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 struct plain_handle_data {
80 struct llog_handle *phd_cat_handle;
81 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 */
87 struct llog_handle *chd_next_log; /* llog to be used next */
90 static inline void logid_to_fid(struct llog_logid *id, struct lu_fid *fid)
92 /* For compatibility purposes we identify pre-OSD (~< 2.3.51 MDS)
93 * logid's by non-zero ogen (inode generation) and convert them
95 if (id->lgl_ogen == 0) {
96 fid->f_seq = id->lgl_oseq;
97 fid->f_oid = id->lgl_oid;
100 lu_igif_build(fid, id->lgl_oid, id->lgl_ogen);
104 static inline void fid_to_logid(struct lu_fid *fid, struct llog_logid *id)
106 id->lgl_oseq = fid->f_seq;
107 id->lgl_oid = fid->f_oid;
113 /* llog.c - general API */
114 typedef int (*llog_cb_t)(const struct lu_env *env, struct llog_handle *lgh,
115 struct llog_rec_hdr *rec, void *data);
116 extern struct llog_handle *llog_alloc_handle(void);
117 int llog_init_handle(const struct lu_env *env, struct llog_handle *handle,
118 int flags, struct obd_uuid *uuid);
119 extern void llog_free_handle(struct llog_handle *handle);
120 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
121 llog_cb_t cb, void *data, void *catdata);
122 int llog_reverse_process(const struct lu_env *env,
123 struct llog_handle *loghandle, llog_cb_t cb,
124 void *data, void *catdata);
125 int llog_cancel_rec(const struct lu_env *env, struct llog_handle *loghandle,
127 int llog_close(const struct lu_env *env, struct llog_handle *cathandle);
128 int llog_get_size(struct llog_handle *loghandle);
130 /* llog_process flags */
131 #define LLOG_FLAG_NODEAMON 0x0001
133 /* llog_cat.c - catalog api */
134 struct llog_process_data {
136 * Any useful data needed while processing catalog. This is
137 * passed later to process callback.
141 * Catalog process callback function, called for each record
146 * Start processing the catalog from startcat/startidx
152 struct llog_process_cat_data {
154 * Temporary stored first_idx while scanning log.
158 * Temporary stored last_idx while scanning log.
163 struct llog_process_cat_args {
165 * Llog context used in recovery thread on OST (recov_thread.c)
167 struct llog_ctxt *lpca_ctxt;
169 * Llog callback used in recovery thread on OST (recov_thread.c)
173 * Data pointer for llog callback.
178 int cat_cancel_cb(const struct lu_env *env, struct llog_handle *cathandle,
179 struct llog_rec_hdr *rec, void *data);
180 int llog_cat_put(const struct lu_env *env, struct llog_handle *cathandle);
181 int llog_cat_add_rec(const struct lu_env *env, struct llog_handle *cathandle,
182 struct llog_rec_hdr *rec, struct llog_cookie *reccookie,
184 int llog_cat_cancel_records(const struct lu_env *env,
185 struct llog_handle *cathandle, int count,
186 struct llog_cookie *cookies);
187 int __llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
188 llog_cb_t cb, void *data, int startcat, int startidx,
190 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
191 llog_cb_t cb, void *data, int startcat, int startidx);
192 int llog_cat_process_thread(void *data);
193 int llog_cat_reverse_process(const struct lu_env *env,
194 struct llog_handle *cat_llh, llog_cb_t cb,
196 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
199 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
200 int index, struct obd_device *disk_obd, int count,
201 struct llog_logid *logid, const char *logname,
202 struct llog_operations *op);
203 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
204 struct obd_device *disk_obd, int count, struct llog_logid *logid,
205 struct llog_operations *op);
206 int __llog_ctxt_put(struct llog_ctxt *ctxt);
207 int llog_cleanup(struct llog_ctxt *);
208 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags);
209 int llog_add(const struct lu_env *env, struct llog_ctxt *ctxt,
