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>
73 #define LOG_NAME_LIMIT(logname, name) \
74 snprintf(logname, sizeof(logname), "LOGS/%s", name)
75 #define LLOG_EEMPTY 4711
77 struct plain_handle_data {
79 struct llog_handle *phd_cat_handle;
80 struct llog_cookie phd_cookie; /* cookie of this log in its cat */
84 struct cat_handle_data {
86 struct llog_handle *chd_current_log; /* currently open log */
89 /* In-memory descriptor for a log object or log catalog */
91 cfs_rw_semaphore_t lgh_lock;
92 struct llog_logid lgh_id; /* id of this log */
93 struct llog_log_hdr *lgh_hdr;
94 struct file *lgh_file;
96 int lgh_cur_idx; /* used during llog_process */
97 __u64 lgh_cur_offset; /* used during llog_process */
98 struct llog_ctxt *lgh_ctxt;
100 struct plain_handle_data phd;
101 struct cat_handle_data chd;
105 /* llog.c - general API */
106 typedef int (*llog_cb_t)(struct llog_handle *, struct llog_rec_hdr *, void *);
107 typedef int (*llog_fill_rec_cb_t)(struct llog_rec_hdr *rec, void *data);
108 extern struct llog_handle *llog_alloc_handle(void);
109 int llog_init_handle(struct llog_handle *handle, int flags,
110 struct obd_uuid *uuid);
111 extern void llog_free_handle(struct llog_handle *handle);
112 int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
113 void *data, void *catdata);
114 int llog_process_flags(struct llog_handle *loghandle, llog_cb_t cb,
115 void *data, void *catdata, int flags);
116 int llog_reverse_process(struct llog_handle *loghandle, llog_cb_t cb,
117 void *data, void *catdata);
118 extern int llog_cancel_rec(struct llog_handle *loghandle, int index);
119 extern int llog_close(struct llog_handle *cathandle);
120 extern int llog_get_size(struct llog_handle *loghandle);
122 /* llog_process flags */
123 #define LLOG_FLAG_NODEAMON 0x0001
125 /* llog_cat.c - catalog api */
126 struct llog_process_data {
128 * Any useful data needed while processing catalog. This is
129 * passed later to process callback.
133 * Catalog process callback function, called for each record
138 * Start processing the catalog from startcat/startidx
142 int lpd_flags; /** llog_process flags */
145 struct llog_process_cat_data {
147 * Temporary stored first_idx while scanning log.
151 * Temporary stored last_idx while scanning log.
156 struct llog_process_cat_args {
158 * Llog context used in recovery thread on OST (recov_thread.c)
160 struct llog_ctxt *lpca_ctxt;
162 * Llog callback used in recovery thread on OST (recov_thread.c)
166 * Data pointer for llog callback.
171 int llog_cat_put(struct llog_handle *cathandle);
172 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
173 struct llog_cookie *reccookie, void *buf);
174 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
175 struct llog_cookie *cookies);
176 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
177 int startcat, int startidx);
178 int llog_cat_process_flags(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
179 int flags, int startcat, int startidx);
180 int llog_cat_process_thread(void *data);
181 int llog_cat_reverse_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data);
182 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
185 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
186 int index, struct obd_device *disk_obd, int count,
187 struct llog_logid *logid, const char *logname,
188 struct llog_operations *op);
189 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
190 struct obd_device *disk_obd, int count, struct llog_logid *logid,
191 struct llog_operations *op);
192 int __llog_ctxt_put(struct llog_ctxt *ctxt);
193 int llog_cleanup(struct llog_ctxt *);
194 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags);
195 int llog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
196 struct lov_stripe_md *lsm, struct llog_cookie *logcookies,
198 int llog_cancel(struct llog_ctxt *, struct lov_stripe_md *lsm,
199 int count, struct llog_cookie *cookies, int flags);
201 int llog_obd_origin_setup(struct obd_device *obd, struct obd_llog_group *olg,
