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/lvfs.h>
42 #include <linux/lustre_fsfilt.h>
43 #include <linux/lustre_log.h>
47 static int llog_lvfs_pad(struct llog_ctxt *ctxt, struct l_file *file,
50 struct llog_rec_hdr rec;
51 struct llog_rec_tail tail;
55 LASSERT(len >= LLOG_MIN_REC_SIZE && (len & 0x7) == 0);
57 tail.lrt_len = rec.lrh_len = cpu_to_le32(len);
58 tail.lrt_index = rec.lrh_index = cpu_to_le32(index);
61 rc = llog_fsfilt_write_record(ctxt, file, &rec, sizeof(rec),
64 CERROR("error writing padding record: rc %d\n", rc);
68 file->f_pos += len - sizeof(rec) - sizeof(tail);
69 rc = llog_fsfilt_write_record(ctxt, file, &tail, sizeof(tail),
72 CERROR("error writing padding record: rc %d\n", rc);
80 static int llog_lvfs_write_blob(struct llog_ctxt *ctxt, struct l_file *file,
81 struct llog_rec_hdr *rec, void *buf, loff_t off)
84 struct llog_rec_tail end;
85 loff_t saved_off = file->f_pos;
86 int buflen = le32_to_cpu(rec->lrh_len);
92 rc = llog_fsfilt_write_record(ctxt, file, rec, buflen,
95 CERROR("error writing log record: rc %d\n", rc);
102 rec->lrh_len = cpu_to_le32(sizeof(*rec) + buflen + sizeof(end));
103 rc = llog_fsfilt_write_record(ctxt, file, rec, sizeof(*rec),
106 CERROR("error writing log hdr: rc %d\n", rc);
110 rc = llog_fsfilt_write_record(ctxt, file, buf, buflen,
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 = llog_fsfilt_write_record(ctxt, file, &end, sizeof(end),
122 CERROR("error writing log tail: rc %d\n", rc);
128 if (saved_off > file->f_pos)
129 file->f_pos = saved_off;
134 static int llog_lvfs_read_blob(struct llog_ctxt *ctxt, struct l_file *file,
135 void *buf, int size, loff_t off)
141 rc = llog_fsfilt_read_record(ctxt, file, buf, size, &offset);
143 CERROR("error reading log record: rc %d\n", rc);
149 static int llog_lvfs_read_header(struct llog_handle *handle)
151 struct llog_ctxt *ctxt = handle->lgh_ctxt;
155 LASSERT(sizeof(*handle->lgh_hdr) == LLOG_CHUNK_SIZE);
156 LASSERT(ctxt != NULL);
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(ctxt, 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,
182 struct llog_log_hdr *llh;
183 int reclen = le32_to_cpu(rec->lrh_len), index, rc;
184 struct llog_rec_tail *lrt;
185 struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
191 llh = loghandle->lgh_hdr;
192 file = loghandle->lgh_file;
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(ctxt, 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(ctxt, 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(ctxt, file, left, loghandle->lgh_last_idx);
249 /* if it's the last idx in log file, then return -ENOSPC */
250 if (loghandle->lgh_last_idx == LLOG_BITMAP_SIZE(llh) - 1)
254 loghandle->lgh_last_idx++;
255 index = loghandle->lgh_last_idx;
256 rec->lrh_index = cpu_to_le32(index);
258 lrt = (void *)rec + le32_to_cpu(rec->lrh_len) - sizeof(*lrt);
259 lrt->lrt_len = rec->lrh_len;
260 lrt->lrt_index = rec->lrh_index;