210 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
211 struct llog_cookie *logcookies, int numcookies);
212 int llog_cancel(const struct lu_env *env, struct llog_ctxt *ctxt,
213 struct lov_stripe_md *lsm, int count,
214 struct llog_cookie *cookies, int flags);
215 int llog_obd_origin_setup(const struct lu_env *env, struct obd_device *obd,
216 struct obd_llog_group *olg, int index,
217 struct obd_device *disk_obd, int count,
218 struct llog_logid *logid, const char *name);
219 int llog_obd_origin_cleanup(const struct lu_env *env, struct llog_ctxt *ctxt);
220 int llog_obd_origin_add(const struct lu_env *env, struct llog_ctxt *ctxt,
221 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
222 struct llog_cookie *logcookies, int numcookies);
224 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
225 struct obd_device *disk_obd, int *idx);
227 int obd_llog_finish(struct obd_device *obd, int count);
230 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, 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, int count,
265 struct llog_logid *logid, const char *name);
266 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
268 int (*lop_cleanup)(const struct lu_env *env, struct llog_ctxt *ctxt);
269 int (*lop_cancel)(const struct lu_env *env, struct llog_ctxt *ctxt,
270 struct lov_stripe_md *lsm, int count,
271 struct llog_cookie *cookies, int flags);
272 int (*lop_connect)(struct llog_ctxt *ctxt, struct llog_logid *logid,
273 struct llog_gen *gen, struct obd_uuid *uuid);
274 int (*lop_create)(const struct lu_env *env, struct llog_ctxt *ctxt,
275 struct llog_handle **h, struct llog_logid *logid,
277 int (*lop_write_rec)(const struct lu_env *env,
278 struct llog_handle *loghandle,
279 struct llog_rec_hdr *rec,
280 struct llog_cookie *logcookies, int numcookies,
282 int (*lop_add)(const struct lu_env *env, struct llog_ctxt *ctxt,
283 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
284 struct llog_cookie *logcookies, int numcookies);
285 int (*lop_close)(const struct lu_env *env, struct llog_handle *handle);
288 /* In-memory descriptor for a log object or log catalog */
290 cfs_rw_semaphore_t lgh_lock;
291 struct llog_logid lgh_id; /* id of this log */
292 struct llog_log_hdr *lgh_hdr;
293 cfs_spinlock_t lgh_hdr_lock; /* protect lgh_hdr data */
295 struct file *lgh_file;
296 struct dt_object *lgh_obj;
299 int lgh_cur_idx; /* used during llog_process */
300 __u64 lgh_cur_offset; /* used during llog_process */
301 struct llog_ctxt *lgh_ctxt;
303 struct plain_handle_data phd;
304 struct cat_handle_data chd;
308 struct llog_operations *lgh_logops;
312 extern struct llog_operations llog_lvfs_ops;
313 int llog_get_cat_list(struct obd_device *disk_obd,
314 char *name, int idx, int count,
315 struct llog_catid *idarray);
317 int llog_put_cat_list(struct obd_device *disk_obd,
318 char *name, int idx, int count, struct llog_catid *idarray);
320 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
321 #define LLOG_CTXT_FLAG_STOP 0x00000002
324 int loc_idx; /* my index the obd array of ctxt's */
325 struct llog_gen loc_gen;
326 struct obd_device *loc_obd; /* points back to the containing obd*/
327 struct obd_llog_group *loc_olg; /* group containing that ctxt */
328 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
329 struct obd_import *loc_imp; /* to use in RPC's: can be backward
331 struct llog_operations *loc_logops;
332 struct llog_handle *loc_handle;
333 struct llog_commit_master *loc_lcm;
334 struct llog_canceld_ctxt *loc_llcd;
335 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
336 cfs_atomic_t loc_refcount;
338 long loc_flags; /* flags, see above defines */
339 struct dt_object *loc_dir;
342 #define LCM_NAME_SIZE 64
344 struct llog_commit_master {
346 * Thread control flags (start, stop, etc.)
350 * Number of llcds onthis lcm.
