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, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/include/lustre_log.h
38 * Generic infrastructure for managing a collection of logs.
39 * These logs are used for:
41 * - orphan recovery: OST adds record on create
42 * - mtime/size consistency: the OST adds a record on first write
43 * - open/unlinked objects: OST adds a record on destroy
45 * - mds unlink log: the MDS adds an entry upon delete
47 * - raid1 replication log between OST's
48 * - MDS replication logs
59 #if defined(__linux__)
60 #include <linux/lustre_log.h>
61 #elif defined(__APPLE__)
62 #include <darwin/lustre_log.h>
63 #elif defined(__WINNT__)
64 #include <winnt/lustre_log.h>
66 #error Unsupported operating system.
69 #include <obd_class.h>
71 #include <lustre/lustre_idl.h>
72 #include <dt_object.h>
74 #define LOG_NAME_LIMIT(logname, name) \
75 snprintf(logname, sizeof(logname), "LOGS/%s", name)
76 #define LLOG_EEMPTY 4711
78 enum llog_open_param {
79 LLOG_OPEN_EXISTS = 0x0000,
80 LLOG_OPEN_NEW = 0x0001,
83 struct plain_handle_data {
85 struct llog_handle *phd_cat_handle;
86 struct llog_cookie phd_cookie; /* cookie of this log in its cat */
89 struct cat_handle_data {
91 struct llog_handle *chd_current_log; /* currently open log */
92 struct llog_handle *chd_next_log; /* llog to be used next */
95 static inline void logid_to_fid(struct llog_logid *id, struct lu_fid *fid)
97 /* For compatibility purposes we identify pre-OSD (~< 2.3.51 MDS)
98 * logid's by non-zero ogen (inode generation) and convert them
100 if (id->lgl_ogen == 0) {
101 fid->f_seq = id->lgl_oseq;
102 fid->f_oid = id->lgl_oid;
105 lu_igif_build(fid, id->lgl_oid, id->lgl_ogen);
109 static inline void fid_to_logid(struct lu_fid *fid, struct llog_logid *id)
111 id->lgl_oseq = fid->f_seq;
112 id->lgl_oid = fid->f_oid;
118 /* llog.c - general API */
119 typedef int (*llog_cb_t)(const struct lu_env *env, struct llog_handle *lgh,
120 struct llog_rec_hdr *rec, void *data);
121 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
122 int flags, struct obd_uuid *uuid);
123 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
124 llog_cb_t cb, void *data, void *catdata);
125 int llog_process_or_fork(const struct lu_env *env,
126 struct llog_handle *loghandle,
127 llog_cb_t cb, void *data, void *catdata, bool fork);
128 int llog_reverse_process(const struct lu_env *env,
129 struct llog_handle *loghandle, llog_cb_t cb,
130 void *data, void *catdata);
131 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
133 int llog_open(const struct lu_env *env, struct llog_ctxt *ctxt,
134 struct llog_handle **lgh, struct llog_logid *logid,
135 char *name, enum llog_open_param open_param);
136 int llog_close(const struct lu_env *env, struct llog_handle *cathandle);
137 int llog_get_size(struct llog_handle *loghandle);
139 /* llog_process flags */
140 #define LLOG_FLAG_NODEAMON 0x0001
142 /* llog_cat.c - catalog api */
143 struct llog_process_data {
145 * Any useful data needed while processing catalog. This is
146 * passed later to process callback.
150 * Catalog process callback function, called for each record
155 * Start processing the catalog from startcat/startidx
161 struct llog_process_cat_data {
163 * Temporary stored first_idx while scanning log.
167 * Temporary stored last_idx while scanning log.
172 struct llog_process_cat_args {
174 * Llog context used in recovery thread on OST (recov_thread.c)
176 struct llog_ctxt *lpca_ctxt;
178 * Llog callback used in recovery thread on OST (recov_thread.c)
182 * Data pointer for llog callback.
