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,
67 struct llog_logid *lid)
69 struct llog_handle *loghandle;
70 struct llog_log_hdr *llh;
71 struct llog_logid_rec rec = { { 0 }, };
72 int rc, index, bitmap_size;
75 llh = cathandle->lgh_hdr;
76 bitmap_size = LLOG_BITMAP_SIZE(llh);
78 index = (cathandle->lgh_last_idx + 1) % bitmap_size;
80 /* maximum number of available slots in catlog is bitmap_size - 2 */
81 if (llh->llh_cat_idx == index) {
82 CERROR("no free catalog slots for log...\n");
83 RETURN(ERR_PTR(-ENOSPC));
86 if (OBD_FAIL_CHECK(OBD_FAIL_MDS_LLOG_CREATE_FAILED))
87 RETURN(ERR_PTR(-ENOSPC));
89 if (lid != NULL && lid->lgl_oid == 0)
92 rc = llog_create(cathandle->lgh_ctxt, &loghandle, lid, NULL);
96 rc = llog_init_handle(loghandle,
97 LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
98 &cathandle->lgh_hdr->llh_tgtuuid);
100 GOTO(out_destroy, rc);
104 if (ext2_set_bit(index, llh->llh_bitmap)) {
105 CERROR("argh, index %u already set in log bitmap?\n",
107 LBUG(); /* should never happen */
109 cathandle->lgh_last_idx = index;
111 llh->llh_tail.lrt_index = index;
113 CDEBUG(D_RPCTRACE,"new recovery log "LPX64":%x for index %u of catalog "
114 LPX64"\n", loghandle->lgh_id.lgl_oid, loghandle->lgh_id.lgl_ogen,
115 index, cathandle->lgh_id.lgl_oid);
116 /* build the record for this log in the catalog */
117 rec.lid_hdr.lrh_len = sizeof(rec);
118 rec.lid_hdr.lrh_index = index;
119 rec.lid_hdr.lrh_type = LLOG_LOGID_MAGIC;
120 rec.lid_id = loghandle->lgh_id;
121 rec.lid_tail.lrt_len = sizeof(rec);
122 rec.lid_tail.lrt_index = index;
124 /* update the catalog: header and record */
125 rc = llog_write_rec(cathandle, &rec.lid_hdr,
126 &loghandle->u.phd.phd_cookie, 1, NULL, index);
128 GOTO(out_destroy, rc);
131 loghandle->lgh_hdr->llh_cat_idx = index;
132 cathandle->u.chd.chd_current_log = loghandle;
133 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
134 list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
138 llog_destroy(loghandle);
143 /* Open an existent log handle and add it to the open list.
144 * This log handle will be closed when all of the records in it are removed.
146 * Assumes caller has already pushed us into the kernel context and is locking.
147 * We return a lock on the handle to ensure nobody yanks it from us.
149 int llog_cat_id2handle(struct llog_handle *cathandle, struct llog_handle **res,
150 struct llog_logid *logid)
152 struct llog_handle *loghandle;
156 if (cathandle == NULL)
159 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
161 struct llog_logid *cgl = &loghandle->lgh_id;
162 if (cgl->lgl_oid == logid->lgl_oid) {
163 if (cgl->lgl_ogen != logid->lgl_ogen) {
164 CERROR("log "LPX64" generation %x != %x\n",
165 logid->lgl_oid, cgl->lgl_ogen,
169 loghandle->u.phd.phd_cat_handle = cathandle;
174 rc = llog_create(cathandle->lgh_ctxt, &loghandle, logid, NULL);
176 CERROR("error opening log id "LPX64":%x: rc %d\n",
177 logid->lgl_oid, logid->lgl_ogen, rc);
179 rc = llog_init_handle(loghandle, LLOG_F_IS_PLAIN, NULL);
181 list_add(&loghandle->u.phd.phd_entry,
182 &cathandle->u.chd.chd_head);
186 loghandle->u.phd.phd_cat_handle = cathandle;
187 loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
188 loghandle->u.phd.phd_cookie.lgc_index =
189 loghandle->lgh_hdr->llh_cat_idx;
197 int llog_cat_put(struct llog_handle *cathandle)
199 struct llog_handle *loghandle, *n;
203 list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
205 int err = llog_close(loghandle);
207 CERROR("error closing loghandle\n");
209 rc = llog_close(cathandle);
212 EXPORT_SYMBOL(llog_cat_put);
214 /* Return the currently active log handle. If the current log handle doesn't
215 * have enough space left for the current record, start a new one.
217 * If reclen is 0, we only want to know what the currently active log is,
218 * otherwise we get a lock on this log so nobody can steal our space.
220 * Assumes caller has already pushed us into the kernel context and is locking.
