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)(struct llog_handle *, struct llog_rec_hdr *, void *);
115 typedef int (*llog_fill_rec_cb_t)(struct llog_rec_hdr *rec, void *data);
116 extern struct llog_handle *llog_alloc_handle(void);
117 int llog_init_handle(struct llog_handle *handle, int flags,
118 struct obd_uuid *uuid);
119 extern void llog_free_handle(struct llog_handle *handle);
120 int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
121 void *data, void *catdata);
122 int llog_process_flags(struct llog_handle *loghandle, llog_cb_t cb,
123 void *data, void *catdata, int flags);
124 int llog_reverse_process(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
150 int lpd_flags; /** llog_process flags */
153 struct llog_process_cat_data {
155 * Temporary stored first_idx while scanning log.
159 * Temporary stored last_idx while scanning log.
164 struct llog_process_cat_args {
166 * Llog context used in recovery thread on OST (recov_thread.c)
168 struct llog_ctxt *lpca_ctxt;
170 * Llog callback used in recovery thread on OST (recov_thread.c)
174 * Data pointer for llog callback.
179 int llog_cat_put(struct llog_handle *cathandle);
180 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
181 struct llog_cookie *reccookie, void *buf);
182 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
183 struct llog_cookie *cookies);
184 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
185 int startcat, int startidx);
186 int llog_cat_process_flags(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
187 int flags, int startcat, int startidx);
188 int llog_cat_process_thread(void *data);
189 int llog_cat_reverse_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data);
190 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
193 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
194 int index, struct obd_device *disk_obd, int count,
195 struct llog_logid *logid, const char *logname,
196 struct llog_operations *op);
197 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
198 struct obd_device *disk_obd, int count, struct llog_logid *logid,
199 struct llog_operations *op);
200 int __llog_ctxt_put(struct llog_ctxt *ctxt);
201 int llog_cleanup(struct llog_ctxt *);
202 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp, int flags);
203 int llog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
204 struct lov_stripe_md *lsm, struct llog_cookie *logcookies,
206 int llog_cancel(struct llog_ctxt *, struct lov_stripe_md *lsm,
207 int count, struct llog_cookie *cookies, int flags);
209 int llog_obd_origin_setup(struct obd_device *obd, struct obd_llog_group *olg,
210 int index, struct obd_device *disk_obd, int count,
211 struct llog_logid *logid, const char *name);
212 int llog_obd_origin_cleanup(struct llog_ctxt *ctxt);
213 int llog_obd_origin_add(struct llog_ctxt *ctxt,
214 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
215 struct llog_cookie *logcookies, int numcookies);
217 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
218 struct obd_device *disk_obd, int *idx);
220 int obd_llog_finish(struct obd_device *obd, int count);
223 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
224 int llog_catalog_list(struct obd_device *obd, int count,
225 struct obd_ioctl_data *data);
228 int llog_initiator_connect(struct llog_ctxt *ctxt);
229 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
230 int llog_origin_connect(struct llog_ctxt *ctxt,
231 struct llog_logid *logid, struct llog_gen *gen,
232 struct obd_uuid *uuid);
233 int llog_handle_connect(struct ptlrpc_request *req);
236 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
237 struct lov_stripe_md *lsm, int count,
238 struct llog_cookie *cookies, int flags);
239 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp,
241 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
242 struct llog_logid *logid, struct llog_gen *gen,
243 struct obd_uuid *uuid);
245 struct llog_operations {
246 int (*lop_write_rec)(struct llog_handle *loghandle,
247 struct llog_rec_hdr *rec,
248 struct llog_cookie *logcookies, int numcookies,
250 int (*lop_destroy)(struct llog_handle *handle);
251 int (*lop_next_block)(struct llog_handle *h, int *curr_idx,
252 int next_idx, __u64 *offset, void *buf, int len);
253 int (*lop_prev_block)(struct llog_handle *h,
254 int prev_idx, void *buf, int len);
