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 the Lustre file system, http://www.lustre.org
8 * Lustre is a trademark of Cluster File Systems, Inc.
10 * You may have signed or agreed to another license before downloading
11 * this software. If so, you are bound by the terms and conditions
12 * of that agreement, and the following does not apply to you. See the
13 * LICENSE file included with this distribution for more information.
15 * If you did not agree to a different license, then this copy of Lustre
16 * is open source software; you can redistribute it and/or modify it
17 * under the terms of version 2 of the GNU General Public License as
18 * published by the Free Software Foundation.
20 * In either case, Lustre is distributed in the hope that it will be
21 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
22 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
23 * license text for more details.
25 * OST<->MDS recovery logging infrastructure.
27 * Invariants in implementation:
28 * - we do not share logs among different OST<->MDS connections, so that
29 * if an OST or MDS fails it need only look at log(s) relevant to itself
32 #define DEBUG_SUBSYSTEM S_LOG
41 #include <liblustre.h>
44 #include <linux/obd_class.h>
45 #include <linux/lustre_log.h>
46 #include <libcfs/list.h>
48 /* Allocate a new log or catalog handle */
49 struct llog_handle *llog_alloc_handle(void)
51 struct llog_handle *loghandle;
54 OBD_ALLOC(loghandle, sizeof(*loghandle));
55 if (loghandle == NULL)
56 RETURN(ERR_PTR(-ENOMEM));
58 init_rwsem(&loghandle->lgh_lock);
62 EXPORT_SYMBOL(llog_alloc_handle);
65 void llog_free_handle(struct llog_handle *loghandle)
70 if (!loghandle->lgh_hdr)
72 if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_PLAIN)
73 list_del_init(&loghandle->u.phd.phd_entry);
74 if (loghandle->lgh_hdr->llh_flags & LLOG_F_IS_CAT)
75 LASSERT(list_empty(&loghandle->u.chd.chd_head));
76 OBD_FREE(loghandle->lgh_hdr, LLOG_CHUNK_SIZE);
79 OBD_FREE(loghandle, sizeof(*loghandle));
81 EXPORT_SYMBOL(llog_free_handle);
83 /* returns negative on error; 0 if success; 1 if success & log destroyed */
84 int llog_cancel_rec(struct llog_handle *loghandle, int index)
86 struct llog_log_hdr *llh = loghandle->lgh_hdr;
90 CDEBUG(D_HA, "canceling %d in log "LPX64"\n",
91 index, loghandle->lgh_id.lgl_oid);
94 CERROR("cannot cancel index 0 (which is header)\n");
98 if (!ext2_clear_bit(index, llh->llh_bitmap)) {
99 CDEBUG(D_HA, "catalog index %u already clear?\n", index);
105 if ((llh->llh_flags & LLOG_F_ZAP_WHEN_EMPTY) &&
106 (llh->llh_count == 1) &&
107 (loghandle->lgh_last_idx == (LLOG_BITMAP_BYTES * 8) - 1)) {
108 rc = llog_destroy(loghandle);
110 CERROR("failure destroying log after last cancel: %d\n",
112 ext2_set_bit(index, llh->llh_bitmap);
120 rc = llog_write_rec(loghandle, &llh->llh_hdr, NULL, 0, NULL, 0);
122 CERROR("failure re-writing header %d\n", rc);
123 ext2_set_bit(index, llh->llh_bitmap);
128 EXPORT_SYMBOL(llog_cancel_rec);
130 int llog_init_handle(struct llog_handle *handle, int flags,
131 struct obd_uuid *uuid)
134 struct llog_log_hdr *llh;
136 LASSERT(handle->lgh_hdr == NULL);
138 OBD_ALLOC(llh, sizeof(*llh));
141 handle->lgh_hdr = llh;
142 /* first assign flags to use llog_client_ops */
143 llh->llh_flags = flags;
144 rc = llog_read_header(handle);
146 flags = llh->llh_flags;
148 LASSERT(obd_uuid_equals(uuid, &llh->llh_tgtuuid));
150 } else if (rc != LLOG_EEMPTY || !flags) {
151 /* set a pesudo flag for initialization */
152 flags = LLOG_F_IS_CAT;
157 handle->lgh_last_idx = 0; /* header is record with index 0 */
