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, 2013, 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>
53 * Infrastructure to support tracking of last committed llog record
55 struct osp_id_tracker {
58 __u32 otr_committed_id;
59 /* callback is register once per diskfs -- that's the whole point */
60 struct dt_txn_callback otr_tx_cb;
61 /* single node can run many clusters */
62 struct list_head otr_wakeup_list;
63 struct list_head otr_list;
64 /* underlying shared device */
65 struct dt_device *otr_dev;
66 /* how many users of this tracker */
67 atomic_t otr_refcount;
70 struct osp_precreate {
74 spinlock_t osp_pre_lock;
76 /* last fid to assign in creation */
77 struct lu_fid osp_pre_used_fid;
78 /* last created id OST reported, next-created - available id's */
79 struct lu_fid osp_pre_last_created_fid;
80 /* how many ids are reserved in declare, we shouldn't block in create */
81 __u64 osp_pre_reserved;
82 /* consumers (who needs new ids) wait here */
83 wait_queue_head_t osp_pre_user_waitq;
84 /* current precreation status: working, failed, stopping? */
86 /* how many to precreate next time */
87 int osp_pre_grow_count;
88 int osp_pre_min_grow_count;
89 int osp_pre_max_grow_count;
90 /* whether to grow precreation window next time or not */
91 int osp_pre_grow_slow;
92 /* cleaning up orphans or recreating missing objects */
93 int osp_pre_recovering;
97 struct dt_device opd_dt_dev;
98 /* corresponded OST index */
101 /* corrsponded MDT index, which will be used when connecting to OST
102 * for validating the connection (see ofd_parse_connect_data) */
104 /* device used to store persistent state (llogs, last ids) */
105 struct obd_export *opd_storage_exp;
106 struct dt_device *opd_storage;
107 struct dt_object *opd_last_used_oid_file;
108 struct dt_object *opd_last_used_seq_file;
110 /* stored persistently in LE format, updated directly to/from disk
111 * and required le64_to_cpu() conversion before use.
112 * Protected by opd_pre_lock */
113 struct lu_fid opd_last_used_fid;
114 struct lu_fid opd_gap_start_fid;
116 /* connection to OST */
117 struct obd_device *opd_obd;
118 struct obd_export *opd_exp;
119 struct obd_uuid opd_cluuid;
120 struct obd_connect_data *opd_connect_data;
122 struct proc_dir_entry *opd_proc_entry;
123 struct lprocfs_stats *opd_stats;
124 /* connection status. */
125 unsigned int opd_new_connection:1,
126 opd_got_disconnected:1,
129 opd_imp_seen_connected:1,
132 /* whether local recovery is completed:
133 * reported via ->ldo_recovery_complete() */
134 int opd_recovery_completed;
136 /* precreate structure for OSP */
137 struct osp_precreate *opd_pre;
138 /* dedicate precreate thread */
139 struct ptlrpc_thread opd_pre_thread;
140 /* thread waits for signals about pool going empty */
141 wait_queue_head_t opd_pre_waitq;
144 * OST synchronization
146 spinlock_t opd_syn_lock;
147 /* unique generation, to recognize start of new records in the llog */
148 struct llog_gen opd_syn_generation;
149 /* number of changes to sync, used to wake up sync thread */
150 unsigned long opd_syn_changes;
151 /* processing of changes from previous mount is done? */
152 int opd_syn_prev_done;
154 struct ptlrpc_thread opd_syn_thread;
155 wait_queue_head_t opd_syn_waitq;
156 /* list of remotely committed rpc */
157 struct list_head opd_syn_committed_there;
158 /* number of changes being under sync */
159 int opd_syn_sync_in_progress;