352 cfs_atomic_t lcm_count;
354 * The refcount for lcm
356 cfs_atomic_t lcm_refcount;
358 * Thread control structure. Used for control commit thread.
360 struct ptlrpcd_ctl lcm_pc;
362 * Lock protecting list of llcds.
364 cfs_spinlock_t lcm_lock;
366 * Llcds in flight for debugging purposes.
368 cfs_list_t lcm_llcds;
370 * Commit thread name buffer. Only used for thread start.
372 char lcm_name[LCM_NAME_SIZE];
375 static inline struct llog_commit_master
376 *lcm_get(struct llog_commit_master *lcm)
378 cfs_atomic_inc(&lcm->lcm_refcount);
383 lcm_put(struct llog_commit_master *lcm)
385 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
386 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
390 struct llog_canceld_ctxt {
392 * Llog context this llcd is attached to. Used for accessing
393 * ->loc_import and others in process of canceling cookies
394 * gathered in this llcd.
396 struct llog_ctxt *llcd_ctxt;
398 * Cancel thread control stucture pointer. Used for accessing
399 * it to see if should stop processing and other needs.
401 struct llog_commit_master *llcd_lcm;
403 * Maximal llcd size. Used in calculations on how much of room
404 * left in llcd to cookie comming cookies.
408 * Link to lcm llcds list.
410 cfs_list_t llcd_list;
412 * Current llcd size while gathering cookies. This should not be
413 * more than ->llcd_size. Used for determining if we need to
414 * send this llcd (if full) and allocate new one. This is also
415 * used for copying new cookie at the end of buffer.
417 int llcd_cookiebytes;
419 * Pointer to the start of cookies buffer.
421 struct llog_cookie llcd_cookies[0];
424 /* ptlrpc/recov_thread.c */
425 extern struct llog_commit_master *llog_recov_thread_init(char *name);
426 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
428 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
429 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
432 static inline void llog_gen_init(struct llog_ctxt *ctxt)
434 struct obd_device *obd = ctxt->loc_exp->exp_obd;
436 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
437 "%s: wrong obt magic %#x\n",
438 obd->obd_name, obd->u.obt.obt_magic);
439 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
440 ctxt->loc_gen.conn_cnt++;
443 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
445 if (a.mnt_cnt < b.mnt_cnt)
447 if (a.mnt_cnt > b.mnt_cnt)
449 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
452 #define LLOG_PROC_BREAK 0x0001
453 #define LLOG_DEL_RECORD 0x0002
455 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
456 struct llog_operations **lop)
461 *lop = ctxt->loc_logops;
468 static inline int llog_handle2ops(struct llog_handle *loghandle,
469 struct llog_operations **lop)
471 if (loghandle == NULL)
474 return llog_obd2ops(loghandle->lgh_ctxt, lop);
477 static inline int llog_data_len(int len)
479 return cfs_size_round(len);
482 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
484 cfs_atomic_inc(&ctxt->loc_refcount);
485 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
486 cfs_atomic_read(&ctxt->loc_refcount));
490 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
494 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
495 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
496 cfs_atomic_read(&ctxt->loc_refcount) - 1);
497 __llog_ctxt_put(ctxt);
500 static inline void llog_group_init(struct obd_llog_group *olg, int group)
502 cfs_waitq_init(&olg->olg_waitq);
503 cfs_spin_lock_init(&olg->olg_lock);
504 cfs_mutex_init(&olg->olg_cat_processing);
505 olg->olg_seq = group;
508 static inline void llog_group_set_export(struct obd_llog_group *olg,
509 struct obd_export *exp)
511 LASSERT(exp != NULL);
513 cfs_spin_lock(&olg->olg_lock);
514 if (olg->olg_exp != NULL && olg->olg_exp != exp)
515 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
516 exp->exp_obd->obd_name, olg->olg_seq,
519 cfs_spin_unlock(&olg->olg_lock);
522 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