158 llh->llh_count = 1; /* for the header record */
159 llh->llh_hdr.lrh_type = LLOG_HDR_MAGIC;
160 llh->llh_hdr.lrh_len = llh->llh_tail.lrt_len = LLOG_CHUNK_SIZE;
161 llh->llh_hdr.lrh_index = llh->llh_tail.lrt_index = 0;
162 llh->llh_timestamp = CURRENT_SECONDS;
164 memcpy(&llh->llh_tgtuuid, uuid, sizeof(llh->llh_tgtuuid));
165 llh->llh_bitmap_offset = offsetof(typeof(*llh),llh_bitmap);
166 ext2_set_bit(0, llh->llh_bitmap);
169 if (flags & LLOG_F_IS_CAT) {
170 INIT_LIST_HEAD(&handle->u.chd.chd_head);
171 llh->llh_size = sizeof(struct llog_logid_rec);
172 } else if (flags & LLOG_F_IS_PLAIN) {
173 INIT_LIST_HEAD(&handle->u.phd.phd_entry);
175 CERROR("Unknown flags: %#x (Expected %#x or %#x\n",
176 flags, LLOG_F_IS_CAT, LLOG_F_IS_PLAIN);
181 OBD_FREE(llh, sizeof(*llh));
182 handle->lgh_hdr = NULL;
186 EXPORT_SYMBOL(llog_init_handle);
188 int llog_close(struct llog_handle *loghandle)
190 struct llog_operations *lop;
194 rc = llog_handle2ops(loghandle, &lop);
197 if (lop->lop_close == NULL)
198 GOTO(out, -EOPNOTSUPP);
199 rc = lop->lop_close(loghandle);
201 llog_free_handle(loghandle);
204 EXPORT_SYMBOL(llog_close);
206 int llog_process(struct llog_handle *loghandle, llog_cb_t cb,
207 void *data, void *catdata)
209 struct llog_log_hdr *llh = loghandle->lgh_hdr;
210 struct llog_process_cat_data *cd = catdata;
212 __u64 cur_offset = LLOG_CHUNK_SIZE;
213 int rc = 0, index = 1, last_index;
219 OBD_ALLOC(buf, LLOG_CHUNK_SIZE);
224 index = cd->first_idx + 1;
225 if (cd != NULL && cd->last_idx)
226 last_index = cd->last_idx;
228 last_index = LLOG_BITMAP_BYTES * 8 - 1;
231 struct llog_rec_hdr *rec;
233 /* skip records not set in bitmap */
234 while (index <= last_index &&
235 !ext2_test_bit(index, llh->llh_bitmap))
238 LASSERT(index <= last_index + 1);
239 if (index == last_index + 1)
242 CDEBUG(D_OTHER, "index: %d last_index %d\n",
245 /* get the buf with our target record; avoid old garbage */
246 memset(buf, 0, LLOG_CHUNK_SIZE);
247 rc = llog_next_block(loghandle, &saved_index, index,
248 &cur_offset, buf, LLOG_CHUNK_SIZE);
252 /* NB: when rec->lrh_len is accessed it is already swabbed
253 * since it is used at the "end" of the loop and the rec
254 * swabbing is done at the beginning of the loop. */
255 for (rec = (struct llog_rec_hdr *)buf;
256 (char *)rec < buf + LLOG_CHUNK_SIZE;
257 rec = (struct llog_rec_hdr *)((char *)rec + rec->lrh_len)){
259 CDEBUG(D_OTHER, "processing rec 0x%p type %#x\n",
262 if (LLOG_REC_HDR_NEEDS_SWABBING(rec))
263 lustre_swab_llog_rec(rec, NULL);
265 CDEBUG(D_OTHER, "after swabbing, type: %#x\n",
268 if (rec->lrh_index == 0)
269 GOTO(out, 0); /* no more records */
271 if (rec->lrh_len == 0 || rec->lrh_len >LLOG_CHUNK_SIZE){
272 CWARN("invalid length %d in llog record for "
273 "index %d\n", rec->lrh_len,
278 if (rec->lrh_index < index) {
279 CDEBUG(D_OTHER, "skipping lrh_index %d\n",
285 "lrh_index: %d lrh_len: %d (%d remains)\n",
286 rec->lrh_index, rec->lrh_len,
287 (int)(buf + LLOG_CHUNK_SIZE - (char *)rec));
289 /* if set, process the callback on this record */
290 if (ext2_test_bit(index, llh->llh_bitmap)) {
291 rc = cb(loghandle, rec, data);
292 if (rc == LLOG_PROC_BREAK) {
293 CWARN("recovery from log: "LPX64":%x"
295 loghandle->lgh_id.lgl_oid,
296 loghandle->lgh_id.lgl_ogen);
302 CDEBUG(D_OTHER, "Skipped index %d\n", index);
305 /* next record, still in buffer? */
307 if (index > last_index)
314 OBD_FREE(buf, LLOG_CHUNK_SIZE);
317 EXPORT_SYMBOL(llog_process);
319 int llog_get_size(struct llog_handle *loghandle)
321 if (loghandle && loghandle->lgh_hdr)
322 return loghandle->lgh_hdr->llh_count;
325 EXPORT_SYMBOL(llog_get_size);