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 2008 Sun Microsystems, Inc. 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
54 #if defined(__linux__)
55 #include <linux/lustre_log.h>
56 #elif defined(__APPLE__)
57 #include <darwin/lustre_log.h>
58 #elif defined(__WINNT__)
59 #include <winnt/lustre_log.h>
61 #error Unsupported operating system.
64 #include <obd_class.h>
66 #include <lustre/lustre_idl.h>
68 #define LOG_NAME_LIMIT(logname, name) \
69 snprintf(logname, sizeof(logname), "LOGS/%s", name)
70 #define LLOG_EEMPTY 4711
72 struct plain_handle_data {
74 struct llog_handle *phd_cat_handle;
75 struct llog_cookie phd_cookie; /* cookie of this log in its cat */
79 struct cat_handle_data {
81 struct llog_handle *chd_current_log; /* currently open log */
84 /* In-memory descriptor for a log object or log catalog */
86 cfs_rw_semaphore_t lgh_lock;
87 struct llog_logid lgh_id; /* id of this log */
88 struct llog_log_hdr *lgh_hdr;
89 struct file *lgh_file;
91 int lgh_cur_idx; /* used during llog_process */
92 __u64 lgh_cur_offset; /* used during llog_process */
93 struct llog_ctxt *lgh_ctxt;
95 struct plain_handle_data phd;
96 struct cat_handle_data chd;
100 /* llog.c - general API */
101 typedef int (*llog_cb_t)(struct llog_handle *, struct llog_rec_hdr *, void *);
102 typedef int (*llog_fill_rec_cb_t)(struct llog_rec_hdr *rec, void *data);
103 extern struct llog_handle *llog_alloc_handle(void);
104 int llog_init_handle(struct llog_handle *handle, int flags,
105 struct obd_uuid *uuid);
106 extern void llog_free_handle(struct llog_handle *handle);
107 int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
108 void *data, void *catdata);
109 int llog_process_flags(struct llog_handle *loghandle, llog_cb_t cb,
110 void *data, void *catdata, int flags);
111 int llog_reverse_process(struct llog_handle *loghandle, llog_cb_t cb,
112 void *data, void *catdata);
113 extern int llog_cancel_rec(struct llog_handle *loghandle, int index);
114 extern int llog_close(struct llog_handle *cathandle);
115 extern int llog_get_size(struct llog_handle *loghandle);
117 /* llog_process flags */
118 #define LLOG_FLAG_NODEAMON 0x0001
120 /* llog_cat.c - catalog api */
121 struct llog_process_data {
123 * Any useful data needed while processing catalog. This is
124 * passed later to process callback.
128 * Catalog process callback function, called for each record
133 * Start processing the catalog from startcat/startidx
137 int lpd_flags; /** llog_process flags */
140 struct llog_process_cat_data {
142 * Temporary stored first_idx while scanning log.
146 * Temporary stored last_idx while scanning log.
151 struct llog_process_cat_args {
153 * Llog context used in recovery thread on OST (recov_thread.c)
155 struct llog_ctxt *lpca_ctxt;
157 * Llog callback used in recovery thread on OST (recov_thread.c)
161 * Data pointer for llog callback.
