1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2001-2003 Cluster File Systems, Inc.
5 * Author: Andreas Dilger <adilger@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 * OST<->MDS recovery logging infrastructure.
24 * Invariants in implementation:
25 * - we do not share logs among different OST<->MDS connections, so that
26 * if an OST or MDS fails it need only look at log(s) relevant to itself
29 #define DEBUG_SUBSYSTEM S_LOG
38 #include <liblustre.h>
41 #include <linux/lustre_log.h>
42 #include <portals/list.h>
44 /* Create a new log handle and add it to the open list.
45 * This log handle will be closed when all of the records in it are removed.
47 * Assumes caller has already pushed us into the kernel context and is locking.
49 static struct llog_handle *llog_cat_new_log(struct llog_handle *cathandle)
51 struct llog_handle *loghandle;
52 struct llog_log_hdr *llh;
53 struct llog_logid_rec rec;
54 int rc, index, bitmap_size;
57 llh = cathandle->lgh_hdr;
58 bitmap_size = sizeof(llh->llh_bitmap) * 8;
60 index = (cathandle->lgh_last_idx + 1) % bitmap_size;
62 /* maximum number of available slots in catalog is bitmap_size - 2 */
63 if (llh->llh_cat_idx == cpu_to_le32(index)) {
64 CERROR("no free catalog slots for log...\n");
65 RETURN(ERR_PTR(-ENOSPC));
69 if (ext2_set_bit(index, llh->llh_bitmap)) {
70 CERROR("argh, index %u already set in log bitmap?\n",
72 LBUG(); /* should never happen */
74 cathandle->lgh_last_idx = index;
75 llh->llh_count = cpu_to_le32(le32_to_cpu(llh->llh_count) + 1);
76 llh->llh_tail.lrt_index = cpu_to_le32(index);
79 rc = llog_create(cathandle->lgh_ctxt, &loghandle, NULL, NULL);
81 CERROR("cannot create new log, error = %d\n", rc);
85 rc = llog_init_handle(loghandle,
86 LLOG_F_IS_PLAIN | LLOG_F_ZAP_WHEN_EMPTY,
87 &cathandle->lgh_hdr->llh_tgtuuid);
89 GOTO(out_destroy, rc);
91 CDEBUG(D_HA, "new recovery log "LPX64":%x for index %u of catalog "
92 LPX64"\n", loghandle->lgh_id.lgl_oid, loghandle->lgh_id.lgl_ogen,
93 index, cathandle->lgh_id.lgl_oid);
94 /* build the record for this log in the catalog */
95 rec.lid_hdr.lrh_len = cpu_to_le32(sizeof(rec));
96 rec.lid_hdr.lrh_index = cpu_to_le32(index);
97 rec.lid_hdr.lrh_type = cpu_to_le32(LLOG_LOGID_MAGIC);
98 rec.lid_id = loghandle->lgh_id;
99 rec.lid_tail.lrt_len = cpu_to_le32(sizeof(rec));
100 rec.lid_tail.lrt_index = cpu_to_le32(index);
102 /* update the catalog: header and record */
103 rc = llog_write_rec(cathandle, &rec.lid_hdr,
104 &loghandle->u.phd.phd_cookie, 1, NULL, index);
106 GOTO(out_destroy, rc);
108 loghandle->lgh_hdr->llh_cat_idx = cpu_to_le32(index);
109 cathandle->u.chd.chd_current_log = loghandle;
110 LASSERT(list_empty(&loghandle->u.phd.phd_entry));
111 list_add_tail(&loghandle->u.phd.phd_entry, &cathandle->u.chd.chd_head);
115 llog_destroy(loghandle);
120 /* Open an existent log handle and add it to the open list.
121 * This log handle will be closed when all of the records in it are removed.
123 * Assumes caller has already pushed us into the kernel context and is locking.
124 * We return a lock on the handle to ensure nobody yanks it from us.
