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(const struct lu_env *env, struct llog_ctxt *ctxt, int cmd,
230 struct obd_ioctl_data *data);
231 int llog_catalog_list(struct obd_device *obd, int count,
232 struct obd_ioctl_data *data);
235 int llog_initiator_connect(struct llog_ctxt *ctxt);
236 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
237 int llog_origin_connect(struct llog_ctxt *ctxt,
238 struct llog_logid *logid, struct llog_gen *gen,
239 struct obd_uuid *uuid);
240 int llog_handle_connect(struct ptlrpc_request *req);
243 int llog_obd_repl_cancel(const struct lu_env *env, struct llog_ctxt *ctxt,
244 struct lov_stripe_md *lsm, int count,
245 struct llog_cookie *cookies, int flags);
246 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp,
248 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
249 struct llog_logid *logid, struct llog_gen *gen,
250 struct obd_uuid *uuid);
252 struct llog_operations {
253 int (*lop_destroy)(const struct lu_env *env,
254 struct llog_handle *handle);
255 int (*lop_next_block)(const struct lu_env *env, struct llog_handle *h,
256 int *curr_idx, int next_idx, __u64 *offset,
258 int (*lop_prev_block)(const struct lu_env *env, struct llog_handle *h,
259 int prev_idx, void *buf, int len);
260 int (*lop_read_header)(const struct lu_env *env,
261 struct llog_handle *handle);
262 int (*lop_setup)(const struct lu_env *env, struct obd_device *obd,
263 struct obd_llog_group *olg, int ctxt_idx,
264 struct obd_device *disk_obd);
265 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
267 int (*lop_cleanup)(const struct lu_env *env, struct llog_ctxt *ctxt);
268 int (*lop_cancel)(const struct lu_env *env, struct llog_ctxt *ctxt,
269 struct lov_stripe_md *lsm, int count,
270 struct llog_cookie *cookies, int flags);
271 int (*lop_connect)(struct llog_ctxt *ctxt, struct llog_logid *logid,
272 struct llog_gen *gen, struct obd_uuid *uuid);
274 * Any llog file must be opened first using llog_open(). Llog can be
275 * opened by name, logid or without both, in last case the new logid
278 int (*lop_open)(const struct lu_env *env, struct llog_handle *lgh,
279 struct llog_logid *logid, char *name,
280 enum llog_open_param);
282 * Opened llog may not exist and this must be checked where needed using
283 * the llog_exist() call.
285 int (*lop_exist)(struct llog_handle *lgh);
287 * Close llog file and calls llog_free_handle() implicitly.
288 * Any opened llog must be closed by llog_close() call.
290 int (*lop_close)(const struct lu_env *env, struct llog_handle *handle);
292 * Create new llog file. The llog must be opened.
293 * Must be used only for local llog operations.
295 int (*lop_declare_create)(const struct lu_env *env,
296 struct llog_handle *handle,
298 int (*lop_create)(const struct lu_env *env, struct llog_handle *handle,
301 * write new record in llog. It appends records usually but can edit
302 * existing records too.
304 int (*lop_declare_write_rec)(const struct lu_env *env,
305 struct llog_handle *lgh,
306 struct llog_rec_hdr *rec,
307 int idx, struct thandle *th);
308 int (*lop_write_rec)(const struct lu_env *env,
309 struct llog_handle *loghandle,
310 struct llog_rec_hdr *rec,
311 struct llog_cookie *cookie, int cookiecount,
312 void *buf, int idx, struct thandle *th);
314 * Add new record in llog catalog. Does the same as llog_write_rec()
315 * but using llog catalog.