187 int cat_cancel_cb(const struct lu_env *env, struct llog_handle *cathandle,
188 struct llog_rec_hdr *rec, void *data);
189 int llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle);
190 int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
191 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
193 int llog_cat_cancel_records(const struct lu_env *env,
194 struct llog_handle *cathandle, int count,
195 struct llog_cookie *cookies);
196 int __llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
197 llog_cb_t cb, void *data, int startcat, int startidx,
199 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
200 llog_cb_t cb, void *data, int startcat, int startidx);
201 int llog_cat_process_thread(void *data);
202 int llog_cat_reverse_process(const struct lu_env *env,
203 struct llog_handle *cat_llh, llog_cb_t cb,
205 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
208 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
209 int index, struct obd_device *disk_obd, int count,
210 struct llog_logid *logid, const char *logname,
211 struct llog_operations *op);
212 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
213 struct obd_device *disk_obd, int count, struct llog_logid *logid,
214 struct llog_operations *op);
215 int __llog_ctxt_put(struct llog_ctxt *ctxt);
216 int llog_cleanup(struct llog_ctxt *);
217 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags);
218 int llog_obd_add(const struct lu_env *env, struct llog_ctxt *ctxt,
219 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
220 struct llog_cookie *logcookies, int numcookies);
221 int llog_cancel(const struct lu_env *env, struct llog_ctxt *ctxt,
222 struct lov_stripe_md *lsm, int count,
223 struct llog_cookie *cookies, int flags);
224 int llog_obd_origin_setup(const struct lu_env *env, struct obd_device *obd,
225 struct obd_llog_group *olg, int index,
226 struct obd_device *disk_obd, int count,
227 struct llog_logid *logid, const char *name);
228 int llog_obd_origin_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt);
229 int llog_obd_origin_add(const struct lu_env *env, struct llog_ctxt *ctxt,
230 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
231 struct llog_cookie *logcookies, int numcookies);
233 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
234 struct obd_device *disk_obd, int *idx);
236 int obd_llog_finish(struct obd_device *obd, int count);
239 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
240 int llog_catalog_list(struct obd_device *obd, int count,
241 struct obd_ioctl_data *data);
244 int llog_initiator_connect(struct llog_ctxt *ctxt);
245 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
246 int llog_origin_connect(struct llog_ctxt *ctxt,
247 struct llog_logid *logid, struct llog_gen *gen,
248 struct obd_uuid *uuid);
249 int llog_handle_connect(struct ptlrpc_request *req);
252 int llog_obd_repl_cancel(const struct lu_env *env, struct llog_ctxt *ctxt,
253 struct lov_stripe_md *lsm, int count,
254 struct llog_cookie *cookies, int flags);
255 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp,
257 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
258 struct llog_logid *logid, struct llog_gen *gen,
259 struct obd_uuid *uuid);
261 struct llog_operations {
262 int (*lop_destroy)(const struct lu_env *env,
263 struct llog_handle *handle);
264 int (*lop_next_block)(const struct lu_env *env, struct llog_handle *h,
265 int *curr_idx, int next_idx, __u64 *offset,
267 int (*lop_prev_block)(const struct lu_env *env, struct llog_handle *h,
268 int prev_idx, void *buf, int len);
269 int (*lop_read_header)(const struct lu_env *env,
270 struct llog_handle *handle);
271 int (*lop_setup)(const struct lu_env *env, struct obd_device *obd,
272 struct obd_llog_group *olg, int ctxt_idx,
273 struct obd_device *disk_obd, int count,
274 struct llog_logid *logid, const char *name);
275 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
277 int (*lop_cleanup)(const struct lu_env *env, struct llog_ctxt *ctxt);
278 int (*lop_cancel)(const struct lu_env *env, struct llog_ctxt *ctxt,
279 struct lov_stripe_md *lsm, int count,
280 struct llog_cookie *cookies, int flags);
281 int (*lop_connect)(struct llog_ctxt *ctxt, struct llog_logid *logid,
282 struct llog_gen *gen, struct obd_uuid *uuid);
284 * Any llog file must be opened first using llog_open(). Llog can be
285 * opened by name, logid or without both, in last case the new logid
288 int (*lop_open)(const struct lu_env *env, struct llog_handle *lgh,
289 struct llog_logid *logid, char *name,
290 enum llog_open_param);
292 * Opened llog may not exist and this must be checked where needed using
293 * the llog_exist() call.
