2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
7 * of the GNU General Public License version 2.
10 #include <linux/sched.h>
11 #include <linux/slab.h>
12 #include <linux/spinlock.h>
13 #include <linux/completion.h>
14 #include <linux/buffer_head.h>
15 #include <linux/gfs2_ondisk.h>
16 #include <linux/bio.h>
32 * gfs2_pin - Pin a buffer in memory
33 * @sdp: The superblock
34 * @bh: The buffer to be pinned
36 * The log lock must be held when calling this function
38 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
40 struct gfs2_bufdata *bd;
42 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
44 clear_buffer_dirty(bh);
45 if (test_set_buffer_pinned(bh))
46 gfs2_assert_withdraw(sdp, 0);
47 if (!buffer_uptodate(bh))
48 gfs2_io_error_bh(sdp, bh);
50 /* If this buffer is in the AIL and it has already been written
51 * to in-place disk block, remove it from the AIL.
54 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
59 * gfs2_unpin - Unpin a buffer
60 * @sdp: the filesystem the buffer belongs to
61 * @bh: The buffer to unpin
66 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
69 struct gfs2_bufdata *bd = bh->b_private;
71 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
73 if (!buffer_pinned(bh))
74 gfs2_assert_withdraw(sdp, 0);
77 mark_buffer_dirty(bh);
78 clear_buffer_pinned(bh);
82 list_del(&bd->bd_ail_st_list);
85 struct gfs2_glock *gl = bd->bd_gl;
86 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
87 atomic_inc(&gl->gl_ail_count);
90 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
91 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
97 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
99 return (struct gfs2_log_descriptor *)bh->b_data;
102 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
104 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
105 return (__force __be64 *)(ld + 1);
108 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
110 return (__force __be64 *)(bh->b_data + bh->b_size);
114 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
116 struct buffer_head *bh = gfs2_log_get_buf(sdp);
117 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
118 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
119 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
120 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
121 ld->ld_type = cpu_to_be32(ld_type);
125 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
129 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
131 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
132 struct gfs2_trans *tr;
134 lock_buffer(bd->bd_bh);
136 if (!list_empty(&bd->bd_list_tr))
138 tr = current->journal_info;
141 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
142 if (!list_empty(&le->le_list))
144 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
145 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
146 gfs2_meta_check(sdp, bd->bd_bh);
147 gfs2_pin(sdp, bd->bd_bh);
148 sdp->sd_log_num_buf++;
149 list_add(&le->le_list, &sdp->sd_log_le_buf);
150 tr->tr_num_buf_new++;
152 gfs2_log_unlock(sdp);
153 unlock_buffer(bd->bd_bh);
156 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
158 struct buffer_head *bh;
159 struct gfs2_log_descriptor *ld;
160 struct gfs2_bufdata *bd1 = NULL, *bd2;
167 limit = buf_limit(sdp);
168 /* for 4k blocks, limit = 503 */
171 total = sdp->sd_log_num_buf;
172 bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
177 gfs2_log_unlock(sdp);
178 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
180 ld = bh_log_desc(bh);
181 ptr = bh_log_ptr(bh);
182 ld->ld_length = cpu_to_be32(num + 1);
183 ld->ld_data1 = cpu_to_be32(num);
186 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
188 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
193 gfs2_log_unlock(sdp);
194 submit_bh(WRITE_SYNC_PLUG, bh);
198 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
201 gfs2_log_unlock(sdp);
202 lock_buffer(bd2->bd_bh);
203 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
204 submit_bh(WRITE_SYNC_PLUG, bh);
213 gfs2_log_unlock(sdp);
216 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
218 struct list_head *head = &sdp->sd_log_le_buf;
219 struct gfs2_bufdata *bd;
221 while (!list_empty(head)) {
222 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
223 list_del_init(&bd->bd_le.le_list);
224 sdp->sd_log_num_buf--;
226 gfs2_unpin(sdp, bd->bd_bh, ai);
228 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
231 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
232 struct gfs2_log_header_host *head, int pass)
234 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
239 sdp->sd_found_blocks = 0;
240 sdp->sd_replayed_blocks = 0;
243 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
244 struct gfs2_log_descriptor *ld, __be64 *ptr,
247 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
248 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
249 struct gfs2_glock *gl = ip->i_gl;
250 unsigned int blks = be32_to_cpu(ld->ld_data1);
251 struct buffer_head *bh_log, *bh_ip;
255 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
258 gfs2_replay_incr_blk(sdp, &start);
