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/obd.h>
42 #include <linux/obd_class.h>
43 #include <linux/lustre_log.h>
44 #include <linux/obd_ost.h>
45 #include <portals/list.h>
46 #include <linux/lvfs.h>
47 #include <linux/lustre_fsfilt.h>
48 #include "llog_internal.h"
52 static int llog_lvfs_pad(struct obd_device *obd, struct l_file *file,
55 struct llog_rec_hdr rec;
56 struct llog_rec_tail tail;
60 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
62 tail.lrt_len = rec.lrh_len = cpu_to_le32(len);
63 tail.lrt_index = rec.lrh_index = cpu_to_le32(index);
66 rc = fsfilt_write_record(obd, file, &rec, sizeof(rec), &file->f_pos, 0);
68 CERROR("error writing padding record: rc %d\n", rc);
72 file->f_pos += len - sizeof(rec) - sizeof(tail);
73 rc = fsfilt_write_record(obd, file, &tail, sizeof(tail),&file->f_pos,0);
75 CERROR("error writing padding record: rc %d\n", rc);
83 static int llog_lvfs_write_blob(struct obd_device *obd, struct l_file *file,
84 struct llog_rec_hdr *rec, void *buf, loff_t off)
87 struct llog_rec_tail end;
88 loff_t saved_off = file->f_pos;
89 int buflen = le32_to_cpu(rec->lrh_len);
95 rc = fsfilt_write_record(obd, file, rec, buflen,&file->f_pos,0);
97 CERROR("error writing log record: rc %d\n", rc);
104 rec->lrh_len = cpu_to_le32(sizeof(*rec) + buflen + sizeof(end));
105 rc = fsfilt_write_record(obd, file, rec, sizeof(*rec), &file->f_pos, 0);
107 CERROR("error writing log hdr: rc %d\n", rc);
111 rc = fsfilt_write_record(obd, file, buf, buflen, &file->f_pos, 0);
113 CERROR("error writing log buffer: rc %d\n", rc);
117 end.lrt_len = rec->lrh_len;
118 end.lrt_index = rec->lrh_index;
119 rc = fsfilt_write_record(obd, file, &end, sizeof(end), &file->f_pos, 0);
121 CERROR("error writing log tail: rc %d\n", rc);
127 if (saved_off > file->f_pos)
128 file->f_pos = saved_off;
133 static int llog_lvfs_read_blob(struct obd_device *obd, struct l_file *file,
134 void *buf, int size, loff_t off)
140 rc = fsfilt_read_record(obd, file, buf, size, &offset);
142 CERROR("error reading log record: rc %d\n", rc);
148 static int llog_lvfs_read_header(struct llog_handle *handle)
150 struct obd_device *obd;
154 LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
156 obd = handle->lgh_ctxt->loc_exp->exp_obd;
158 if (handle->lgh_file->f_dentry->d_inode->i_size == 0) {
159 CDEBUG(D_HA, "not reading header from 0-byte log\n");
163 rc = llog_lvfs_read_blob(obd, handle->lgh_file, handle->lgh_hdr,
166 CERROR("error reading log header\n");
168 handle->lgh_last_idx = le32_to_cpu(handle->lgh_hdr->llh_tail.lrt_index);
169 handle->lgh_file->f_pos = handle->lgh_file->f_dentry->d_inode->i_size;
174 /* returns negative in on error; 0 if success && reccookie == 0; 1 otherwise */
175 /* appends if idx == -1, otherwise overwrites record idx. */
176 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
177 struct llog_rec_hdr *rec,
178 struct llog_cookie *reccookie, int cookiecount,
181 struct llog_log_hdr *llh;
182 int reclen = le32_to_cpu(rec->lrh_len), index, rc;
183 struct llog_rec_tail *lrt;
184 struct obd_device *obd;
190 llh = loghandle->lgh_hdr;
191 file = loghandle->lgh_file;
192 obd = loghandle->lgh_ctxt->loc_exp->exp_obd;
194 /* record length should not bigger than LLOG_CHUNK_SIZE */
196 rc = (reclen > LLOG_CHUNK_SIZE - sizeof(struct llog_rec_hdr)
197 - sizeof(struct llog_rec_tail)) ? -E2BIG : 0;
199 rc = (reclen > LLOG_CHUNK_SIZE) ? -E2BIG : 0;
206 /* no header: only allowed to insert record 1 */
207 if (idx != 1 && !file->f_dentry->d_inode->i_size) {
208 CERROR("idx != -1 in empty log\n");
212 if (idx && llh->llh_size && llh->llh_size != reclen)
215 rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
216 /* we are done if we only write the header or on error */
220 saved_offset = sizeof(*llh) + (idx-1)*le32_to_cpu(rec->lrh_len);
221 rc = llog_lvfs_write_blob(obd, file, rec, buf, saved_offset);
222 if (rc == 0 && reccookie) {
223 reccookie->lgc_lgl = loghandle->lgh_id;
224 reccookie->lgc_index = idx;
230 /* Make sure that records don't cross a chunk boundary, so we can
231 * process them page-at-a-time if needed. If it will cross a chunk
232 * boundary, write in a fake (but referenced) entry to pad the chunk.