166 int llog_cat_put(struct llog_handle *cathandle);
167 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
168 struct llog_cookie *reccookie, void *buf);
169 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
170 struct llog_cookie *cookies);
171 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
172 int startcat, int startidx);
173 int llog_cat_process_flags(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
174 int flags, int startcat, int startidx);
175 int llog_cat_process_thread(void *data);
176 int llog_cat_reverse_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data);
177 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index);
180 int llog_setup_named(struct obd_device *obd, struct obd_llog_group *olg,
181 int index, struct obd_device *disk_obd, int count,
182 struct llog_logid *logid, const char *logname,
183 struct llog_operations *op);
184 int llog_setup(struct obd_device *obd, struct obd_llog_group *olg, int index,
185 struct obd_device *disk_obd, int count, struct llog_logid *logid,
186 struct llog_operations *op);
187 int __llog_ctxt_put(struct llog_ctxt *ctxt);
188 int llog_cleanup(struct llog_ctxt *);
189 int llog_sync(struct llog_ctxt *ctxt, struct obd_export *exp);
190 int llog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
191 struct lov_stripe_md *lsm, struct llog_cookie *logcookies,
193 int llog_cancel(struct llog_ctxt *, struct lov_stripe_md *lsm,
194 int count, struct llog_cookie *cookies, int flags);
196 int llog_obd_origin_setup(struct obd_device *obd, struct obd_llog_group *olg,
197 int index, struct obd_device *disk_obd, int count,
198 struct llog_logid *logid, const char *name);
199 int llog_obd_origin_cleanup(struct llog_ctxt *ctxt);
200 int llog_obd_origin_add(struct llog_ctxt *ctxt,
201 struct llog_rec_hdr *rec, struct lov_stripe_md *lsm,
202 struct llog_cookie *logcookies, int numcookies);
204 int obd_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
205 struct obd_device *disk_obd, int *idx);
207 int obd_llog_finish(struct obd_device *obd, int count);
210 int llog_ioctl(struct llog_ctxt *ctxt, int cmd, struct obd_ioctl_data *data);
211 int llog_catalog_list(struct obd_device *obd, int count,
212 struct obd_ioctl_data *data);
215 int llog_initiator_connect(struct llog_ctxt *ctxt);
216 int llog_receptor_accept(struct llog_ctxt *ctxt, struct obd_import *imp);
217 int llog_origin_connect(struct llog_ctxt *ctxt,
218 struct llog_logid *logid, struct llog_gen *gen,
219 struct obd_uuid *uuid);
220 int llog_handle_connect(struct ptlrpc_request *req);
223 int llog_obd_repl_cancel(struct llog_ctxt *ctxt,
224 struct lov_stripe_md *lsm, int count,
225 struct llog_cookie *cookies, int flags);
226 int llog_obd_repl_sync(struct llog_ctxt *ctxt, struct obd_export *exp);
227 int llog_obd_repl_connect(struct llog_ctxt *ctxt,
228 struct llog_logid *logid, struct llog_gen *gen,
229 struct obd_uuid *uuid);
231 struct llog_operations {
232 int (*lop_write_rec)(struct llog_handle *loghandle,
233 struct llog_rec_hdr *rec,
234 struct llog_cookie *logcookies, int numcookies,
236 int (*lop_destroy)(struct llog_handle *handle);
237 int (*lop_next_block)(struct llog_handle *h, int *curr_idx,
238 int next_idx, __u64 *offset, void *buf, int len);
239 int (*lop_prev_block)(struct llog_handle *h,
240 int prev_idx, void *buf, int len);
241 int (*lop_create)(struct llog_ctxt *ctxt, struct llog_handle **,
242 struct llog_logid *logid, char *name);
243 int (*lop_close)(struct llog_handle *handle);
244 int (*lop_read_header)(struct llog_handle *handle);