202 int index, struct obd_device *disk_obd, int count,
203 struct llog_logid *logid, const char *name);
204 int llog_obd_origin_cleanup(struct llog_ctxt *ctxt);
205 int llog_obd_origin_add(struct llog_ctxt *ctxt,
206 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
207 struct llog_cookie *logcookies, int numcookies);
209 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
210 struct obd_device *disk_obd, int *idx);
212 int obd_llog_finish(struct obd_device *obd, int count);
215 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
216 int llog_catalog_list(struct obd_device *obd, int count,
217 struct obd_ioctl_data *data);
220 int llog_initiator_connect(struct llog_ctxt *ctxt);
221 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
222 int llog_origin_connect(struct llog_ctxt *ctxt,
223 struct llog_logid *logid, struct llog_gen *gen,
224 struct obd_uuid *uuid);
225 int llog_handle_connect(struct ptlrpc_request *req);
228 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
229 struct lov_stripe_md *lsm, int count,
230 struct llog_cookie *cookies, int flags);
231 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp,
233 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
234 struct llog_logid *logid, struct llog_gen *gen,
235 struct obd_uuid *uuid);
237 struct llog_operations {
238 int (*lop_write_rec)(struct llog_handle *loghandle,
239 struct llog_rec_hdr *rec,
240 struct llog_cookie *logcookies, int numcookies,
242 int (*lop_destroy)(struct llog_handle *handle);
243 int (*lop_next_block)(struct llog_handle *h, int *curr_idx,
244 int next_idx, __u64 *offset, void *buf, int len);
245 int (*lop_prev_block)(struct llog_handle *h,
246 int prev_idx, void *buf, int len);
247 int (*lop_create)(struct llog_ctxt *ctxt, struct llog_handle **,
248 struct llog_logid *logid, char *name);
249 int (*lop_close)(struct llog_handle *handle);
250 int (*lop_read_header)(struct llog_handle *handle);
252 int (*lop_setup)(struct obd_device *obd, struct obd_llog_group *olg,
253 int ctxt_idx, struct obd_device *disk_obd, int count,
254 struct llog_logid *logid, const char *name);
255 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
257 int (*lop_cleanup)(struct llog_ctxt *ctxt);
258 int (*lop_add)(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
259 struct lov_stripe_md *lsm,
260 struct llog_cookie *logcookies, int numcookies);
261 int (*lop_cancel)(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
262 int count, struct llog_cookie *cookies, int flags);
263 int (*lop_connect)(struct llog_ctxt *ctxt,
264 struct llog_logid *logid, struct llog_gen *gen,
265 struct obd_uuid *uuid);
266 /* XXX add 2 more: commit callbacks and llog recovery functions */
270 extern struct llog_operations llog_lvfs_ops;
271 int llog_get_cat_list(struct obd_device *disk_obd,
272 char *name, int idx, int count,
273 struct llog_catid *idarray);
275 int llog_put_cat_list(struct obd_device *disk_obd,
276 char *name, int idx, int count, struct llog_catid *idarray);
278 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
279 #define LLOG_CTXT_FLAG_STOP 0x00000002
282 int loc_idx; /* my index the obd array of ctxt's */
283 struct llog_gen loc_gen;
284 struct obd_device *loc_obd; /* points back to the containing obd*/
285 struct obd_llog_group *loc_olg; /* group containing that ctxt */
286 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
287 struct obd_import *loc_imp; /* to use in RPC's: can be backward
289 struct llog_operations *loc_logops;
290 struct llog_handle *loc_handle;
291 struct llog_commit_master *loc_lcm;
292 struct llog_canceld_ctxt *loc_llcd;
293 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
294 cfs_atomic_t loc_refcount;
296 long loc_flags; /* flags, see above defines */
299 #define LCM_NAME_SIZE 64
301 struct llog_commit_master {
303 * Thread control flags (start, stop, etc.)
307 * Number of llcds onthis lcm.
309 cfs_atomic_t lcm_count;
311 * The refcount for lcm
313 cfs_atomic_t lcm_refcount;
315 * Thread control structure. Used for control commit thread.
