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 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
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.
55 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
57 trace_gfs2_pin(bd, 1);
61 * gfs2_unpin - Unpin a buffer
62 * @sdp: the filesystem the buffer belongs to
63 * @bh: The buffer to unpin
68 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
71 struct gfs2_bufdata *bd = bh->b_private;
73 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
75 if (!buffer_pinned(bh))
76 gfs2_assert_withdraw(sdp, 0);
79 mark_buffer_dirty(bh);
80 clear_buffer_pinned(bh);
84 list_del(&bd->bd_ail_st_list);
87 struct gfs2_glock *gl = bd->bd_gl;
88 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
89 atomic_inc(&gl->gl_ail_count);
92 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
93 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
94 trace_gfs2_pin(bd, 0);
100 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
102 return (struct gfs2_log_descriptor *)bh->b_data;
105 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
107 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
108 return (__force __be64 *)(ld + 1);
111 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
113 return (__force __be64 *)(bh->b_data + bh->b_size);
117 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
119 struct buffer_head *bh = gfs2_log_get_buf(sdp);
120 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
121 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
122 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
123 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
124 ld->ld_type = cpu_to_be32(ld_type);
128 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
132 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
134 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
135 struct gfs2_meta_header *mh;
136 struct gfs2_trans *tr;
138 lock_buffer(bd->bd_bh);
140 if (!list_empty(&bd->bd_list_tr))
142 tr = current->journal_info;
145 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
146 if (!list_empty(&le->le_list))
148 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
149 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
150 gfs2_meta_check(sdp, bd->bd_bh);
151 gfs2_pin(sdp, bd->bd_bh);
152 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
153 mh->__pad0 = cpu_to_be64(0);
154 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
155 sdp->sd_log_num_buf++;
156 list_add(&le->le_list, &sdp->sd_log_le_buf);
157 tr->tr_num_buf_new++;
159 gfs2_log_unlock(sdp);
160 unlock_buffer(bd->bd_bh);
163 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
165 struct buffer_head *bh;
166 struct gfs2_log_descriptor *ld;
167 struct gfs2_bufdata *bd1 = NULL, *bd2;
174 limit = buf_limit(sdp);
175 /* for 4k blocks, limit = 503 */
178 total = sdp->sd_log_num_buf;
179 bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
184 gfs2_log_unlock(sdp);
185 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
187 ld = bh_log_desc(bh);
188 ptr = bh_log_ptr(bh);
189 ld->ld_length = cpu_to_be32(num + 1);
190 ld->ld_data1 = cpu_to_be32(num);
193 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
195 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
200 gfs2_log_unlock(sdp);
201 submit_bh(WRITE_SYNC_PLUG, bh);
205 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
208 gfs2_log_unlock(sdp);
209 lock_buffer(bd2->bd_bh);
210 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
211 submit_bh(WRITE_SYNC_PLUG, bh);
220 gfs2_log_unlock(sdp);
223 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
225 struct list_head *head = &sdp->sd_log_le_buf;
226 struct gfs2_bufdata *bd;
228 while (!list_empty(head)) {
229 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
230 list_del_init(&bd->bd_le.le_list);
231 sdp->sd_log_num_buf--;
233 gfs2_unpin(sdp, bd->bd_bh, ai);
235 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
238 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
239 struct gfs2_log_header_host *head, int pass)
241 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
246 sdp->sd_found_blocks = 0;
247 sdp->sd_replayed_blocks = 0;
250 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
251 struct gfs2_log_descriptor *ld, __be64 *ptr,
254 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
255 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
256 struct gfs2_glock *gl = ip->i_gl;
257 unsigned int blks = be32_to_cpu(ld->ld_data1);
258 struct buffer_head *bh_log, *bh_ip;
262 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