246 int (*lop_setup)(struct obd_device *obd, struct obd_llog_group *olg,
247 int ctxt_idx, struct obd_device *disk_obd, int count,
248 struct llog_logid *logid, const char *name);
249 int (*lop_sync)(struct llog_ctxt *ctxt, struct obd_export *exp);
250 int (*lop_cleanup)(struct llog_ctxt *ctxt);
251 int (*lop_add)(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
252 struct lov_stripe_md *lsm,
253 struct llog_cookie *logcookies, int numcookies);
254 int (*lop_cancel)(struct llog_ctxt *ctxt, struct lov_stripe_md *lsm,
255 int count, struct llog_cookie *cookies, int flags);
256 int (*lop_connect)(struct llog_ctxt *ctxt,
257 struct llog_logid *logid, struct llog_gen *gen,
258 struct obd_uuid *uuid);
259 /* XXX add 2 more: commit callbacks and llog recovery functions */
263 extern struct llog_operations llog_lvfs_ops;
264 int llog_get_cat_list(struct obd_device *disk_obd,
265 char *name, int idx, int count,
266 struct llog_catid *idarray);
268 int llog_put_cat_list(struct obd_device *disk_obd,
269 char *name, int idx, int count, struct llog_catid *idarray);
271 #define LLOG_CTXT_FLAG_UNINITIALIZED 0x00000001
274 int loc_idx; /* my index the obd array of ctxt's */
275 struct llog_gen loc_gen;
276 struct obd_device *loc_obd; /* points back to the containing obd*/
277 struct obd_llog_group *loc_olg; /* group containing that ctxt */
278 struct obd_export *loc_exp; /* parent "disk" export (e.g. MDS) */
279 struct obd_import *loc_imp; /* to use in RPC's: can be backward
281 struct llog_operations *loc_logops;
282 struct llog_handle *loc_handle;
283 struct llog_commit_master *loc_lcm;
284 struct llog_canceld_ctxt *loc_llcd;
285 cfs_semaphore_t loc_sem; /* protects loc_llcd and loc_imp */
286 cfs_atomic_t loc_refcount;
288 long loc_flags; /* flags, see above defines */
291 #define LCM_NAME_SIZE 64
293 struct llog_commit_master {
295 * Thread control flags (start, stop, etc.)
299 * Number of llcds onthis lcm.
301 cfs_atomic_t lcm_count;
303 * The refcount for lcm
305 cfs_atomic_t lcm_refcount;
307 * Thread control structure. Used for control commit thread.
309 struct ptlrpcd_ctl lcm_pc;
311 * Lock protecting list of llcds.
313 cfs_spinlock_t lcm_lock;
315 * Llcds in flight for debugging purposes.
317 cfs_list_t lcm_llcds;
319 * Commit thread name buffer. Only used for thread start.
321 char lcm_name[LCM_NAME_SIZE];
324 static inline struct llog_commit_master
325 *lcm_get(struct llog_commit_master *lcm)
327 LASSERT(cfs_atomic_read(&lcm->lcm_refcount) > 0);
328 cfs_atomic_inc(&lcm->lcm_refcount);
333 lcm_put(struct llog_commit_master *lcm)
335 if (!cfs_atomic_dec_and_test(&lcm->lcm_refcount)) {
341 struct llog_canceld_ctxt {
343 * Llog context this llcd is attached to. Used for accessing
344 * ->loc_import and others in process of canceling cookies
345 * gathered in this llcd.
347 struct llog_ctxt *llcd_ctxt;
349 * Cancel thread control stucture pointer. Used for accessing
350 * it to see if should stop processing and other needs.
352 struct llog_commit_master *llcd_lcm;
354 * Maximal llcd size. Used in calculations on how much of room
355 * left in llcd to cookie comming cookies.
359 * Link to lcm llcds list.
361 cfs_list_t llcd_list;
363 * Current llcd size while gathering cookies. This should not be
364 * more than ->llcd_size. Used for determining if we need to
365 * send this llcd (if full) and allocate new one. This is also
366 * used for copying new cookie at the end of buffer.
368 int llcd_cookiebytes;
370 * Pointer to the start of cookies buffer.
