1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
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);
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);
232 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
233 struct llog_logid *logid, struct llog_gen *gen,
234 struct obd_uuid *uuid);
236 struct llog_operations {
237 int (*lop_write_rec)(struct llog_handle *loghandle,
238 struct llog_rec_hdr *rec,
239 struct llog_cookie *logcookies, int numcookies,
241 int (*lop_destroy)(struct llog_handle *handle);
242 int (*lop_next_block)(struct llog_handle *h, int *curr_idx,
243 int next_idx, __u64 *offset, void *buf, int len);
244 int (*lop_prev_block)(struct llog_handle *h,
245 int prev_idx, void *buf, int len);
246 int (*lop_create)(struct llog_ctxt *ctxt, struct llog_handle **,
247 struct llog_logid *logid, char *name);
248 int (*lop_close)(struct llog_handle *handle);
249 int (*lop_read_header)(struct llog_handle *handle);
251 int (*lop_setup)(struct obd_device *obd, struct obd_llog_group *olg,
252 int ctxt_idx, struct obd_device *disk_obd, int count,
253 struct llog_logid *logid, const char *name);
254 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp);
255 int (*lop_cleanup)(struct llog_ctxt *ctxt);
256 int (*lop_add)(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
257 struct lov_stripe_md *lsm,
258 struct llog_cookie *logcookies, int numcookies);
259 int (*lop_cancel)(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
260 int count, struct llog_cookie *cookies, int flags);
261 int (*lop_connect)(struct llog_ctxt *ctxt,
262 struct llog_logid *logid, struct llog_gen *gen,
263 struct obd_uuid *uuid);
264 /* XXX add 2 more: commit callbacks and llog recovery functions */
268 extern struct llog_operations llog_lvfs_ops;
269 int llog_get_cat_list(struct obd_device *disk_obd,
270 char *name, int idx, int count,
271 struct llog_catid *idarray);
273 int llog_put_cat_list(struct obd_device *disk_obd,
274 char *name, int idx, int count, struct llog_catid *idarray);
276 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
279 int loc_idx; /* my index the obd array of ctxt's */
280 struct llog_gen loc_gen;
281 struct obd_device *loc_obd; /* points back to the containing obd*/
282 struct obd_llog_group *loc_olg; /* group containing that ctxt */
283 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
284 struct obd_import *loc_imp; /* to use in RPC's: can be backward
286 struct llog_operations *loc_logops;
287 struct llog_handle *loc_handle;
288 struct llog_commit_master *loc_lcm;
289 struct llog_canceld_ctxt *loc_llcd;
290 cfs_semaphore_t loc_sem; /* protects loc_llcd and loc_imp */
291 cfs_atomic_t loc_refcount;
293 long loc_flags; /* flags, see above defines */
296 #define LCM_NAME_SIZE 64
298 struct llog_commit_master {
300 * Thread control flags (start, stop, etc.)
304 * Number of llcds onthis lcm.
306 cfs_atomic_t lcm_count;
308 * The refcount for lcm
310 cfs_atomic_t lcm_refcount;
312 * Thread control structure. Used for control commit thread.
314 struct ptlrpcd_ctl lcm_pc;
316 * Lock protecting list of llcds.
318 cfs_spinlock_t lcm_lock;
320 * Llcds in flight for debugging purposes.
322 cfs_list_t lcm_llcds;
324 * Commit thread name buffer. Only used for thread start.
326 char lcm_name[LCM_NAME_SIZE];
329 static inline struct llog_commit_master
330 *lcm_get(struct llog_commit_master *lcm)
332 LASSERT(cfs_atomic_read(&lcm->lcm_refcount) > 0);
333 cfs_atomic_inc(&lcm->lcm_refcount);
338 lcm_put(struct llog_commit_master *lcm)
340 if (!cfs_atomic_dec_and_test(&lcm->lcm_refcount)) {
346 struct llog_canceld_ctxt {
348 * Llog context this llcd is attached to. Used for accessing
349 * ->loc_import and others in process of canceling cookies
350 * gathered in this llcd.
352 struct llog_ctxt *llcd_ctxt;
354 * Cancel thread control stucture pointer. Used for accessing
355 * it to see if should stop processing and other needs.
357 struct llog_commit_master *llcd_lcm;
359 * Maximal llcd size. Used in calculations on how much of room
360 * left in llcd to cookie comming cookies.
364 * Link to lcm llcds list.
366 cfs_list_t llcd_list;
368 * Current llcd size while gathering cookies. This should not be
369 * more than ->llcd_size. Used for determining if we need to
370 * send this llcd (if full) and allocate new one. This is also
371 * used for copying new cookie at the end of buffer.
373 int llcd_cookiebytes;
375 * Pointer to the start of cookies buffer.
