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 llog_cat_close(const struct lu_env *env, struct llog_handle *cathandle);
188 int llog_cat_add(const struct lu_env *env, struct llog_handle *cathandle,
189 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
191 int llog_cat_cancel_records(const struct lu_env *env,
192 struct llog_handle *cathandle, int count,
193 struct llog_cookie *cookies);
194 int llog_cat_process_or_fork(const struct lu_env *env,
195 struct llog_handle *cat_llh, llog_cb_t cb,
196 void *data, int startcat, int startidx, bool fork);
197 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
198 llog_cb_t cb, void *data, int startcat, int startidx);
199 int llog_cat_process_thread(void *data);
200 int llog_cat_reverse_process(const struct lu_env *env,
201 struct llog_handle *cat_llh, llog_cb_t cb,
203 int llog_cat_init_and_process(const struct lu_env *env,
204 struct llog_handle *llh);
207 int llog_setup(const struct lu_env *env, struct obd_device *obd,
208 struct obd_llog_group *olg, int index,
209 struct obd_device *disk_obd, struct llog_operations *op);
210 int __llog_ctxt_put(const struct lu_env *env, struct llog_ctxt *ctxt);
211 int llog_cleanup(const struct lu_env *env, struct llog_ctxt *);
212 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags);
213 int llog_obd_add(const struct lu_env *env, struct llog_ctxt *ctxt,
214 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
215 struct llog_cookie *logcookies, int numcookies);
216 int llog_cancel(const struct lu_env *env, struct llog_ctxt *ctxt,
217 struct lov_stripe_md *lsm, int count,
218 struct llog_cookie *cookies, int flags);
219 int llog_obd_origin_add(const struct lu_env *env, struct llog_ctxt *ctxt,
220 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
221 struct llog_cookie *logcookies, int numcookies);
223 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
224 struct obd_device *disk_obd, int *idx);
226 int obd_llog_finish(struct obd_device *obd, int count);
229 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
230 int llog_catalog_list(struct obd_device *obd, int count,
231 struct obd_ioctl_data *data);
234 int llog_initiator_connect(struct llog_ctxt *ctxt);
235 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
236 int llog_origin_connect(struct llog_ctxt *ctxt,
237 struct llog_logid *logid, struct llog_gen *gen,
238 struct obd_uuid *uuid);
239 int llog_handle_connect(struct ptlrpc_request *req);
242 int llog_obd_repl_cancel(const struct lu_env *env, struct llog_ctxt *ctxt,
243 struct lov_stripe_md *lsm, int count,
244 struct llog_cookie *cookies, int flags);
245 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp,
247 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
248 struct llog_logid *logid, struct llog_gen *gen,
249 struct obd_uuid *uuid);
251 struct llog_operations {
252 int (*lop_destroy)(const struct lu_env *env,
253 struct llog_handle *handle);
254 int (*lop_next_block)(const struct lu_env *env, struct llog_handle *h,
255 int *curr_idx, int next_idx, __u64 *offset,
257 int (*lop_prev_block)(const struct lu_env *env, struct llog_handle *h,
258 int prev_idx, void *buf, int len);
259 int (*lop_read_header)(const struct lu_env *env,
260 struct llog_handle *handle);
261 int (*lop_setup)(const struct lu_env *env, struct obd_device *obd,
262 struct obd_llog_group *olg, int ctxt_idx,
263 struct obd_device *disk_obd);
264 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
266 int (*lop_cleanup)(const struct lu_env *env, struct llog_ctxt *ctxt);
267 int (*lop_cancel)(const struct lu_env *env, struct llog_ctxt *ctxt,
268 struct lov_stripe_md *lsm, int count,
269 struct llog_cookie *cookies, int flags);
270 int (*lop_connect)(struct llog_ctxt *ctxt, struct llog_logid *logid,
271 struct llog_gen *gen, struct obd_uuid *uuid);
273 * Any llog file must be opened first using llog_open(). Llog can be
274 * opened by name, logid or without both, in last case the new logid
277 int (*lop_open)(const struct lu_env *env, struct llog_handle *lgh,
278 struct llog_logid *logid, char *name,
279 enum llog_open_param);
281 * Opened llog may not exist and this must be checked where needed using
282 * the llog_exist() call.