262 if (ext2_set_bit(index, llh->llh_bitmap)) {
263 CERROR("argh, index %u already set in log bitmap?\n", index);
264 LBUG(); /* should never happen */
266 llh->llh_count = cpu_to_le32(le32_to_cpu(llh->llh_count) + 1);
267 llh->llh_tail.lrt_index = cpu_to_le32(index);
270 rc = llog_lvfs_write_blob(ctxt, file, &llh->llh_hdr, NULL, 0);
274 CDEBUG(D_HA, "adding record "LPX64": idx: %u, %u bytes off: %lld\n",
275 loghandle->lgh_id.lgl_oid, index, le32_to_cpu(rec->lrh_len),
278 rc = llog_lvfs_write_blob(ctxt, file, rec, buf, file->f_pos);
282 if (rc == 0 && reccookie) {
283 if (llog_cookie_get_flags(reccookie) & LLOG_COOKIE_REPLAY) {
284 LASSERTF(EQ_LOGID(reccookie->lgc_lgl,loghandle->lgh_id),
285 "lgc_lgl.oid/gr "LPU64"/"LPU64" lgh_id.oid/gr"
287 reccookie->lgc_lgl.lgl_oid,
288 reccookie->lgc_lgl.lgl_ogr,
289 loghandle->lgh_id.lgl_oid,
290 loghandle->lgh_id.lgl_oid);
291 LASSERTF(reccookie->lgc_index == index,
292 "lgc_index %u != index %u\n",
293 reccookie->lgc_index, index);
295 reccookie->lgc_lgl = loghandle->lgh_id;
296 reccookie->lgc_index = index;
297 llog_cookie_add_flags(reccookie, LLOG_COOKIE_REPLAY);
300 if (le32_to_cpu(rec->lrh_type) == MDS_UNLINK_REC)
301 reccookie->lgc_subsys = LLOG_UNLINK_ORIG_CTXT;
302 else if (le32_to_cpu(rec->lrh_type) == OST_SZ_REC)
303 reccookie->lgc_subsys = LLOG_SIZE_ORIG_CTXT;
304 else if (le32_to_cpu(rec->lrh_type) == OST_RAID1_REC)
305 reccookie->lgc_subsys = LLOG_RD1_ORIG_CTXT;
307 reccookie->lgc_subsys = -1;
310 if (rc == 0 && (le32_to_cpu(rec->lrh_type) == LLOG_GEN_REC ||
311 le32_to_cpu(rec->lrh_type) == SMFS_UPDATE_REC))
317 /* We can skip reading at least as many log blocks as the number of
318 * minimum sized log records we are skipping. If it turns out
319 * that we are not far enough along the log (because the
320 * actual records are larger than minimum size) we just skip
321 * some more records. */
323 static void llog_skip_over(__u64 *off, int curr, int goal)
327 *off = (*off + (goal-curr-1) * LLOG_MIN_REC_SIZE) &
328 ~(LLOG_CHUNK_SIZE - 1);
332 * - curr_offset to the furthest point read in the log file
333 * - curr_idx to the log index preceeding curr_offset
334 * returns -EIO/-EINVAL on error
336 static int llog_lvfs_next_block(struct llog_handle *loghandle, int *curr_idx,
337 int next_idx, __u64 *curr_offset, void *buf,
340 struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
343 if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
346 CDEBUG(D_OTHER, "looking for log index %u (cur idx %u off "LPU64")\n",
347 next_idx, *curr_idx, *curr_offset);
349 while (*curr_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
350 struct llog_rec_hdr *rec;
351 struct llog_rec_tail *tail;
355 llog_skip_over(curr_offset, *curr_idx, next_idx);
358 rc = llog_fsfilt_read_record(ctxt, loghandle->lgh_file,
362 CERROR("Cant read llog block at log id "LPU64
363 "/%u offset "LPU64"\n",
364 loghandle->lgh_id.lgl_oid,
365 loghandle->lgh_id.lgl_ogen,