126 int llog_cat_id2handle(struct llog_handle *cathandle, struct llog_handle **res,
127 struct llog_logid *logid)
129 struct llog_handle *loghandle;
133 if (cathandle == NULL)
136 list_for_each_entry(loghandle, &cathandle->u.chd.chd_head,
138 struct llog_logid *cgl = &loghandle->lgh_id;
139 if (cgl->lgl_oid == logid->lgl_oid) {
140 if (cgl->lgl_ogen != logid->lgl_ogen) {
141 CERROR("log "LPX64" generation %x != %x\n",
142 logid->lgl_oid, cgl->lgl_ogen,
146 loghandle->u.phd.phd_cat_handle = cathandle;
151 rc = llog_create(cathandle->lgh_ctxt, &loghandle, logid, NULL);
153 CERROR("error opening log id "LPX64":%x: rc %d\n",
154 logid->lgl_oid, logid->lgl_ogen, rc);
156 rc = llog_init_handle(loghandle, LLOG_F_IS_PLAIN, NULL);
158 list_add(&loghandle->u.phd.phd_entry,
159 &cathandle->u.chd.chd_head);
163 loghandle->u.phd.phd_cat_handle = cathandle;
164 loghandle->u.phd.phd_cookie.lgc_lgl = cathandle->lgh_id;
165 loghandle->u.phd.phd_cookie.lgc_index =
166 le32_to_cpu(loghandle->lgh_hdr->llh_cat_idx);
173 EXPORT_SYMBOL(llog_cat_id2handle);
175 int llog_cat_put(struct llog_handle *cathandle)
177 struct llog_handle *loghandle, *n;
181 list_for_each_entry_safe(loghandle, n, &cathandle->u.chd.chd_head,
183 int err = llog_close(loghandle);
185 CERROR("error closing loghandle\n");
187 rc = llog_close(cathandle);
190 EXPORT_SYMBOL(llog_cat_put);
192 /* Return the currently active log handle. If the current log handle doesn't
193 * have enough space left for the current record, start a new one.
195 * If reclen is 0, we only want to know what the currently active log is,
196 * otherwise we get a lock on this log so nobody can steal our space.
198 * Assumes caller has already pushed us into the kernel context and is locking.
200 * NOTE: loghandle is write-locked upon successful return
202 static struct llog_handle *llog_cat_current_log(struct llog_handle *cathandle,
205 struct llog_handle *loghandle = NULL;
208 down_read(&cathandle->lgh_lock);
209 loghandle = cathandle->u.chd.chd_current_log;
211 struct llog_log_hdr *llh = loghandle->lgh_hdr;
212 if (loghandle->lgh_last_idx < (sizeof(llh->llh_bitmap)*8) - 1) {
213 down_write(&loghandle->lgh_lock);
214 up_read(&cathandle->lgh_lock);
220 down_write(&loghandle->lgh_lock);
221 up_read(&cathandle->lgh_lock);
224 up_read(&cathandle->lgh_lock);
226 /* time to create new log */
228 /* first, we have to make sure the state hasn't changed */
229 down_write(&cathandle->lgh_lock);
230 loghandle = cathandle->u.chd.chd_current_log;
232 struct llog_log_hdr *llh = loghandle->lgh_hdr;
233 if (loghandle->lgh_last_idx < (sizeof(llh->llh_bitmap)*8) - 1) {
234 down_write(&loghandle->lgh_lock);
235 up_write(&cathandle->lgh_lock);
240 CDEBUG(D_INODE, "creating new log\n");
241 loghandle = llog_cat_new_log(cathandle);
242 if (!IS_ERR(loghandle))
243 down_write(&loghandle->lgh_lock);
244 up_write(&cathandle->lgh_lock);
248 /* Add a single record to the recovery log(s) using a catalog
249 * Returns as llog_write_record
251 * Assumes caller has already pushed us into the kernel context.
