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 typedef int (*llog_fill_rec_cb_t)(struct llog_rec_hdr *rec, void *data);
117 extern struct llog_handle *llog_alloc_handle(void);
118 int llog_init_handle(struct llog_handle *handle, int flags,
119 struct obd_uuid *uuid);
120 extern void llog_free_handle(struct llog_handle *handle);
121 int llog_process(const struct lu_env *env, struct llog_handle *loghandle,
122 llog_cb_t cb, void *data, void *catdata);
123 int llog_reverse_process(const struct lu_env *env,
124 struct llog_handle *loghandle, llog_cb_t cb,
125 void *data, void *catdata);
126 extern int llog_cancel_rec(struct llog_handle *loghandle, int index);
127 extern int llog_close(struct llog_handle *cathandle);
128 extern 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(struct llog_handle *cathandle);
181 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
182 struct llog_cookie *reccookie, void *buf);
183 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
184 struct llog_cookie *cookies);
185 int __llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
186 llog_cb_t cb, void *data, int startcat, int startidx,
188 int llog_cat_process(const struct lu_env *env, struct llog_handle *cat_llh,
189 llog_cb_t cb, void *data, int startcat, int startidx);
190 int llog_cat_process_thread(void *data);
191 int llog_cat_reverse_process(const struct lu_env *env,
192 struct llog_handle *cat_llh, llog_cb_t cb,
194 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
197 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
198 int index, struct obd_device *disk_obd, int count,
199 struct llog_logid *logid, const char *logname,
200 struct llog_operations *op);
201 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
202 struct obd_device *disk_obd, int count, struct llog_logid *logid,
203 struct llog_operations *op);
204 int __llog_ctxt_put(struct llog_ctxt *ctxt);
205 int llog_cleanup(struct llog_ctxt *);
206 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags);
207 int llog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
208 struct lov_stripe_md *lsm, struct llog_cookie *logcookies,
210 int llog_cancel(struct llog_ctxt *, struct lov_stripe_md *lsm,
211 int count, struct llog_cookie *cookies, int flags);
213 int llog_obd_origin_setup(struct obd_device *obd, struct obd_llog_group *olg,
214 int index, struct obd_device *disk_obd, int count,
215 struct llog_logid *logid, const char *name);
216 int llog_obd_origin_cleanup(struct llog_ctxt *ctxt);
217 int llog_obd_origin_add(struct llog_ctxt *ctxt,
218 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
219 struct llog_cookie *logcookies, int numcookies);
221 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
222 struct obd_device *disk_obd, int *idx);
224 int obd_llog_finish(struct obd_device *obd, int count);
227 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
228 int llog_catalog_list(struct obd_device *obd, int count,
229 struct obd_ioctl_data *data);
232 int llog_initiator_connect(struct llog_ctxt *ctxt);
233 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
234 int llog_origin_connect(struct llog_ctxt *ctxt,
235 struct llog_logid *logid, struct llog_gen *gen,
236 struct obd_uuid *uuid);
237 int llog_handle_connect(struct ptlrpc_request *req);
240 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
241 struct lov_stripe_md *lsm, int count,
242 struct llog_cookie *cookies, int flags);
243 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp,
245 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
246 struct llog_logid *logid, struct llog_gen *gen,
247 struct obd_uuid *uuid);
249 struct llog_operations {
250 int (*lop_write_rec)(struct llog_handle *loghandle,
251 struct llog_rec_hdr *rec,
252 struct llog_cookie *logcookies, int numcookies,
254 int (*lop_destroy)(struct llog_handle *handle);
255 int (*lop_next_block)(struct llog_handle *h, int *curr_idx,
256 int next_idx, __u64 *offset, void *buf, int len);
257 int (*lop_prev_block)(struct llog_handle *h,
258 int prev_idx, void *buf, int len);
259 int (*lop_create)(struct llog_ctxt *ctxt, struct llog_handle **,
260 struct llog_logid *logid, char *name);
261 int (*lop_close)(struct llog_handle *handle);