370 nbytes = ppos - *curr_offset;
373 if (nbytes == 0) /* end of file, nothing to do */
376 if (nbytes < sizeof(*tail)) {
377 CERROR("Invalid llog block at log id "LPU64"/%u offset "
378 LPU64"\n", loghandle->lgh_id.lgl_oid,
379 loghandle->lgh_id.lgl_ogen, *curr_offset);
383 tail = buf + nbytes - sizeof(struct llog_rec_tail);
384 *curr_idx = le32_to_cpu(tail->lrt_index);
386 /* this shouldn't happen */
387 if (tail->lrt_index == 0) {
388 CERROR("Invalid llog tail at log id "LPU64"/%u offset "
389 LPU64"\n", loghandle->lgh_id.lgl_oid,
390 loghandle->lgh_id.lgl_ogen, *curr_offset);
393 if (le32_to_cpu(tail->lrt_index) < next_idx)
396 /* sanity check that the start of the new buffer is no farther
397 * than the record that we wanted. This shouldn't happen. */
399 if (le32_to_cpu(rec->lrh_index) > next_idx) {
400 CERROR("missed desired record? %u > %u\n",
401 le32_to_cpu(rec->lrh_index), next_idx);
409 static int llog_lvfs_prev_block(struct llog_handle *loghandle,
410 int prev_idx, void *buf, int len)
412 struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
417 if (len == 0 || len & (LLOG_CHUNK_SIZE - 1))
420 CDEBUG(D_OTHER, "looking for log index %u n", prev_idx);
422 curr_offset = LLOG_CHUNK_SIZE;
423 llog_skip_over(&curr_offset, 0, prev_idx);
425 while (curr_offset < loghandle->lgh_file->f_dentry->d_inode->i_size) {
426 struct llog_rec_hdr *rec;
427 struct llog_rec_tail *tail;
431 rc = llog_fsfilt_read_record(ctxt, loghandle->lgh_file,
435 CERROR("Cant read llog block at log id "LPU64
436 "/%u offset "LPU64"\n",
437 loghandle->lgh_id.lgl_oid,
438 loghandle->lgh_id.lgl_ogen,
443 /* put number of bytes read into rc to make code simpler */
444 rc = ppos - curr_offset;
447 if (rc == 0) /* end of file, nothing to do */
450 if (rc < sizeof(*tail)) {
451 CERROR("Invalid llog block at log id "LPU64"/%u offset "
452 LPU64"\n", loghandle->lgh_id.lgl_oid,
453 loghandle->lgh_id.lgl_ogen, curr_offset);
457 tail = buf + rc - sizeof(struct llog_rec_tail);
459 /* this shouldn't happen */
460 if (tail->lrt_index == 0) {
461 CERROR("Invalid llog tail at log id "LPU64"/%u offset "
462 LPU64"\n", loghandle->lgh_id.lgl_oid,
463 loghandle->lgh_id.lgl_ogen, curr_offset);
466 if (le32_to_cpu(tail->lrt_index) < prev_idx)
469 /* sanity check that the start of the new buffer is no farther
470 * than the record that we wanted. This shouldn't happen. */
472 if (le32_to_cpu(rec->lrh_index) > prev_idx) {
473 CERROR("missed desired record? %u > %u\n",
474 le32_to_cpu(rec->lrh_index), prev_idx);
482 static struct file *llog_filp_open(char *name, int flags, int mode)
488 OBD_ALLOC(logname, PATH_MAX);
490 return ERR_PTR(-ENOMEM);
492 len = snprintf(logname, PATH_MAX, "LOGS/%s", name);
493 if (len >= PATH_MAX - 1) {
494 filp = ERR_PTR(-ENAMETOOLONG);
496 filp = l_filp_open(logname, flags, mode);
498 CERROR("logfile %s(%s): %ld\n",
499 flags & O_CREAT ? "create" : "open", logname,