265 gfs2_replay_incr_blk(sdp, &start);
267 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
268 blkno = be64_to_cpu(*ptr++);
270 sdp->sd_found_blocks++;
272 if (gfs2_revoke_check(sdp, blkno, start))
275 error = gfs2_replay_read_block(jd, start, &bh_log);
279 bh_ip = gfs2_meta_new(gl, blkno);
280 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
282 if (gfs2_meta_check(sdp, bh_ip))
285 mark_buffer_dirty(bh_ip);
293 sdp->sd_replayed_blocks++;
299 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
301 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
302 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
305 gfs2_meta_sync(ip->i_gl);
311 gfs2_meta_sync(ip->i_gl);
313 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
314 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
317 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
319 struct gfs2_trans *tr;
321 tr = current->journal_info;
324 sdp->sd_log_num_revoke++;
325 list_add(&le->le_list, &sdp->sd_log_le_revoke);
328 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
330 struct gfs2_log_descriptor *ld;
331 struct gfs2_meta_header *mh;
332 struct buffer_head *bh;
334 struct list_head *head = &sdp->sd_log_le_revoke;
335 struct gfs2_bufdata *bd;
337 if (!sdp->sd_log_num_revoke)
340 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
341 ld = bh_log_desc(bh);
342 ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
344 ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
345 offset = sizeof(struct gfs2_log_descriptor);
347 while (!list_empty(head)) {
348 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
349 list_del_init(&bd->bd_le.le_list);
350 sdp->sd_log_num_revoke--;
352 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
353 submit_bh(WRITE_SYNC_PLUG, bh);
355 bh = gfs2_log_get_buf(sdp);
356 mh = (struct gfs2_meta_header *)bh->b_data;
357 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
358 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
359 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
360 offset = sizeof(struct gfs2_meta_header);
363 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
364 kmem_cache_free(gfs2_bufdata_cachep, bd);
366 offset += sizeof(u64);
368 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
370 submit_bh(WRITE_SYNC_PLUG, bh);
373 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
374 struct gfs2_log_header_host *head, int pass)
376 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
381 sdp->sd_found_revokes = 0;
382 sdp->sd_replay_tail = head->lh_tail;
385 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
386 struct gfs2_log_descriptor *ld, __be64 *ptr,
389 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
390 unsigned int blks = be32_to_cpu(ld->ld_length);
391 unsigned int revokes = be32_to_cpu(ld->ld_data1);
392 struct buffer_head *bh;
398 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
401 offset = sizeof(struct gfs2_log_descriptor);
403 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
404 error = gfs2_replay_read_block(jd, start, &bh);
409 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
411 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
412 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
414 error = gfs2_revoke_add(sdp, blkno, start);
420 sdp->sd_found_revokes++;
424 offset += sizeof(u64);
428 offset = sizeof(struct gfs2_meta_header);
435 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
437 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
440 gfs2_revoke_clean(sdp);
446 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
447 jd->jd_jid, sdp->sd_found_revokes);
449 gfs2_revoke_clean(sdp);
452 static void rg_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
454 struct gfs2_rgrpd *rgd;
455 struct gfs2_trans *tr = current->journal_info;
459 rgd = container_of(le, struct gfs2_rgrpd, rd_le);
462 if (!list_empty(&le->le_list)){
463 gfs2_log_unlock(sdp);
466 gfs2_rgrp_bh_hold(rgd);
467 sdp->sd_log_num_rg++;
468 list_add(&le->le_list, &sdp->sd_log_le_rg);
469 gfs2_log_unlock(sdp);
472 static void rg_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
474 struct list_head *head = &sdp->sd_log_le_rg;
475 struct gfs2_rgrpd *rgd;
477 while (!list_empty(head)) {
478 rgd = list_entry(head->next, struct gfs2_rgrpd, rd_le.le_list);
479 list_del_init(&rgd->rd_le.le_list);
480 sdp->sd_log_num_rg--;
482 gfs2_rgrp_repolish_clones(rgd);
483 gfs2_rgrp_bh_put(rgd);
485 gfs2_assert_warn(sdp, !sdp->sd_log_num_rg);
489 * databuf_lo_add - Add a databuf to the transaction.