255 int (*lop_create)(struct llog_ctxt *ctxt, struct llog_handle **,
256 struct llog_logid *logid, char *name);
257 int (*lop_close)(struct llog_handle *handle);
258 int (*lop_read_header)(struct llog_handle *handle);
260 int (*lop_setup)(struct obd_device *obd, struct obd_llog_group *olg,
261 int ctxt_idx, struct obd_device *disk_obd, int count,
262 struct llog_logid *logid, const char *name);
263 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp,
265 int (*lop_cleanup)(struct llog_ctxt *ctxt);
266 int (*lop_add)(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
267 struct lov_stripe_md *lsm,
268 struct llog_cookie *logcookies, int numcookies);
269 int (*lop_cancel)(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
270 int count, struct llog_cookie *cookies, int flags);
271 int (*lop_connect)(struct llog_ctxt *ctxt,
272 struct llog_logid *logid, struct llog_gen *gen,
273 struct obd_uuid *uuid);
274 /* XXX add 2 more: commit callbacks and llog recovery functions */
277 /* In-memory descriptor for a log object or log catalog */
279 cfs_rw_semaphore_t lgh_lock;
280 struct llog_logid lgh_id; /* id of this log */
281 struct llog_log_hdr *lgh_hdr;
282 cfs_spinlock_t lgh_hdr_lock; /* protect lgh_hdr data */
284 struct file *lgh_file;
285 struct dt_object *lgh_obj;
288 int lgh_cur_idx; /* used during llog_process */
289 __u64 lgh_cur_offset; /* used during llog_process */
290 struct llog_ctxt *lgh_ctxt;
292 struct plain_handle_data phd;
293 struct cat_handle_data chd;
297 struct llog_operations *lgh_logops;
301 extern struct llog_operations llog_lvfs_ops;
302 int llog_get_cat_list(struct obd_device *disk_obd,
303 char *name, int idx, int count,
304 struct llog_catid *idarray);
306 int llog_put_cat_list(struct obd_device *disk_obd,
307 char *name, int idx, int count, struct llog_catid *idarray);
309 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
310 #define LLOG_CTXT_FLAG_STOP 0x00000002
313 int loc_idx; /* my index the obd array of ctxt's */
314 struct llog_gen loc_gen;
315 struct obd_device *loc_obd; /* points back to the containing obd*/
316 struct obd_llog_group *loc_olg; /* group containing that ctxt */
317 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
318 struct obd_import *loc_imp; /* to use in RPC's: can be backward
320 struct llog_operations *loc_logops;
321 struct llog_handle *loc_handle;
322 struct llog_commit_master *loc_lcm;
323 struct llog_canceld_ctxt *loc_llcd;
324 cfs_mutex_t loc_mutex; /* protects loc_llcd and loc_imp */
325 cfs_atomic_t loc_refcount;
327 long loc_flags; /* flags, see above defines */
328 struct dt_object *loc_dir;
331 #define LCM_NAME_SIZE 64
333 struct llog_commit_master {
335 * Thread control flags (start, stop, etc.)
339 * Number of llcds onthis lcm.
341 cfs_atomic_t lcm_count;
343 * The refcount for lcm
345 cfs_atomic_t lcm_refcount;
347 * Thread control structure. Used for control commit thread.
349 struct ptlrpcd_ctl lcm_pc;
351 * Lock protecting list of llcds.
353 cfs_spinlock_t lcm_lock;
355 * Llcds in flight for debugging purposes.
357 cfs_list_t lcm_llcds;
359 * Commit thread name buffer. Only used for thread start.
361 char lcm_name[LCM_NAME_SIZE];
364 static inline struct llog_commit_master
365 *lcm_get(struct llog_commit_master *lcm)
367 cfs_atomic_inc(&lcm->lcm_refcount);
372 lcm_put(struct llog_commit_master *lcm)
374 LASSERT_ATOMIC_POS(&lcm->lcm_refcount);
375 if (cfs_atomic_dec_and_test(&lcm->lcm_refcount))
379 struct llog_canceld_ctxt {
381 * Llog context this llcd is attached to. Used for accessing
382 * ->loc_import and others in process of canceling cookies
383 * gathered in this llcd.
385 struct llog_ctxt *llcd_ctxt;
387 * Cancel thread control stucture pointer. Used for accessing
388 * it to see if should stop processing and other needs.
390 struct llog_commit_master *llcd_lcm;
392 * Maximal llcd size. Used in calculations on how much of room
393 * left in llcd to cookie comming cookies.
397 * Link to lcm llcds list.
399 cfs_list_t llcd_list;
401 * Current llcd size while gathering cookies. This should not be
402 * more than ->llcd_size. Used for determining if we need to
403 * send this llcd (if full) and allocate new one. This is also
404 * used for copying new cookie at the end of buffer.
406 int llcd_cookiebytes;
408 * Pointer to the start of cookies buffer.
