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>
30 #include "trace_gfs2.h"
33 * gfs2_pin - Pin a buffer in memory
34 * @sdp: The superblock
35 * @bh: The buffer to be pinned
37 * The log lock must be held when calling this function
39 static void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
41 struct gfs2_bufdata *bd;
43 BUG_ON(!current->journal_info);
45 clear_buffer_dirty(bh);
46 if (test_set_buffer_pinned(bh))
47 gfs2_assert_withdraw(sdp, 0);
48 if (!buffer_uptodate(bh))
49 gfs2_io_error_bh(sdp, bh);
51 /* If this buffer is in the AIL and it has already been written
52 * to in-place disk block, remove it from the AIL.
54 spin_lock(&sdp->sd_ail_lock);
56 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
57 spin_unlock(&sdp->sd_ail_lock);
59 atomic_inc(&sdp->sd_log_pinned);
60 trace_gfs2_pin(bd, 1);
64 * gfs2_unpin - Unpin a buffer
65 * @sdp: the filesystem the buffer belongs to
66 * @bh: The buffer to unpin
68 * @flags: The inode dirty flags
72 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
75 struct gfs2_bufdata *bd = bh->b_private;
77 BUG_ON(!buffer_uptodate(bh));
78 BUG_ON(!buffer_pinned(bh));
81 mark_buffer_dirty(bh);
82 clear_buffer_pinned(bh);
84 spin_lock(&sdp->sd_ail_lock);
86 list_del(&bd->bd_ail_st_list);
89 struct gfs2_glock *gl = bd->bd_gl;
90 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
91 atomic_inc(&gl->gl_ail_count);
94 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
95 spin_unlock(&sdp->sd_ail_lock);
97 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
98 trace_gfs2_pin(bd, 0);
100 atomic_dec(&sdp->sd_log_pinned);
104 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
106 return (struct gfs2_log_descriptor *)bh->b_data;
109 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
111 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
112 return (__force __be64 *)(ld + 1);
115 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
117 return (__force __be64 *)(bh->b_data + bh->b_size);
121 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
123 struct buffer_head *bh = gfs2_log_get_buf(sdp);
124 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
125 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
126 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
127 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
128 ld->ld_type = cpu_to_be32(ld_type);
132 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
136 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
138 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
139 struct gfs2_meta_header *mh;
140 struct gfs2_trans *tr;
142 if (!list_empty(&bd->bd_list_tr))
144 tr = current->journal_info;
147 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
148 if (!list_empty(&le->le_list))
150 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
151 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
152 gfs2_meta_check(sdp, bd->bd_bh);
153 gfs2_pin(sdp, bd->bd_bh);
154 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
155 mh->__pad0 = cpu_to_be64(0);
156 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
157 sdp->sd_log_num_buf++;
158 list_add(&le->le_list, &sdp->sd_log_le_buf);
159 tr->tr_num_buf_new++;
162 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
164 struct buffer_head *bh;
165 struct gfs2_log_descriptor *ld;
166 struct gfs2_bufdata *bd1 = NULL, *bd2;
173 limit = buf_limit(sdp);
174 /* for 4k blocks, limit = 503 */
177 total = sdp->sd_log_num_buf;
178 bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
183 gfs2_log_unlock(sdp);
184 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
186 ld = bh_log_desc(bh);
187 ptr = bh_log_ptr(bh);
188 ld->ld_length = cpu_to_be32(num + 1);
189 ld->ld_data1 = cpu_to_be32(num);
192 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
194 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
199 gfs2_log_unlock(sdp);
200 submit_bh(WRITE_SYNC, bh);
204 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
207 gfs2_log_unlock(sdp);
208 lock_buffer(bd2->bd_bh);
209 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
210 submit_bh(WRITE_SYNC, bh);
219 gfs2_log_unlock(sdp);
222 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