295 int (*lop_exist)(struct llog_handle *lgh);
297 * Close llog file and calls llog_free_handle() implicitly.
298 * Any opened llog must be closed by llog_close() call.
300 int (*lop_close)(const struct lu_env *env, struct llog_handle *handle);
302 * Create new llog file. The llog must be opened.
303 * Must be used only for local llog operations.
305 int (*lop_declare_create)(const struct lu_env *env,
306 struct llog_handle *handle,
308 int (*lop_create)(const struct lu_env *env, struct llog_handle *handle,
311 * write new record in llog. It appends records usually but can edit
312 * existing records too.
314 int (*lop_declare_write_rec)(const struct lu_env *env,
315 struct llog_handle *lgh,
316 struct llog_rec_hdr *rec,
317 int idx, struct thandle *th);
318 int (*lop_write_rec)(const struct lu_env *env,
319 struct llog_handle *loghandle,
320 struct llog_rec_hdr *rec,
321 struct llog_cookie *cookie, int cookiecount,
322 void *buf, int idx, struct thandle *th);
324 * Add new record in llog catalog. Does the same as llog_write_rec()
325 * but using llog catalog.
327 int (*lop_declare_add)(const struct lu_env *env,
328 struct llog_handle *lgh,
329 struct llog_rec_hdr *rec, struct thandle *th);
330 int (*lop_add)(const struct lu_env *env, struct llog_handle *lgh,
331 struct llog_rec_hdr *rec, struct llog_cookie *cookie,
332 void *buf, struct thandle *th);
333 /* Old llog_add version, used in MDS-LOV-OSC now and will gone with
334 * LOD/OSP replacement */
335 int (*lop_obd_add)(const struct lu_env *env, struct llog_ctxt *ctxt,
336 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
337 struct llog_cookie *logcookies, int numcookies);
340 /* In-memory descriptor for a log object or log catalog */
342 cfs_rw_semaphore_t lgh_lock;
343 cfs_spinlock_t lgh_hdr_lock; /* protect lgh_hdr data */
344 struct llog_logid lgh_id; /* id of this log */
345 struct llog_log_hdr *lgh_hdr;
346 struct file *lgh_file;
347 struct dt_object *lgh_obj;
349 int lgh_cur_idx; /* used during llog_process */
350 __u64 lgh_cur_offset; /* used during llog_process */
351 struct llog_ctxt *lgh_ctxt;
353 struct plain_handle_data phd;
354 struct cat_handle_data chd;
358 struct llog_operations *lgh_logops;
362 extern struct llog_operations llog_lvfs_ops;
363 int llog_get_cat_list(struct obd_device *disk_obd,
364 char *name, int idx, int count,
365 struct llog_catid *idarray);
367 int llog_put_cat_list(struct obd_device *disk_obd,
368 char *name, int idx, int count, struct llog_catid *idarray);
370 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
371 #define LLOG_CTXT_FLAG_STOP 0x00000002
374 int loc_idx; /* my index the obd array of ctxt's */
375 struct llog_gen loc_gen;
376 struct obd_device *loc_obd; /* points back to the containing obd*/
377 struct obd_llog_group *loc_olg; /* group containing that ctxt */
378 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
379 struct obd_import *loc_imp; /* to use in RPC's: can be backward
381 struct llog_operations *loc_logops;
382 struct llog_handle *loc_handle;
383 struct llog_commit_master *loc_lcm;
384 struct llog_canceld_ctxt *loc_llcd;
385 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
386 cfs_atomic_t loc_refcount;
388 long loc_flags; /* flags, see above defines */
389 struct dt_object *loc_dir;
392 #define LCM_NAME_SIZE 64
394 struct llog_commit_master {
396 * Thread control flags (start, stop, etc.)
