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/mempool.h>
16 #include <linux/gfs2_ondisk.h>
17 #include <linux/bio.h>
19 #include <linux/list_sort.h>
20 #include <linux/blkdev.h>
35 #include "trace_gfs2.h"
38 * gfs2_pin - Pin a buffer in memory
39 * @sdp: The superblock
40 * @bh: The buffer to be pinned
42 * The log lock must be held when calling this function
44 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
46 struct gfs2_bufdata *bd;
48 BUG_ON(!current->journal_info);
50 clear_buffer_dirty(bh);
51 if (test_set_buffer_pinned(bh))
52 gfs2_assert_withdraw(sdp, 0);
53 if (!buffer_uptodate(bh))
54 gfs2_io_error_bh_wd(sdp, bh);
56 /* If this buffer is in the AIL and it has already been written
57 * to in-place disk block, remove it from the AIL.
59 spin_lock(&sdp->sd_ail_lock);
61 list_move(&bd->bd_ail_st_list, &bd->bd_tr->tr_ail2_list);
62 spin_unlock(&sdp->sd_ail_lock);
64 atomic_inc(&sdp->sd_log_pinned);
65 trace_gfs2_pin(bd, 1);
68 static bool buffer_is_rgrp(const struct gfs2_bufdata *bd)
70 return bd->bd_gl->gl_name.ln_type == LM_TYPE_RGRP;
73 static void maybe_release_space(struct gfs2_bufdata *bd)
75 struct gfs2_glock *gl = bd->bd_gl;
76 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
77 struct gfs2_rgrpd *rgd = gfs2_glock2rgrp(gl);
78 unsigned int index = bd->bd_bh->b_blocknr - gl->gl_name.ln_number;
79 struct gfs2_bitmap *bi = rgd->rd_bits + index;
81 if (bi->bi_clone == NULL)
83 if (sdp->sd_args.ar_discard)
84 gfs2_rgrp_send_discards(sdp, rgd->rd_data0, bd->bd_bh, bi, 1, NULL);
85 memcpy(bi->bi_clone + bi->bi_offset,
86 bd->bd_bh->b_data + bi->bi_offset, bi->bi_bytes);
87 clear_bit(GBF_FULL, &bi->bi_flags);
88 rgd->rd_free_clone = rgd->rd_free;
89 rgd->rd_extfail_pt = rgd->rd_free;
93 * gfs2_unpin - Unpin a buffer
94 * @sdp: the filesystem the buffer belongs to
95 * @bh: The buffer to unpin
97 * @flags: The inode dirty flags
101 static void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
102 struct gfs2_trans *tr)
104 struct gfs2_bufdata *bd = bh->b_private;
106 BUG_ON(!buffer_uptodate(bh));
107 BUG_ON(!buffer_pinned(bh));
110 mark_buffer_dirty(bh);
111 clear_buffer_pinned(bh);
113 if (buffer_is_rgrp(bd))
114 maybe_release_space(bd);
116 spin_lock(&sdp->sd_ail_lock);
118 list_del(&bd->bd_ail_st_list);
121 struct gfs2_glock *gl = bd->bd_gl;
122 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
123 atomic_inc(&gl->gl_ail_count);
126 list_add(&bd->bd_ail_st_list, &tr->tr_ail1_list);
127 spin_unlock(&sdp->sd_ail_lock);
129 clear_bit(GLF_LFLUSH, &bd->bd_gl->gl_flags);
130 trace_gfs2_pin(bd, 0);
132 atomic_dec(&sdp->sd_log_pinned);
135 static void gfs2_log_incr_head(struct gfs2_sbd *sdp)
137 BUG_ON((sdp->sd_log_flush_head == sdp->sd_log_tail) &&
138 (sdp->sd_log_flush_head != sdp->sd_log_head));
140 if (++sdp->sd_log_flush_head == sdp->sd_jdesc->jd_blocks)
141 sdp->sd_log_flush_head = 0;
144 u64 gfs2_log_bmap(struct gfs2_sbd *sdp)
146 unsigned int lbn = sdp->sd_log_flush_head;
147 struct gfs2_journal_extent *je;
150 list_for_each_entry(je, &sdp->sd_jdesc->extent_list, list) {
151 if ((lbn >= je->lblock) && (lbn < (je->lblock + je->blocks))) {
152 block = je->dblock + lbn - je->lblock;
153 gfs2_log_incr_head(sdp);
162 * gfs2_end_log_write_bh - end log write of pagecache data with buffers
163 * @sdp: The superblock
165 * @error: The i/o status
167 * This finds the relevant buffers and unlocks them and sets the
168 * error flag according to the status of the i/o request. This is
169 * used when the log is writing data which has an in-place version
170 * that is pinned in the pagecache.