234 * We know that llog_current_log() will return a loghandle that is
235 * big enough to hold reclen, so all we care about is padding here.
237 left = LLOG_CHUNK_SIZE - (file->f_pos & (LLOG_CHUNK_SIZE - 1));
239 reclen = sizeof(*rec) + le32_to_cpu(rec->lrh_len) +
240 sizeof(struct llog_rec_tail);
242 /* NOTE: padding is a record, but no bit is set */
243 if (left != 0 && left != reclen &&
244 left < (reclen + LLOG_MIN_REC_SIZE)) {
245 loghandle->lgh_last_idx++;
246 rc = llog_lvfs_pad(obd, file, left, loghandle->lgh_last_idx);
251 loghandle->lgh_last_idx++;
252 index = loghandle->lgh_last_idx;
253 rec->lrh_index = cpu_to_le32(index);
255 lrt = (void *)rec + le32_to_cpu(rec->lrh_len) - sizeof(*lrt);
256 lrt->lrt_len = rec->lrh_len;
257 lrt->lrt_index = rec->lrh_index;
259 if (ext2_set_bit(index, llh->llh_bitmap)) {
260 CERROR("argh, index %u already set in log bitmap?\n", index);
261 LBUG(); /* should never happen */
263 llh->llh_count = cpu_to_le32(le32_to_cpu(llh->llh_count) + 1);
264 llh->llh_tail.lrt_index = cpu_to_le32(index);
267 rc = llog_lvfs_write_blob(obd, file, &llh->llh_hdr, NULL, 0);
271 rc = llog_lvfs_write_blob(obd, file, rec, buf, file->f_pos);
275 CDEBUG(D_HA, "added record "LPX64": idx: %u, %u bytes\n",
276 loghandle->lgh_id.lgl_oid, index, le32_to_cpu(rec->lrh_len));
277 if (rc == 0 && reccookie) {
278 reccookie->lgc_lgl = loghandle->lgh_id;
279 reccookie->lgc_index = index;
280 if (le32_to_cpu(rec->lrh_type) == MDS_UNLINK_REC)
281 reccookie->lgc_subsys = LLOG_UNLINK_ORIG_CTXT;
282 else if (le32_to_cpu(rec->lrh_type) == OST_SZ_REC)
283 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
284 else if (le32_to_cpu(rec->lrh_type) == OST_RAID1_REC)
285 reccookie->lgc_subsys = LLOG_RD1_ORIG_CTXT;
287 reccookie->lgc_subsys = -1;
290 if (rc == 0 && le32_to_cpu(rec->lrh_type) == LLOG_GEN_REC)
296 /* We can skip reading at least as many log blocks as the number of
297 * minimum sized log records we are skipping. If it turns out
298 * that we are not far enough along the log (because the
299 * actual records are larger than minimum size) we just skip
300 * some more records. */
302 static void llog_skip_over(__u64 *off, int curr, int goal)
306 *off = (*off + (goal-curr-1) * LLOG_MIN_REC_SIZE) &
307 ~(LLOG_CHUNK_SIZE - 1);
312 * - cur_offset to the furthest point read in the log file
313 * - cur_idx to the log index preceeding cur_offset
314 * returns -EIO/-EINVAL on error
316 static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
317 int next_idx, __u64 *cur_offset, void *buf,
323 if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
326 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
327 next_idx, *cur_idx, *cur_offset);
329 while (*cur_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
330 struct llog_rec_hdr *rec;
331 struct llog_rec_tail *tail;
334 llog_skip_over(cur_offset, *cur_idx, next_idx);
337 rc = fsfilt_read_record(loghandle->lgh_ctxt->loc_exp->exp_obd,
338 loghandle->lgh_file, buf, len,
342 CERROR("Cant read llog block at log id "LPU64
343 "/%u offset "LPU64"\n",
344 loghandle->lgh_id.lgl_oid,
345 loghandle->lgh_id.lgl_ogen,
350 /* put number of readed bytes in rc to make code simpler */
351 rc = *cur_offset - ppos;
353 if (rc == 0) /* end of file, nothing to do */
356 if (rc < sizeof(*tail)) {
357 CERROR("Invalid llog block at log id "LPU64"/%u offset "
358 LPU64"\n", loghandle->lgh_id.lgl_oid,
359 loghandle->lgh_id.lgl_ogen, *cur_offset);
363 tail = buf + rc - sizeof(struct llog_rec_tail);
364 *cur_idx = le32_to_cpu(tail->lrt_index);