262 int (*lop_read_header)(struct llog_handle *handle);
264 int (*lop_setup)(struct obd_device *obd, struct obd_llog_group *olg,
265 int ctxt_idx, struct obd_device *disk_obd, int count,
266 struct llog_logid *logid, const char *name);
267 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
269 int (*lop_cleanup)(struct llog_ctxt *ctxt);
270 int (*lop_add)(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
271 struct lov_stripe_md *lsm,
272 struct llog_cookie *logcookies, int numcookies);
273 int (*lop_cancel)(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
274 int count, struct llog_cookie *cookies, int flags);
275 int (*lop_connect)(struct llog_ctxt *ctxt,
276 struct llog_logid *logid, struct llog_gen *gen,
277 struct obd_uuid *uuid);
278 /* XXX add 2 more: commit callbacks and llog recovery functions */
281 /* In-memory descriptor for a log object or log catalog */
283 cfs_rw_semaphore_t lgh_lock;
284 struct llog_logid lgh_id; /* id of this log */
285 struct llog_log_hdr *lgh_hdr;
286 cfs_spinlock_t lgh_hdr_lock; /* protect lgh_hdr data */
288 struct file *lgh_file;
289 struct dt_object *lgh_obj;
292 int lgh_cur_idx; /* used during llog_process */
293 __u64 lgh_cur_offset; /* used during llog_process */
294 struct llog_ctxt *lgh_ctxt;
296 struct plain_handle_data phd;
297 struct cat_handle_data chd;
301 struct llog_operations *lgh_logops;
305 extern struct llog_operations llog_lvfs_ops;
306 int llog_get_cat_list(struct obd_device *disk_obd,
307 char *name, int idx, int count,
308 struct llog_catid *idarray);
310 int llog_put_cat_list(struct obd_device *disk_obd,
311 char *name, int idx, int count, struct llog_catid *idarray);
313 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
314 #define LLOG_CTXT_FLAG_STOP 0x00000002
317 int loc_idx; /* my index the obd array of ctxt's */
318 struct llog_gen loc_gen;
319 struct obd_device *loc_obd; /* points back to the containing obd*/
320 struct obd_llog_group *loc_olg; /* group containing that ctxt */
321 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
322 struct obd_import *loc_imp; /* to use in RPC's: can be backward
324 struct llog_operations *loc_logops;
325 struct llog_handle *loc_handle;
326 struct llog_commit_master *loc_lcm;
327 struct llog_canceld_ctxt *loc_llcd;
328 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
329 cfs_atomic_t loc_refcount;
331 long loc_flags; /* flags, see above defines */
332 struct dt_object *loc_dir;
335 #define LCM_NAME_SIZE 64
337 struct llog_commit_master {
339 * Thread control flags (start, stop, etc.)
343 * Number of llcds onthis lcm.
345 cfs_atomic_t lcm_count;
347 * The refcount for lcm
349 cfs_atomic_t lcm_refcount;
351 * Thread control structure. Used for control commit thread.
353 struct ptlrpcd_ctl lcm_pc;
355 * Lock protecting list of llcds.
357 cfs_spinlock_t lcm_lock;
359 * Llcds in flight for debugging purposes.
361 cfs_list_t lcm_llcds;
363 * Commit thread name buffer. Only used for thread start.
365 char lcm_name[LCM_NAME_SIZE];
368 static inline struct llog_commit_master
369 *lcm_get(struct llog_commit_master *lcm)
371 cfs_atomic_inc(&lcm->lcm_refcount);
376 lcm_put(struct llog_commit_master *lcm)
378 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
379 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
383 struct llog_canceld_ctxt {
385 * Llog context this llcd is attached to. Used for accessing
386 * ->loc_import and others in process of canceling cookies
387 * gathered in this llcd.
389 struct llog_ctxt *llcd_ctxt;
391 * Cancel thread control stucture pointer. Used for accessing
392 * it to see if should stop processing and other needs.
394 struct llog_commit_master *llcd_lcm;
396 * Maximal llcd size. Used in calculations on how much of room
397 * left in llcd to cookie comming cookies.
401 * Link to lcm llcds list.
403 cfs_list_t llcd_list;
405 * Current llcd size while gathering cookies. This should not be
406 * more than ->llcd_size. Used for determining if we need to
407 * send this llcd (if full) and allocate new one. This is also
408 * used for copying new cookie at the end of buffer.
410 int llcd_cookiebytes;
412 * Pointer to the start of cookies buffer.