173 static void gfs2_end_log_write_bh(struct gfs2_sbd *sdp,
174 struct bio_vec *bvec,
177 struct buffer_head *bh, *next;
178 struct page *page = bvec->bv_page;
181 bh = page_buffers(page);
183 while (bh_offset(bh) < bvec->bv_offset)
184 bh = bh->b_this_page;
187 mark_buffer_write_io_error(bh);
189 next = bh->b_this_page;
197 * gfs2_end_log_write - end of i/o to the log
200 * Each bio_vec contains either data from the pagecache or data
201 * relating to the log itself. Here we iterate over the bio_vec
202 * array, processing both kinds of data.
206 static void gfs2_end_log_write(struct bio *bio)
208 struct gfs2_sbd *sdp = bio->bi_private;
209 struct bio_vec *bvec;
212 struct bvec_iter_all iter_all;
214 if (bio->bi_status) {
215 fs_err(sdp, "Error %d writing to journal, jid=%u\n",
216 bio->bi_status, sdp->sd_jdesc->jd_jid);
217 wake_up(&sdp->sd_logd_waitq);
220 bio_for_each_segment_all(bvec, bio, i, iter_all) {
221 page = bvec->bv_page;
222 if (page_has_buffers(page))
223 gfs2_end_log_write_bh(sdp, bvec, bio->bi_status);
225 mempool_free(page, gfs2_page_pool);
229 if (atomic_dec_and_test(&sdp->sd_log_in_flight))
230 wake_up(&sdp->sd_log_flush_wait);
234 * gfs2_log_submit_bio - Submit any pending log bio
235 * @biop: Address of the bio pointer
236 * @opf: REQ_OP | op_flags
238 * Submit any pending part-built or full bio to the block device. If
239 * there is no pending bio, then this is a no-op.
242 void gfs2_log_submit_bio(struct bio **biop, int opf)
244 struct bio *bio = *biop;
246 struct gfs2_sbd *sdp = bio->bi_private;
247 atomic_inc(&sdp->sd_log_in_flight);
255 * gfs2_log_alloc_bio - Allocate a bio
256 * @sdp: The super block
257 * @blkno: The device block number we want to write to
258 * @end_io: The bi_end_io callback
260 * Allocate a new bio, initialize it with the given parameters and return it.
262 * Returns: The newly allocated bio
265 static struct bio *gfs2_log_alloc_bio(struct gfs2_sbd *sdp, u64 blkno,
266 bio_end_io_t *end_io)
268 struct super_block *sb = sdp->sd_vfs;
269 struct bio *bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
271 bio->bi_iter.bi_sector = blkno * (sb->s_blocksize >> 9);
272 bio_set_dev(bio, sb->s_bdev);
273 bio->bi_end_io = end_io;
274 bio->bi_private = sdp;
280 * gfs2_log_get_bio - Get cached log bio, or allocate a new one
281 * @sdp: The super block
282 * @blkno: The device block number we want to write to
283 * @bio: The bio to get or allocate
285 * @end_io: The bi_end_io callback
286 * @flush: Always flush the current bio and allocate a new one?
288 * If there is a cached bio, then if the next block number is sequential
289 * with the previous one, return it, otherwise flush the bio to the
290 * device. If there is no cached bio, or we just flushed it, then
291 * allocate a new one.