253 int llog_cat_add_rec(struct llog_handle *cathandle, struct llog_rec_hdr *rec,
254 struct llog_cookie *reccookie, void *buf)
256 struct llog_handle *loghandle;
260 LASSERT(le32_to_cpu(rec->lrh_len) <= LLOG_CHUNK_SIZE);
261 loghandle = llog_cat_current_log(cathandle, 1);
262 if (IS_ERR(loghandle))
263 RETURN(PTR_ERR(loghandle));
264 /* loghandle is already locked by llog_cat_current_log() for us */
265 rc = llog_write_rec(loghandle, rec, reccookie, 1, buf, -1);
266 up_write(&loghandle->lgh_lock);
270 EXPORT_SYMBOL(llog_cat_add_rec);
272 /* For each cookie in the cookie array, we clear the log in-use bit and either:
273 * - the log is empty, so mark it free in the catalog header and delete it
274 * - the log is not empty, just write out the log header
276 * The cookies may be in different log files, so we need to get new logs
279 * Assumes caller has already pushed us into the kernel context.
281 int llog_cat_cancel_records(struct llog_handle *cathandle, int count,
282 struct llog_cookie *cookies)
284 int i, index, rc = 0;
287 down_write(&cathandle->lgh_lock);
288 for (i = 0; i < count; i++, cookies++) {
289 struct llog_handle *loghandle;
290 struct llog_logid *lgl = &cookies->lgc_lgl;
292 rc = llog_cat_id2handle(cathandle, &loghandle, lgl);
294 CERROR("Cannot find log "LPX64"\n", lgl->lgl_oid);
298 down_write(&loghandle->lgh_lock);
299 rc = llog_cancel_rec(loghandle, cookies->lgc_index);
300 up_write(&loghandle->lgh_lock);
302 if (rc == 1) { /* log has been destroyed */
303 index = loghandle->u.phd.phd_cookie.lgc_index;
304 if (cathandle->u.chd.chd_current_log == loghandle)
305 cathandle->u.chd.chd_current_log = NULL;
306 llog_free_handle(loghandle);
309 llog_cat_set_first_idx(cathandle, index);
310 rc = llog_cancel_rec(cathandle, index);
312 CDEBUG(D_HA, "cancel plain log at index %u "
313 "of catalog "LPX64"\n",
314 index, cathandle->lgh_id.lgl_oid);
317 up_write(&cathandle->lgh_lock);
321 EXPORT_SYMBOL(llog_cat_cancel_records);
323 static int llog_cat_process_cb(struct llog_handle *cat_llh,
324 struct llog_rec_hdr *rec, void *data)
326 struct llog_process_data *d = data;
327 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
328 struct llog_handle *llh;
331 if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
332 CERROR("invalid record in catalog\n");
335 CWARN("processing log "LPX64":%x at index %u of catalog "LPX64"\n",
336 lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
337 le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid);
339 rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
341 CERROR("Cannot find handle for log "LPX64"\n",
342 lir->lid_id.lgl_oid);
346 rc = llog_process(llh, d->lpd_cb, d->lpd_data, NULL);
350 int llog_cat_process(struct llog_handle *cat_llh, llog_cb_t cb, void *data)
352 struct llog_process_data d;
353 struct llog_process_cat_data cd;
354 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
358 LASSERT(llh->llh_flags &cpu_to_le32(LLOG_F_IS_CAT));
362 if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