222 * NOTE: loghandle is write-locked upon successful return
224 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
225 struct llog_logid *lid,
228 struct llog_handle *loghandle = NULL;
231 down_read(&cathandle->lgh_lock);
232 loghandle = cathandle->u.chd.chd_current_log;
234 struct llog_log_hdr *llh = loghandle->lgh_hdr;
235 down_write(&loghandle->lgh_lock);
236 if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) {
237 up_read(&cathandle->lgh_lock);
241 up_write(&loghandle->lgh_lock);
246 down_write(&loghandle->lgh_lock);
247 up_read(&cathandle->lgh_lock);
250 up_read(&cathandle->lgh_lock);
252 /* time to create new log */
254 /* first, we have to make sure the state hasn't changed */
255 down_write(&cathandle->lgh_lock);
256 loghandle = cathandle->u.chd.chd_current_log;
258 struct llog_log_hdr *llh = loghandle->lgh_hdr;
259 down_write(&loghandle->lgh_lock);
260 if (loghandle->lgh_last_idx < LLOG_BITMAP_SIZE(llh) - 1) {
261 up_write(&cathandle->lgh_lock);
265 up_write(&loghandle->lgh_lock);
269 CDEBUG(D_INODE, "creating new log\n");
270 loghandle = llog_cat_new_log(cathandle, lid);
271 if (!IS_ERR(loghandle))
272 down_write(&loghandle->lgh_lock);
273 up_write(&cathandle->lgh_lock);
277 /* Add a single record to the recovery log(s) using a catalog
278 * Returns as llog_write_record
280 * Assumes caller has already pushed us into the kernel context.
282 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
283 struct llog_cookie *reccookie, void *buf)
285 struct llog_handle *loghandle;
289 LASSERT(rec->lrh_len <= LLOG_CHUNK_SIZE);
290 loghandle = llog_cat_current_log(cathandle, &reccookie->lgc_lgl, 1);
291 if (IS_ERR(loghandle))
292 RETURN(PTR_ERR(loghandle));
293 /* loghandle is already locked by llog_cat_current_log() for us */
294 rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
295 up_write(&loghandle->lgh_lock);
297 /* to create a new plain log */
298 loghandle = llog_cat_current_log(cathandle, &reccookie->lgc_lgl, 1);
299 if (IS_ERR(loghandle))
300 RETURN(PTR_ERR(loghandle));
301 rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
302 up_write(&loghandle->lgh_lock);
307 EXPORT_SYMBOL(llog_cat_add_rec);
309 /* For each cookie in the cookie array, we clear the log in-use bit and either:
310 * - the log is empty, so mark it free in the catalog header and delete it
311 * - the log is not empty, just write out the log header
313 * The cookies may be in different log files, so we need to get new logs
316 * Assumes caller has already pushed us into the kernel context.
318 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
319 struct llog_cookie *cookies)
321 int i, index, rc = 0;
324 down_write(&cathandle->lgh_lock);
325 for (i = 0; i < count; i++, cookies++) {
326 struct llog_handle *loghandle;
327 struct llog_logid *lgl = &cookies->lgc_lgl;
329 rc = llog_cat_id2handle(cathandle, &loghandle, lgl);
331 CERROR("Cannot find log "LPX64"\n", lgl->lgl_oid);
335 down_write(&loghandle->lgh_lock);
336 rc = llog_cancel_rec(loghandle, cookies->lgc_index);
337 up_write(&loghandle->lgh_lock);
339 if (rc == 1) { /* log has been destroyed */
340 index = loghandle->u.phd.phd_cookie.lgc_index;
341 if (cathandle->u.chd.chd_current_log == loghandle)
342 cathandle->u.chd.chd_current_log = NULL;
343 llog_free_handle(loghandle);
346 llog_cat_set_first_idx(cathandle, index);
347 rc = llog_cancel_rec(cathandle, index);
349 CDEBUG(D_RPCTRACE,"cancel plain log at index %u"
350 " of catalog "LPX64"\n",
351 index, cathandle->lgh_id.lgl_oid);
354 up_write(&cathandle->lgh_lock);
358 EXPORT_SYMBOL(llog_cat_cancel_records);
360 int llog_cat_process_cb(struct llog_handle *cat_llh, struct llog_rec_hdr *rec,
363 struct llog_process_data *d = data;
364 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
365 struct llog_handle *llh;
369 if (rec->lrh_type != LLOG_LOGID_MAGIC) {
370 CERROR("invalid record in catalog\n");
373 CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
374 LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
375 rec->lrh_index, cat_llh->lgh_id.lgl_oid);
377 rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
379 CERROR("Cannot find handle for log "LPX64"\n",
380 lir->lid_id.lgl_oid);
384 rc = llog_process(llh, d->lpd_cb, d->lpd_data, NULL);
388 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data)
390 struct llog_process_data d;
391 struct llog_process_cat_data cd;
392 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
396 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
400 if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
401 CWARN("catlog "LPX64" crosses index zero\n",
402 cat_llh->lgh_id.lgl_oid);
404 cd.lpcd_first_idx = llh->llh_cat_idx;
405 cd.lpcd_last_idx = 0;
406 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
410 cd.lpcd_first_idx = 0;
411 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
412 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
414 rc = llog_process(cat_llh, llog_cat_process_cb, &d, NULL);
419 EXPORT_SYMBOL(llog_cat_process);
422 int llog_cat_process_thread(void *data)
424 struct llog_process_cat_args *args = data;
425 struct llog_ctxt *ctxt = args->lpca_ctxt;
426 struct llog_handle *llh = NULL;
427 void *cb = args->lpca_cb;
428 struct llog_logid logid;
432 cfs_daemonize_ctxt("ll_log_process");
434 logid = *(struct llog_logid *)(args->lpca_arg);
435 rc = llog_create(ctxt, &llh, &logid, NULL);
437 CERROR("llog_create() failed %d\n", rc);
440 rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
442 CERROR("llog_init_handle failed %d\n", rc);
443 GOTO(release_llh, rc);
447 rc = llog_cat_process(llh, (llog_cb_t)cb, NULL);
448 if (rc != LLOG_PROC_BREAK)
449 CERROR("llog_cat_process() failed %d\n", rc);
451 CWARN("No callback function for recovery\n");
455 * Make sure that all cached data is sent.
