,
332 __u32 *id, int index)
334 /* Note: through id is only 32 bits, it will also write 64 bits
335 * for oid to keep compatibility with the previous version. */
336 buf->lb_buf = (void *)id;
337 buf->lb_len = sizeof(u64);
338 *off = sizeof(u64) * index;
341 static inline void osp_objseq_buf_prep(struct lu_buf *buf, loff_t *off,
342 __u64 *seq, int index)
344 buf->lb_buf = (void *)seq;
345 buf->lb_len = sizeof(u64);
346 *off = sizeof(u64) * index;
349 static inline void osp_buf_prep(struct lu_buf *lb, void *buf, int buf_len)
352 lb->lb_len = buf_len;
355 extern struct lu_context_key osp_thread_key;
357 static inline struct osp_thread_info *osp_env_info(const struct lu_env *env)
359 struct osp_thread_info *info;
361 info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
363 lu_env_refill((struct lu_env *)env);
364 info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
370 struct osp_txn_info {
371 __u32 oti_current_id;
374 extern struct lu_context_key osp_txn_key;
376 static inline struct osp_txn_info *osp_txn_info(struct lu_context *ctx)
378 struct osp_txn_info *info;
380 info = lu_context_key_get(ctx, &osp_txn_key);
384 extern const struct lu_device_operations osp_lu_ops;
386 static inline int lu_device_is_osp(struct lu_device *d)
388 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osp_lu_ops);
391 static inline struct osp_device *lu2osp_dev(struct lu_device *d)
393 LASSERT(lu_device_is_osp(d));
394 return container_of0(d, struct osp_device, opd_dt_dev.dd_lu_dev);
397 static inline struct lu_device *osp2lu_dev(struct osp_device *d)
399 return &d->opd_dt_dev.dd_lu_dev;
402 static inline struct osp_device *dt2osp_dev(struct dt_device *d)
404 LASSERT(lu_device_is_osp(&d->dd_lu_dev));
405 return container_of0(d, struct osp_device, opd_dt_dev);
408 static inline struct osp_object *lu2osp_obj(struct lu_object *o)
410 LASSERT(ergo(o != NULL, lu_device_is_osp(o->lo_dev)));
411 return container_of0(o, struct osp_object, opo_obj.do_lu);
414 static inline struct lu_object *osp2lu_obj(struct osp_object *obj)
416 return &obj->opo_obj.do_lu;
419 static inline struct osp_object *osp_obj(const struct lu_object *o)
421 LASSERT(lu_device_is_osp(o->lo_dev));
422 return container_of0(o, struct osp_object, opo_obj.do_lu);
425 static inline struct osp_object *dt2osp_obj(const struct dt_object *d)
427 return osp_obj(&d->do_lu);
430 static inline struct dt_object *osp_object_child(struct osp_object *o)
432 return container_of0(lu_object_next(osp2lu_obj(o)),
433 struct dt_object, do_lu);
436 static inline struct seq_server_site *osp_seq_site(struct osp_device *osp)
438 return osp->opd_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
441 #define osp_init_rpc_lock(lck) mdc_init_rpc_lock(lck)
443 static inline void osp_get_rpc_lock(struct osp_device *osp)
445 struct mdc_rpc_lock *rpc_lock = osp->opd_obd->u.cli.cl_rpc_lock;
447 mdc_get_rpc_lock(rpc_lock, NULL);
450 static inline void osp_put_rpc_lock(struct osp_device *osp)
452 struct mdc_rpc_lock *rpc_lock = osp->opd_obd->u.cli.cl_rpc_lock;
454 mdc_put_rpc_lock(rpc_lock, NULL);
457 static inline int osp_fid_diff(const struct lu_fid *fid1,
458 const struct lu_fid *fid2)
460 /* In 2.6+ ost_idx is packed into IDIF FID, while in 2.4 and 2.5 IDIF
461 * is always FID_SEQ_IDIF(0x100000000ULL), which does not include OST
462 * index in the seq. So we can not compare IDIF FID seq here */
463 if (fid_is_idif(fid1) && fid_is_idif(fid2)) {
464 __u32 ost_idx1 = fid_idif_ost_idx(fid1);
465 __u32 ost_idx2 = fid_idif_ost_idx(fid2);
467 LASSERTF(ost_idx1 == 0 || ost_idx2 == 0 || ost_idx1 == ost_idx2,
468 "fid1: "DFID", fid2: "DFID"\n", PFID(fid1),
471 return fid_idif_id(fid1->f_seq, fid1->f_oid, 0) -
472 fid_idif_id(fid2->f_seq, fid2->f_oid, 0);
475 LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID
476 ", fid2:"DFID"\n", PFID(fid1), PFID(fid2));
478 return fid_oid(fid1) - fid_oid(fid2);
482 static inline void osp_update_last_fid(struct osp_device *d, struct lu_fid *fid)
484 int diff = osp_fid_diff(fid, &d->opd_last_used_fid);
486 * we might have lost precreated objects due to VBR and precreate
487 * orphans, the gap in objid can be calculated properly only here
491 d->opd_gap_start_fid = d->opd_last_used_fid;
492 d->opd_gap_start_fid.f_oid++;
493 d->opd_gap_count = diff - 1;
494 CDEBUG(D_HA, "Gap in objids: start="DFID", count =%d\n",
495 PFID(&d->opd_gap_start_fid), d->opd_gap_count);
497 d->opd_last_used_fid = *fid;
501 static int osp_fid_end_seq(const struct lu_env *env, struct lu_fid *fid)
503 if (fid_is_idif(fid)) {
504 struct osp_thread_info *info = osp_env_info(env);
505 struct ost_id *oi = &info->osi_oi;
507 fid_to_ostid(fid, oi);
508 return ostid_id(oi) == IDIF_MAX_OID;
510 return fid_oid(fid) == LUSTRE_DATA_SEQ_MAX_WIDTH;
514 static inline int osp_precreate_end_seq_nolock(const struct lu_env *env,
515 struct osp_device *osp)
517 struct lu_fid *fid = &osp->opd_pre_last_created_fid;
519 return osp_fid_end_seq(env, fid);
522 static inline int osp_precreate_end_seq(const struct lu_env *env,
523 struct osp_device *osp)
527 spin_lock(&osp->opd_pre_lock);
528 rc = osp_precreate_end_seq_nolock(env, osp);
529 spin_unlock(&osp->opd_pre_lock);
533 static inline int osp_is_fid_client(struct osp_device *osp)
535 struct obd_import *imp = osp->opd_obd->u.cli.cl_import;
537 return imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_FID;
540 struct object_update *
541 update_buffer_get_update(struct object_update_request *request,
544 int osp_extend_update_buffer(const struct lu_env *env,
545 struct update_buffer *ubuf);
547 #define osp_update_rpc_pack(env, name, update, op, ...) \
549 struct object_update *object_update; \
550 size_t max_update_length; \
551 struct object_update_request *ureq; \
555 ureq = update->dur_buf.ub_req; \
556 max_update_length = update->dur_buf.ub_req_size - \
557 object_update_request_size(ureq); \
559 object_update = update_buffer_get_update(ureq, \
561 ret = out_##name##_pack(env, object_update, max_update_length, \
563 if (ret == -E2BIG) { \
565 /* extend the buffer and retry */ \
566 rc1 = osp_extend_update_buffer(env, &update->dur_buf); \
573 object_update->ou_flags |= update->dur_flags; \
574 ureq->ourq_count++; \
582 typedef int (*osp_update_interpreter_t)(const struct lu_env *env,
583 struct object_update_reply *rep,
584 struct ptlrpc_request *req,
585 struct osp_object *obj,
586 void *data, int index, int rc);
589 void osp_update_last_id(struct osp_device *d, u64 objid);
590 extern struct llog_operations osp_mds_ost_orig_logops;
593 int osp_insert_async_request(const struct lu_env *env, enum update_type op,
594 struct osp_object *obj, int count, __u16 *lens,
595 const void **bufs, void *data,
596 osp_update_interpreter_t interpreter);
598 int osp_unplug_async_request(const struct lu_env *env,
599 struct osp_device *osp,
600 struct dt_update_request *update);
601 int osp_trans_update_request_create(struct thandle *th);
602 struct thandle *osp_trans_create(const struct lu_env *env,
603 struct dt_device *d);
604 int osp_trans_start(const struct lu_env *env, struct dt_device *dt,
606 int osp_insert_update_callback(const struct lu_env *env,
607 struct dt_update_request *update,
608 struct osp_object *obj, void *data,
609 osp_update_interpreter_t interpreter);
610 int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
611 const struct object_update_request *ureq,
612 struct ptlrpc_request **reqp);
613 struct dt_update_request *dt_update_request_create(struct dt_device *dt);
614 void dt_update_request_destroy(struct dt_update_request *dt_update);
616 int osp_prep_update_req(const struct lu_env *env, struct obd_import *imp,