293 * Returns: The bio to use for log writes
296 static struct bio *gfs2_log_get_bio(struct gfs2_sbd *sdp, u64 blkno,
297 struct bio **biop, int op,
298 bio_end_io_t *end_io, bool flush)
300 struct bio *bio = *biop;
305 nblk = bio_end_sector(bio);
306 nblk >>= sdp->sd_fsb2bb_shift;
307 if (blkno == nblk && !flush)
309 gfs2_log_submit_bio(biop, op);
312 *biop = gfs2_log_alloc_bio(sdp, blkno, end_io);
317 * gfs2_log_write - write to log
318 * @sdp: the filesystem
319 * @page: the page to write
320 * @size: the size of the data to write
321 * @offset: the offset within the page
322 * @blkno: block number of the log entry
324 * Try and add the page segment to the current bio. If that fails,
325 * submit the current bio to the device and create a new one, and
326 * then add the page segment to that.
329 void gfs2_log_write(struct gfs2_sbd *sdp, struct page *page,
330 unsigned size, unsigned offset, u64 blkno)
335 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio, REQ_OP_WRITE,
336 gfs2_end_log_write, false);
337 ret = bio_add_page(bio, page, size, offset);
339 bio = gfs2_log_get_bio(sdp, blkno, &sdp->sd_log_bio,
340 REQ_OP_WRITE, gfs2_end_log_write, true);
341 ret = bio_add_page(bio, page, size, offset);
347 * gfs2_log_write_bh - write a buffer's content to the log
348 * @sdp: The super block
349 * @bh: The buffer pointing to the in-place location
351 * This writes the content of the buffer to the next available location
352 * in the log. The buffer will be unlocked once the i/o to the log has
356 static void gfs2_log_write_bh(struct gfs2_sbd *sdp, struct buffer_head *bh)
358 gfs2_log_write(sdp, bh->b_page, bh->b_size, bh_offset(bh),
363 * gfs2_log_write_page - write one block stored in a page, into the log
364 * @sdp: The superblock
365 * @page: The struct page
367 * This writes the first block-sized part of the page into the log. Note
368 * that the page must have been allocated from the gfs2_page_pool mempool
369 * and that after this has been called, ownership has been transferred and
370 * the page may be freed at any time.
373 void gfs2_log_write_page(struct gfs2_sbd *sdp, struct page *page)
375 struct super_block *sb = sdp->sd_vfs;
376 gfs2_log_write(sdp, page, sb->s_blocksize, 0,
381 * gfs2_end_log_read - end I/O callback for reads from the log
384 * Simply unlock the pages in the bio. The main thread will wait on them and
385 * process them in order as necessary.
388 static void gfs2_end_log_read(struct bio *bio)
391 struct bio_vec *bvec;
393 struct bvec_iter_all iter_all;
395 bio_for_each_segment_all(bvec, bio, i, iter_all) {
396 page = bvec->bv_page;
397 if (bio->bi_status) {
398 int err = blk_status_to_errno(bio->bi_status);
401 mapping_set_error(page->mapping, err);
410 * gfs2_jhead_pg_srch - Look for the journal head in a given page.
411 * @jd: The journal descriptor
412 * @page: The page to look in
414 * Returns: 1 if found, 0 otherwise.
417 static bool gfs2_jhead_pg_srch(struct gfs2_jdesc *jd,
418 struct gfs2_log_header_host *head,
421 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
422 struct gfs2_log_header_host uninitialized_var(lh);
423 void *kaddr = kmap_atomic(page);
427 for (offset = 0; offset < PAGE_SIZE; offset += sdp->sd_sb.sb_bsize) {
428 if (!__get_log_header(sdp, kaddr + offset, 0, &lh)) {
429 if (lh.lh_sequence > head->lh_sequence)
437 kunmap_atomic(kaddr);
442 * gfs2_jhead_process_page - Search/cleanup a page
443 * @jd: The journal descriptor
444 * @index: Index of the page to look into
445 * @done: If set, perform only cleanup, else search and set if found.
447 * Find the page with 'index' in the journal's mapping. Search the page for
448 * the journal head if requested (cleanup == false). Release refs on the
449 * page so the page cache can reclaim it (put_page() twice). We grabbed a
450 * reference on this page two times, first when we did a find_or_create_page()
451 * to obtain the page to add it to the bio and second when we do a
452 * find_get_page() here to get the page to wait on while I/O on it is being
454 * This function is also used to free up a page we might've grabbed but not
455 * used. Maybe we added it to a bio, but not submitted it for I/O. Or we
456 * submitted the I/O, but we already found the jhead so we only need to drop
457 * our references to the page.