224 struct list_head *head = &sdp->sd_log_le_buf;
225 struct gfs2_bufdata *bd;
227 while (!list_empty(head)) {
228 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
229 list_del_init(&bd->bd_le.le_list);
230 sdp->sd_log_num_buf--;
232 gfs2_unpin(sdp, bd->bd_bh, ai);
234 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
237 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
238 struct gfs2_log_header_host *head, int pass)
240 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
245 sdp->sd_found_blocks = 0;
246 sdp->sd_replayed_blocks = 0;
249 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
250 struct gfs2_log_descriptor *ld, __be64 *ptr,
253 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
254 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
255 struct gfs2_glock *gl = ip->i_gl;
256 unsigned int blks = be32_to_cpu(ld->ld_data1);
257 struct buffer_head *bh_log, *bh_ip;
261 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
264 gfs2_replay_incr_blk(sdp, &start);
266 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
267 blkno = be64_to_cpu(*ptr++);
269 sdp->sd_found_blocks++;
271 if (gfs2_revoke_check(sdp, blkno, start))
274 error = gfs2_replay_read_block(jd, start, &bh_log);
278 bh_ip = gfs2_meta_new(gl, blkno);
279 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
281 if (gfs2_meta_check(sdp, bh_ip))
284 mark_buffer_dirty(bh_ip);
292 sdp->sd_replayed_blocks++;
298 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
300 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
301 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
304 gfs2_meta_sync(ip->i_gl);
310 gfs2_meta_sync(ip->i_gl);
312 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
313 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
316 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
318 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
319 struct gfs2_glock *gl = bd->bd_gl;
320 struct gfs2_trans *tr;
322 tr = current->journal_info;
325 sdp->sd_log_num_revoke++;
326 atomic_inc(&gl->gl_revokes);
327 set_bit(GLF_LFLUSH, &gl->gl_flags);
328 list_add(&le->le_list, &sdp->sd_log_le_revoke);
331 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
333 struct gfs2_log_descriptor *ld;
334 struct gfs2_meta_header *mh;
335 struct buffer_head *bh;
337 struct list_head *head = &sdp->sd_log_le_revoke;
338 struct gfs2_bufdata *bd;
340 if (!sdp->sd_log_num_revoke)
343 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
344 ld = bh_log_desc(bh);
345 ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
347 ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
348 offset = sizeof(struct gfs2_log_descriptor);
350 list_for_each_entry(bd, head, bd_le.le_list) {
351 sdp->sd_log_num_revoke--;
353 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
354 submit_bh(WRITE_SYNC, bh);
356 bh = gfs2_log_get_buf(sdp);
357 mh = (struct gfs2_meta_header *)bh->b_data;
358 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
359 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
360 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
361 offset = sizeof(struct gfs2_meta_header);
364 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
365 offset += sizeof(u64);
367 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
369 submit_bh(WRITE_SYNC, bh);
372 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
374 struct list_head *head = &sdp->sd_log_le_revoke;
375 struct gfs2_bufdata *bd;
376 struct gfs2_glock *gl;
378 while (!list_empty(head)) {
379 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
380 list_del_init(&bd->bd_le.le_list);
382 atomic_dec(&gl->gl_revokes);
383 clear_bit(GLF_LFLUSH, &gl->gl_flags);
384 kmem_cache_free(gfs2_bufdata_cachep, bd);
388 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
389 struct gfs2_log_header_host *head, int pass)
391 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
396 sdp->sd_found_revokes = 0;
397 sdp->sd_replay_tail = head->lh_tail;
400 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
401 struct gfs2_log_descriptor *ld, __be64 *ptr,
404 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
405 unsigned int blks = be32_to_cpu(ld->ld_length);