317 struct ptlrpcd_ctl lcm_pc;
319 * Lock protecting list of llcds.
321 cfs_spinlock_t lcm_lock;
323 * Llcds in flight for debugging purposes.
325 cfs_list_t lcm_llcds;
327 * Commit thread name buffer. Only used for thread start.
329 char lcm_name[LCM_NAME_SIZE];
332 static inline struct llog_commit_master
333 *lcm_get(struct llog_commit_master *lcm)
335 cfs_atomic_inc(&lcm->lcm_refcount);
340 lcm_put(struct llog_commit_master *lcm)
342 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
343 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
347 struct llog_canceld_ctxt {
349 * Llog context this llcd is attached to. Used for accessing
350 * ->loc_import and others in process of canceling cookies
351 * gathered in this llcd.
353 struct llog_ctxt *llcd_ctxt;
355 * Cancel thread control stucture pointer. Used for accessing
356 * it to see if should stop processing and other needs.
358 struct llog_commit_master *llcd_lcm;
360 * Maximal llcd size. Used in calculations on how much of room
361 * left in llcd to cookie comming cookies.
365 * Link to lcm llcds list.
367 cfs_list_t llcd_list;
369 * Current llcd size while gathering cookies. This should not be
370 * more than ->llcd_size. Used for determining if we need to
371 * send this llcd (if full) and allocate new one. This is also
372 * used for copying new cookie at the end of buffer.
374 int llcd_cookiebytes;
376 * Pointer to the start of cookies buffer.
378 struct llog_cookie llcd_cookies[0];
381 /* ptlrpc/recov_thread.c */
382 extern struct llog_commit_master *llog_recov_thread_init(char *name);
383 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
385 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
386 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
389 static inline void llog_gen_init(struct llog_ctxt *ctxt)
391 struct obd_device *obd = ctxt->loc_exp->exp_obd;
393 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
394 "%s: wrong obt magic %#x\n",
395 obd->obd_name, obd->u.obt.obt_magic);
396 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
397 ctxt->loc_gen.conn_cnt++;
400 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
402 if (a.mnt_cnt < b.mnt_cnt)
404 if (a.mnt_cnt > b.mnt_cnt)
406 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
409 #define LLOG_PROC_BREAK 0x0001
410 #define LLOG_DEL_RECORD 0x0002
412 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
413 struct llog_operations **lop)
418 *lop = ctxt->loc_logops;
425 static inline int llog_handle2ops(struct llog_handle *loghandle,
426 struct llog_operations **lop)
428 if (loghandle == NULL)
431 return llog_obd2ops(loghandle->lgh_ctxt, lop);
434 static inline int llog_data_len(int len)
436 return cfs_size_round(len);
439 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
441 cfs_atomic_inc(&ctxt->loc_refcount);
442 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
443 cfs_atomic_read(&ctxt->loc_refcount));
447 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
451 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
452 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
453 cfs_atomic_read(&ctxt->loc_refcount) - 1);
454 __llog_ctxt_put(ctxt);
457 static inline void llog_group_init(struct obd_llog_group *olg, int group)
459 cfs_waitq_init(&olg->olg_waitq);
460 cfs_spin_lock_init(&olg->olg_lock);
461 cfs_mutex_init(&olg->olg_cat_processing);
462 olg->olg_seq = group;
465 static inline void llog_group_set_export(struct obd_llog_group *olg,
466 struct obd_export *exp)
468 LASSERT(exp != NULL);
470 cfs_spin_lock(&olg->olg_lock);
471 if (olg->olg_exp != NULL && olg->olg_exp != exp)
472 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
473 exp->exp_obd->obd_name, olg->olg_seq,
476 cfs_spin_unlock(&olg->olg_lock);
479 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
480 struct llog_ctxt *ctxt, int index)
482 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