372 struct llog_cookie llcd_cookies[0];
375 /* ptlrpc/recov_thread.c */
376 extern struct llog_commit_master *llog_recov_thread_init(char *name);
377 extern void llog_recov_thread_fini(struct llog_commit_master *lcm,
379 extern int llog_recov_thread_start(struct llog_commit_master *lcm);
380 extern void llog_recov_thread_stop(struct llog_commit_master *lcm,
383 static inline void llog_gen_init(struct llog_ctxt *ctxt)
385 struct obd_device *obd = ctxt->loc_exp->exp_obd;
387 if (!strcmp(obd->obd_type->typ_name, LUSTRE_MDS_NAME))
388 ctxt->loc_gen.mnt_cnt = obd->u.mds.mds_mount_count;
389 else if (!strstr(obd->obd_type->typ_name, LUSTRE_OST_NAME))
390 ctxt->loc_gen.mnt_cnt = obd->u.filter.fo_mount_count;
392 ctxt->loc_gen.mnt_cnt = 0;
393 ctxt->loc_gen.conn_cnt++;
396 static inline int llog_gen_lt(struct llog_gen a, struct llog_gen b)
398 if (a.mnt_cnt < b.mnt_cnt)
400 if (a.mnt_cnt > b.mnt_cnt)
402 return(a.conn_cnt < b.conn_cnt ? 1 : 0);
405 #define LLOG_PROC_BREAK 0x0001
406 #define LLOG_DEL_RECORD 0x0002
408 static inline int llog_obd2ops(struct llog_ctxt *ctxt,
409 struct llog_operations **lop)
414 *lop = ctxt->loc_logops;
421 static inline int llog_handle2ops(struct llog_handle *loghandle,
422 struct llog_operations **lop)
424 if (loghandle == NULL)
427 return llog_obd2ops(loghandle->lgh_ctxt, lop);
430 static inline int llog_data_len(int len)
432 return cfs_size_round(len);
435 static inline struct llog_ctxt *llog_ctxt_get(struct llog_ctxt *ctxt)
437 LASSERT(cfs_atomic_read(&ctxt->loc_refcount) > 0);
438 cfs_atomic_inc(&ctxt->loc_refcount);
439 CDEBUG(D_INFO, "GETting ctxt %p : new refcount %d\n", ctxt,
440 cfs_atomic_read(&ctxt->loc_refcount));
444 static inline void llog_ctxt_put(struct llog_ctxt *ctxt)
448 LASSERT(cfs_atomic_read(&ctxt->loc_refcount) > 0);
449 LASSERT(cfs_atomic_read(&ctxt->loc_refcount) < 0x5a5a5a);
450 CDEBUG(D_INFO, "PUTting ctxt %p : new refcount %d\n", ctxt,
451 cfs_atomic_read(&ctxt->loc_refcount) - 1);
452 __llog_ctxt_put(ctxt);
455 static inline void llog_group_init(struct obd_llog_group *olg, int group)
457 cfs_waitq_init(&olg->olg_waitq);
458 cfs_spin_lock_init(&olg->olg_lock);
459 cfs_sema_init(&olg->olg_cat_processing, 1);
460 olg->olg_group = group;
463 static inline void llog_group_set_export(struct obd_llog_group *olg,
464 struct obd_export *exp)
466 LASSERT(exp != NULL);
468 cfs_spin_lock(&olg->olg_lock);
469 if (olg->olg_exp != NULL && olg->olg_exp != exp)
470 CWARN("%s: export for group %d is changed: 0x%p -> 0x%p\n",
471 exp->exp_obd->obd_name, olg->olg_group,
474 cfs_spin_unlock(&olg->olg_lock);
477 static inline int llog_group_set_ctxt(struct obd_llog_group *olg,
478 struct llog_ctxt *ctxt, int index)
480 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
482 cfs_spin_lock(&olg->olg_lock);
483 if (olg->olg_ctxts[index] != NULL) {
484 cfs_spin_unlock(&olg->olg_lock);
487 olg->olg_ctxts[index] = ctxt;
488 cfs_spin_unlock(&olg->olg_lock);
492 static inline struct llog_ctxt *llog_group_get_ctxt(struct obd_llog_group *olg,
495 struct llog_ctxt *ctxt;
497 LASSERT(index >= 0 && index < LLOG_MAX_CTXTS);
499 cfs_spin_lock(&olg->olg_lock);
500 if (olg->olg_ctxts[index] == NULL) {
503 ctxt = llog_ctxt_get(olg->olg_ctxts[index]);
505 cfs_spin_unlock(&olg->olg_lock);
509 static inline struct llog_ctxt *llog_get_context(struct obd_device *obd,