406 unsigned int revokes = be32_to_cpu(ld->ld_data1);
407 struct buffer_head *bh;
413 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
416 offset = sizeof(struct gfs2_log_descriptor);
418 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
419 error = gfs2_replay_read_block(jd, start, &bh);
424 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
426 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
427 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
429 error = gfs2_revoke_add(sdp, blkno, start);
435 sdp->sd_found_revokes++;
439 offset += sizeof(u64);
443 offset = sizeof(struct gfs2_meta_header);
450 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
452 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
455 gfs2_revoke_clean(sdp);
461 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
462 jd->jd_jid, sdp->sd_found_revokes);
464 gfs2_revoke_clean(sdp);
467 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
469 struct gfs2_rgrpd *rgd;
470 struct gfs2_trans *tr = current->journal_info;
474 rgd = container_of(le, struct gfs2_rgrpd, rd_le);
477 if (!list_empty(&le->le_list)){
478 gfs2_log_unlock(sdp);
481 gfs2_rgrp_bh_hold(rgd);
482 sdp->sd_log_num_rg++;
483 list_add(&le->le_list, &sdp->sd_log_le_rg);
484 gfs2_log_unlock(sdp);
487 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
489 struct list_head *head = &sdp->sd_log_le_rg;
490 struct gfs2_rgrpd *rgd;
492 while (!list_empty(head)) {
493 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
494 list_del_init(&rgd->rd_le.le_list);
495 sdp->sd_log_num_rg--;
497 gfs2_rgrp_repolish_clones(rgd);
498 gfs2_rgrp_bh_put(rgd);
500 gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
504 * databuf_lo_add - Add a databuf to the transaction.
506 * This is used in two distinct cases:
507 * i) In ordered write mode
508 * We put the data buffer on a list so that we can ensure that its
509 * synced to disk at the right time
510 * ii) In journaled data mode
511 * We need to journal the data block in the same way as metadata in
512 * the functions above. The difference is that here we have a tag
513 * which is two __be64's being the block number (as per meta data)
514 * and a flag which says whether the data block needs escaping or
515 * not. This means we need a new log entry for each 251 or so data
516 * blocks, which isn't an enormous overhead but twice as much as
517 * for normal metadata blocks.
519 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
521 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
522 struct gfs2_trans *tr = current->journal_info;
523 struct address_space *mapping = bd->bd_bh->b_page->mapping;
524 struct gfs2_inode *ip = GFS2_I(mapping->host);
527 if (!list_empty(&bd->bd_list_tr))
530 if (gfs2_is_jdata(ip)) {
532 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
535 if (!list_empty(&le->le_list))
538 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
539 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
540 if (gfs2_is_jdata(ip)) {
541 gfs2_pin(sdp, bd->bd_bh);
542 tr->tr_num_databuf_new++;
543 sdp->sd_log_num_databuf++;
544 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
546 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
550 static void gfs2_check_magic(struct buffer_head *bh)
555 clear_buffer_escaped(bh);
556 kaddr = kmap_atomic(bh->b_page, KM_USER0);
557 ptr = kaddr + bh_offset(bh);
558 if (*ptr == cpu_to_be32(GFS2_MAGIC))
559 set_buffer_escaped(bh);
560 kunmap_atomic(kaddr, KM_USER0);
563 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
564 struct list_head *list, struct list_head *done,
567 struct buffer_head *bh1;
568 struct gfs2_log_descriptor *ld;
569 struct gfs2_bufdata *bd;
575 ld = bh_log_desc(bh);
576 ld->ld_length = cpu_to_be32(n + 1);
577 ld->ld_data1 = cpu_to_be32(n);
579 ptr = bh_log_ptr(bh);
582 submit_bh(WRITE_SYNC, bh);
584 while(!list_empty(list)) {
585 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
586 list_move_tail(&bd->bd_le.le_list, done);
588 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
589 gfs2_log_incr_head(sdp);
592 gfs2_log_unlock(sdp);
593 lock_buffer(bd->bd_bh);
594 if (buffer_escaped(bd->bd_bh)) {
596 bh1 = gfs2_log_get_buf(sdp);