460 static void gfs2_jhead_process_page(struct gfs2_jdesc *jd, unsigned long index,
461 struct gfs2_log_header_host *head,
466 page = find_get_page(jd->jd_inode->i_mapping, index);
467 wait_on_page_locked(page);
473 *done = gfs2_jhead_pg_srch(jd, head, page);
475 put_page(page); /* Once for find_get_page */
476 put_page(page); /* Once more for find_or_create_page */
480 * gfs2_find_jhead - find the head of a log
481 * @jd: The journal descriptor
482 * @head: The log descriptor for the head of the log is returned here
484 * Do a search of a journal by reading it in large chunks using bios and find
485 * the valid log entry with the highest sequence number. (i.e. the log head)
487 * Returns: 0 on success, errno otherwise
489 int gfs2_find_jhead(struct gfs2_jdesc *jd, struct gfs2_log_header_host *head,
492 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
493 struct address_space *mapping = jd->jd_inode->i_mapping;
494 unsigned int block = 0, blocks_submitted = 0, blocks_read = 0;
495 unsigned int bsize = sdp->sd_sb.sb_bsize;
496 unsigned int bsize_shift = sdp->sd_sb.sb_bsize_shift;
497 unsigned int shift = PAGE_SHIFT - bsize_shift;
498 unsigned int readhead_blocks = BIO_MAX_PAGES << shift;
499 struct gfs2_journal_extent *je;
501 struct bio *bio = NULL;
502 struct page *page = NULL;
506 memset(head, 0, sizeof(*head));
507 if (list_empty(&jd->extent_list))
508 gfs2_map_journal_extents(sdp, jd);
510 since = filemap_sample_wb_err(mapping);
511 list_for_each_entry(je, &jd->extent_list, list) {
512 for (; block < je->lblock + je->blocks; block++) {
516 page = find_or_create_page(mapping,
517 block >> shift, GFP_NOFS);
528 off = (block << bsize_shift) & ~PAGE_MASK;
529 sz = bio_add_page(bio, page, bsize, off);
530 if (sz == bsize) { /* block added */
531 if (off + bsize == PAGE_SIZE) {
537 blocks_submitted = block + 1;
542 dblock = je->dblock + (block - je->lblock);
543 bio = gfs2_log_alloc_bio(sdp, dblock, gfs2_end_log_read);
544 bio->bi_opf = REQ_OP_READ;
545 sz = bio_add_page(bio, page, bsize, 0);
546 gfs2_assert_warn(sdp, sz == bsize);
547 if (bsize == PAGE_SIZE)
551 if (blocks_submitted < blocks_read + readhead_blocks) {
552 /* Keep at least one bio in flight */
556 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
557 blocks_read += PAGE_SIZE >> bsize_shift;
559 goto out; /* found */
566 while (blocks_read < block) {
567 gfs2_jhead_process_page(jd, blocks_read >> shift, head, &done);
568 blocks_read += PAGE_SIZE >> bsize_shift;
572 ret = filemap_check_wb_err(mapping, since);
575 truncate_inode_pages(mapping, 0);
580 static struct page *gfs2_get_log_desc(struct gfs2_sbd *sdp, u32 ld_type,
581 u32 ld_length, u32 ld_data1)
583 struct page *page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
584 struct gfs2_log_descriptor *ld = page_address(page);
586 ld->ld_header.mh_magic = cpu_to_be32(GFS2_MAGIC);
587 ld->ld_header.mh_type = cpu_to_be32(GFS2_METATYPE_LD);
588 ld->ld_header.mh_format = cpu_to_be32(GFS2_FORMAT_LD);
589 ld->ld_type = cpu_to_be32(ld_type);
590 ld->ld_length = cpu_to_be32(ld_length);
591 ld->ld_data1 = cpu_to_be32(ld_data1);
596 static void gfs2_check_magic(struct buffer_head *bh)
601 clear_buffer_escaped(bh);
602 kaddr = kmap_atomic(bh->b_page);
603 ptr = kaddr + bh_offset(bh);
604 if (*ptr == cpu_to_be32(GFS2_MAGIC))
605 set_buffer_escaped(bh);
606 kunmap_atomic(kaddr);
609 static int blocknr_cmp(void *priv, struct list_head *a, struct list_head *b)