160 /* number of RPCs in flight - flow control */
161 int opd_syn_rpc_in_flight;
162 int opd_syn_max_rpc_in_flight;
163 /* number of RPC in processing (including non-committed by OST) */
164 int opd_syn_rpc_in_progress;
165 int opd_syn_max_rpc_in_progress;
166 /* osd api's commit cb control structure */
167 struct dt_txn_callback opd_syn_txn_cb;
168 /* last used change number -- semantically similar to transno */
169 unsigned long opd_syn_last_used_id;
170 /* last committed change number -- semantically similar to
172 unsigned long opd_syn_last_committed_id;
173 /* last processed (taken from llog) id */
174 unsigned long opd_syn_last_processed_id;
175 struct osp_id_tracker *opd_syn_tracker;
176 struct list_head opd_syn_ontrack;
177 /* stop processing new requests until barrier=0 */
178 atomic_t opd_syn_barrier;
179 wait_queue_head_t opd_syn_barrier_waitq;
182 * statfs related fields: OSP maintains it on its own
184 struct obd_statfs opd_statfs;
185 cfs_time_t opd_statfs_fresh_till;
186 struct timer_list opd_statfs_timer;
187 int opd_statfs_update_in_progress;
188 /* how often to update statfs data */
189 int opd_statfs_maxage;
191 struct proc_dir_entry *opd_symlink;
193 /* If the caller wants to do some idempotent async operations on
194 * remote server, it can append the async remote requests on the
195 * osp_device::opd_async_requests via declare() functions, these
196 * requests can be packed together and sent to the remote server
197 * via single OUT RPC later. */
198 struct dt_update_request *opd_async_requests;
199 /* Protect current operations on opd_async_requests. */
200 struct mutex opd_async_requests_mutex;
203 #define opd_pre_lock opd_pre->osp_pre_lock
204 #define opd_pre_used_fid opd_pre->osp_pre_used_fid
205 #define opd_pre_last_created_fid opd_pre->osp_pre_last_created_fid
206 #define opd_pre_reserved opd_pre->osp_pre_reserved
207 #define opd_pre_user_waitq opd_pre->osp_pre_user_waitq
208 #define opd_pre_status opd_pre->osp_pre_status
209 #define opd_pre_grow_count opd_pre->osp_pre_grow_count
210 #define opd_pre_min_grow_count opd_pre->osp_pre_min_grow_count
211 #define opd_pre_max_grow_count opd_pre->osp_pre_max_grow_count
212 #define opd_pre_grow_slow opd_pre->osp_pre_grow_slow
213 #define opd_pre_recovering opd_pre->osp_pre_recovering
215 extern struct kmem_cache *osp_object_kmem;
217 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
218 * The left part is for value, binary mode. */
219 struct osp_xattr_entry {
220 struct list_head oxe_list;
226 unsigned int oxe_exist:1,
231 struct osp_object_attr {
232 struct lu_attr ooa_attr;
233 struct list_head ooa_xattr_list;
236 /* this is a top object */
238 struct lu_object_header opo_header;
239 struct dt_object opo_obj;
240 unsigned int opo_reserved:1,
244 /* read/write lock for md osp object */
245 struct rw_semaphore opo_sem;
246 const struct lu_env *opo_owner;
247 struct osp_object_attr *opo_ooa;
248 /* Protect opo_ooa. */
252 extern struct lu_object_operations osp_lu_obj_ops;
253 extern const struct dt_device_operations osp_dt_ops;
254 extern struct dt_object_operations osp_md_obj_ops;
255 extern struct dt_body_operations osp_md_body_ops;
257 struct osp_thread_info {
258 struct lu_buf osi_lb;
259 struct lu_buf osi_lb2;
260 struct lu_fid osi_fid;
261 struct lu_attr osi_attr;
262 struct ost_id osi_oi;
263 struct ost_id osi_oi2;
267 struct llog_rec_hdr osi_hdr;
268 struct llog_unlink64_rec osi_unlink;