457 llog_sync(ctxt, NULL);
460 rc = llog_cat_put(llh);
462 CERROR("llog_cat_put() failed %d\n", rc);
465 OBD_FREE(args, sizeof(*args));
468 EXPORT_SYMBOL(llog_cat_process_thread);
471 static int llog_cat_reverse_process_cb(struct llog_handle *cat_llh,
472 struct llog_rec_hdr *rec, void *data)
474 struct llog_process_data *d = data;
475 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
476 struct llog_handle *llh;
479 if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
480 CERROR("invalid record in catalog\n");
483 CDEBUG(D_HA, "processing log "LPX64":%x at index %u of catalog "
484 LPX64"\n", lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
485 le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid);
487 rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
489 CERROR("Cannot find handle for log "LPX64"\n",
490 lir->lid_id.lgl_oid);
494 rc = llog_reverse_process(llh, d->lpd_cb, d->lpd_data, NULL);
498 int llog_cat_reverse_process(struct llog_handle *cat_llh,
499 llog_cb_t cb, void *data)
501 struct llog_process_data d;
502 struct llog_process_cat_data cd;
503 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
507 LASSERT(llh->llh_flags & LLOG_F_IS_CAT);
511 if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
512 CWARN("catalog "LPX64" crosses index zero\n",
513 cat_llh->lgh_id.lgl_oid);
515 cd.lpcd_first_idx = 0;
516 cd.lpcd_last_idx = cat_llh->lgh_last_idx;
517 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
522 cd.lpcd_first_idx = le32_to_cpu(llh->llh_cat_idx);
523 cd.lpcd_last_idx = 0;
524 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
527 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
533 EXPORT_SYMBOL(llog_cat_reverse_process);
535 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
537 struct llog_log_hdr *llh = cathandle->lgh_hdr;
538 int i, bitmap_size, idx;
541 bitmap_size = LLOG_BITMAP_SIZE(llh);
542 if (llh->llh_cat_idx == (index - 1)) {
543 idx = llh->llh_cat_idx + 1;
544 llh->llh_cat_idx = idx;
545 if (idx == cathandle->lgh_last_idx)
547 for (i = (index + 1) % bitmap_size;
548 i != cathandle->lgh_last_idx;
549 i = (i + 1) % bitmap_size) {
550 if (!ext2_test_bit(i, llh->llh_bitmap)) {
551 idx = llh->llh_cat_idx + 1;
552 llh->llh_cat_idx = idx;
554 llh->llh_cat_idx = 0;
560 CDEBUG(D_RPCTRACE, "set catlog "LPX64" first idx %u\n",
561 cathandle->lgh_id.lgl_oid, llh->llh_cat_idx);
568 /* Assumes caller has already pushed us into the kernel context. */
569 int llog_cat_init(struct llog_handle *cathandle, struct obd_uuid *tgtuuid)
571 struct llog_log_hdr *llh;
576 LASSERT(sizeof(*llh) == LLOG_CHUNK_SIZE);
578 down(&cathandle->lgh_lock);
579 llh = cathandle->lgh_hdr;
581 if (i_size_read(cathandle->lgh_file->f_dentry->d_inode) == 0) {
582 llog_write_rec(cathandle, &llh->llh_hdr, NULL, 0, NULL, 0);
585 rc = lustre_fwrite(cathandle->lgh_file, llh, LLOG_CHUNK_SIZE,
587 if (rc != LLOG_CHUNK_SIZE) {
588 CERROR("error writing catalog header: rc %d\n", rc);
589 OBD_FREE(llh, sizeof(*llh));
595 rc = lustre_fread(cathandle->lgh_file, llh, LLOG_CHUNK_SIZE,
597 if (rc != LLOG_CHUNK_SIZE) {
598 CERROR("error reading catalog header: rc %d\n", rc);
599 /* Can we do much else if the header is bad? */
605 cathandle->lgh_tgtuuid = &llh->llh_tgtuuid;
606 up(&cathandle->lgh_lock);
609 EXPORT_SYMBOL(llog_cat_init);