611 struct gfs2_bufdata *bda, *bdb;
613 bda = list_entry(a, struct gfs2_bufdata, bd_list);
614 bdb = list_entry(b, struct gfs2_bufdata, bd_list);
616 if (bda->bd_bh->b_blocknr < bdb->bd_bh->b_blocknr)
618 if (bda->bd_bh->b_blocknr > bdb->bd_bh->b_blocknr)
623 static void gfs2_before_commit(struct gfs2_sbd *sdp, unsigned int limit,
624 unsigned int total, struct list_head *blist,
627 struct gfs2_log_descriptor *ld;
628 struct gfs2_bufdata *bd1 = NULL, *bd2;
635 list_sort(NULL, blist, blocknr_cmp);
636 bd1 = bd2 = list_prepare_entry(bd1, blist, bd_list);
641 gfs2_log_unlock(sdp);
642 page = gfs2_get_log_desc(sdp,
643 is_databuf ? GFS2_LOG_DESC_JDATA :
644 GFS2_LOG_DESC_METADATA, num + 1, num);
645 ld = page_address(page);
647 ptr = (__be64 *)(ld + 1);
650 list_for_each_entry_continue(bd1, blist, bd_list) {
651 *ptr++ = cpu_to_be64(bd1->bd_bh->b_blocknr);
653 gfs2_check_magic(bd1->bd_bh);
654 *ptr++ = cpu_to_be64(buffer_escaped(bd1->bd_bh) ? 1 : 0);
660 gfs2_log_unlock(sdp);
661 gfs2_log_write_page(sdp, page);
665 list_for_each_entry_continue(bd2, blist, bd_list) {
667 gfs2_log_unlock(sdp);
668 lock_buffer(bd2->bd_bh);
670 if (buffer_escaped(bd2->bd_bh)) {
672 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
673 ptr = page_address(page);
674 kaddr = kmap_atomic(bd2->bd_bh->b_page);
675 memcpy(ptr, kaddr + bh_offset(bd2->bd_bh),
677 kunmap_atomic(kaddr);
679 clear_buffer_escaped(bd2->bd_bh);
680 unlock_buffer(bd2->bd_bh);
682 gfs2_log_write_page(sdp, page);
684 gfs2_log_write_bh(sdp, bd2->bd_bh);
694 gfs2_log_unlock(sdp);
697 static void buf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
699 unsigned int limit = buf_limit(sdp); /* 503 for 4k blocks */
703 nbuf = tr->tr_num_buf_new - tr->tr_num_buf_rm;
704 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_buf, 0);
707 static void buf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
709 struct list_head *head;
710 struct gfs2_bufdata *bd;
716 while (!list_empty(head)) {
717 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
718 list_del_init(&bd->bd_list);
719 gfs2_unpin(sdp, bd->bd_bh, tr);
723 static void buf_lo_before_scan(struct gfs2_jdesc *jd,
724 struct gfs2_log_header_host *head, int pass)
729 jd->jd_found_blocks = 0;
730 jd->jd_replayed_blocks = 0;
733 static int buf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
734 struct gfs2_log_descriptor *ld, __be64 *ptr,
737 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
738 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
739 struct gfs2_glock *gl = ip->i_gl;
740 unsigned int blks = be32_to_cpu(ld->ld_data1);
741 struct buffer_head *bh_log, *bh_ip;
745 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_METADATA)
748 gfs2_replay_incr_blk(jd, &start);
750 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
751 blkno = be64_to_cpu(*ptr++);
753 jd->jd_found_blocks++;
755 if (gfs2_revoke_check(jd, blkno, start))
758 error = gfs2_replay_read_block(jd, start, &bh_log);
762 bh_ip = gfs2_meta_new(gl, blkno);
763 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
765 if (gfs2_meta_check(sdp, bh_ip))
768 mark_buffer_dirty(bh_ip);
776 jd->jd_replayed_blocks++;
783 * gfs2_meta_sync - Sync all buffers associated with a glock
788 static void gfs2_meta_sync(struct gfs2_glock *gl)
790 struct address_space *mapping = gfs2_glock2aspace(gl);
791 struct gfs2_sbd *sdp = gl->gl_name.ln_sbd;
795 mapping = &sdp->sd_aspace;