523 struct llog_ctxt *ctxt, int index)
525 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
527 cfs_spin_lock(&olg->olg_lock);
528 if (olg->olg_ctxts[index] != NULL) {
529 cfs_spin_unlock(&olg->olg_lock);
532 olg->olg_ctxts[index] = ctxt;
533 cfs_spin_unlock(&olg->olg_lock);
537 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
540 struct llog_ctxt *ctxt;
542 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
544 cfs_spin_lock(&olg->olg_lock);
545 if (olg->olg_ctxts[index] == NULL) {
548 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
550 cfs_spin_unlock(&olg->olg_lock);
554 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
557 return llog_group_get_ctxt(&obd->obd_olg, index);
560 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
562 return (olg->olg_ctxts[index] == NULL);
565 static inline int llog_ctxt_null(struct obd_device *obd, int index)
567 return (llog_group_ctxt_null(&obd->obd_olg, index));
570 static inline int llog_write_rec(const struct lu_env *env,
571 struct llog_handle *handle,
572 struct llog_rec_hdr *rec,
573 struct llog_cookie *logcookies,
574 int numcookies, void *buf, int idx)
576 struct llog_operations *lop;
577 int raised, rc, buflen;
581 rc = llog_handle2ops(handle, &lop);
585 if (lop->lop_write_rec == NULL)
588 /* FIXME: Why doesn't caller just set the right lrh_len itself? */
590 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr) +
591 sizeof(struct llog_rec_tail);
593 buflen = rec->lrh_len;
594 LASSERT(cfs_size_round(buflen) == buflen);
596 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
598 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
599 rc = lop->lop_write_rec(env, handle, rec, logcookies, numcookies,
602 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
606 static inline int llog_read_header(const struct lu_env *env,
607 struct llog_handle *handle)
609 struct llog_operations *lop;
614 rc = llog_handle2ops(handle, &lop);
617 if (lop->lop_read_header == NULL)
620 rc = lop->lop_read_header(env, handle);
624 static inline int llog_destroy(const struct lu_env *env,
625 struct llog_handle *handle)
627 struct llog_operations *lop;
632 rc = llog_handle2ops(handle, &lop);
635 if (lop->lop_destroy == NULL)
638 rc = lop->lop_destroy(env, handle);
642 static inline int llog_next_block(const struct lu_env *env,
643 struct llog_handle *loghandle, int *cur_idx,
644 int next_idx, __u64 *cur_offset, void *buf,
647 struct llog_operations *lop;
652 rc = llog_handle2ops(loghandle, &lop);
655 if (lop->lop_next_block == NULL)
658 rc = lop->lop_next_block(env, loghandle, cur_idx, next_idx,
659 cur_offset, buf, len);
663 static inline int llog_prev_block(const struct lu_env *env,
664 struct llog_handle *loghandle,
665 int prev_idx, void *buf, int len)
667 struct llog_operations *lop;
672 rc = llog_handle2ops(loghandle, &lop);
675 if (lop->lop_prev_block == NULL)
678 rc = lop->lop_prev_block(env, loghandle, prev_idx, buf, len);
682 static inline int llog_create(const struct lu_env *env, struct llog_ctxt *ctxt,
683 struct llog_handle **res,
684 struct llog_logid *logid, char *name)
686 struct llog_operations *lop;
691 rc = llog_obd2ops(ctxt, &lop);
694 if (lop->lop_create == NULL)
697 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
699 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
700 rc = lop->lop_create(env, ctxt, res, logid, name);
702 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
706 static inline int llog_connect(struct llog_ctxt *ctxt,
707 struct llog_logid *logid, struct llog_gen *gen,
708 struct obd_uuid *uuid)
710 struct llog_operations *lop;
714 rc = llog_obd2ops(ctxt, &lop);
717 if (lop->lop_connect == NULL)
720 rc = lop->lop_connect(ctxt, logid, gen, uuid);
724 int lustre_process_log(struct super_block *sb, char *logname,
725 struct config_llog_instance *cfg);
726 int lustre_end_log(struct super_block *sb, char *logname,
727 struct config_llog_instance *cfg);