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/obdclass/llog_cat.c
38 * OST<->MDS recovery logging infrastructure.
40 * Invariants in implementation:
41 * - we do not share logs among different OST<->MDS connections, so that
42 * if an OST or MDS fails it need only look at log(s) relevant to itself
44 * Author: Andreas Dilger <adilger@clusterfs.com>
47 #define DEBUG_SUBSYSTEM S_LOG
54 #include <liblustre.h>
57 #include <obd_class.h>
58 #include <lustre_log.h>
59 #include <libcfs/list.h>
61 /* Create a new log handle and add it to the open list.
62 * This log handle will be closed when all of the records in it are removed.
64 * Assumes caller has already pushed us into the kernel context and is locking.
66 static struct llog_handle *llog_cat_new_log(struct llog_handle *cathandle)
68 struct llog_handle *loghandle;
69 struct llog_log_hdr *llh;
70 struct llog_logid_rec rec = { { 0 }, };
71 int rc, index, bitmap_size;
74 llh = cathandle->lgh_hdr;
75 bitmap_size = LLOG_BITMAP_SIZE(llh);
77 index = (cathandle->lgh_last_idx + 1) % bitmap_size;
79 /* maximum number of available slots in catlog is bitmap_size - 2 */
80 if (llh->llh_cat_idx == index) {
81 CERROR("no free catalog slots for log...\n");
82 RETURN(ERR_PTR(-ENOSPC));
85 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
86 RETURN(ERR_PTR(-ENOSPC));
88 rc = llog_create(cathandle->lgh_ctxt, &loghandle, NULL, NULL);
92 rc = llog_init_handle(loghandle,
93 LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
94 &cathandle->lgh_hdr->llh_tgtuuid);
96 GOTO(out_destroy, rc);
100 if (ext2_set_bit(index, llh->llh_bitmap)) {
101 CERROR("argh, index %u already set in log bitmap?\n",
103 LBUG(); /* should never happen */
105 cathandle->lgh_last_idx = index;
107 llh->llh_tail.lrt_index = index;
109 CDEBUG(D_RPCTRACE,"new recovery log "LPX64":%x for index %u of catalog "
110 LPX64"\n", loghandle->lgh_id.lgl_oid, loghandle->lgh_id.lgl_ogen,
111 index, cathandle->lgh_id.lgl_oid);
112 /* build the record for this log in the catalog */
113 rec.lid_hdr.lrh_len = sizeof(rec);
114 rec.lid_hdr.lrh_index = index;
115 rec.lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
116 rec.lid_id = loghandle->lgh_id;
117 rec.lid_tail.lrt_len = sizeof(rec);
118 rec.lid_tail.lrt_index = index;
120 /* update the catalog: header and record */
121 rc = llog_write_rec(cathandle, &rec.lid_hdr,
122 &loghandle->u.phd.phd_cookie, 1, NULL, index);
124 GOTO(out_destroy, rc);
127 loghandle->lgh_hdr->llh_cat_idx = index;
128 cathandle->u.chd.chd_current_log = loghandle;
129 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
130 list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
134 llog_destroy(loghandle);
139 /* Open an existent log handle and add it to the open list.
140 * This log handle will be closed when all of the records in it are removed.
142 * Assumes caller has already pushed us into the kernel context and is locking.
143 * We return a lock on the handle to ensure nobody yanks it from us.
145 int llog_cat_id2handle(struct llog_handle *cathandle, struct llog_handle **res,
146 struct llog_logid *logid)
148 struct llog_handle *loghandle;
152 if (cathandle == NULL)
155 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
157 struct llog_logid *cgl = &loghandle->lgh_id;
158 if (cgl->lgl_oid == logid->lgl_oid) {
159 if (cgl->lgl_ogen != logid->lgl_ogen) {
160 CERROR("log "LPX64" generation %x != %x\n",
161 logid->lgl_oid, cgl->lgl_ogen,
165 loghandle->u.phd.phd_cat_handle = cathandle;
170 rc = llog_create(cathandle->lgh_ctxt, &loghandle, logid, NULL);
172 CERROR("error opening log id "LPX64":%x: rc %d\n",
173 logid->lgl_oid, logid->lgl_ogen, rc);
175 rc = llog_init_handle(loghandle, LLOG_F_IS_PLAIN, NULL);
177 list_add(&loghandle->u.phd.phd_entry,
178 &cathandle->u.chd.chd_head);
182 loghandle->u.phd.phd_cat_handle = cathandle;
183 loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
184 loghandle->u.phd.phd_cookie.lgc_index =
185 loghandle->lgh_hdr->llh_cat_idx;
193 int llog_cat_put(struct llog_handle *cathandle)
195 struct llog_handle *loghandle, *n;
199 list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
201 int err = llog_close(loghandle);
203 CERROR("error closing loghandle\n");
205 rc = llog_close(cathandle);
208 EXPORT_SYMBOL(llog_cat_put);
211 * lockdep markers for nested struct llog_handle::lgh_lock locking.