284 int (*lop_exist)(struct llog_handle *lgh);
286 * Close llog file and calls llog_free_handle() implicitly.
287 * Any opened llog must be closed by llog_close() call.
289 int (*lop_close)(const struct lu_env *env, struct llog_handle *handle);
291 * Create new llog file. The llog must be opened.
292 * Must be used only for local llog operations.
294 int (*lop_declare_create)(const struct lu_env *env,
295 struct llog_handle *handle,
297 int (*lop_create)(const struct lu_env *env, struct llog_handle *handle,
300 * write new record in llog. It appends records usually but can edit
301 * existing records too.
303 int (*lop_declare_write_rec)(const struct lu_env *env,
304 struct llog_handle *lgh,
305 struct llog_rec_hdr *rec,
306 int idx, struct thandle *th);
307 int (*lop_write_rec)(const struct lu_env *env,
308 struct llog_handle *loghandle,
309 struct llog_rec_hdr *rec,
310 struct llog_cookie *cookie, int cookiecount,
311 void *buf, int idx, struct thandle *th);
313 * Add new record in llog catalog. Does the same as llog_write_rec()
314 * but using llog catalog.
316 int (*lop_declare_add)(const struct lu_env *env,
317 struct llog_handle *lgh,
318 struct llog_rec_hdr *rec, struct thandle *th);
319 int (*lop_add)(const struct lu_env *env, struct llog_handle *lgh,
320 struct llog_rec_hdr *rec, struct llog_cookie *cookie,
321 void *buf, struct thandle *th);
322 /* Old llog_add version, used in MDS-LOV-OSC now and will gone with
323 * LOD/OSP replacement */
324 int (*lop_obd_add)(const struct lu_env *env, struct llog_ctxt *ctxt,
325 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
326 struct llog_cookie *logcookies, int numcookies);
329 /* In-memory descriptor for a log object or log catalog */
331 cfs_rw_semaphore_t lgh_lock;
332 cfs_spinlock_t lgh_hdr_lock; /* protect lgh_hdr data */
333 struct llog_logid lgh_id; /* id of this log */
334 struct llog_log_hdr *lgh_hdr;
335 struct file *lgh_file;
336 struct dt_object *lgh_obj;
338 int lgh_cur_idx; /* used during llog_process */
339 __u64 lgh_cur_offset; /* used during llog_process */
340 struct llog_ctxt *lgh_ctxt;
342 struct plain_handle_data phd;
343 struct cat_handle_data chd;
347 struct llog_operations *lgh_logops;
351 extern struct llog_operations llog_lvfs_ops;
352 int llog_get_cat_list(struct obd_device *disk_obd,
353 char *name, int idx, int count,
354 struct llog_catid *idarray);
356 int llog_put_cat_list(struct obd_device *disk_obd,
357 char *name, int idx, int count, struct llog_catid *idarray);
359 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
360 #define LLOG_CTXT_FLAG_STOP 0x00000002
363 int loc_idx; /* my index the obd array of ctxt's */
364 struct llog_gen loc_gen;
365 struct obd_device *loc_obd; /* points back to the containing obd*/
366 struct obd_llog_group *loc_olg; /* group containing that ctxt */
367 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
368 struct obd_import *loc_imp; /* to use in RPC's: can be backward
370 struct llog_operations *loc_logops;
371 struct llog_handle *loc_handle;
372 struct llog_commit_master *loc_lcm;
373 struct llog_canceld_ctxt *loc_llcd;
374 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
375 cfs_atomic_t loc_refcount;
377 long loc_flags; /* flags, see above defines */
378 struct dt_object *loc_dir;
381 #define LCM_NAME_SIZE 64
383 struct llog_commit_master {
385 * Thread control flags (start, stop, etc.)
