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);
63 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
65 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
68 static void maybe_release_space(struct gfs2_bufdata *bd)
70 struct gfs2_glock *gl = bd->bd_gl;
71 struct gfs2_sbd *sdp = gl->gl_sbd;
72 struct gfs2_rgrpd *rgd = gl->gl_object;
73 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
74 struct gfs2_bitmap *bi = rgd->rd_bits + index;
76 if (bi->bi_clone == 0)
78 if (sdp->sd_args.ar_discard)
79 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi);
80 memcpy(bi->bi_clone + bi->bi_offset,
81 bd->bd_bh->b_data + bi->bi_offset, bi->bi_len);
82 clear_bit(GBF_FULL, &bi->bi_flags);
83 rgd->rd_free_clone = rgd->rd_free;
87 * gfs2_unpin - Unpin a buffer
88 * @sdp: the filesystem the buffer belongs to
89 * @bh: The buffer to unpin
91 * @flags: The inode dirty flags
95 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
98 struct gfs2_bufdata *bd = bh->b_private;
100 BUG_ON(!buffer_uptodate(bh));
101 BUG_ON(!buffer_pinned(bh));
104 mark_buffer_dirty(bh);
105 clear_buffer_pinned(bh);
107 if (buffer_is_rgrp(bd))
108 maybe_release_space(bd);
110 spin_lock(&sdp->sd_ail_lock);
112 list_del(&bd->bd_ail_st_list);
115 struct gfs2_glock *gl = bd->bd_gl;
116 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
117 atomic_inc(&gl->gl_ail_count);
120 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
121 spin_unlock(&sdp->sd_ail_lock);
123 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
124 trace_gfs2_pin(bd, 0);
126 atomic_dec(&sdp->sd_log_pinned);
130 static inline struct gfs2_log_descriptor *bh_log_desc(struct buffer_head *bh)
132 return (struct gfs2_log_descriptor *)bh->b_data;
135 static inline __be64 *bh_log_ptr(struct buffer_head *bh)
137 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
138 return (__force __be64 *)(ld + 1);
141 static inline __be64 *bh_ptr_end(struct buffer_head *bh)
143 return (__force __be64 *)(bh->b_data + bh->b_size);
147 static struct buffer_head *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type)
149 struct buffer_head *bh = gfs2_log_get_buf(sdp);
150 struct gfs2_log_descriptor *ld = bh_log_desc(bh);
151 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
152 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
153 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
154 ld->ld_type = cpu_to_be32(ld_type);
158 memset(ld->ld_reserved, 0, sizeof(ld->ld_reserved));
162 static void buf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
164 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
165 struct gfs2_meta_header *mh;
166 struct gfs2_trans *tr;
168 if (!list_empty(&bd->bd_list_tr))
170 tr = current->journal_info;
173 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
174 if (!list_empty(&le->le_list))
176 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
177 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
178 gfs2_meta_check(sdp, bd->bd_bh);
179 gfs2_pin(sdp, bd->bd_bh);
180 mh = (struct gfs2_meta_header *)bd->bd_bh->b_data;
181 mh->__pad0 = cpu_to_be64(0);
182 mh->mh_jid = cpu_to_be32(sdp->sd_jdesc->jd_jid);
183 sdp->sd_log_num_buf++;
184 list_add(&le->le_list, &sdp->sd_log_le_buf);
185 tr->tr_num_buf_new++;
188 static void buf_lo_before_commit(struct gfs2_sbd *sdp)
190 struct buffer_head *bh;
191 struct gfs2_log_descriptor *ld;
192 struct gfs2_bufdata *bd1 = NULL, *bd2;
199 limit = buf_limit(sdp);
200 /* for 4k blocks, limit = 503 */
203 total = sdp->sd_log_num_buf;
204 bd1 = bd2 = list_prepare_entry(bd1, &sdp->sd_log_le_buf, bd_le.le_list);
209 gfs2_log_unlock(sdp);
210 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_METADATA);
212 ld = bh_log_desc(bh);
213 ptr = bh_log_ptr(bh);
214 ld->ld_length = cpu_to_be32(num + 1);
215 ld->ld_data1 = cpu_to_be32(num);
218 list_for_each_entry_continue(bd1, &sdp->sd_log_le_buf,
220 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
225 gfs2_log_unlock(sdp);
226 submit_bh(WRITE_SYNC, bh);
230 list_for_each_entry_continue(bd2, &sdp->sd_log_le_buf,
233 gfs2_log_unlock(sdp);
234 lock_buffer(bd2->bd_bh);
235 bh = gfs2_log_fake_buf(sdp, bd2->bd_bh);
236 submit_bh(WRITE_SYNC, bh);
245 gfs2_log_unlock(sdp);
248 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
250 struct list_head *head = &sdp->sd_log_le_buf;
251 struct gfs2_bufdata *bd;
253 while (!list_empty(head)) {
254 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
255 list_del_init(&bd->bd_le.le_list);
256 sdp->sd_log_num_buf--;
258 gfs2_unpin(sdp, bd->bd_bh, ai);
260 gfs2_assert_warn(sdp, !sdp->sd_log_num_buf);
263 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
264 struct gfs2_log_header_host *head, int pass)
266 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