617 const struct object_update_request *ureq,
618 struct ptlrpc_request **reqp);
619 int osp_remote_sync(const struct lu_env *env, struct osp_device *osp,
620 struct dt_update_request *update,
621 struct ptlrpc_request **reqp);
623 struct thandle *osp_get_storage_thandle(const struct lu_env *env,
625 struct osp_device *osp);
627 int osp_attr_get(const struct lu_env *env, struct dt_object *dt,
628 struct lu_attr *attr);
629 int osp_xattr_get(const struct lu_env *env, struct dt_object *dt,
630 struct lu_buf *buf, const char *name);
631 int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
632 const struct lu_buf *buf, const char *name,
633 int flag, struct thandle *th);
634 int osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
635 const struct lu_buf *buf, const char *name, int fl,
637 int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
638 const char *name, struct thandle *th);
639 int osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
640 const char *name, struct thandle *th);
642 int osp_trans_stop(const struct lu_env *env, struct dt_device *dt,
644 int osp_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb);
646 struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt,
648 void osp_it_fini(const struct lu_env *env, struct dt_it *di);
649 int osp_it_get(const struct lu_env *env, struct dt_it *di,
650 const struct dt_key *key);
651 void osp_it_put(const struct lu_env *env, struct dt_it *di);
652 __u64 osp_it_store(const struct lu_env *env, const struct dt_it *di);
653 int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di,
655 int osp_it_next_page(const struct lu_env *env, struct dt_it *di);
656 int osp_oac_init(struct osp_object *obj);
657 /* osp_md_object.c */
658 int osp_md_declare_object_create(const struct lu_env *env,
659 struct dt_object *dt,
660 struct lu_attr *attr,
661 struct dt_allocation_hint *hint,
662 struct dt_object_format *dof,
664 int osp_md_object_create(const struct lu_env *env, struct dt_object *dt,
665 struct lu_attr *attr, struct dt_allocation_hint *hint,
666 struct dt_object_format *dof, struct thandle *th);
667 int osp_md_declare_attr_set(const struct lu_env *env, struct dt_object *dt,
668 const struct lu_attr *attr, struct thandle *th);
669 int osp_md_attr_set(const struct lu_env *env, struct dt_object *dt,
670 const struct lu_attr *attr, struct thandle *th);
671 extern const struct dt_index_operations osp_md_index_ops;
673 /* osp_precreate.c */
674 int osp_init_precreate(struct osp_device *d);
675 int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d);
676 __u64 osp_precreate_get_id(struct osp_device *d);
677 int osp_precreate_get_fid(const struct lu_env *env, struct osp_device *d,
679 void osp_precreate_fini(struct osp_device *d);
680 int osp_object_truncate(const struct lu_env *env, struct dt_object *dt, __u64);
681 void osp_pre_update_status(struct osp_device *d, int rc);
682 void osp_statfs_need_now(struct osp_device *d);
683 int osp_reset_last_used(const struct lu_env *env, struct osp_device *osp);
684 int osp_write_last_oid_seq_files(struct lu_env *env, struct osp_device *osp,
685 struct lu_fid *fid, int sync);
686 int osp_init_pre_fid(struct osp_device *osp);
689 void osp_lprocfs_init(struct osp_device *osp);
692 int osp_sync_declare_add(const struct lu_env *env, struct osp_object *o,
693 llog_op_type type, struct thandle *th);
694 int osp_sync_add(const struct lu_env *env, struct osp_object *o,
695 llog_op_type type, struct thandle *th,
696 const struct lu_attr *attr);
697 int osp_sync_init(const struct lu_env *env, struct osp_device *d);
698 int osp_sync_fini(struct osp_device *d);
699 void __osp_sync_check_for_work(struct osp_device *d);
702 extern struct obd_ops lwp_obd_device_ops;
703 extern struct lu_device_type lwp_device_type;