504 OBD_FREE(logname, PATH_MAX);
508 /* creates object for the case when we have no obd (smfs). */
510 llog_object_create_alone(struct llog_ctxt *ctxt, struct llog_logid *lgh_id)
516 LASSERT(lgh_id != NULL);
517 if (lgh_id->lgl_oid) {
518 struct dentry *dchild;
519 char fidname[LL_FID_NAMELEN];
522 down(&ctxt->loc_objects_dir->d_inode->i_sem);
523 fidlen = ll_fid2str(fidname, lgh_id->lgl_oid, lgh_id->lgl_ogen);
524 dchild = lookup_one_len(fidname, ctxt->loc_objects_dir, fidlen);
525 if (IS_ERR(dchild)) {
526 up(&ctxt->loc_objects_dir->d_inode->i_sem);
527 RETURN((struct file *)dchild);
529 if (dchild->d_inode == NULL) {
530 struct dentry_params dp;
533 dchild->d_fsdata = (void *) &dp;
535 dp.p_inum = lgh_id->lgl_oid;
536 rc = ll_vfs_create(ctxt->loc_objects_dir->d_inode,
537 dchild, S_IFREG, NULL);
538 if (dchild->d_fsdata == (void *)(unsigned long)lgh_id->lgl_oid)
539 dchild->d_fsdata = NULL;
541 CDEBUG(D_INODE, "err during create: %d\n", rc);
543 up(&ctxt->loc_objects_dir->d_inode->i_sem);
546 inode = dchild->d_inode;
547 LASSERT(inode->i_ino == lgh_id->lgl_oid);
548 inode->i_generation = lgh_id->lgl_ogen;
549 CDEBUG(D_HA, "recreated ino %lu with gen %u\n",
550 inode->i_ino, inode->i_generation);
551 mark_inode_dirty(inode);
554 mntget(ctxt->loc_lvfs_ctxt->pwdmnt);
555 filp = dentry_open(dchild, ctxt->loc_lvfs_ctxt->pwdmnt,
556 O_RDWR | O_LARGEFILE);
559 up(&ctxt->loc_objects_dir->d_inode->i_sem);
562 if (!S_ISREG(filp->f_dentry->d_inode->i_mode)) {
563 CERROR("%s is not a regular file!: mode = %o\n", fidname,
564 filp->f_dentry->d_inode->i_mode);
566 up(&ctxt->loc_objects_dir->d_inode->i_sem);
567 RETURN(ERR_PTR(-ENOENT));
570 up(&ctxt->loc_objects_dir->d_inode->i_sem);
574 unsigned int tmpname = ll_insecure_random_int();
575 char fidname[LL_FID_NAMELEN];
576 struct dentry *new_child, *parent;
580 sprintf(fidname, "OBJECTS/%u", tmpname);
581 filp = filp_open(fidname, O_CREAT | O_EXCL, 0644);
585 CERROR("impossible object name collision %u\n",
589 CERROR("error creating tmp object %u: rc %d\n", tmpname, rc);
593 namelen = ll_fid2str(fidname, filp->f_dentry->d_inode->i_ino,
594 filp->f_dentry->d_inode->i_generation);
595 parent = filp->f_dentry->d_parent;
596 down(&parent->d_inode->i_sem);
597 new_child = lookup_one_len(fidname, parent, namelen);
598 if (IS_ERR(new_child)) {
599 CERROR("getting neg dentry for obj rename: %d\n", rc);
600 GOTO(out_close, rc = PTR_ERR(new_child));
602 if (new_child->d_inode != NULL) {
603 CERROR("impossible non-negative obj dentry %lu:%u!\n",
604 filp->f_dentry->d_inode->i_ino,
605 filp->f_dentry->d_inode->i_generation);
609 handle = llog_fsfilt_start(ctxt, parent->d_inode, FSFILT_OP_RENAME, NULL);
611 GOTO(out_dput, rc = PTR_ERR(handle));
614 rc = vfs_rename(parent->d_inode, filp->f_dentry,
615 parent->d_inode, new_child);
618 CERROR("error renaming new object %lu:%u: rc %d\n",
619 filp->f_dentry->d_inode->i_ino,
620 filp->f_dentry->d_inode->i_generation, rc);