366 /* this shouldn't happen */
367 if (tail->lrt_index == 0) {
368 CERROR("Invalid llog tail at log id "LPU64"/%u offset "
369 LPU64"\n", loghandle->lgh_id.lgl_oid,
370 loghandle->lgh_id.lgl_ogen, *cur_offset);
373 if (le32_to_cpu(tail->lrt_index) < next_idx)
376 /* sanity check that the start of the new buffer is no farther
377 * than the record that we wanted. This shouldn't happen. */
379 if (le32_to_cpu(rec->lrh_index) > next_idx) {
380 CERROR("missed desired record? %u > %u\n",
381 le32_to_cpu(rec->lrh_index), next_idx);
389 static struct file *llog_filp_open(char *name, int flags, int mode)
395 OBD_ALLOC(logname, PATH_MAX);
397 return ERR_PTR(-ENOMEM);
399 len = snprintf(logname, PATH_MAX, "LOGS/%s", name);
400 if (len >= PATH_MAX - 1) {
401 filp = ERR_PTR(-ENAMETOOLONG);
403 filp = l_filp_open(logname, flags, mode);
405 CERROR("logfile creation %s: %ld\n", logname,
409 OBD_FREE(logname, PATH_MAX);
413 /* This is a callback from the llog_* functions.
414 * Assumes caller has already pushed us into the kernel context. */
415 static int llog_lvfs_create(struct llog_ctxt *ctxt, struct llog_handle **res,
416 struct llog_logid *logid, char *name)
418 struct llog_handle *handle;
419 struct obd_device *obd;
420 struct l_dentry *dchild = NULL;
421 struct obdo *oa = NULL;
422 int rc = 0, cleanup_phase = 1;
423 int open_flags = O_RDWR | O_CREAT | O_LARGEFILE;
426 handle = llog_alloc_handle();
432 LASSERT(ctxt->loc_exp);
433 obd = ctxt->loc_exp->exp_obd;
436 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, logid->lgl_oid,
437 logid->lgl_ogen, logid->lgl_ogr);
439 if (IS_ERR(dchild)) {
440 rc = PTR_ERR(dchild);
441 CERROR("error looking up logfile "LPX64":0x%x: rc %d\n",
442 logid->lgl_oid, logid->lgl_ogen, rc);
447 if (dchild->d_inode == NULL) {
449 CERROR("nonexistent log file "LPX64":"LPX64": rc %d\n",
450 logid->lgl_oid, logid->lgl_ogr, rc);
454 handle->lgh_file = l_dentry_open(&obd->obd_ctxt, dchild,
455 O_RDWR | O_LARGEFILE);
456 if (IS_ERR(handle->lgh_file)) {
457 rc = PTR_ERR(handle->lgh_file);
458 CERROR("error opening logfile "LPX64"0x%x: rc %d\n",
459 logid->lgl_oid, logid->lgl_ogen, rc);
463 /* assign the value of lgh_id for handle directly */
464 handle->lgh_id = *logid;
467 handle->lgh_file = llog_filp_open(name, open_flags, 0644);
468 if (IS_ERR(handle->lgh_file))
469 GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
471 handle->lgh_id.lgl_ogr = 1;
472 handle->lgh_id.lgl_oid =
473 handle->lgh_file->f_dentry->d_inode->i_ino;
474 handle->lgh_id.lgl_ogen =
475 handle->lgh_file->f_dentry->d_inode->i_generation;
479 GOTO(cleanup, rc = -ENOMEM);
480 /* XXX get some filter group constants */
482 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
483 rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
487 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, oa->o_id,
488 oa->o_generation, oa->o_gr);
491 GOTO(cleanup, rc = PTR_ERR(dchild));
493 handle->lgh_file = l_dentry_open(&obd->obd_ctxt, dchild,
495 if (IS_ERR(handle->lgh_file))
496 GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
498 handle->lgh_id.lgl_ogr = oa->o_gr;
499 handle->lgh_id.lgl_oid = oa->o_id;
500 handle->lgh_id.lgl_ogen = oa->o_generation;
503 handle->lgh_ctxt = ctxt;
509 switch (cleanup_phase) {
513 llog_free_handle(handle);
518 static int llog_lvfs_close(struct llog_handle *handle)
523 rc = filp_close(handle->lgh_file, 0);
525 CERROR("error closing log: rc %d\n", rc);
529 static int llog_lvfs_destroy(struct llog_handle *handle)