218 /** Return the currently active log handle. If the current log handle doesn't
219 * have enough space left for the current record, start a new one.
221 * If reclen is 0, we only want to know what the currently active log is,
222 * otherwise we get a lock on this log so nobody can steal our space.
224 * Assumes caller has already pushed us into the kernel context and is locking.
226 * NOTE: loghandle is write-locked upon successful return
228 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
231 struct llog_handle *loghandle = NULL;
234 down_read_nested(&cathandle->lgh_lock, LLOGH_CAT);
235 loghandle = cathandle->u.chd.chd_current_log;
237 struct llog_log_hdr *llh = loghandle->lgh_hdr;
238 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
239 if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) {
240 up_read(&cathandle->lgh_lock);
243 up_write(&loghandle->lgh_lock);
248 down_write(&loghandle->lgh_lock);
249 up_read(&cathandle->lgh_lock);
252 up_read(&cathandle->lgh_lock);
254 /* time to create new log */
256 /* first, we have to make sure the state hasn't changed */
257 down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
258 loghandle = cathandle->u.chd.chd_current_log;
260 struct llog_log_hdr *llh = loghandle->lgh_hdr;
261 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
262 if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) {
263 up_write(&cathandle->lgh_lock);
266 up_write(&loghandle->lgh_lock);
270 CDEBUG(D_INODE, "creating new log\n");
271 loghandle = llog_cat_new_log(cathandle);
272 if (!IS_ERR(loghandle))
273 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
274 up_write(&cathandle->lgh_lock);
278 /* Add a single record to the recovery log(s) using a catalog
279 * Returns as llog_write_record
281 * Assumes caller has already pushed us into the kernel context.
283 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
284 struct llog_cookie *reccookie, void *buf)
286 struct llog_handle *loghandle;
290 LASSERT(rec->lrh_len <= LLOG_CHUNK_SIZE);
291 loghandle = llog_cat_current_log(cathandle, 1);
292 if (IS_ERR(loghandle))
293 RETURN(PTR_ERR(loghandle));
294 /* loghandle is already locked by llog_cat_current_log() for us */
295 rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
297 CERROR("llog_write_rec %d: lh=%p\n", rc, loghandle);
298 up_write(&loghandle->lgh_lock);
300 /* to create a new plain log */
301 loghandle = llog_cat_current_log(cathandle, 1);
302 if (IS_ERR(loghandle))
303 RETURN(PTR_ERR(loghandle));
304 rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
305 up_write(&loghandle->lgh_lock);
310 EXPORT_SYMBOL(llog_cat_add_rec);
312 /* For each cookie in the cookie array, we clear the log in-use bit and either:
313 * - the log is empty, so mark it free in the catalog header and delete it
314 * - the log is not empty, just write out the log header
316 * The cookies may be in different log files, so we need to get new logs
319 * Assumes caller has already pushed us into the kernel context.
321 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
322 struct llog_cookie *cookies)
324 int i, index, rc = 0;
327 down_write_nested(&cathandle->lgh_lock, LLOGH_CAT);
328 for (i = 0; i < count; i++, cookies++) {
329 struct llog_handle *loghandle;
330 struct llog_logid *lgl = &cookies->lgc_lgl;
332 rc = llog_cat_id2handle(cathandle, &loghandle, lgl);
334 CERROR("Cannot find log "LPX64"\n", lgl->lgl_oid);
338 down_write_nested(&loghandle->lgh_lock, LLOGH_LOG);
339 rc = llog_cancel_rec(loghandle, cookies->lgc_index);
340 up_write(&loghandle->lgh_lock);
342 if (rc == 1) { /* log has been destroyed */
343 index = loghandle->u.phd.phd_cookie.lgc_index;
344 if (cathandle->u.chd.chd_current_log == loghandle)
345 cathandle->u.chd.chd_current_log = NULL;
346 llog_free_handle(loghandle);
349 llog_cat_set_first_idx(cathandle, index);
350 rc = llog_cancel_rec(cathandle, index);
352 CDEBUG(D_RPCTRACE,"cancel plain log at index %u"
353 " of catalog "LPX64"\n",
354 index, cathandle->lgh_id.lgl_oid);
357 up_write(&cathandle->lgh_lock);
361 EXPORT_SYMBOL(llog_cat_cancel_records);
363 int llog_cat_process_cb(struct llog_handle *cat_llh, struct llog_rec_hdr *rec,
366 struct llog_process_data *d = data;
367 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
368 struct llog_handle *llh;
372 if (rec->lrh_type != LLOG_LOGID_MAGIC) {
373 CERROR("invalid record in catalog\n");
376 CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
377 LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
378 rec->lrh_index, cat_llh->lgh_id.lgl_oid);