622 err = llog_fsfilt_commit(ctxt, parent->d_inode, handle, 0);
629 up(&parent->d_inode->i_sem);
634 /* FIXME: is this group 1 is correct? */
636 lgh_id->lgl_oid = filp->f_dentry->d_inode->i_ino;
637 lgh_id->lgl_ogen = filp->f_dentry->d_inode->i_generation;
643 /* creates object for generic case (obd exists) */
645 llog_object_create_generic(struct llog_ctxt *ctxt, struct llog_logid *lgh_id)
647 struct file *filp = NULL;
648 struct dentry *dchild;
649 struct obd_device *obd;
650 struct obdo *oa = NULL;
651 int open_flags = O_RDWR | O_LARGEFILE;
655 obd = ctxt->loc_exp->exp_obd;
656 LASSERT(obd != NULL);
658 if (lgh_id->lgl_oid) {
659 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, lgh_id->lgl_oid,
660 lgh_id->lgl_ogen, lgh_id->lgl_ogr);
661 if (IS_ERR(dchild) == -ENOENT) {
662 OBD_ALLOC(oa, sizeof(*oa));
664 RETURN(ERR_PTR(-ENOMEM));
666 oa->o_id = lgh_id->lgl_oid;
667 oa->o_generation = lgh_id->lgl_ogen;
668 oa->o_gr = lgh_id->lgl_ogr;
669 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
670 rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
672 CDEBUG(D_INODE, "err during create: %d\n", rc);
673 GOTO(out_free_oa, rc);
675 CDEBUG(D_HA, "re-create log object "LPX64":0x%x:"LPX64"\n",
676 lgh_id->lgl_oid, lgh_id->lgl_ogen, lgh_id->lgl_ogr);
678 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, lgh_id->lgl_oid,
679 lgh_id->lgl_ogen, lgh_id->lgl_ogr);
680 } else if (IS_ERR(dchild)) {
681 CERROR("error looking up logfile "LPX64":0x%x: rc %d\n",
682 lgh_id->lgl_oid, lgh_id->lgl_ogen, rc);
683 RETURN((struct file *)dchild);
686 filp = l_dentry_open(&obd->obd_lvfs_ctxt, dchild, open_flags);
690 CERROR("error opening logfile "LPX64"0x%x: rc %d\n",
691 lgh_id->lgl_oid, lgh_id->lgl_ogen, rc);
693 GOTO(out_free_oa, rc);
695 /* this is important to work here over obd_create() as it manages
696 groups and we need it. Yet another reason is that mds_obd_create()
697 is fully the same as old version of this function and this helps
698 us to avoid code duplicating and layering violating. */
699 OBD_ALLOC(oa, sizeof(*oa));
701 RETURN(ERR_PTR(-ENOMEM));
703 oa->o_gr = FILTER_GROUP_LLOG;
704 oa->o_valid = OBD_MD_FLGENER | OBD_MD_FLGROUP;
705 rc = obd_create(ctxt->loc_exp, oa, NULL, NULL);
707 GOTO(out_free_oa, rc);
709 dchild = obd_lvfs_fid2dentry(ctxt->loc_exp, oa->o_id,
710 oa->o_generation, oa->o_gr);
712 GOTO(out_free_oa, rc = PTR_ERR(dchild));
714 filp = l_dentry_open(&obd->obd_lvfs_ctxt, dchild,
718 GOTO(out_free_oa, rc = PTR_ERR(filp));
721 /* group 1 is not longer valid, we use the group which is set
722 by obd_create()->mds_obd_create(). */
723 lgh_id->lgl_ogr = oa->o_gr;
724 lgh_id->lgl_oid = oa->o_id;
725 lgh_id->lgl_ogen = oa->o_generation;
732 OBD_FREE(oa, sizeof(*oa));
737 llog_object_create(struct llog_ctxt *ctxt, struct llog_logid *lgh_id)
740 return llog_object_create_alone(ctxt, lgh_id);
742 return llog_object_create_generic(ctxt, lgh_id);