484 cfs_spin_lock(&olg->olg_lock);
485 if (olg->olg_ctxts[index] != NULL) {
486 cfs_spin_unlock(&olg->olg_lock);
489 olg->olg_ctxts[index] = ctxt;
490 cfs_spin_unlock(&olg->olg_lock);
494 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
497 struct llog_ctxt *ctxt;
499 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
501 cfs_spin_lock(&olg->olg_lock);
502 if (olg->olg_ctxts[index] == NULL) {
505 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
507 cfs_spin_unlock(&olg->olg_lock);
511 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
514 return llog_group_get_ctxt(&obd->obd_olg, index);
517 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
519 return (olg->olg_ctxts[index] == NULL);
522 static inline int llog_ctxt_null(struct obd_device *obd, int index)
524 return (llog_group_ctxt_null(&obd->obd_olg, index));
527 static inline int llog_write_rec(struct llog_handle *handle,
528 struct llog_rec_hdr *rec,
529 struct llog_cookie *logcookies,
530 int numcookies, void *buf, int idx)
532 struct llog_operations *lop;
533 int raised, rc, buflen;
536 rc = llog_handle2ops(handle, &lop);
540 if (lop->lop_write_rec == NULL)
543 /* FIXME: Why doesn't caller just set the right lrh_len itself? */
545 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr)
546 + sizeof(struct llog_rec_tail);
548 buflen = rec->lrh_len;
549 LASSERT(cfs_size_round(buflen) == buflen);
551 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
553 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
554 rc = lop->lop_write_rec(handle, rec, logcookies, numcookies, buf, idx);
556 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
560 static inline int llog_read_header(struct llog_handle *handle)
562 struct llog_operations *lop;
566 rc = llog_handle2ops(handle, &lop);
569 if (lop->lop_read_header == NULL)
572 rc = lop->lop_read_header(handle);
576 static inline int llog_destroy(struct llog_handle *handle)
578 struct llog_operations *lop;
582 rc = llog_handle2ops(handle, &lop);
585 if (lop->lop_destroy == NULL)
588 rc = lop->lop_destroy(handle);
593 static inline int llog_cancel(struct obd_export *exp,
594 struct lov_stripe_md *lsm, int count,
595 struct llog_cookie *cookies, int flags)
597 struct llog_operations *lop;
601 rc = llog_handle2ops(loghandle, &lop);
604 if (lop->lop_cancel == NULL)
607 rc = lop->lop_cancel(exp, lsm, count, cookies, flags);
612 static inline int llog_next_block(struct llog_handle *loghandle, int *cur_idx,
613 int next_idx, __u64 *cur_offset, void *buf,
616 struct llog_operations *lop;
620 rc = llog_handle2ops(loghandle, &lop);
623 if (lop->lop_next_block == NULL)
626 rc = lop->lop_next_block(loghandle, cur_idx, next_idx, cur_offset, buf,
631 static inline int llog_prev_block(struct llog_handle *loghandle,
632 int prev_idx, void *buf, int len)
634 struct llog_operations *lop;
638 rc = llog_handle2ops(loghandle, &lop);
641 if (lop->lop_prev_block == NULL)
644 rc = lop->lop_prev_block(loghandle, prev_idx, buf, len);
648 static inline int llog_create(struct llog_ctxt *ctxt, struct llog_handle **res,
649 struct llog_logid *logid, char *name)
651 struct llog_operations *lop;
655 rc = llog_obd2ops(ctxt, &lop);
658 if (lop->lop_create == NULL)
661 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
663 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
664 rc = lop->lop_create(ctxt, res, logid, name);
666 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
670 static inline int llog_connect(struct llog_ctxt *ctxt,
671 struct llog_logid *logid, struct llog_gen *gen,
672 struct obd_uuid *uuid)
674 struct llog_operations *lop;
678 rc = llog_obd2ops(ctxt, &lop);
681 if (lop->lop_connect == NULL)
684 rc = lop->lop_connect(ctxt, logid, gen, uuid);
688 int lustre_process_log(struct super_block *sb, char *logname,
689 struct config_llog_instance *cfg);
690 int lustre_end_log(struct super_block *sb, char *logname,
691 struct config_llog_instance *cfg);