269 struct llog_setattr64_rec osi_setattr;
270 struct llog_gen_rec osi_gen;
272 struct llog_cookie osi_cookie;
273 struct llog_catid osi_cid;
274 struct lu_seq_range osi_seq;
275 struct ldlm_res_id osi_resid;
276 struct obdo osi_obdo;
279 /* Iterator for OSP */
282 __u32 ooi_pos_lu_page;
285 int ooi_total_npages;
286 int ooi_valid_npages;
287 unsigned int ooi_swab:1;
289 struct dt_object *ooi_obj;
291 struct page *ooi_cur_page;
292 struct lu_idxpage *ooi_cur_idxpage;
293 struct page **ooi_pages;
296 /* The transaction only include the updates on the remote node, and
297 * no local updates at all */
298 static inline bool is_only_remote_trans(struct thandle *th)
300 return th->th_dev != NULL && th->th_dev->dd_ops == &osp_dt_ops;
303 static inline void osp_objid_buf_prep(struct lu_buf *buf, loff_t *off,
304 __u32 *id, int index)
306 buf->lb_buf = (void *)id;
307 buf->lb_len = sizeof(obd_id);
308 *off = sizeof(obd_id) * index;
311 static inline void osp_objseq_buf_prep(struct lu_buf *buf, loff_t *off,
312 __u64 *seq, int index)
314 buf->lb_buf = (void *)seq;
315 buf->lb_len = sizeof(obd_id);
316 *off = sizeof(obd_id) * index;
319 static inline void osp_buf_prep(struct lu_buf *lb, void *buf, int buf_len)
322 lb->lb_len = buf_len;
325 extern struct lu_context_key osp_thread_key;
327 static inline struct osp_thread_info *osp_env_info(const struct lu_env *env)
329 struct osp_thread_info *info;
331 info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
333 lu_env_refill((struct lu_env *)env);
334 info = lu_context_key_get(&env->le_ctx, &osp_thread_key);
340 struct osp_txn_info {
341 __u32 oti_current_id;
344 extern struct lu_context_key osp_txn_key;
346 static inline struct osp_txn_info *osp_txn_info(struct lu_context *ctx)
348 struct osp_txn_info *info;
350 info = lu_context_key_get(ctx, &osp_txn_key);
354 extern const struct lu_device_operations osp_lu_ops;
356 static inline int lu_device_is_osp(struct lu_device *d)
358 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &osp_lu_ops);
361 static inline struct osp_device *lu2osp_dev(struct lu_device *d)
363 LASSERT(lu_device_is_osp(d));
364 return container_of0(d, struct osp_device, opd_dt_dev.dd_lu_dev);
367 static inline struct lu_device *osp2lu_dev(struct osp_device *d)
369 return &d->opd_dt_dev.dd_lu_dev;
372 static inline struct osp_device *dt2osp_dev(struct dt_device *d)
374 LASSERT(lu_device_is_osp(&d->dd_lu_dev));
375 return container_of0(d, struct osp_device, opd_dt_dev);
378 static inline struct osp_object *lu2osp_obj(struct lu_object *o)
380 LASSERT(ergo(o != NULL, lu_device_is_osp(o->lo_dev)));
381 return container_of0(o, struct osp_object, opo_obj.do_lu);
384 static inline struct lu_object *osp2lu_obj(struct osp_object *obj)
386 return &obj->opo_obj.do_lu;
389 static inline struct osp_object *osp_obj(const struct lu_object *o)
391 LASSERT(lu_device_is_osp(o->lo_dev));
392 return container_of0(o, struct osp_object, opo_obj.do_lu);
395 static inline struct osp_object *dt2osp_obj(const struct dt_object *d)
397 return osp_obj(&d->do_lu);
400 static inline struct dt_object *osp_object_child(struct osp_object *o)
402 return container_of0(lu_object_next(osp2lu_obj(o)),
403 struct dt_object, do_lu);
406 static inline struct seq_server_site *osp_seq_site(struct osp_device *osp)
408 return osp->opd_dt_dev.dd_lu_dev.ld_site->ld_seq_site;
411 #define osp_init_rpc_lock(lck) mdc_init_rpc_lock(lck)
412 #define osp_get_rpc_lock(lck, it) mdc_get_rpc_lock(lck, it)