539 oa->o_id = handle->lgh_id.lgl_oid;
540 oa->o_gr = handle->lgh_id.lgl_ogr;
541 oa->o_generation = handle->lgh_id.lgl_ogen;
542 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLGENER;
544 rc = llog_lvfs_close(handle);
548 rc = obd_destroy(handle->lgh_ctxt->loc_exp, oa, NULL, NULL);
554 /* reads the catalog list */
555 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
556 char *name, int count, struct llog_logid *idarray)
558 struct obd_run_ctxt saved;
561 int size = sizeof(*idarray) * count;
566 push_ctxt(&saved, &obd->obd_ctxt, NULL);
567 file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
568 if (!file || IS_ERR(file)) {
570 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
575 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
576 CERROR("%s is not a regular file!: mode = %o\n", name,
577 file->f_dentry->d_inode->i_mode);
578 GOTO(out, rc = -ENOENT);
581 rc = fsfilt_read_record(disk_obd, file, idarray, size, &off);
583 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
589 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
590 if (file && !IS_ERR(file))
591 rc = filp_close(file, 0);
594 EXPORT_SYMBOL(llog_get_cat_list);
596 /* writes the cat list */
597 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
598 char *name, int count, struct llog_logid *idarray)
600 struct obd_run_ctxt saved;
603 int size = sizeof(*idarray) * count;
608 push_ctxt(&saved, &obd->obd_ctxt, NULL);
609 file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
610 if (!file || IS_ERR(file)) {
612 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
617 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
618 CERROR("%s is not a regular file!: mode = %o\n", name,
619 file->f_dentry->d_inode->i_mode);
620 GOTO(out, rc = -ENOENT);
623 rc = fsfilt_write_record(disk_obd, file, idarray, size, &off, 1);
625 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
631 pop_ctxt(&saved, &obd->obd_ctxt, NULL);
632 if (file && !IS_ERR(file))
633 rc = filp_close(file, 0);
637 struct llog_operations llog_lvfs_ops = {
638 lop_write_rec: llog_lvfs_write_rec,
639 lop_next_block: llog_lvfs_next_block,
640 lop_read_header: llog_lvfs_read_header,
641 lop_create: llog_lvfs_create,
642 lop_destroy: llog_lvfs_destroy,
643 lop_close: llog_lvfs_close,
644 // lop_cancel: llog_lvfs_cancel,
647 EXPORT_SYMBOL(llog_lvfs_ops);
649 #else /* !__KERNEL__ */
651 static int llog_lvfs_read_header(struct llog_handle *handle)
657 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
658 struct llog_rec_hdr *rec,
659 struct llog_cookie *reccookie, int cookiecount,
666 static int llog_lvfs_next_block(struct llog_handle *loghandle, int *cur_idx,
667 int next_idx, __u64 *cur_offset, void *buf,
674 static int llog_lvfs_create(struct llog_obd_ctxt *ctxt,struct llog_handle **res,
675 struct llog_logid *logid, char *name)
681 static int llog_lvfs_close(struct llog_handle *handle)
687 static int llog_lvfs_destroy(struct llog_handle *handle)
693 int llog_get_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
694 char *name, int count, struct llog_logid *idarray)
700 int llog_put_cat_list(struct obd_device *obd, struct obd_device *disk_obd,
701 char *name, int count, struct llog_logid *idarray)
707 struct llog_operations llog_lvfs_ops = {
708 lop_write_rec: llog_lvfs_write_rec,
709 lop_next_block: llog_lvfs_next_block,
710 lop_read_header: llog_lvfs_read_header,
711 lop_create: llog_lvfs_create,
712 lop_destroy: llog_lvfs_destroy,
713 lop_close: llog_lvfs_close,
714 // lop_cancel: llog_lvfs_cancel,