317 int (*lop_declare_add)(const struct lu_env *env,
318 struct llog_handle *lgh,
319 struct llog_rec_hdr *rec, struct thandle *th);
320 int (*lop_add)(const struct lu_env *env, struct llog_handle *lgh,
321 struct llog_rec_hdr *rec, struct llog_cookie *cookie,
322 void *buf, struct thandle *th);
323 /* Old llog_add version, used in MDS-LOV-OSC now and will gone with
324 * LOD/OSP replacement */
325 int (*lop_obd_add)(const struct lu_env *env, struct llog_ctxt *ctxt,
326 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
327 struct llog_cookie *logcookies, int numcookies);
330 /* In-memory descriptor for a log object or log catalog */
332 cfs_rw_semaphore_t lgh_lock;
333 cfs_spinlock_t lgh_hdr_lock; /* protect lgh_hdr data */
334 struct llog_logid lgh_id; /* id of this log */
335 struct llog_log_hdr *lgh_hdr;
336 struct file *lgh_file;
337 struct dt_object *lgh_obj;
339 int lgh_cur_idx; /* used during llog_process */
340 __u64 lgh_cur_offset; /* used during llog_process */
341 struct llog_ctxt *lgh_ctxt;
343 struct plain_handle_data phd;
344 struct cat_handle_data chd;
348 struct llog_operations *lgh_logops;
352 extern struct llog_operations llog_lvfs_ops;
353 int llog_get_cat_list(struct obd_device *disk_obd,
354 char *name, int idx, int count,
355 struct llog_catid *idarray);
357 int llog_put_cat_list(struct obd_device *disk_obd,
358 char *name, int idx, int count, struct llog_catid *idarray);
361 extern struct llog_operations llog_osd_ops;
362 int llog_osd_get_cat_list(const struct lu_env *env, struct dt_device *d,
363 int idx, int count, struct llog_catid *idarray);
364 int llog_osd_put_cat_list(const struct lu_env *env, struct dt_device *d,
365 int idx, int count, struct llog_catid *idarray);
367 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
368 #define LLOG_CTXT_FLAG_STOP 0x00000002
371 int loc_idx; /* my index the obd array of ctxt's */
372 struct llog_gen loc_gen;
373 struct obd_device *loc_obd; /* points back to the containing obd*/
374 struct obd_llog_group *loc_olg; /* group containing that ctxt */
375 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
376 struct obd_import *loc_imp; /* to use in RPC's: can be backward
378 struct llog_operations *loc_logops;
379 struct llog_handle *loc_handle;
380 struct llog_commit_master *loc_lcm;
381 struct llog_canceld_ctxt *loc_llcd;
382 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
383 cfs_atomic_t loc_refcount;
385 long loc_flags; /* flags, see above defines */
386 struct dt_object *loc_dir;
389 #define LCM_NAME_SIZE 64
391 struct llog_commit_master {
393 * Thread control flags (start, stop, etc.)
397 * Number of llcds onthis lcm.
399 cfs_atomic_t lcm_count;
401 * The refcount for lcm
403 cfs_atomic_t lcm_refcount;
405 * Thread control structure. Used for control commit thread.
407 struct ptlrpcd_ctl lcm_pc;
409 * Lock protecting list of llcds.
411 cfs_spinlock_t lcm_lock;
413 * Llcds in flight for debugging purposes.
415 cfs_list_t lcm_llcds;
417 * Commit thread name buffer. Only used for thread start.
419 char lcm_name[LCM_NAME_SIZE];
422 static inline struct llog_commit_master
423 *lcm_get(struct llog_commit_master *lcm)
425 cfs_atomic_inc(&lcm->lcm_refcount);
430 lcm_put(struct llog_commit_master *lcm)
432 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
433 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
437 struct llog_canceld_ctxt {
439 * Llog context this llcd is attached to. Used for accessing
440 * ->loc_import and others in process of canceling cookies
441 * gathered in this llcd.
443 struct llog_ctxt *llcd_ctxt;
445 * Cancel thread control stucture pointer. Used for accessing
446 * it to see if should stop processing and other needs.
448 struct llog_commit_master *llcd_lcm;
450 * Maximal llcd size. Used in calculations on how much of room
451 * left in llcd to cookie comming cookies.
455 * Link to lcm llcds list.
457 cfs_list_t llcd_list;
459 * Current llcd size while gathering cookies. This should not be
460 * more than ->llcd_size. Used for determining if we need to
461 * send this llcd (if full) and allocate new one. This is also
462 * used for copying new cookie at the end of buffer.
464 int llcd_cookiebytes;
466 * Pointer to the start of cookies buffer.
468 struct llog_cookie llcd_cookies[0];
471 /* ptlrpc/recov_thread.c */
472 extern struct llog_commit_master *llog_recov_thread_init(char *name);
473 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
475 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
476 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
479 static inline void llog_gen_init(struct llog_ctxt *ctxt)
481 struct obd_device *obd = ctxt->loc_exp->exp_obd;
483 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
484 "%s: wrong obt magic %#x\n",
485 obd->obd_name, obd->u.obt.obt_magic);
486 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
487 ctxt->loc_gen.conn_cnt++;
490 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
492 if (a.mnt_cnt < b.mnt_cnt)
494 if (a.mnt_cnt > b.mnt_cnt)
496 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
499 #define LLOG_PROC_BREAK 0x0001
500 #define LLOG_DEL_RECORD 0x0002
502 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
503 struct llog_operations **lop)
508 *lop = ctxt->loc_logops;