271 sdp->sd_found_blocks = 0;
272 sdp->sd_replayed_blocks = 0;
275 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
276 struct gfs2_log_descriptor *ld, __be64 *ptr,
279 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
280 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
281 struct gfs2_glock *gl = ip->i_gl;
282 unsigned int blks = be32_to_cpu(ld->ld_data1);
283 struct buffer_head *bh_log, *bh_ip;
287 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
290 gfs2_replay_incr_blk(sdp, &start);
292 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
293 blkno = be64_to_cpu(*ptr++);
295 sdp->sd_found_blocks++;
297 if (gfs2_revoke_check(sdp, blkno, start))
300 error = gfs2_replay_read_block(jd, start, &bh_log);
304 bh_ip = gfs2_meta_new(gl, blkno);
305 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
307 if (gfs2_meta_check(sdp, bh_ip))
310 mark_buffer_dirty(bh_ip);
318 sdp->sd_replayed_blocks++;
324 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
326 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
327 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
330 gfs2_meta_sync(ip->i_gl);
336 gfs2_meta_sync(ip->i_gl);
338 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
339 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
342 static void revoke_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
344 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
345 struct gfs2_glock *gl = bd->bd_gl;
346 struct gfs2_trans *tr;
348 tr = current->journal_info;
351 sdp->sd_log_num_revoke++;
352 atomic_inc(&gl->gl_revokes);
353 set_bit(GLF_LFLUSH, &gl->gl_flags);
354 list_add(&le->le_list, &sdp->sd_log_le_revoke);
357 static void revoke_lo_before_commit(struct gfs2_sbd *sdp)
359 struct gfs2_log_descriptor *ld;
360 struct gfs2_meta_header *mh;
361 struct buffer_head *bh;
363 struct list_head *head = &sdp->sd_log_le_revoke;
364 struct gfs2_bufdata *bd;
366 if (!sdp->sd_log_num_revoke)
369 bh = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE);
370 ld = bh_log_desc(bh);
371 ld->ld_length = cpu_to_be32(gfs2_struct2blk(sdp, sdp->sd_log_num_revoke,
373 ld->ld_data1 = cpu_to_be32(sdp->sd_log_num_revoke);
374 offset = sizeof(struct gfs2_log_descriptor);
376 list_for_each_entry(bd, head, bd_le.le_list) {
377 sdp->sd_log_num_revoke--;
379 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
380 submit_bh(WRITE_SYNC, bh);
382 bh = gfs2_log_get_buf(sdp);
383 mh = (struct gfs2_meta_header *)bh->b_data;
384 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
385 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
386 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
387 offset = sizeof(struct gfs2_meta_header);
390 *(__be64 *)(bh->b_data + offset) = cpu_to_be64(bd->bd_blkno);
391 offset += sizeof(u64);
393 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
395 submit_bh(WRITE_SYNC, bh);
398 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
400 struct list_head *head = &sdp->sd_log_le_revoke;
401 struct gfs2_bufdata *bd;
402 struct gfs2_glock *gl;
404 while (!list_empty(head)) {
405 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
406 list_del_init(&bd->bd_le.le_list);
408 atomic_dec(&gl->gl_revokes);
409 clear_bit(GLF_LFLUSH, &gl->gl_flags);
410 kmem_cache_free(gfs2_bufdata_cachep, bd);
414 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
415 struct gfs2_log_header_host *head, int pass)
417 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
422 sdp->sd_found_revokes = 0;
423 sdp->sd_replay_tail = head->lh_tail;
426 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, unsigned int start,
427 struct gfs2_log_descriptor *ld, __be64 *ptr,
430 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
431 unsigned int blks = be32_to_cpu(ld->ld_length);
432 unsigned int revokes = be32_to_cpu(ld->ld_data1);
433 struct buffer_head *bh;
439 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
442 offset = sizeof(struct gfs2_log_descriptor);
444 for (; blks; gfs2_replay_incr_blk(sdp, &start), blks--) {
445 error = gfs2_replay_read_block(jd, start, &bh);
450 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
452 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
453 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
455 error = gfs2_revoke_add(sdp, blkno, start);
461 sdp->sd_found_revokes++;
465 offset += sizeof(u64);
469 offset = sizeof(struct gfs2_meta_header);
476 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
478 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
481 gfs2_revoke_clean(sdp);
487 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
488 jd->jd_jid, sdp->sd_found_revokes);
490 gfs2_revoke_clean(sdp);
494 * databuf_lo_add - Add a databuf to the transaction.