400 * Number of llcds onthis lcm.
402 cfs_atomic_t lcm_count;
404 * The refcount for lcm
406 cfs_atomic_t lcm_refcount;
408 * Thread control structure. Used for control commit thread.
410 struct ptlrpcd_ctl lcm_pc;
412 * Lock protecting list of llcds.
414 cfs_spinlock_t lcm_lock;
416 * Llcds in flight for debugging purposes.
418 cfs_list_t lcm_llcds;
420 * Commit thread name buffer. Only used for thread start.
422 char lcm_name[LCM_NAME_SIZE];
425 static inline struct llog_commit_master
426 *lcm_get(struct llog_commit_master *lcm)
428 cfs_atomic_inc(&lcm->lcm_refcount);
433 lcm_put(struct llog_commit_master *lcm)
435 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
436 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
440 struct llog_canceld_ctxt {
442 * Llog context this llcd is attached to. Used for accessing
443 * ->loc_import and others in process of canceling cookies
444 * gathered in this llcd.
446 struct llog_ctxt *llcd_ctxt;
448 * Cancel thread control stucture pointer. Used for accessing
449 * it to see if should stop processing and other needs.
451 struct llog_commit_master *llcd_lcm;
453 * Maximal llcd size. Used in calculations on how much of room
454 * left in llcd to cookie comming cookies.
458 * Link to lcm llcds list.
460 cfs_list_t llcd_list;
462 * Current llcd size while gathering cookies. This should not be
463 * more than ->llcd_size. Used for determining if we need to
464 * send this llcd (if full) and allocate new one. This is also
465 * used for copying new cookie at the end of buffer.
467 int llcd_cookiebytes;
469 * Pointer to the start of cookies buffer.
471 struct llog_cookie llcd_cookies[0];
474 /* ptlrpc/recov_thread.c */
475 extern struct llog_commit_master *llog_recov_thread_init(char *name);
476 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
478 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
479 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
482 static inline void llog_gen_init(struct llog_ctxt *ctxt)
484 struct obd_device *obd = ctxt->loc_exp->exp_obd;
486 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
487 "%s: wrong obt magic %#x\n",
488 obd->obd_name, obd->u.obt.obt_magic);
489 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
490 ctxt->loc_gen.conn_cnt++;
493 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
495 if (a.mnt_cnt < b.mnt_cnt)
497 if (a.mnt_cnt > b.mnt_cnt)
499 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
502 #define LLOG_PROC_BREAK 0x0001
503 #define LLOG_DEL_RECORD 0x0002
505 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
506 struct llog_operations **lop)
511 *lop = ctxt->loc_logops;
518 static inline int llog_handle2ops(struct llog_handle *loghandle,
519 struct llog_operations **lop)
521 if (loghandle == NULL || loghandle->lgh_logops == NULL)
524 *lop = loghandle->lgh_logops;
528 static inline int llog_data_len(int len)
530 return cfs_size_round(len);
533 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
535 cfs_atomic_inc(&ctxt->loc_refcount);
536 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
537 cfs_atomic_read(&ctxt->loc_refcount));
541 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
545 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
546 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
547 cfs_atomic_read(&ctxt->loc_refcount) - 1);
548 __llog_ctxt_put(ctxt);
551 static inline void llog_group_init(struct obd_llog_group *olg, int group)
553 cfs_waitq_init(&olg->olg_waitq);
554 cfs_spin_lock_init(&olg->olg_lock);
555 cfs_mutex_init(&olg->olg_cat_processing);
556 olg->olg_seq = group;
559 static inline void llog_group_set_export(struct obd_llog_group *olg,
560 struct obd_export *exp)
562 LASSERT(exp != NULL);
564 cfs_spin_lock(&olg->olg_lock);