260 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
261 blkno = be64_to_cpu(*ptr++);
263 sdp->sd_found_blocks++;
265 if (gfs2_revoke_check(sdp, blkno, start))
268 error = gfs2_replay_read_block(jd, start, &bh_log);
272 bh_ip = gfs2_meta_new(gl, blkno);
273 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
275 if (gfs2_meta_check(sdp, bh_ip))
278 mark_buffer_dirty(bh_ip);
286 sdp->sd_replayed_blocks++;
292 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
294 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
295 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
298 gfs2_meta_sync(ip->i_gl);
304 gfs2_meta_sync(ip->i_gl);
306 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
307 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
310 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
312 struct gfs2_trans *tr;
314 tr = current->journal_info;
317 sdp->sd_log_num_revoke++;
318 list_add(&le->le_list, &sdp->sd_log_le_revoke);
321 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
323 struct gfs2_log_descriptor *ld;
324 struct gfs2_meta_header *mh;
325 struct buffer_head *bh;
327 struct list_head *head = &sdp->sd_log_le_revoke;
328 struct gfs2_bufdata *bd;
330 if (!sdp->sd_log_num_revoke)
333 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
334 ld = bh_log_desc(bh);
335 ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
337 ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
338 offset = sizeof(struct gfs2_log_descriptor);
340 while (!list_empty(head)) {
341 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
342 list_del_init(&bd->bd_le.le_list);
343 sdp->sd_log_num_revoke--;
345 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
346 submit_bh(WRITE_SYNC_PLUG, bh);
348 bh = gfs2_log_get_buf(sdp);
349 mh = (struct gfs2_meta_header *)bh->b_data;
350 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
351 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
352 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
353 offset = sizeof(struct gfs2_meta_header);
356 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
357 kmem_cache_free(gfs2_bufdata_cachep, bd);
359 offset += sizeof(u64);
361 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
363 submit_bh(WRITE_SYNC_PLUG, bh);
366 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
367 struct gfs2_log_header_host *head, int pass)
369 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
374 sdp->sd_found_revokes = 0;
375 sdp->sd_replay_tail = head->lh_tail;
378 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
379 struct gfs2_log_descriptor *ld, __be64 *ptr,
382 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
383 unsigned int blks = be32_to_cpu(ld->ld_length);
384 unsigned int revokes = be32_to_cpu(ld->ld_data1);
385 struct buffer_head *bh;
391 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
394 offset = sizeof(struct gfs2_log_descriptor);
396 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
397 error = gfs2_replay_read_block(jd, start, &bh);
402 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
404 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
405 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
407 error = gfs2_revoke_add(sdp, blkno, start);
413 sdp->sd_found_revokes++;
417 offset += sizeof(u64);
421 offset = sizeof(struct gfs2_meta_header);
428 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
430 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
433 gfs2_revoke_clean(sdp);
439 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
440 jd->jd_jid, sdp->sd_found_revokes);
442 gfs2_revoke_clean(sdp);
445 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
447 struct gfs2_rgrpd *rgd;
448 struct gfs2_trans *tr = current->journal_info;
452 rgd = container_of(le, struct gfs2_rgrpd, rd_le);
455 if (!list_empty(&le->le_list)){
456 gfs2_log_unlock(sdp);
459 gfs2_rgrp_bh_hold(rgd);
460 sdp->sd_log_num_rg++;
461 list_add(&le->le_list, &sdp->sd_log_le_rg);
462 gfs2_log_unlock(sdp);
465 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
467 struct list_head *head = &sdp->sd_log_le_rg;
468 struct gfs2_rgrpd *rgd;
470 while (!list_empty(head)) {
471 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
472 list_del_init(&rgd->rd_le.le_list);
473 sdp->sd_log_num_rg--;
475 gfs2_rgrp_repolish_clones(rgd);
476 gfs2_rgrp_bh_put(rgd);
478 gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
482 * databuf_lo_add - Add a databuf to the transaction.
484 * This is used in two distinct cases:
485 * i) In ordered write mode
486 * We put the data buffer on a list so that we can ensure that its
487 * synced to disk at the right time
488 * ii) In journaled data mode
489 * We need to journal the data block in the same way as metadata in
490 * the functions above. The difference is that here we have a tag
491 * which is two __be64's being the block number (as per meta data)
492 * and a flag which says whether the data block needs escaping or
493 * not. This means we need a new log entry for each 251 or so data
494 * blocks, which isn't an enormous overhead but twice as much as
495 * for normal metadata blocks.