410 struct llog_cookie llcd_cookies[0];
413 /* ptlrpc/recov_thread.c */
414 extern struct llog_commit_master *llog_recov_thread_init(char *name);
415 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
417 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
418 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
421 static inline void llog_gen_init(struct llog_ctxt *ctxt)
423 struct obd_device *obd = ctxt->loc_exp->exp_obd;
425 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
426 "%s: wrong obt magic %#x\n",
427 obd->obd_name, obd->u.obt.obt_magic);
428 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
429 ctxt->loc_gen.conn_cnt++;
432 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
434 if (a.mnt_cnt < b.mnt_cnt)
436 if (a.mnt_cnt > b.mnt_cnt)
438 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
441 #define LLOG_PROC_BREAK 0x0001
442 #define LLOG_DEL_RECORD 0x0002
444 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
445 struct llog_operations **lop)
450 *lop = ctxt->loc_logops;
457 static inline int llog_handle2ops(struct llog_handle *loghandle,
458 struct llog_operations **lop)
460 if (loghandle == NULL)
463 return llog_obd2ops(loghandle->lgh_ctxt, lop);
466 static inline int llog_data_len(int len)
468 return cfs_size_round(len);
471 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
473 cfs_atomic_inc(&ctxt->loc_refcount);
474 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
475 cfs_atomic_read(&ctxt->loc_refcount));
479 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
483 LASSERT_ATOMIC_GT_LT(&ctxt->loc_refcount, 0, LI_POISON);
484 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
485 cfs_atomic_read(&ctxt->loc_refcount) - 1);
486 __llog_ctxt_put(ctxt);
489 static inline void llog_group_init(struct obd_llog_group *olg, int group)
491 cfs_waitq_init(&olg->olg_waitq);
492 cfs_spin_lock_init(&olg->olg_lock);
493 cfs_mutex_init(&olg->olg_cat_processing);
494 olg->olg_seq = group;
497 static inline void llog_group_set_export(struct obd_llog_group *olg,
498 struct obd_export *exp)
500 LASSERT(exp != NULL);
502 cfs_spin_lock(&olg->olg_lock);
503 if (olg->olg_exp != NULL && olg->olg_exp != exp)
504 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
505 exp->exp_obd->obd_name, olg->olg_seq,
508 cfs_spin_unlock(&olg->olg_lock);
511 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
512 struct llog_ctxt *ctxt, int index)
514 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
516 cfs_spin_lock(&olg->olg_lock);
517 if (olg->olg_ctxts[index] != NULL) {
518 cfs_spin_unlock(&olg->olg_lock);
521 olg->olg_ctxts[index] = ctxt;
522 cfs_spin_unlock(&olg->olg_lock);
526 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
529 struct llog_ctxt *ctxt;
531 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
533 cfs_spin_lock(&olg->olg_lock);
534 if (olg->olg_ctxts[index] == NULL) {
537 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
539 cfs_spin_unlock(&olg->olg_lock);
543 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
546 return llog_group_get_ctxt(&obd->obd_olg, index);
549 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
551 return (olg->olg_ctxts[index] == NULL);
554 static inline int llog_ctxt_null(struct obd_device *obd, int index)
556 return (llog_group_ctxt_null(&obd->obd_olg, index));
559 static inline int llog_write_rec(struct llog_handle *handle,
560 struct llog_rec_hdr *rec,
561 struct llog_cookie *logcookies,
562 int numcookies, void *buf, int idx)
564 struct llog_operations *lop;
565 int raised, rc, buflen;
568 rc = llog_handle2ops(handle, &lop);
572 if (lop->lop_write_rec == NULL)
575 /* FIXME: Why doesn't caller just set the right lrh_len itself? */
577 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr)
578 + sizeof(struct llog_rec_tail);
580 buflen = rec->lrh_len;
581 LASSERT(cfs_size_round(buflen) == buflen);
583 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
585 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
586 rc = lop->lop_write_rec(handle, rec, logcookies, numcookies, buf, idx);
588 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
592 static inline int llog_read_header(struct llog_handle *handle)
594 struct llog_operations *lop;
598 rc = llog_handle2ops(handle, &lop);
601 if (lop->lop_read_header == NULL)
604 rc = lop->lop_read_header(handle);
608 static inline int llog_destroy(struct llog_handle *handle)
610 struct llog_operations *lop;
614 rc = llog_handle2ops(handle, &lop);
617 if (lop->lop_destroy == NULL)
620 rc = lop->lop_destroy(handle);
625 static inline int llog_cancel(struct obd_export *exp,
626 struct lov_stripe_md *lsm, int count,
627 struct llog_cookie *cookies, int flags)
629 struct llog_operations *lop;
633 rc = llog_handle2ops(loghandle, &lop);
636 if (lop->lop_cancel == NULL)
639 rc = lop->lop_cancel(exp, lsm, count, cookies, flags);
644 static inline int llog_next_block(struct llog_handle *loghandle, int *cur_idx,
645 int next_idx, __u64 *cur_offset, void *buf,
648 struct llog_operations *lop;
652 rc = llog_handle2ops(loghandle, &lop);
655 if (lop->lop_next_block == NULL)
658 rc = lop->lop_next_block(loghandle, cur_idx, next_idx, cur_offset, buf,
663 static inline int llog_prev_block(struct llog_handle *loghandle,
664 int prev_idx, void *buf, int len)
666 struct llog_operations *lop;
670 rc = llog_handle2ops(loghandle, &lop);
673 if (lop->lop_prev_block == NULL)
676 rc = lop->lop_prev_block(loghandle, prev_idx, buf, len);
680 static inline int llog_create(struct llog_ctxt *ctxt, struct llog_handle **res,
681 struct llog_logid *logid, char *name)
683 struct llog_operations *lop;
687 rc = llog_obd2ops(ctxt, &lop);
690 if (lop->lop_create == NULL)
693 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
695 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
696 rc = lop->lop_create(ctxt, res, logid, name);
698 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
702 static inline int llog_connect(struct llog_ctxt *ctxt,
703 struct llog_logid *logid, struct llog_gen *gen,
704 struct obd_uuid *uuid)
706 struct llog_operations *lop;
710 rc = llog_obd2ops(ctxt, &lop);
713 if (lop->lop_connect == NULL)
716 rc = lop->lop_connect(ctxt, logid, gen, uuid);
720 int lustre_process_log(struct super_block *sb, char *logname,
721 struct config_llog_instance *cfg);
722 int lustre_end_log(struct super_block *sb, char *logname,
723 struct config_llog_instance *cfg);