389 * Number of llcds onthis lcm.
391 cfs_atomic_t lcm_count;
393 * The refcount for lcm
395 cfs_atomic_t lcm_refcount;
397 * Thread control structure. Used for control commit thread.
399 struct ptlrpcd_ctl lcm_pc;
401 * Lock protecting list of llcds.
403 cfs_spinlock_t lcm_lock;
405 * Llcds in flight for debugging purposes.
407 cfs_list_t lcm_llcds;
409 * Commit thread name buffer. Only used for thread start.
411 char lcm_name[LCM_NAME_SIZE];
414 static inline struct llog_commit_master
415 *lcm_get(struct llog_commit_master *lcm)
417 cfs_atomic_inc(&lcm->lcm_refcount);
422 lcm_put(struct llog_commit_master *lcm)
424 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
425 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
429 struct llog_canceld_ctxt {
431 * Llog context this llcd is attached to. Used for accessing
432 * ->loc_import and others in process of canceling cookies
433 * gathered in this llcd.
435 struct llog_ctxt *llcd_ctxt;
437 * Cancel thread control stucture pointer. Used for accessing
438 * it to see if should stop processing and other needs.
440 struct llog_commit_master *llcd_lcm;
442 * Maximal llcd size. Used in calculations on how much of room
443 * left in llcd to cookie comming cookies.
447 * Link to lcm llcds list.
449 cfs_list_t llcd_list;
451 * Current llcd size while gathering cookies. This should not be
452 * more than ->llcd_size. Used for determining if we need to
453 * send this llcd (if full) and allocate new one. This is also
454 * used for copying new cookie at the end of buffer.
456 int llcd_cookiebytes;
458 * Pointer to the start of cookies buffer.
460 struct llog_cookie llcd_cookies[0];
463 /* ptlrpc/recov_thread.c */
464 extern struct llog_commit_master *llog_recov_thread_init(char *name);
465 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
467 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
468 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
471 static inline void llog_gen_init(struct llog_ctxt *ctxt)
473 struct obd_device *obd = ctxt->loc_exp->exp_obd;
475 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
476 "%s: wrong obt magic %#x\n",
477 obd->obd_name, obd->u.obt.obt_magic);
478 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
479 ctxt->loc_gen.conn_cnt++;
482 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
484 if (a.mnt_cnt < b.mnt_cnt)
486 if (a.mnt_cnt > b.mnt_cnt)
488 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
491 #define LLOG_PROC_BREAK 0x0001
492 #define LLOG_DEL_RECORD 0x0002
494 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
495 struct llog_operations **lop)
500 *lop = ctxt->loc_logops;
507 static inline int llog_handle2ops(struct llog_handle *loghandle,
508 struct llog_operations **lop)
510 if (loghandle == NULL || loghandle->lgh_logops == NULL)
513 *lop = loghandle->lgh_logops;
517 static inline int llog_data_len(int len)
519 return cfs_size_round(len);
522 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
524 cfs_atomic_inc(&ctxt->loc_refcount);
525 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
526 cfs_atomic_read(&ctxt->loc_refcount));
530 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
534 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
535 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
536 cfs_atomic_read(&ctxt->loc_refcount) - 1);
537 __llog_ctxt_put(NULL, ctxt);
540 static inline void llog_group_init(struct obd_llog_group *olg, int group)
542 cfs_waitq_init(&olg->olg_waitq);
543 cfs_spin_lock_init(&olg->olg_lock);
544 cfs_mutex_init(&olg->olg_cat_processing);
545 olg->olg_seq = group;
548 static inline void llog_group_set_export(struct obd_llog_group *olg,
549 struct obd_export *exp)
551 LASSERT(exp != NULL);
553 cfs_spin_lock(&olg->olg_lock);
554 if (olg->olg_exp != NULL && olg->olg_exp != exp)
555 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
556 exp->exp_obd->obd_name, olg->olg_seq,