363 CWARN("catalog "LPX64" crosses index zero\n",
364 cat_llh->lgh_id.lgl_oid);
366 cd.first_idx = le32_to_cpu(llh->llh_cat_idx);
368 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
373 cd.last_idx = cat_llh->lgh_last_idx;
374 rc = llog_process(cat_llh, llog_cat_process_cb, &d, &cd);
376 rc = llog_process(cat_llh, llog_cat_process_cb, &d, NULL);
381 EXPORT_SYMBOL(llog_cat_process);
383 static int llog_cat_reverse_process_cb(struct llog_handle *cat_llh,
384 struct llog_rec_hdr *rec, void *data)
386 struct llog_process_data *d = data;
387 struct llog_logid_rec *lir = (struct llog_logid_rec *)rec;
388 struct llog_handle *llh;
391 if (le32_to_cpu(rec->lrh_type) != LLOG_LOGID_MAGIC) {
392 CERROR("invalid record in catalog\n");
395 CWARN("processing log "LPX64":%x at index %u of catalog "LPX64"\n",
396 lir->lid_id.lgl_oid, lir->lid_id.lgl_ogen,
397 le32_to_cpu(rec->lrh_index), cat_llh->lgh_id.lgl_oid);
399 rc = llog_cat_id2handle(cat_llh, &llh, &lir->lid_id);
401 CERROR("Cannot find handle for log "LPX64"\n",
402 lir->lid_id.lgl_oid);
406 rc = llog_reverse_process(llh, d->lpd_cb, d->lpd_data, NULL);
410 int llog_cat_reverse_process(struct llog_handle *cat_llh,
411 llog_cb_t cb, void *data)
413 struct llog_process_data d;
414 struct llog_process_cat_data cd;
415 struct llog_log_hdr *llh = cat_llh->lgh_hdr;
419 LASSERT(llh->llh_flags &cpu_to_le32(LLOG_F_IS_CAT));
423 if (llh->llh_cat_idx > cat_llh->lgh_last_idx) {
424 CWARN("catalog "LPX64" crosses index zero\n",
425 cat_llh->lgh_id.lgl_oid);
428 cd.last_idx = cat_llh->lgh_last_idx;
429 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
434 cd.first_idx = le32_to_cpu(llh->llh_cat_idx);
436 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
439 rc = llog_reverse_process(cat_llh, llog_cat_reverse_process_cb,
445 EXPORT_SYMBOL(llog_cat_reverse_process);
447 int llog_cat_set_first_idx(struct llog_handle *cathandle, int index)
449 struct llog_log_hdr *llh = cathandle->lgh_hdr;
450 int i, bitmap_size, idx;
453 bitmap_size = sizeof(llh->llh_bitmap) * 8;
454 if (llh->llh_cat_idx == cpu_to_le32(index - 1)) {
455 idx = le32_to_cpu(llh->llh_cat_idx) + 1;
456 llh->llh_cat_idx = cpu_to_le32(idx);
457 if (idx == cathandle->lgh_last_idx)
459 for (i = (index + 1) % bitmap_size;
460 i != cathandle->lgh_last_idx;
461 i = (i + 1) % bitmap_size) {
462 if (!ext2_test_bit(i, llh->llh_bitmap)) {
463 idx = le32_to_cpu(llh->llh_cat_idx) + 1;
464 llh->llh_cat_idx = cpu_to_le32(idx);
466 llh->llh_cat_idx = 0;
472 CDEBUG(D_HA, "set catalog "LPX64" first idx %u\n",
473 cathandle->lgh_id.lgl_oid,le32_to_cpu(llh->llh_cat_idx));
478 EXPORT_SYMBOL(llog_cat_set_first_idx);
480 int llog_catalog_add(struct llog_ctxt *ctxt, struct llog_rec_hdr *rec,
481 void *buf, struct llog_cookie *logcookies,
482 int numcookies, void *data)
484 struct llog_handle *cathandle;
488 cathandle = ctxt->loc_handle;
489 LASSERT(cathandle != NULL);
491 rc = llog_cat_add_rec(cathandle, rec, logcookies, buf);
493 CERROR("write one catalog record failed: %d\n", rc);