745 static int llog_add_link_object(struct llog_ctxt *ctxt, struct llog_logid logid,
746 struct dentry *dentry)
748 struct dentry *new_child;
749 char fidname[LL_FID_NAMELEN];
751 int namelen, rc = 0, err;
754 namelen = ll_fid2str(fidname, logid.lgl_oid, logid.lgl_ogen);
755 down(&ctxt->loc_objects_dir->d_inode->i_sem);
756 new_child = lookup_one_len(fidname, ctxt->loc_objects_dir, namelen);
757 if (IS_ERR(new_child)) {
758 CERROR("getting neg dentry for obj rename: %d\n", rc);
759 GOTO(out, rc = PTR_ERR(new_child));
761 if (new_child->d_inode == dentry->d_inode)
763 if (new_child->d_inode != NULL) {
764 CERROR("impossible non-negative obj dentry "LPX64":%u!\n",
765 logid.lgl_oid, logid.lgl_ogen);
768 handle = llog_fsfilt_start(ctxt, ctxt->loc_objects_dir->d_inode,
769 FSFILT_OP_LINK, NULL);
771 GOTO(out_dput, rc = PTR_ERR(handle));
774 rc = vfs_link(dentry, ctxt->loc_objects_dir->d_inode, new_child);
777 CERROR("error link new object "LPX64":%u: rc %d\n",
778 logid.lgl_oid, logid.lgl_ogen, rc);
779 err = llog_fsfilt_commit(ctxt, ctxt->loc_objects_dir->d_inode, handle, 0);
783 up(&ctxt->loc_objects_dir->d_inode->i_sem);
787 static int llog_lvfs_open(struct llog_ctxt *ctxt, struct llog_handle **res,
788 struct llog_logid *logid, char *name, int flags)
790 struct llog_handle *handle;
791 struct lvfs_run_ctxt saved;
793 int open_flags = O_RDWR | O_LARGEFILE;
796 if (flags & OBD_LLOG_FL_CREATE)
797 open_flags |= O_CREAT;
799 handle = llog_alloc_handle();
805 if (ctxt->loc_lvfs_ctxt)
806 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
809 handle->lgh_file = llog_object_create(ctxt, logid);
810 if (IS_ERR(handle->lgh_file)) {
811 CERROR("cannot create/open llog object "LPX64":%x "
812 "error = %ld", logid->lgl_oid, logid->lgl_ogen,
813 PTR_ERR(handle->lgh_file));
814 GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
816 handle->lgh_id = *logid;
819 handle->lgh_file = llog_filp_open(name, open_flags, 0644);
820 if (IS_ERR(handle->lgh_file)) {
821 CERROR("cannot open %s file, error = %ld\n",
822 name, PTR_ERR(handle->lgh_file));
823 GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
825 LASSERT(handle->lgh_file->f_dentry->d_parent == ctxt->loc_logs_dir);
827 handle->lgh_id.lgl_ogr = 1;
828 handle->lgh_id.lgl_oid = handle->lgh_file->f_dentry->d_inode->i_ino;
829 handle->lgh_id.lgl_ogen = handle->lgh_file->f_dentry->d_inode->i_generation;
830 rc = llog_add_link_object(ctxt, handle->lgh_id, handle->lgh_file->f_dentry);
835 handle->lgh_file = llog_object_create(ctxt, &handle->lgh_id);
836 if (IS_ERR(handle->lgh_file)) {
837 CERROR("cannot create llog object, error = %ld\n",
838 PTR_ERR(handle->lgh_file));
839 GOTO(cleanup, rc = PTR_ERR(handle->lgh_file));
843 handle->lgh_ctxt = ctxt;
845 if (ctxt->loc_lvfs_ctxt)
846 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
849 llog_free_handle(handle);
853 static int llog_lvfs_close(struct llog_handle *handle)