559 cfs_spin_unlock(&olg->olg_lock);
562 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
563 struct llog_ctxt *ctxt, int index)
565 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
567 cfs_spin_lock(&olg->olg_lock);
568 if (olg->olg_ctxts[index] != NULL) {
569 cfs_spin_unlock(&olg->olg_lock);
572 olg->olg_ctxts[index] = ctxt;
573 cfs_spin_unlock(&olg->olg_lock);
577 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
580 struct llog_ctxt *ctxt;
582 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
584 cfs_spin_lock(&olg->olg_lock);
585 if (olg->olg_ctxts[index] == NULL) {
588 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
590 cfs_spin_unlock(&olg->olg_lock);
594 static inline void llog_group_clear_ctxt(struct obd_llog_group *olg, int index)
596 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
597 cfs_spin_lock(&olg->olg_lock);
598 olg->olg_ctxts[index] = NULL;
599 cfs_spin_unlock(&olg->olg_lock);
602 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
605 return llog_group_get_ctxt(&obd->obd_olg, index);
608 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
610 return (olg->olg_ctxts[index] == NULL);
613 static inline int llog_ctxt_null(struct obd_device *obd, int index)
615 return (llog_group_ctxt_null(&obd->obd_olg, index));
618 static inline int llog_destroy(const struct lu_env *env,
619 struct llog_handle *handle)
621 struct llog_operations *lop;
626 rc = llog_handle2ops(handle, &lop);
629 if (lop->lop_destroy == NULL)
632 rc = lop->lop_destroy(env, handle);
636 static inline int llog_next_block(const struct lu_env *env,
637 struct llog_handle *loghandle, int *cur_idx,
638 int next_idx, __u64 *cur_offset, void *buf,
641 struct llog_operations *lop;
646 rc = llog_handle2ops(loghandle, &lop);
649 if (lop->lop_next_block == NULL)
652 rc = lop->lop_next_block(env, loghandle, cur_idx, next_idx,
653 cur_offset, buf, len);
657 static inline int llog_prev_block(const struct lu_env *env,
658 struct llog_handle *loghandle,
659 int prev_idx, void *buf, int len)
661 struct llog_operations *lop;
666 rc = llog_handle2ops(loghandle, &lop);
669 if (lop->lop_prev_block == NULL)
672 rc = lop->lop_prev_block(env, loghandle, prev_idx, buf, len);
676 static inline int llog_connect(struct llog_ctxt *ctxt,
677 struct llog_logid *logid, struct llog_gen *gen,
678 struct obd_uuid *uuid)
680 struct llog_operations *lop;
685 rc = llog_obd2ops(ctxt, &lop);
688 if (lop->lop_connect == NULL)
691 rc = lop->lop_connect(ctxt, logid, gen, uuid);
696 int llog_exist(struct llog_handle *loghandle);
697 int llog_declare_create(const struct lu_env *env,
698 struct llog_handle *loghandle, struct thandle *th);
699 int llog_create(const struct lu_env *env, struct llog_handle *handle,
701 int llog_declare_write_rec(const struct lu_env *env,
702 struct llog_handle *handle,
703 struct llog_rec_hdr *rec, int idx,
705 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
706 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
707 int numcookies, void *buf, int idx, struct thandle *th);
708 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
709 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
710 void *buf, struct thandle *th);
711 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
712 struct llog_rec_hdr *rec, struct thandle *th);
713 int lustre_process_log(struct super_block *sb, char *logname,
714 struct config_llog_instance *cfg);
715 int lustre_end_log(struct super_block *sb, char *logname,
716 struct config_llog_instance *cfg);
717 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
718 struct llog_handle **res, struct llog_logid *logid,
720 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
721 struct llog_logid *logid, char *name);
722 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
723 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
724 int cookiecount, void *buf, int idx);