497 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
499 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
500 struct gfs2_trans *tr = current->journal_info;
501 struct address_space *mapping = bd->bd_bh->b_page->mapping;
502 struct gfs2_inode *ip = GFS2_I(mapping->host);
504 lock_buffer(bd->bd_bh);
507 if (!list_empty(&bd->bd_list_tr))
510 if (gfs2_is_jdata(ip)) {
512 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
515 if (!list_empty(&le->le_list))
518 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
519 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
520 if (gfs2_is_jdata(ip)) {
521 gfs2_pin(sdp, bd->bd_bh);
522 tr->tr_num_databuf_new++;
523 sdp->sd_log_num_databuf++;
524 list_add(&le->le_list, &sdp->sd_log_le_databuf);
526 list_add(&le->le_list, &sdp->sd_log_le_ordered);
529 gfs2_log_unlock(sdp);
530 unlock_buffer(bd->bd_bh);
533 static void gfs2_check_magic(struct buffer_head *bh)
538 clear_buffer_escaped(bh);
539 kaddr = kmap_atomic(bh->b_page, KM_USER0);
540 ptr = kaddr + bh_offset(bh);
541 if (*ptr == cpu_to_be32(GFS2_MAGIC))
542 set_buffer_escaped(bh);
543 kunmap_atomic(kaddr, KM_USER0);
546 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
547 struct list_head *list, struct list_head *done,
550 struct buffer_head *bh1;
551 struct gfs2_log_descriptor *ld;
552 struct gfs2_bufdata *bd;
558 ld = bh_log_desc(bh);
559 ld->ld_length = cpu_to_be32(n + 1);
560 ld->ld_data1 = cpu_to_be32(n);
562 ptr = bh_log_ptr(bh);
565 submit_bh(WRITE_SYNC_PLUG, bh);
567 while(!list_empty(list)) {
568 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
569 list_move_tail(&bd->bd_le.le_list, done);
571 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
572 gfs2_log_incr_head(sdp);
575 gfs2_log_unlock(sdp);
576 lock_buffer(bd->bd_bh);
577 if (buffer_escaped(bd->bd_bh)) {
579 bh1 = gfs2_log_get_buf(sdp);
580 kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
581 memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
583 kunmap_atomic(kaddr, KM_USER0);
584 *(__be32 *)bh1->b_data = 0;
585 clear_buffer_escaped(bd->bd_bh);
586 unlock_buffer(bd->bd_bh);
589 bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
591 submit_bh(WRITE_SYNC_PLUG, bh1);
595 gfs2_log_unlock(sdp);
600 * databuf_lo_before_commit - Scan the data buffers, writing as we go
604 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
606 struct gfs2_bufdata *bd = NULL;
607 struct buffer_head *bh = NULL;
609 __be64 *ptr = NULL, *end = NULL;
610 LIST_HEAD(processed);
611 LIST_HEAD(in_progress);
614 while (!list_empty(&sdp->sd_log_le_databuf)) {
616 gfs2_log_unlock(sdp);
617 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
619 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
620 ptr = bh_log_ptr(bh);
621 end = bh_ptr_end(bh) - 1;
625 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
626 list_move_tail(&bd->bd_le.le_list, &in_progress);
627 gfs2_check_magic(bd->bd_bh);
628 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
629 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
632 gfs2_log_unlock(sdp);
633 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
635 list_splice(&processed, &sdp->sd_log_le_databuf);
636 gfs2_log_unlock(sdp);
639 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
640 struct gfs2_log_descriptor *ld,
641 __be64 *ptr, int pass)
643 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
644 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
645 struct gfs2_glock *gl = ip->i_gl;
646 unsigned int blks = be32_to_cpu(ld->ld_data1);
647 struct buffer_head *bh_log, *bh_ip;
652 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
655 gfs2_replay_incr_blk(sdp, &start);
656 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
657 blkno = be64_to_cpu(*ptr++);
658 esc = be64_to_cpu(*ptr++);
660 sdp->sd_found_blocks++;
662 if (gfs2_revoke_check(sdp, blkno, start))
665 error = gfs2_replay_read_block(jd, start, &bh_log);
669 bh_ip = gfs2_meta_new(gl, blkno);
670 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
674 __be32 *eptr = (__be32 *)bh_ip->b_data;
675 *eptr = cpu_to_be32(GFS2_MAGIC);
677 mark_buffer_dirty(bh_ip);
684 sdp->sd_replayed_blocks++;
690 /* FIXME: sort out accounting for log blocks etc. */
692 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
694 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
695 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
698 gfs2_meta_sync(ip->i_gl);
705 gfs2_meta_sync(ip->i_gl);
707 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
708 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
711 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
713 struct list_head *head = &sdp->sd_log_le_databuf;
714 struct gfs2_bufdata *bd;
716 while (!list_empty(head)) {
717 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
718 list_del_init(&bd->bd_le.le_list);
719 sdp->sd_log_num_databuf--;
720 gfs2_unpin(sdp, bd->bd_bh, ai);
722 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
726 const struct gfs2_log_operations gfs2_buf_lops = {
727 .lo_add = buf_lo_add,
728 .lo_before_commit = buf_lo_before_commit,
729 .lo_after_commit = buf_lo_after_commit,
730 .lo_before_scan = buf_lo_before_scan,
731 .lo_scan_elements = buf_lo_scan_elements,
732 .lo_after_scan = buf_lo_after_scan,
736 const struct gfs2_log_operations gfs2_revoke_lops = {
737 .lo_add = revoke_lo_add,
738 .lo_before_commit = revoke_lo_before_commit,
739 .lo_before_scan = revoke_lo_before_scan,
740 .lo_scan_elements = revoke_lo_scan_elements,
741 .lo_after_scan = revoke_lo_after_scan,
745 const struct gfs2_log_operations gfs2_rg_lops = {
747 .lo_after_commit = rg_lo_after_commit,
751 const struct gfs2_log_operations gfs2_databuf_lops = {
752 .lo_add = databuf_lo_add,
753 .lo_before_commit = databuf_lo_before_commit,
754 .lo_after_commit = databuf_lo_after_commit,
755 .lo_scan_elements = databuf_lo_scan_elements,
756 .lo_after_scan = databuf_lo_after_scan,
757 .lo_name = "databuf",
760 const struct gfs2_log_operations *gfs2_log_ops[] = {