597 kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
598 memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
600 kunmap_atomic(kaddr, KM_USER0);
601 *(__be32 *)bh1->b_data = 0;
602 clear_buffer_escaped(bd->bd_bh);
603 unlock_buffer(bd->bd_bh);
606 bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
608 submit_bh(WRITE_SYNC, bh1);
612 gfs2_log_unlock(sdp);
617 * databuf_lo_before_commit - Scan the data buffers, writing as we go
621 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
623 struct gfs2_bufdata *bd = NULL;
624 struct buffer_head *bh = NULL;
626 __be64 *ptr = NULL, *end = NULL;
627 LIST_HEAD(processed);
628 LIST_HEAD(in_progress);
631 while (!list_empty(&sdp->sd_log_le_databuf)) {
633 gfs2_log_unlock(sdp);
634 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
636 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
637 ptr = bh_log_ptr(bh);
638 end = bh_ptr_end(bh) - 1;
642 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
643 list_move_tail(&bd->bd_le.le_list, &in_progress);
644 gfs2_check_magic(bd->bd_bh);
645 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
646 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
649 gfs2_log_unlock(sdp);
650 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
652 list_splice(&processed, &sdp->sd_log_le_databuf);
653 gfs2_log_unlock(sdp);
656 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
657 struct gfs2_log_descriptor *ld,
658 __be64 *ptr, int pass)
660 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
661 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
662 struct gfs2_glock *gl = ip->i_gl;
663 unsigned int blks = be32_to_cpu(ld->ld_data1);
664 struct buffer_head *bh_log, *bh_ip;
669 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
672 gfs2_replay_incr_blk(sdp, &start);
673 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
674 blkno = be64_to_cpu(*ptr++);
675 esc = be64_to_cpu(*ptr++);
677 sdp->sd_found_blocks++;
679 if (gfs2_revoke_check(sdp, blkno, start))
682 error = gfs2_replay_read_block(jd, start, &bh_log);
686 bh_ip = gfs2_meta_new(gl, blkno);
687 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
691 __be32 *eptr = (__be32 *)bh_ip->b_data;
692 *eptr = cpu_to_be32(GFS2_MAGIC);
694 mark_buffer_dirty(bh_ip);
701 sdp->sd_replayed_blocks++;
707 /* FIXME: sort out accounting for log blocks etc. */
709 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
711 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
712 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
715 gfs2_meta_sync(ip->i_gl);
722 gfs2_meta_sync(ip->i_gl);
724 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
725 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
728 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
730 struct list_head *head = &sdp->sd_log_le_databuf;
731 struct gfs2_bufdata *bd;
733 while (!list_empty(head)) {
734 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
735 list_del_init(&bd->bd_le.le_list);
736 sdp->sd_log_num_databuf--;
737 gfs2_unpin(sdp, bd->bd_bh, ai);
739 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
743 const struct gfs2_log_operations gfs2_buf_lops = {
744 .lo_add = buf_lo_add,
745 .lo_before_commit = buf_lo_before_commit,
746 .lo_after_commit = buf_lo_after_commit,
747 .lo_before_scan = buf_lo_before_scan,
748 .lo_scan_elements = buf_lo_scan_elements,
749 .lo_after_scan = buf_lo_after_scan,
753 const struct gfs2_log_operations gfs2_revoke_lops = {
754 .lo_add = revoke_lo_add,
755 .lo_before_commit = revoke_lo_before_commit,
756 .lo_after_commit = revoke_lo_after_commit,
757 .lo_before_scan = revoke_lo_before_scan,
758 .lo_scan_elements = revoke_lo_scan_elements,
759 .lo_after_scan = revoke_lo_after_scan,
763 const struct gfs2_log_operations gfs2_rg_lops = {
765 .lo_after_commit = rg_lo_after_commit,
769 const struct gfs2_log_operations gfs2_databuf_lops = {
770 .lo_add = databuf_lo_add,
771 .lo_before_commit = databuf_lo_before_commit,
772 .lo_after_commit = databuf_lo_after_commit,
773 .lo_scan_elements = databuf_lo_scan_elements,
774 .lo_after_scan = databuf_lo_after_scan,
775 .lo_name = "databuf",
778 const struct gfs2_log_operations *gfs2_log_ops[] = {