797 filemap_fdatawrite(mapping);
798 error = filemap_fdatawait(mapping);
801 gfs2_io_error(gl->gl_name.ln_sbd);
804 static void buf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
806 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
807 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
810 gfs2_meta_sync(ip->i_gl);
816 gfs2_meta_sync(ip->i_gl);
818 fs_info(sdp, "jid=%u: Replayed %u of %u blocks\n",
819 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
822 static void revoke_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
824 struct gfs2_meta_header *mh;
826 struct list_head *head = &sdp->sd_log_revokes;
827 struct gfs2_bufdata *bd;
831 gfs2_write_revokes(sdp);
832 if (!sdp->sd_log_num_revoke)
835 length = gfs2_struct2blk(sdp, sdp->sd_log_num_revoke, sizeof(u64));
836 page = gfs2_get_log_desc(sdp, GFS2_LOG_DESC_REVOKE, length, sdp->sd_log_num_revoke);
837 offset = sizeof(struct gfs2_log_descriptor);
839 list_for_each_entry(bd, head, bd_list) {
840 sdp->sd_log_num_revoke--;
842 if (offset + sizeof(u64) > sdp->sd_sb.sb_bsize) {
844 gfs2_log_write_page(sdp, page);
845 page = mempool_alloc(gfs2_page_pool, GFP_NOIO);
846 mh = page_address(page);
848 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
849 mh->mh_type = cpu_to_be32(GFS2_METATYPE_LB);
850 mh->mh_format = cpu_to_be32(GFS2_FORMAT_LB);
851 offset = sizeof(struct gfs2_meta_header);
854 *(__be64 *)(page_address(page) + offset) = cpu_to_be64(bd->bd_blkno);
855 offset += sizeof(u64);
857 gfs2_assert_withdraw(sdp, !sdp->sd_log_num_revoke);
859 gfs2_log_write_page(sdp, page);
862 static void revoke_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
864 struct list_head *head = &sdp->sd_log_revokes;
865 struct gfs2_bufdata *bd, *tmp;
868 * Glocks can be referenced repeatedly on the revoke list, but the list
869 * only holds one reference. All glocks on the list will have the
870 * GLF_REVOKES flag set initially.
873 list_for_each_entry_safe(bd, tmp, head, bd_list) {
874 struct gfs2_glock *gl = bd->bd_gl;
876 if (test_bit(GLF_REVOKES, &gl->gl_flags)) {
877 /* Keep each glock on the list exactly once. */
878 clear_bit(GLF_REVOKES, &gl->gl_flags);
881 list_del(&bd->bd_list);
882 kmem_cache_free(gfs2_bufdata_cachep, bd);
884 list_for_each_entry_safe(bd, tmp, head, bd_list) {
885 struct gfs2_glock *gl = bd->bd_gl;
887 list_del(&bd->bd_list);
888 kmem_cache_free(gfs2_bufdata_cachep, bd);
889 clear_bit(GLF_LFLUSH, &gl->gl_flags);
890 gfs2_glock_queue_put(gl);
892 /* the list is empty now */
895 static void revoke_lo_before_scan(struct gfs2_jdesc *jd,
896 struct gfs2_log_header_host *head, int pass)
901 jd->jd_found_revokes = 0;
902 jd->jd_replay_tail = head->lh_tail;
905 static int revoke_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
906 struct gfs2_log_descriptor *ld, __be64 *ptr,
909 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
910 unsigned int blks = be32_to_cpu(ld->ld_length);
911 unsigned int revokes = be32_to_cpu(ld->ld_data1);
912 struct buffer_head *bh;
918 if (pass != 0 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_REVOKE)
921 offset = sizeof(struct gfs2_log_descriptor);
923 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
924 error = gfs2_replay_read_block(jd, start, &bh);
929 gfs2_metatype_check(sdp, bh, GFS2_METATYPE_LB);
931 while (offset + sizeof(u64) <= sdp->sd_sb.sb_bsize) {
932 blkno = be64_to_cpu(*(__be64 *)(bh->b_data + offset));