515 static inline int llog_handle2ops(struct llog_handle *loghandle,
516 struct llog_operations **lop)
518 if (loghandle == NULL || loghandle->lgh_logops == NULL)
521 *lop = loghandle->lgh_logops;
525 static inline int llog_data_len(int len)
527 return cfs_size_round(len);
530 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
532 cfs_atomic_inc(&ctxt->loc_refcount);
533 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
534 cfs_atomic_read(&ctxt->loc_refcount));
538 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
542 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
543 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
544 cfs_atomic_read(&ctxt->loc_refcount) - 1);
545 __llog_ctxt_put(NULL, ctxt);
548 static inline void llog_group_init(struct obd_llog_group *olg, int group)
550 cfs_waitq_init(&olg->olg_waitq);
551 cfs_spin_lock_init(&olg->olg_lock);
552 cfs_mutex_init(&olg->olg_cat_processing);
553 olg->olg_seq = group;
556 static inline void llog_group_set_export(struct obd_llog_group *olg,
557 struct obd_export *exp)
559 LASSERT(exp != NULL);
561 cfs_spin_lock(&olg->olg_lock);
562 if (olg->olg_exp != NULL && olg->olg_exp != exp)
563 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
564 exp->exp_obd->obd_name, olg->olg_seq,
567 cfs_spin_unlock(&olg->olg_lock);
570 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
571 struct llog_ctxt *ctxt, int index)
573 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
575 cfs_spin_lock(&olg->olg_lock);
576 if (olg->olg_ctxts[index] != NULL) {
577 cfs_spin_unlock(&olg->olg_lock);
580 olg->olg_ctxts[index] = ctxt;
581 cfs_spin_unlock(&olg->olg_lock);
585 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
588 struct llog_ctxt *ctxt;
590 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
592 cfs_spin_lock(&olg->olg_lock);
593 if (olg->olg_ctxts[index] == NULL) {
596 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
598 cfs_spin_unlock(&olg->olg_lock);
602 static inline void llog_group_clear_ctxt(struct obd_llog_group *olg, int index)
604 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
605 cfs_spin_lock(&olg->olg_lock);
606 olg->olg_ctxts[index] = NULL;
607 cfs_spin_unlock(&olg->olg_lock);
610 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
613 return llog_group_get_ctxt(&obd->obd_olg, index);
616 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
618 return (olg->olg_ctxts[index] == NULL);
621 static inline int llog_ctxt_null(struct obd_device *obd, int index)
623 return (llog_group_ctxt_null(&obd->obd_olg, index));
626 static inline int llog_destroy(const struct lu_env *env,
627 struct llog_handle *handle)
629 struct llog_operations *lop;
634 rc = llog_handle2ops(handle, &lop);
637 if (lop->lop_destroy == NULL)
640 rc = lop->lop_destroy(env, handle);
644 static inline int llog_next_block(const struct lu_env *env,
645 struct llog_handle *loghandle, int *cur_idx,
646 int next_idx, __u64 *cur_offset, void *buf,
649 struct llog_operations *lop;
654 rc = llog_handle2ops(loghandle, &lop);
657 if (lop->lop_next_block == NULL)
660 rc = lop->lop_next_block(env, loghandle, cur_idx, next_idx,
661 cur_offset, buf, len);
665 static inline int llog_prev_block(const struct lu_env *env,
666 struct llog_handle *loghandle,
667 int prev_idx, void *buf, int len)
669 struct llog_operations *lop;
674 rc = llog_handle2ops(loghandle, &lop);
677 if (lop->lop_prev_block == NULL)
680 rc = lop->lop_prev_block(env, loghandle, prev_idx, buf, len);
684 static inline int llog_connect(struct llog_ctxt *ctxt,
685 struct llog_logid *logid, struct llog_gen *gen,
686 struct obd_uuid *uuid)
688 struct llog_operations *lop;
693 rc = llog_obd2ops(ctxt, &lop);
696 if (lop->lop_connect == NULL)
699 rc = lop->lop_connect(ctxt, logid, gen, uuid);
704 int llog_exist(struct llog_handle *loghandle);
705 int llog_declare_create(const struct lu_env *env,
706 struct llog_handle *loghandle, struct thandle *th);
707 int llog_create(const struct lu_env *env, struct llog_handle *handle,
709 int llog_declare_write_rec(const struct lu_env *env,
710 struct llog_handle *handle,
711 struct llog_rec_hdr *rec, int idx,
713 int llog_write_rec(const struct lu_env *env, struct llog_handle *handle,
714 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
715 int numcookies, void *buf, int idx, struct thandle *th);
716 int llog_add(const struct lu_env *env, struct llog_handle *lgh,
717 struct llog_rec_hdr *rec, struct llog_cookie *logcookies,
718 void *buf, struct thandle *th);
719 int llog_declare_add(const struct lu_env *env, struct llog_handle *lgh,
720 struct llog_rec_hdr *rec, struct thandle *th);
721 int lustre_process_log(struct super_block *sb, char *logname,
722 struct config_llog_instance *cfg);
723 int lustre_end_log(struct super_block *sb, char *logname,
724 struct config_llog_instance *cfg);
725 int llog_open_create(const struct lu_env *env, struct llog_ctxt *ctxt,
726 struct llog_handle **res, struct llog_logid *logid,
728 int llog_erase(const struct lu_env *env, struct llog_ctxt *ctxt,
729 struct llog_logid *logid, char *name);
730 int llog_write(const struct lu_env *env, struct llog_handle *loghandle,
731 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
732 int cookiecount, void *buf, int idx);