512 return llog_group_get_ctxt(&obd->obd_olg, index);
515 static inline int llog_group_ctxt_null(struct obd_llog_group *olg, int index)
517 return (olg->olg_ctxts[index] == NULL);
520 static inline int llog_ctxt_null(struct obd_device *obd, int index)
522 return (llog_group_ctxt_null(&obd->obd_olg, index));
525 static inline int llog_write_rec(struct llog_handle *handle,
526 struct llog_rec_hdr *rec,
527 struct llog_cookie *logcookies,
528 int numcookies, void *buf, int idx)
530 struct llog_operations *lop;
531 int raised, rc, buflen;
534 rc = llog_handle2ops(handle, &lop);
538 if (lop->lop_write_rec == NULL)
541 /* FIXME: Why doesn't caller just set the right lrh_len itself? */
543 buflen = rec->lrh_len + sizeof(struct llog_rec_hdr)
544 + sizeof(struct llog_rec_tail);
546 buflen = rec->lrh_len;
547 LASSERT(cfs_size_round(buflen) == buflen);
549 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
551 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
552 rc = lop->lop_write_rec(handle, rec, logcookies, numcookies, buf, idx);
554 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
558 static inline int llog_read_header(struct llog_handle *handle)
560 struct llog_operations *lop;
564 rc = llog_handle2ops(handle, &lop);
567 if (lop->lop_read_header == NULL)
570 rc = lop->lop_read_header(handle);
574 static inline int llog_destroy(struct llog_handle *handle)
576 struct llog_operations *lop;
580 rc = llog_handle2ops(handle, &lop);
583 if (lop->lop_destroy == NULL)
586 rc = lop->lop_destroy(handle);
591 static inline int llog_cancel(struct obd_export *exp,
592 struct lov_stripe_md *lsm, int count,
593 struct llog_cookie *cookies, int flags)
595 struct llog_operations *lop;
599 rc = llog_handle2ops(loghandle, &lop);
602 if (lop->lop_cancel == NULL)
605 rc = lop->lop_cancel(exp, lsm, count, cookies, flags);
610 static inline int llog_next_block(struct llog_handle *loghandle, int *cur_idx,
611 int next_idx, __u64 *cur_offset, void *buf,
614 struct llog_operations *lop;
618 rc = llog_handle2ops(loghandle, &lop);
621 if (lop->lop_next_block == NULL)
624 rc = lop->lop_next_block(loghandle, cur_idx, next_idx, cur_offset, buf,
629 static inline int llog_prev_block(struct llog_handle *loghandle,
630 int prev_idx, void *buf, int len)
632 struct llog_operations *lop;
636 rc = llog_handle2ops(loghandle, &lop);
639 if (lop->lop_prev_block == NULL)
642 rc = lop->lop_prev_block(loghandle, prev_idx, buf, len);
646 static inline int llog_create(struct llog_ctxt *ctxt, struct llog_handle **res,
647 struct llog_logid *logid, char *name)
649 struct llog_operations *lop;
653 rc = llog_obd2ops(ctxt, &lop);
656 if (lop->lop_create == NULL)
659 raised = cfs_cap_raised(CFS_CAP_SYS_RESOURCE);
661 cfs_cap_raise(CFS_CAP_SYS_RESOURCE);
662 rc = lop->lop_create(ctxt, res, logid, name);
664 cfs_cap_lower(CFS_CAP_SYS_RESOURCE);
668 static inline int llog_connect(struct llog_ctxt *ctxt,
669 struct llog_logid *logid, struct llog_gen *gen,
670 struct obd_uuid *uuid)
672 struct llog_operations *lop;
676 rc = llog_obd2ops(ctxt, &lop);
679 if (lop->lop_connect == NULL)
682 rc = lop->lop_connect(ctxt, logid, gen, uuid);
686 int lustre_process_log(struct super_block *sb, char *logname,
687 struct config_llog_instance *cfg);
688 int lustre_end_log(struct super_block *sb, char *logname,
689 struct config_llog_instance *cfg);