858 rc = filp_close(handle->lgh_file, 0);
860 CERROR("error closing log: rc %d\n", rc);
864 static int llog_lvfs_destroy(struct llog_handle *loghandle)
866 struct llog_ctxt *ctxt = loghandle->lgh_ctxt;
867 struct lvfs_run_ctxt saved;
868 struct dentry *fdentry;
869 struct inode *parent_inode;
870 char fidname[LL_FID_NAMELEN];
872 int rc = -EINVAL, err, namelen;
875 if (ctxt->loc_lvfs_ctxt)
876 push_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
878 fdentry = loghandle->lgh_file->f_dentry;
879 parent_inode = fdentry->d_parent->d_inode;
881 if (!strcmp(fdentry->d_parent->d_name.name, "LOGS")) {
882 LASSERT(parent_inode == ctxt->loc_logs_dir->d_inode);
884 namelen = ll_fid2str(fidname, fdentry->d_inode->i_ino,
885 fdentry->d_inode->i_generation);
887 rc = llog_lvfs_close(loghandle);
893 handle = llog_fsfilt_start(ctxt, parent_inode,
894 FSFILT_OP_UNLINK, NULL);
895 if (IS_ERR(handle)) {
897 GOTO(out, rc = PTR_ERR(handle));
900 down(&parent_inode->i_sem);
901 rc = vfs_unlink(parent_inode, fdentry);
902 up(&parent_inode->i_sem);
906 down(&ctxt->loc_objects_dir->d_inode->i_sem);
907 fdentry = lookup_one_len(fidname, ctxt->loc_objects_dir,
909 if (fdentry == NULL || fdentry->d_inode == NULL) {
910 CERROR("destroy non_existent object %s\n", fidname);
911 GOTO(out_err, rc = IS_ERR(fdentry) ?
912 PTR_ERR(fdentry) : -ENOENT);
914 rc = vfs_unlink(ctxt->loc_objects_dir->d_inode, fdentry);
917 up(&ctxt->loc_objects_dir->d_inode->i_sem);
919 err = llog_fsfilt_commit(ctxt, parent_inode, handle, 0);
925 if (ctxt->loc_alone) {
926 if (!strcmp(fdentry->d_parent->d_name.name, "OBJECTS")) {
927 LASSERT(parent_inode == ctxt->loc_objects_dir->d_inode);
930 rc = llog_lvfs_close(loghandle);
932 down(&parent_inode->i_sem);
933 rc = vfs_unlink(parent_inode, fdentry);
934 up(&parent_inode->i_sem);
939 struct obdo *oa = NULL;
941 OBD_ALLOC(oa, sizeof(*oa));
943 GOTO(out, rc = -ENOMEM);
945 oa->o_id = loghandle->lgh_id.lgl_oid;
946 oa->o_gr = loghandle->lgh_id.lgl_ogr;
947 oa->o_generation = loghandle->lgh_id.lgl_ogen;
948 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP | OBD_MD_FLGENER;
950 rc = llog_lvfs_close(loghandle);
952 GOTO(out_free_oa, rc);
954 rc = obd_destroy(loghandle->lgh_ctxt->loc_exp, oa, NULL, NULL);
956 OBD_FREE(oa, sizeof(*oa));
959 if (ctxt->loc_lvfs_ctxt)
960 pop_ctxt(&saved, ctxt->loc_lvfs_ctxt, NULL);
964 /* reads the catalog list */
965 int llog_get_cat_list(struct lvfs_run_ctxt *ctxt,
966 struct fsfilt_operations *fsops, char *name,
967 int count, struct llog_catid *idarray)
969 struct lvfs_run_ctxt saved;
971 int size = sizeof(*idarray) * count;
978 push_ctxt(&saved, ctxt, NULL);
979 file = l_filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
980 if (!file || IS_ERR(file)) {
982 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
987 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
988 CERROR("%s is not a regular file!: mode = %o\n", name,
989 file->f_dentry->d_inode->i_mode);