934 error = gfs2_revoke_add(jd, blkno, start);
940 jd->jd_found_revokes++;
944 offset += sizeof(u64);
948 offset = sizeof(struct gfs2_meta_header);
955 static void revoke_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
957 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
960 gfs2_revoke_clean(jd);
966 fs_info(sdp, "jid=%u: Found %u revoke tags\n",
967 jd->jd_jid, jd->jd_found_revokes);
969 gfs2_revoke_clean(jd);
973 * databuf_lo_before_commit - Scan the data buffers, writing as we go
977 static void databuf_lo_before_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
979 unsigned int limit = databuf_limit(sdp);
983 nbuf = tr->tr_num_databuf_new - tr->tr_num_databuf_rm;
984 gfs2_before_commit(sdp, limit, nbuf, &tr->tr_databuf, 1);
987 static int databuf_lo_scan_elements(struct gfs2_jdesc *jd, u32 start,
988 struct gfs2_log_descriptor *ld,
989 __be64 *ptr, int pass)
991 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
992 struct gfs2_glock *gl = ip->i_gl;
993 unsigned int blks = be32_to_cpu(ld->ld_data1);
994 struct buffer_head *bh_log, *bh_ip;
999 if (pass != 1 || be32_to_cpu(ld->ld_type) != GFS2_LOG_DESC_JDATA)
1002 gfs2_replay_incr_blk(jd, &start);
1003 for (; blks; gfs2_replay_incr_blk(jd, &start), blks--) {
1004 blkno = be64_to_cpu(*ptr++);
1005 esc = be64_to_cpu(*ptr++);
1007 jd->jd_found_blocks++;
1009 if (gfs2_revoke_check(jd, blkno, start))
1012 error = gfs2_replay_read_block(jd, start, &bh_log);
1016 bh_ip = gfs2_meta_new(gl, blkno);
1017 memcpy(bh_ip->b_data, bh_log->b_data, bh_log->b_size);
1021 __be32 *eptr = (__be32 *)bh_ip->b_data;
1022 *eptr = cpu_to_be32(GFS2_MAGIC);
1024 mark_buffer_dirty(bh_ip);
1029 jd->jd_replayed_blocks++;
1035 /* FIXME: sort out accounting for log blocks etc. */
1037 static void databuf_lo_after_scan(struct gfs2_jdesc *jd, int error, int pass)
1039 struct gfs2_inode *ip = GFS2_I(jd->jd_inode);
1040 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode);
1043 gfs2_meta_sync(ip->i_gl);
1050 gfs2_meta_sync(ip->i_gl);
1052 fs_info(sdp, "jid=%u: Replayed %u of %u data blocks\n",
1053 jd->jd_jid, jd->jd_replayed_blocks, jd->jd_found_blocks);
1056 static void databuf_lo_after_commit(struct gfs2_sbd *sdp, struct gfs2_trans *tr)
1058 struct list_head *head;
1059 struct gfs2_bufdata *bd;
1064 head = &tr->tr_databuf;
1065 while (!list_empty(head)) {
1066 bd = list_entry(head->next, struct gfs2_bufdata, bd_list);
1067 list_del_init(&bd->bd_list);
1068 gfs2_unpin(sdp, bd->bd_bh, tr);
1073 static const struct gfs2_log_operations gfs2_buf_lops = {
1074 .lo_before_commit = buf_lo_before_commit,
1075 .lo_after_commit = buf_lo_after_commit,
1076 .lo_before_scan = buf_lo_before_scan,
1077 .lo_scan_elements = buf_lo_scan_elements,
1078 .lo_after_scan = buf_lo_after_scan,
1082 static const struct gfs2_log_operations gfs2_revoke_lops = {
1083 .lo_before_commit = revoke_lo_before_commit,
1084 .lo_after_commit = revoke_lo_after_commit,
1085 .lo_before_scan = revoke_lo_before_scan,
1086 .lo_scan_elements = revoke_lo_scan_elements,
1087 .lo_after_scan = revoke_lo_after_scan,
1088 .lo_name = "revoke",
1091 static const struct gfs2_log_operations gfs2_databuf_lops = {
1092 .lo_before_commit = databuf_lo_before_commit,
1093 .lo_after_commit = databuf_lo_after_commit,
1094 .lo_scan_elements = databuf_lo_scan_elements,
1095 .lo_after_scan = databuf_lo_after_scan,
1096 .lo_name = "databuf",
1099 const struct gfs2_log_operations *gfs2_log_ops[] = {