491 * This is used in two distinct cases:
492 * i) In ordered write mode
493 * We put the data buffer on a list so that we can ensure that its
494 * synced to disk at the right time
495 * ii) In journaled data mode
496 * We need to journal the data block in the same way as metadata in
497 * the functions above. The difference is that here we have a tag
498 * which is two __be64's being the block number (as per meta data)
499 * and a flag which says whether the data block needs escaping or
500 * not. This means we need a new log entry for each 251 or so data
501 * blocks, which isn't an enormous overhead but twice as much as
502 * for normal metadata blocks.
504 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
506 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
507 struct gfs2_trans *tr = current->journal_info;
508 struct address_space *mapping = bd->bd_bh->b_page->mapping;
509 struct gfs2_inode *ip = GFS2_I(mapping->host);
511 lock_buffer(bd->bd_bh);
514 if (!list_empty(&bd->bd_list_tr))
517 if (gfs2_is_jdata(ip)) {
519 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
522 if (!list_empty(&le->le_list))
525 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
526 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
527 if (gfs2_is_jdata(ip)) {
528 gfs2_pin(sdp, bd->bd_bh);
529 tr->tr_num_databuf_new++;
530 sdp->sd_log_num_databuf++;
531 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
533 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
536 gfs2_log_unlock(sdp);
537 unlock_buffer(bd->bd_bh);
540 static void gfs2_check_magic(struct buffer_head *bh)
545 clear_buffer_escaped(bh);
546 kaddr = kmap_atomic(bh->b_page, KM_USER0);
547 ptr = kaddr + bh_offset(bh);
548 if (*ptr == cpu_to_be32(GFS2_MAGIC))
549 set_buffer_escaped(bh);
550 kunmap_atomic(kaddr, KM_USER0);
553 static void gfs2_write_blocks(struct gfs2_sbd *sdp, struct buffer_head *bh,
554 struct list_head *list, struct list_head *done,
557 struct buffer_head *bh1;
558 struct gfs2_log_descriptor *ld;
559 struct gfs2_bufdata *bd;
565 ld = bh_log_desc(bh);
566 ld->ld_length = cpu_to_be32(n + 1);
567 ld->ld_data1 = cpu_to_be32(n);
569 ptr = bh_log_ptr(bh);
572 submit_bh(WRITE_SYNC_PLUG, bh);
574 while(!list_empty(list)) {
575 bd = list_entry(list->next, struct gfs2_bufdata, bd_le.le_list);
576 list_move_tail(&bd->bd_le.le_list, done);
578 while (be64_to_cpu(*ptr) != bd->bd_bh->b_blocknr) {
579 gfs2_log_incr_head(sdp);
582 gfs2_log_unlock(sdp);
583 lock_buffer(bd->bd_bh);
584 if (buffer_escaped(bd->bd_bh)) {
586 bh1 = gfs2_log_get_buf(sdp);
587 kaddr = kmap_atomic(bd->bd_bh->b_page, KM_USER0);
588 memcpy(bh1->b_data, kaddr + bh_offset(bd->bd_bh),
590 kunmap_atomic(kaddr, KM_USER0);
591 *(__be32 *)bh1->b_data = 0;
592 clear_buffer_escaped(bd->bd_bh);
593 unlock_buffer(bd->bd_bh);
596 bh1 = gfs2_log_fake_buf(sdp, bd->bd_bh);
598 submit_bh(WRITE_SYNC_PLUG, bh1);
602 gfs2_log_unlock(sdp);
607 * databuf_lo_before_commit - Scan the data buffers, writing as we go
611 static void databuf_lo_before_commit(struct gfs2_sbd *sdp)
613 struct gfs2_bufdata *bd = NULL;
614 struct buffer_head *bh = NULL;
616 __be64 *ptr = NULL, *end = NULL;
617 LIST_HEAD(processed);
618 LIST_HEAD(in_progress);
621 while (!list_empty(&sdp->sd_log_le_databuf)) {
623 gfs2_log_unlock(sdp);
624 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