990 GOTO(out, rc = -ENOENT);
993 rc = fsops->fs_read_record(file, idarray, size, &off);
995 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
1001 if (file && !IS_ERR(file))
1002 rc = filp_close(file, 0);
1004 pop_ctxt(&saved, ctxt, NULL);
1007 EXPORT_SYMBOL(llog_get_cat_list);
1009 /* writes the cat list */
1010 int llog_put_cat_list(struct lvfs_run_ctxt *ctxt,
1011 struct fsfilt_operations *fsops, char *name,
1012 int count, struct llog_catid *idarray)
1014 struct lvfs_run_ctxt saved;
1015 struct l_file *file;
1016 int size = sizeof(*idarray) * count;
1023 push_ctxt(&saved, ctxt, NULL);
1024 file = filp_open(name, O_RDWR | O_CREAT | O_LARGEFILE, 0700);
1025 if (!file || IS_ERR(file)) {
1027 CERROR("OBD filter: cannot open/create %s: rc = %d\n",
1032 if (!S_ISREG(file->f_dentry->d_inode->i_mode)) {
1033 CERROR("%s is not a regular file!: mode = %o\n", name,
1034 file->f_dentry->d_inode->i_mode);
1035 GOTO(out, rc = -ENOENT);
1038 rc = fsops->fs_write_record(file, idarray, size, &off, 1);
1040 CDEBUG(D_INODE,"OBD filter: error reading %s: rc %d\n",
1046 if (file && !IS_ERR(file))
1047 rc = filp_close(file, 0);
1049 pop_ctxt(&saved, ctxt, NULL);
1052 EXPORT_SYMBOL(llog_put_cat_list);
1054 struct llog_operations llog_lvfs_ops = {
1055 lop_open: llog_lvfs_open,
1056 lop_destroy: llog_lvfs_destroy,
1057 lop_close: llog_lvfs_close,
1058 lop_read_header: llog_lvfs_read_header,
1059 lop_write_rec: llog_lvfs_write_rec,
1060 lop_next_block: llog_lvfs_next_block,
1061 lop_prev_block: llog_lvfs_prev_block,
1063 EXPORT_SYMBOL(llog_lvfs_ops);
1065 #else /* !__KERNEL__ */
1067 static int llog_lvfs_read_header(struct llog_handle *handle)
1073 static int llog_lvfs_write_rec(struct llog_handle *loghandle,
1074 struct llog_rec_hdr *rec,
1075 struct llog_cookie *reccookie, int cookiecount,
1082 static int llog_lvfs_open(struct llog_ctxt *ctxt, struct llog_handle **res,
1083 struct llog_logid *logid, char *name, int flags)
1089 static int llog_lvfs_close(struct llog_handle *handle)
1095 static int llog_lvfs_destroy(struct llog_handle *handle)
1101 int llog_get_cat_list(struct lvfs_run_ctxt *ctxt,
1102 struct fsfilt_operations *fsops, char *name,
1103 int count, struct llog_catid *idarray)
1109 int llog_put_cat_list(struct lvfs_run_ctxt *ctxt,
1110 struct fsfilt_operations *fsops, char *name,
1111 int count, struct llog_catid *idarray)
1117 int llog_lvfs_prev_block(struct llog_handle *loghandle,
1118 int prev_idx, void *buf, int len)
1124 int llog_lvfs_next_block(struct llog_handle *loghandle, int *curr_idx,
1125 int next_idx, __u64 *offset, void *buf, int len)
1131 struct llog_operations llog_lvfs_ops = {
1132 lop_open: llog_lvfs_open,
1133 lop_destroy: llog_lvfs_destroy,
1134 lop_close: llog_lvfs_close,
1135 lop_read_header: llog_lvfs_read_header,
1136 lop_write_rec: llog_lvfs_write_rec,
1137 lop_next_block: llog_lvfs_next_block,
1138 lop_prev_block: llog_lvfs_prev_block,