496 EXPORT_SYMBOL(llog_catalog_add);
498 int llog_catalog_cancel(struct llog_ctxt *ctxt, int count,
499 struct llog_cookie *cookies, int flags, void *data)
501 struct llog_handle *cathandle;
505 if (cookies == NULL || count == 0)
507 cathandle = ctxt->loc_handle;
508 LASSERT(cathandle != NULL);
509 rc = llog_cat_cancel_records(cathandle, count, cookies);
512 EXPORT_SYMBOL(llog_catalog_cancel);
514 int llog_catalog_setup(struct llog_ctxt **res, char *name,
515 struct obd_export *exp,
516 struct lvfs_run_ctxt *lvfs_ctxt,
517 struct fsfilt_operations *fsops,
518 struct dentry *logs_de,
519 struct dentry *objects_de)
521 struct llog_ctxt *ctxt;
522 struct llog_catid catid;
523 struct llog_handle *handle;
528 OBD_ALLOC(ctxt, sizeof(*ctxt));
534 /* marking this ctxt alone. */
536 ctxt->loc_fsops = fsops;
537 ctxt->loc_lvfs_ctxt = lvfs_ctxt;
539 ctxt->loc_logs_dir = logs_de;
540 ctxt->loc_objects_dir = objects_de;
541 ctxt->loc_logops = &llog_lvfs_ops;
542 ctxt->loc_logops->lop_add = llog_catalog_add;
543 ctxt->loc_logops->lop_cancel = llog_catalog_cancel;
545 memset(&catid, 0, sizeof(struct llog_catid));
546 rc = llog_get_cat_list(lvfs_ctxt, fsops, name, 1, &catid);
548 CERROR("error llog_get_cat_list rc: %d\n", rc);
551 if (catid.lci_logid.lgl_oid)
552 rc = llog_create(ctxt, &handle, &catid.lci_logid, 0);
554 rc = llog_create(ctxt, &handle, NULL, NULL);
556 catid.lci_logid = handle->lgh_id;
561 ctxt->loc_handle = handle;
562 rc = llog_init_handle(handle, LLOG_F_IS_CAT, NULL);
566 rc = llog_put_cat_list(lvfs_ctxt, fsops, name, 1, &catid);
568 CERROR("error llog_get_cat_list rc: %d\n", rc);
571 OBD_FREE(ctxt, sizeof(*ctxt));
574 EXPORT_SYMBOL(llog_catalog_setup);
576 int llog_catalog_cleanup(struct llog_ctxt *ctxt)
578 struct llog_handle *cathandle, *n, *loghandle;
579 struct llog_log_hdr *llh;
586 cathandle = ctxt->loc_handle;
588 list_for_each_entry_safe(loghandle, n,
589 &cathandle->u.chd.chd_head,
591 llh = loghandle->lgh_hdr;
592 if ((le32_to_cpu(llh->llh_flags) &
593 LLOG_F_ZAP_WHEN_EMPTY) &&
594 (le32_to_cpu(llh->llh_count) == 1)) {
595 rc = llog_destroy(loghandle);
597 CERROR("failure destroying log during "
598 "cleanup: %d\n", rc);
600 index = loghandle->u.phd.phd_cookie.lgc_index;
601 llog_free_handle(loghandle);
604 llog_cat_set_first_idx(cathandle, index);
605 rc = llog_cancel_rec(cathandle, index);
607 CDEBUG(D_HA, "cancel plain log at index"
608 " %u of catalog "LPX64"\n",
609 index,cathandle->lgh_id.lgl_oid);
612 llog_cat_put(ctxt->loc_handle);
616 EXPORT_SYMBOL(llog_catalog_cleanup);
618 int llog_cat_half_bottom(struct llog_cookie *cookie, struct llog_handle *handle)
620 struct llog_handle *loghandle;
621 struct llog_logid *lgl = &cookie->lgc_lgl;
624 down_read(&handle->lgh_lock);
625 rc = llog_cat_id2handle(handle, &loghandle, lgl);
628 if (2 * loghandle->lgh_hdr->llh_cat_idx <=
629 handle->lgh_last_idx + handle->lgh_hdr->llh_cat_idx + 1)
634 up_read(&handle->lgh_lock);
637 EXPORT_SYMBOL(llog_cat_half_bottom);