414 struct llog_cookie llcd_cookies[0];
417 /* ptlrpc/recov_thread.c */
418 extern struct llog_commit_master *llog_recov_thread_init(char *name);
419 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
421 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
422 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
425 static inline void llog_gen_init(struct llog_ctxt *ctxt)
427 struct obd_device *obd = ctxt->loc_exp->exp_obd;
429 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
430 "%s: wrong obt magic %#x\n",
431 obd->obd_name, obd->u.obt.obt_magic);
432 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
433 ctxt->loc_gen.conn_cnt++;
436 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
438 if (a.mnt_cnt < b.mnt_cnt)
440 if (a.mnt_cnt > b.mnt_cnt)
442 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
445 #define LLOG_PROC_BREAK 0x0001
446 #define LLOG_DEL_RECORD 0x0002
448 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
449 struct llog_operations **lop)
454 *lop = ctxt->loc_logops;
461 static inline int llog_handle2ops(struct llog_handle *loghandle,
462 struct llog_operations **lop)
464 if (loghandle == NULL)
467 return llog_obd2ops(loghandle->lgh_ctxt, lop);
470 static inline int llog_data_len(int len)
472 return cfs_size_round(len);
475 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
477 cfs_atomic_inc(&ctxt->loc_refcount);
478 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
479 cfs_atomic_read(&ctxt->loc_refcount));
483 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
487 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
488 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
489 cfs_atomic_read(&ctxt->loc_refcount) - 1);
490 __llog_ctxt_put(ctxt);
493 static inline void llog_group_init(struct obd_llog_group *olg, int group)
495 cfs_waitq_init(&olg->olg_waitq);
496 cfs_spin_lock_init(&olg->olg_lock);
497 cfs_mutex_init(&olg->olg_cat_processing);
498 olg->olg_seq = group;
501 static inline void llog_group_set_export(struct obd_llog_group *olg,
502 struct obd_export *exp)
504 LASSERT(exp != NULL);
506 cfs_spin_lock(&olg->olg_lock);
507 if (olg->olg_exp != NULL && olg->olg_exp != exp)
508 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
509 exp->exp_obd->obd_name, olg->olg_seq,
512 cfs_spin_unlock(&olg->olg_lock);
515 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
516 struct llog_ctxt *ctxt, int index)
518 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
520 cfs_spin_lock(&olg->olg_lock);
521 if (olg->olg_ctxts[index] != NULL) {
522 cfs_spin_unlock(&olg->olg_lock);
525 olg->olg_ctxts[index] = ctxt;
526 cfs_spin_unlock(&olg->olg_lock);
530 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
533 struct llog_ctxt *ctxt;
535 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
537 cfs_spin_lock(&olg->olg_lock);
538 if (olg->olg_ctxts[index] == NULL) {
541 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
543 cfs_spin_unlock(&olg->olg_lock);
547 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
550 return llog_group_get_ctxt(&obd->obd_olg, index);
553 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
555 return (olg->olg_ctxts[index] == NULL);
558 static inline int llog_ctxt_null(struct obd_device *obd, int index)
560 return (llog_group_ctxt_null(&obd->obd_olg, index));
563 static inline int llog_write_rec(struct llog_handle *handle,
564 struct llog_rec_hdr *rec,
565 struct llog_cookie *logcookies,
566 int numcookies, void *buf, int idx)
568 struct llog_operations *lop;
569 int raised, rc, buflen;
572 rc = llog_handle2ops(handle, &lop);
576 if (lop->lop_write_rec == NULL)
579 /* FIXME: Why doesn't caller just set the right lrh_len itself? */
581 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr)
582 + sizeof(struct llog_rec_tail);
584 buflen = rec->lrh_len;
585 LASSERT(cfs_size_round(buflen) == buflen);
587 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
589 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
590 rc = lop->lop_write_rec(handle, rec, logcookies, numcookies, buf, idx);
592 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
596 static inline int llog_read_header(struct llog_handle *handle)
598 struct llog_operations *lop;
602 rc = llog_handle2ops(handle, &lop);
605 if (lop->lop_read_header == NULL)
608 rc = lop->lop_read_header(handle);
612 static inline int llog_destroy(struct llog_handle *handle)
614 struct llog_operations *lop;
618 rc = llog_handle2ops(handle, &lop);
621 if (lop->lop_destroy == NULL)
624 rc = lop->lop_destroy(handle);
629 static inline int llog_cancel(struct obd_export *exp,
630 struct lov_stripe_md *lsm, int count,
631 struct llog_cookie *cookies, int flags)
633 struct llog_operations *lop;
637 rc = llog_handle2ops(loghandle, &lop);
640 if (lop->lop_cancel == NULL)
643 rc = lop->lop_cancel(exp, lsm, count, cookies, flags);
648 static inline int llog_next_block(struct llog_handle *loghandle, int *cur_idx,
649 int next_idx, __u64 *cur_offset, void *buf,
652 struct llog_operations *lop;
656 rc = llog_handle2ops(loghandle, &lop);
659 if (lop->lop_next_block == NULL)
662 rc = lop->lop_next_block(loghandle, cur_idx, next_idx, cur_offset, buf,
667 static inline int llog_prev_block(struct llog_handle *loghandle,
668 int prev_idx, void *buf, int len)
670 struct llog_operations *lop;
674 rc = llog_handle2ops(loghandle, &lop);
677 if (lop->lop_prev_block == NULL)
680 rc = lop->lop_prev_block(loghandle, prev_idx, buf, len);
684 static inline int llog_create(struct llog_ctxt *ctxt, struct llog_handle **res,
685 struct llog_logid *logid, char *name)
687 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(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);