380 rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
382 CERROR("Cannot find handle for log "LPX64"\n",
383 lir->lid_id.lgl_oid);
387 if (rec->lrh_index < d->lpd_startcat)
388 /* Skip processing of the logs until startcat */
391 if (d->lpd_startidx > 0) {
392 struct llog_process_cat_data cd;
394 cd.lpcd_first_idx = d->lpd_startidx;
395 cd.lpcd_last_idx = 0;
396 rc = llog_process(llh, d->lpd_cb, d->lpd_data, &cd);
397 /* Continue processing the next log from idx 0 */
400 rc = llog_process(llh, d->lpd_cb, d->lpd_data, NULL);
406 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data,
407 int startcat, int startidx)
409 struct llog_process_data d;
410 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
414 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
417 d.lpd_startcat = startcat;
418 d.lpd_startidx = startidx;
420 if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
421 struct llog_process_cat_data cd;
423 CWARN("catlog "LPX64" crosses index zero\n",
424 cat_llh->lgh_id.lgl_oid);
426 cd.lpcd_first_idx = llh->llh_cat_idx;
427 cd.lpcd_last_idx = 0;
428 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
432 cd.lpcd_first_idx = 0;
433 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
434 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
436 rc = llog_process(cat_llh, llog_cat_process_cb, &d, NULL);
441 EXPORT_SYMBOL(llog_cat_process);
444 int llog_cat_process_thread(void *data)
446 struct llog_process_cat_args *args = data;
447 struct llog_ctxt *ctxt = args->lpca_ctxt;
448 struct llog_handle *llh = NULL;
449 llog_cb_t cb = args->lpca_cb;
450 struct llog_logid logid;
454 cfs_daemonize_ctxt("ll_log_process");
456 logid = *(struct llog_logid *)(args->lpca_arg);
457 rc = llog_create(ctxt, &llh, &logid, NULL);
459 CERROR("llog_create() failed %d\n", rc);
462 rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
464 CERROR("llog_init_handle failed %d\n", rc);
465 GOTO(release_llh, rc);
469 rc = llog_cat_process(llh, cb, NULL, 0, 0);
470 if (rc != LLOG_PROC_BREAK && rc != 0)
471 CERROR("llog_cat_process() failed %d\n", rc);
474 CWARN("No callback function for recovery\n");
478 * Make sure that all cached data is sent.
480 llog_sync(ctxt, NULL);
481 GOTO(release_llh, rc);
483 rc = llog_cat_put(llh);
485 CERROR("llog_cat_put() failed %d\n", rc);
491 EXPORT_SYMBOL(llog_cat_process_thread);
494 static int llog_cat_reverse_process_cb(struct llog_handle *cat_llh,
495 struct llog_rec_hdr *rec, void *data)
497 struct llog_process_data *d = data;
498 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
499 struct llog_handle *llh;
502 if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
503 CERROR("invalid record in catalog\n");
506 CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
507 LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
508 le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid);
510 rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
512 CERROR("Cannot find handle for log "LPX64"\n",
513 lir->lid_id.lgl_oid);
517 rc = llog_reverse_process(llh, d->lpd_cb, d->lpd_data, NULL);
521 int llog_cat_reverse_process(struct llog_handle *cat_llh,
522 llog_cb_t cb, void *data)
524 struct llog_process_data d;
525 struct llog_process_cat_data cd;
526 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
530 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
534 if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
535 CWARN("catalog "LPX64" crosses index zero\n",
536 cat_llh->lgh_id.lgl_oid);
538 cd.lpcd_first_idx = 0;
539 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
540 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
545 cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
546 cd.lpcd_last_idx = 0;
547 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
550 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
556 EXPORT_SYMBOL(llog_cat_reverse_process);
558 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
560 struct llog_log_hdr *llh = cathandle->lgh_hdr;
561 int i, bitmap_size, idx;
564 bitmap_size = LLOG_BITMAP_SIZE(llh);
565 if (llh->llh_cat_idx == (index - 1)) {
566 idx = llh->llh_cat_idx + 1;
567 llh->llh_cat_idx = idx;
568 if (idx == cathandle->lgh_last_idx)
570 for (i = (index + 1) % bitmap_size;
571 i != cathandle->lgh_last_idx;
572 i = (i + 1) % bitmap_size) {
573 if (!ext2_test_bit(i, llh->llh_bitmap)) {
574 idx = llh->llh_cat_idx + 1;
575 llh->llh_cat_idx = idx;
577 llh->llh_cat_idx = 0;
583 CDEBUG(D_RPCTRACE, "set catlog "LPX64" first idx %u\n",
584 cathandle->lgh_id.lgl_oid, llh->llh_cat_idx);