377 struct llog_cookie llcd_cookies[0];
380 /* ptlrpc/recov_thread.c */
381 extern struct llog_commit_master *llog_recov_thread_init(char *name);
382 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
384 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
385 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
388 static inline void llog_gen_init(struct llog_ctxt *ctxt)
390 struct obd_device *obd = ctxt->loc_exp->exp_obd;
392 LASSERTF(obd->u.obt.obt_magic == OBT_MAGIC,
393 "%s: wrong obt magic %#x\n",
394 obd->obd_name, obd->u.obt.obt_magic);
395 ctxt->loc_gen.mnt_cnt = obd->u.obt.obt_mount_count;
396 ctxt->loc_gen.conn_cnt++;
399 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
401 if (a.mnt_cnt < b.mnt_cnt)
403 if (a.mnt_cnt > b.mnt_cnt)
405 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
408 #define LLOG_PROC_BREAK 0x0001
409 #define LLOG_DEL_RECORD 0x0002
411 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
412 struct llog_operations **lop)
417 *lop = ctxt->loc_logops;
424 static inline int llog_handle2ops(struct llog_handle *loghandle,
425 struct llog_operations **lop)
427 if (loghandle == NULL)
430 return llog_obd2ops(loghandle->lgh_ctxt, lop);
433 static inline int llog_data_len(int len)
435 return cfs_size_round(len);
438 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
440 LASSERT(cfs_atomic_read(&ctxt->loc_refcount) > 0);
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(cfs_atomic_read(&ctxt->loc_refcount) > 0);
452 LASSERT(cfs_atomic_read(&ctxt->loc_refcount) < 0x5a5a5a);
453 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
454 cfs_atomic_read(&ctxt->loc_refcount) - 1);
455 __llog_ctxt_put(ctxt);
458 static inline void llog_group_init(struct obd_llog_group *olg, int group)
460 cfs_waitq_init(&olg->olg_waitq);
461 cfs_spin_lock_init(&olg->olg_lock);
462 cfs_sema_init(&olg->olg_cat_processing, 1);
463 olg->olg_seq = group;
466 static inline void llog_group_set_export(struct obd_llog_group *olg,
467 struct obd_export *exp)
469 LASSERT(exp != NULL);
471 cfs_spin_lock(&olg->olg_lock);
472 if (olg->olg_exp != NULL && olg->olg_exp != exp)
473 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
474 exp->exp_obd->obd_name, olg->olg_seq,
477 cfs_spin_unlock(&olg->olg_lock);
480 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
481 struct llog_ctxt *ctxt, int index)
483 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
485 cfs_spin_lock(&olg->olg_lock);
486 if (olg->olg_ctxts[index] != NULL) {
487 cfs_spin_unlock(&olg->olg_lock);
490 olg->olg_ctxts[index] = ctxt;
491 cfs_spin_unlock(&olg->olg_lock);
495 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
498 struct llog_ctxt *ctxt;
500 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
502 cfs_spin_lock(&olg->olg_lock);
503 if (olg->olg_ctxts[index] == NULL) {
506 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
508 cfs_spin_unlock(&olg->olg_lock);
512 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
515 return llog_group_get_ctxt(&obd->obd_olg, index);
518 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
520 return (olg->olg_ctxts[index] == NULL);
523 static inline int llog_ctxt_null(struct obd_device *obd, int index)
525 return (llog_group_ctxt_null(&obd->obd_olg, index));
528 static inline int llog_write_rec(struct llog_handle *handle,
529 struct llog_rec_hdr *rec,
530 struct llog_cookie *logcookies,
531 int numcookies, void *buf, int idx)
533 struct llog_operations *lop;
534 int raised, rc, buflen;
537 rc = llog_handle2ops(handle, &lop);
541 if (lop->lop_write_rec == NULL)
544 /* FIXME: Why doesn't caller just set the right lrh_len itself? */
546 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr)
547 + sizeof(struct llog_rec_tail);
549 buflen = rec->lrh_len;
550 LASSERT(cfs_size_round(buflen) == buflen);
552 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
554 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
555 rc = lop->lop_write_rec(handle, rec, logcookies, numcookies, buf, idx);
557 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
561 static inline int llog_read_header(struct llog_handle *handle)
563 struct llog_operations *lop;
567 rc = llog_handle2ops(handle, &lop);
570 if (lop->lop_read_header == NULL)
573 rc = lop->lop_read_header(handle);
577 static inline int llog_destroy(struct llog_handle *handle)
579 struct llog_operations *lop;
583 rc = llog_handle2ops(handle, &lop);
586 if (lop->lop_destroy == NULL)
589 rc = lop->lop_destroy(handle);
594 static inline int llog_cancel(struct obd_export *exp,
595 struct lov_stripe_md *lsm, int count,
596 struct llog_cookie *cookies, int flags)
598 struct llog_operations *lop;
602 rc = llog_handle2ops(loghandle, &lop);
605 if (lop->lop_cancel == NULL)
608 rc = lop->lop_cancel(exp, lsm, count, cookies, flags);
613 static inline int llog_next_block(struct llog_handle *loghandle, int *cur_idx,
614 int next_idx, __u64 *cur_offset, void *buf,
617 struct llog_operations *lop;
621 rc = llog_handle2ops(loghandle, &lop);
624 if (lop->lop_next_block == NULL)
627 rc = lop->lop_next_block(loghandle, cur_idx, next_idx, cur_offset, buf,
632 static inline int llog_prev_block(struct llog_handle *loghandle,
633 int prev_idx, void *buf, int len)
635 struct llog_operations *lop;
639 rc = llog_handle2ops(loghandle, &lop);
642 if (lop->lop_prev_block == NULL)
645 rc = lop->lop_prev_block(loghandle, prev_idx, buf, len);
649 static inline int llog_create(struct llog_ctxt *ctxt, struct llog_handle **res,
650 struct llog_logid *logid, char *name)
652 struct llog_operations *lop;
656 rc = llog_obd2ops(ctxt, &lop);
659 if (lop->lop_create == NULL)
662 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
664 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
665 rc = lop->lop_create(ctxt, res, logid, name);
667 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
671 static inline int llog_connect(struct llog_ctxt *ctxt,
672 struct llog_logid *logid, struct llog_gen *gen,
673 struct obd_uuid *uuid)
675 struct llog_operations *lop;
679 rc = llog_obd2ops(ctxt, &lop);
682 if (lop->lop_connect == NULL)
685 rc = lop->lop_connect(ctxt, logid, gen, uuid);
689 int lustre_process_log(struct super_block *sb, char *logname,
690 struct config_llog_instance *cfg);
691 int lustre_end_log(struct super_block *sb, char *logname,
692 struct config_llog_instance *cfg);