626 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_JDATA);
627 ptr = bh_log_ptr(bh);
628 end = bh_ptr_end(bh) - 1;
632 bd = list_entry(sdp->sd_log_le_databuf.next, struct gfs2_bufdata, bd_le.le_list);
633 list_move_tail(&bd->bd_le.le_list, &in_progress);
634 gfs2_check_magic(bd->bd_bh);
635 *ptr++ = cpu_to_be64(bd->bd_bh->b_blocknr);
636 *ptr++ = cpu_to_be64(buffer_escaped(bh) ? 1 : 0);
639 gfs2_log_unlock(sdp);
640 gfs2_write_blocks(sdp, bh, &in_progress, &processed, n);
642 list_splice(&processed, &sdp->sd_log_le_databuf);
643 gfs2_log_unlock(sdp);
646 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
647 struct gfs2_log_descriptor *ld,
648 __be64 *ptr, int pass)
650 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
651 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
652 struct gfs2_glock *gl = ip->i_gl;
653 unsigned int blks = be32_to_cpu(ld->ld_data1);
654 struct buffer_head *bh_log, *bh_ip;
659 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
662 gfs2_replay_incr_blk(sdp, &start);
663 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
664 blkno = be64_to_cpu(*ptr++);
665 esc = be64_to_cpu(*ptr++);
667 sdp->sd_found_blocks++;
669 if (gfs2_revoke_check(sdp, blkno, start))
672 error = gfs2_replay_read_block(jd, start, &bh_log);
676 bh_ip = gfs2_meta_new(gl, blkno);
677 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
681 __be32 *eptr = (__be32 *)bh_ip->b_data;
682 *eptr = cpu_to_be32(GFS2_MAGIC);
684 mark_buffer_dirty(bh_ip);
691 sdp->sd_replayed_blocks++;
697 /* FIXME: sort out accounting for log blocks etc. */
699 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
701 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
702 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
705 gfs2_meta_sync(ip->i_gl);
712 gfs2_meta_sync(ip->i_gl);
714 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
715 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
718 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
720 struct list_head *head = &sdp->sd_log_le_databuf;
721 struct gfs2_bufdata *bd;
723 while (!list_empty(head)) {
724 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
725 list_del_init(&bd->bd_le.le_list);
726 sdp->sd_log_num_databuf--;
727 gfs2_unpin(sdp, bd->bd_bh, ai);
729 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
733 const struct gfs2_log_operations gfs2_buf_lops = {
734 .lo_add = buf_lo_add,
735 .lo_before_commit = buf_lo_before_commit,
736 .lo_after_commit = buf_lo_after_commit,
737 .lo_before_scan = buf_lo_before_scan,
738 .lo_scan_elements = buf_lo_scan_elements,
739 .lo_after_scan = buf_lo_after_scan,
743 const struct gfs2_log_operations gfs2_revoke_lops = {
744 .lo_add = revoke_lo_add,
745 .lo_before_commit = revoke_lo_before_commit,
746 .lo_before_scan = revoke_lo_before_scan,
747 .lo_scan_elements = revoke_lo_scan_elements,
748 .lo_after_scan = revoke_lo_after_scan,
752 const struct gfs2_log_operations gfs2_rg_lops = {
754 .lo_after_commit = rg_lo_after_commit,
758 const struct gfs2_log_operations gfs2_databuf_lops = {
759 .lo_add = databuf_lo_add,
760 .lo_before_commit = databuf_lo_before_commit,
761 .lo_after_commit = databuf_lo_after_commit,
762 .lo_scan_elements = databuf_lo_scan_elements,
763 .lo_after_scan = databuf_lo_after_scan,
764 .lo_name = "databuf",
767 const struct gfs2_log_operations *gfs2_log_ops[] = {