565 if (olg->olg_exp != NULL && olg->olg_exp != exp)
566 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
567 exp->exp_obd->obd_name, olg->olg_seq,
570 cfs_spin_unlock(&olg->olg_lock);
573 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
574 struct llog_ctxt *ctxt, int index)
576 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
578 cfs_spin_lock(&olg->olg_lock);
579 if (olg->olg_ctxts[index] != NULL) {
580 cfs_spin_unlock(&olg->olg_lock);
583 olg->olg_ctxts[index] = ctxt;
584 cfs_spin_unlock(&olg->olg_lock);
588 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
591 struct llog_ctxt *ctxt;
593 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
595 cfs_spin_lock(&olg->olg_lock);
596 if (olg->olg_ctxts[index] == NULL) {
599 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
601 cfs_spin_unlock(&olg->olg_lock);
605 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
608 return llog_group_get_ctxt(&obd->obd_olg, index);
611 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
613 return (olg->olg_ctxts[index] == NULL);
616 static inline int llog_ctxt_null(struct obd_device *obd, int index)
618 return (llog_group_ctxt_null(&obd->obd_olg, index));
621 static inline int llog_destroy(const struct lu_env *env,
622 struct llog_handle *handle)
624 struct llog_operations *lop;
629 rc = llog_handle2ops(handle, &lop);
632 if (lop->lop_destroy == NULL)
635 rc = lop->lop_destroy(env, handle);
639 static inline int llog_next_block(const struct lu_env *env,
640 struct llog_handle *loghandle, int *cur_idx,
641 int next_idx, __u64 *cur_offset, void *buf,
644 struct llog_operations *lop;
649 rc = llog_handle2ops(loghandle, &lop);
652 if (lop->lop_next_block == NULL)
655 rc = lop->lop_next_block(env, loghandle, cur_idx, next_idx,
656 cur_offset, buf, len);
660 static inline int llog_prev_block(const struct lu_env *env,
661 struct llog_handle *loghandle,
662 int prev_idx, void *buf, int len)
664 struct llog_operations *lop;
669 rc = llog_handle2ops(loghandle, &lop);
672 if (lop->lop_prev_block == NULL)
675 rc = lop->lop_prev_block(env, loghandle, prev_idx, buf, len);
679 static inline int llog_connect(struct llog_ctxt *ctxt,
680 struct llog_logid *logid, struct llog_gen *gen,
681 struct obd_uuid *uuid)
683 struct llog_operations *lop;
688 rc = llog_obd2ops(ctxt, &lop);
691 if (lop->lop_connect == NULL)
694 rc = lop->lop_connect(ctxt, logid, gen, uuid);
699 int llog_exist(struct llog_handle *loghandle);
700 int llog_declare_create(const struct lu_env *env,
701 struct llog_handle *loghandle, struct thandle *th);
702 int llog_create(const struct lu_env *env, struct llog_handle *handle,
704 int llog_declare_write_rec(const struct lu_env *env,
705 struct llog_handle *handle,
706 struct llog_rec_hdr *rec, int idx,
708 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
709 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
710 int numcookies, void *buf, int idx, struct thandle *th);
711 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
712 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
713 void *buf, struct thandle *th);
714 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
715 struct llog_rec_hdr *rec, struct thandle *th);
716 int lustre_process_log(struct super_block *sb, char *logname,
717 struct config_llog_instance *cfg);
718 int lustre_end_log(struct super_block *sb, char *logname,
719 struct config_llog_instance *cfg);
720 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
721 struct llog_handle **res, struct llog_logid *logid,
723 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
724 struct llog_logid *logid, char *name);
725 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
726 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
727 int cookiecount, void *buf, int idx);