413 #define osp_put_rpc_lock(lck, it) mdc_put_rpc_lock(lck, it)
415 static inline int osp_fid_diff(const struct lu_fid *fid1,
416 const struct lu_fid *fid2)
418 /* In 2.6+ ost_idx is packed into IDIF FID, while in 2.4 and 2.5 IDIF
419 * is always FID_SEQ_IDIF(0x100000000ULL), which does not include OST
420 * index in the seq. So we can not compare IDIF FID seq here */
421 if (fid_is_idif(fid1) && fid_is_idif(fid2)) {
422 __u32 ost_idx1 = fid_idif_ost_idx(fid1);
423 __u32 ost_idx2 = fid_idif_ost_idx(fid2);
425 LASSERTF(ost_idx1 == 0 || ost_idx2 == 0 || ost_idx1 == ost_idx2,
426 "fid1: "DFID", fid2: "DFID"\n", PFID(fid1),
429 return fid_idif_id(fid1->f_seq, fid1->f_oid, 0) -
430 fid_idif_id(fid2->f_seq, fid2->f_oid, 0);
433 LASSERTF(fid_seq(fid1) == fid_seq(fid2), "fid1:"DFID
434 ", fid2:"DFID"\n", PFID(fid1), PFID(fid2));
436 return fid_oid(fid1) - fid_oid(fid2);
440 static inline void osp_update_last_fid(struct osp_device *d, struct lu_fid *fid)
442 int diff = osp_fid_diff(fid, &d->opd_last_used_fid);
444 * we might have lost precreated objects due to VBR and precreate
445 * orphans, the gap in objid can be calculated properly only here
449 d->opd_gap_start_fid = d->opd_last_used_fid;
450 d->opd_gap_start_fid.f_oid++;
451 d->opd_gap_count = diff - 1;
452 CDEBUG(D_HA, "Gap in objids: start="DFID", count =%d\n",
453 PFID(&d->opd_gap_start_fid), d->opd_gap_count);
455 d->opd_last_used_fid = *fid;
459 static int osp_fid_end_seq(const struct lu_env *env, struct lu_fid *fid)
461 if (fid_is_idif(fid)) {
462 struct osp_thread_info *info = osp_env_info(env);
463 struct ost_id *oi = &info->osi_oi;
465 fid_to_ostid(fid, oi);
466 return ostid_id(oi) == IDIF_MAX_OID;
468 return fid_oid(fid) == LUSTRE_DATA_SEQ_MAX_WIDTH;
472 static inline int osp_precreate_end_seq_nolock(const struct lu_env *env,
473 struct osp_device *osp)
475 struct lu_fid *fid = &osp->opd_pre_last_created_fid;
477 return osp_fid_end_seq(env, fid);
480 static inline int osp_precreate_end_seq(const struct lu_env *env,
481 struct osp_device *osp)
485 spin_lock(&osp->opd_pre_lock);
486 rc = osp_precreate_end_seq_nolock(env, osp);
487 spin_unlock(&osp->opd_pre_lock);
491 static inline int osp_is_fid_client(struct osp_device *osp)
493 struct obd_import *imp = osp->opd_obd->u.cli.cl_import;
495 return imp->imp_connect_data.ocd_connect_flags & OBD_CONNECT_FID;
498 typedef int (*osp_async_request_interpreter_t)(const struct lu_env *env,
499 struct object_update_reply *rep,
500 struct ptlrpc_request *req,
501 struct osp_object *obj,
502 void *data, int index, int rc);
505 void osp_update_last_id(struct osp_device *d, obd_id objid);
506 extern struct llog_operations osp_mds_ost_orig_logops;
509 int osp_insert_async_request(const struct lu_env *env, enum update_type op,
510 struct osp_object *obj, int count, __u16 *lens,
511 const void **bufs, void *data,
512 osp_async_request_interpreter_t interpreter);
513 int osp_unplug_async_request(const struct lu_env *env,
514 struct osp_device *osp,
515 struct dt_update_request *update);
516 struct thandle *osp_trans_create(const struct lu_env *env,
517 struct dt_device *d);
518 int osp_trans_start(const struct lu_env *env, struct dt_device *dt,
522 int osp_attr_get(const struct lu_env *env, struct dt_object *dt,
523 struct lu_attr *attr, struct lustre_capa *capa);
524 int osp_xattr_get(const struct lu_env *env, struct dt_object *dt,
525 struct lu_buf *buf, const char *name,