496 * This is used in two distinct cases:
497 * i) In ordered write mode
498 * We put the data buffer on a list so that we can ensure that its
499 * synced to disk at the right time
500 * ii) In journaled data mode
501 * We need to journal the data block in the same way as metadata in
502 * the functions above. The difference is that here we have a tag
503 * which is two __be64's being the block number (as per meta data)
504 * and a flag which says whether the data block needs escaping or
505 * not. This means we need a new log entry for each 251 or so data
506 * blocks, which isn't an enormous overhead but twice as much as
507 * for normal metadata blocks.
509 static void databuf_lo_add(struct gfs2_sbd *sdp, struct gfs2_log_element *le)
511 struct gfs2_bufdata *bd = container_of(le, struct gfs2_bufdata, bd_le);
512 struct gfs2_trans *tr = current->journal_info;
513 struct address_space *mapping = bd->bd_bh->b_page->mapping;
514 struct gfs2_inode *ip = GFS2_I(mapping->host);
517 if (!list_empty(&bd->bd_list_tr))
520 if (gfs2_is_jdata(ip)) {
522 list_add(&bd->bd_list_tr, &tr->tr_list_buf);
525 if (!list_empty(&le->le_list))
528 set_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
529 set_bit(GLF_DIRTY, &bd->bd_gl->gl_flags);
530 if (gfs2_is_jdata(ip)) {
531 gfs2_pin(sdp, bd->bd_bh);
532 tr->tr_num_databuf_new++;
533 sdp->sd_log_num_databuf++;
534 list_add_tail(&le->le_list, &sdp->sd_log_le_databuf);
536 list_add_tail(&le->le_list, &sdp->sd_log_le_ordered);
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, 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, 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);
689 sdp->sd_replayed_blocks++;
695 /* FIXME: sort out accounting for log blocks etc. */
697 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
699 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
700 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
703 gfs2_meta_sync(ip->i_gl);
710 gfs2_meta_sync(ip->i_gl);
712 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
713 jd->jd_jid, sdp->sd_replayed_blocks, sdp->sd_found_blocks);
716 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
718 struct list_head *head = &sdp->sd_log_le_databuf;
719 struct gfs2_bufdata *bd;
721 while (!list_empty(head)) {
722 bd = list_entry(head->next, struct gfs2_bufdata, bd_le.le_list);
723 list_del_init(&bd->bd_le.le_list);
724 sdp->sd_log_num_databuf--;
725 gfs2_unpin(sdp, bd->bd_bh, ai);
727 gfs2_assert_warn(sdp, !sdp->sd_log_num_databuf);
731 const struct gfs2_log_operations gfs2_buf_lops = {
732 .lo_add = buf_lo_add,
733 .lo_before_commit = buf_lo_before_commit,
734 .lo_after_commit = buf_lo_after_commit,
735 .lo_before_scan = buf_lo_before_scan,
736 .lo_scan_elements = buf_lo_scan_elements,
737 .lo_after_scan = buf_lo_after_scan,
741 const struct gfs2_log_operations gfs2_revoke_lops = {
742 .lo_add = revoke_lo_add,
743 .lo_before_commit = revoke_lo_before_commit,
744 .lo_after_commit = revoke_lo_after_commit,
745 .lo_before_scan = revoke_lo_before_scan,
746 .lo_scan_elements = revoke_lo_scan_elements,
747 .lo_after_scan = revoke_lo_after_scan,
751 const struct gfs2_log_operations gfs2_rg_lops = {
755 const struct gfs2_log_operations gfs2_databuf_lops = {
756 .lo_add = databuf_lo_add,
757 .lo_before_commit = databuf_lo_before_commit,
758 .lo_after_commit = databuf_lo_after_commit,
759 .lo_scan_elements = databuf_lo_scan_elements,
760 .lo_after_scan = databuf_lo_after_scan,
761 .lo_name = "databuf",
764 const struct gfs2_log_operations *gfs2_log_ops[] = {