526 struct lustre_capa *capa);
527 int osp_declare_xattr_set(const struct lu_env *env, struct dt_object *dt,
528 const struct lu_buf *buf, const char *name,
529 int flag, struct thandle *th);
530 int osp_xattr_set(const struct lu_env *env, struct dt_object *dt,
531 const struct lu_buf *buf, const char *name, int fl,
532 struct thandle *th, struct lustre_capa *capa);
533 int osp_declare_xattr_del(const struct lu_env *env, struct dt_object *dt,
534 const char *name, struct thandle *th);
535 int osp_xattr_del(const struct lu_env *env, struct dt_object *dt,
536 const char *name, struct thandle *th,
537 struct lustre_capa *capa);
539 int osp_declare_object_destroy(const struct lu_env *env,
540 struct dt_object *dt, struct thandle *th);
541 int osp_object_destroy(const struct lu_env *env, struct dt_object *dt,
544 int osp_trans_stop(const struct lu_env *env, struct dt_device *dt,
547 struct dt_it *osp_it_init(const struct lu_env *env, struct dt_object *dt,
548 __u32 attr, struct lustre_capa *capa);
549 void osp_it_fini(const struct lu_env *env, struct dt_it *di);
550 int osp_it_get(const struct lu_env *env, struct dt_it *di,
551 const struct dt_key *key);
552 void osp_it_put(const struct lu_env *env, struct dt_it *di);
553 __u64 osp_it_store(const struct lu_env *env, const struct dt_it *di);
554 int osp_it_key_rec(const struct lu_env *env, const struct dt_it *di,
556 int osp_it_next_page(const struct lu_env *env, struct dt_it *di);
557 /* osp_md_object.c */
558 int osp_md_declare_object_create(const struct lu_env *env,
559 struct dt_object *dt,
560 struct lu_attr *attr,
561 struct dt_allocation_hint *hint,
562 struct dt_object_format *dof,
564 int osp_md_object_create(const struct lu_env *env, struct dt_object *dt,
565 struct lu_attr *attr, struct dt_allocation_hint *hint,
566 struct dt_object_format *dof, struct thandle *th);
567 int __osp_md_attr_set(const struct lu_env *env, struct dt_object *dt,
568 const struct lu_attr *attr, struct thandle *th);
569 extern const struct dt_index_operations osp_md_index_ops;
571 /* osp_precreate.c */
572 int osp_init_precreate(struct osp_device *d);
573 int osp_precreate_reserve(const struct lu_env *env, struct osp_device *d);
574 __u64 osp_precreate_get_id(struct osp_device *d);
575 int osp_precreate_get_fid(const struct lu_env *env, struct osp_device *d,
577 void osp_precreate_fini(struct osp_device *d);
578 int osp_object_truncate(const struct lu_env *env, struct dt_object *dt, __u64);
579 void osp_pre_update_status(struct osp_device *d, int rc);
580 void osp_statfs_need_now(struct osp_device *d);
581 int osp_reset_last_used(const struct lu_env *env, struct osp_device *osp);
582 int osp_write_last_oid_seq_files(struct lu_env *env, struct osp_device *osp,
583 struct lu_fid *fid, int sync);
584 int osp_init_pre_fid(struct osp_device *osp);
587 void osp_lprocfs_init(struct osp_device *osp);
590 int osp_sync_declare_add(const struct lu_env *env, struct osp_object *o,
591 llog_op_type type, struct thandle *th);
592 int osp_sync_add(const struct lu_env *env, struct osp_object *o,
593 llog_op_type type, struct thandle *th,
594 const struct lu_attr *attr);
595 int osp_sync_init(const struct lu_env *env, struct osp_device *d);
596 int osp_sync_fini(struct osp_device *d);
597 void __osp_sync_check_for_work(struct osp_device *d);
600 extern struct obd_ops lwp_obd_device_ops;
601 extern struct lu_device_type lwp_device_type;