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 v.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>
16 #include <linux/pagemap.h>
17 #include <linux/writeback.h>
18 #include <linux/swap.h>
19 #include <linux/delay.h>
20 #include <linux/gfs2_ondisk.h>
23 #include "lm_interface.h"
35 #define buffer_busy(bh) \
36 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
37 #define buffer_in_io(bh) \
38 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
40 static int aspace_get_block(struct inode *inode, sector_t lblock,
41 struct buffer_head *bh_result, int create)
43 gfs2_assert_warn(inode->i_sb->s_fs_info, 0);
47 static int gfs2_aspace_writepage(struct page *page,
48 struct writeback_control *wbc)
50 return block_write_full_page(page, aspace_get_block, wbc);
54 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
55 * @bh: the buffer we're stuck on
59 static void stuck_releasepage(struct buffer_head *bh)
61 struct inode *inode = bh->b_page->mapping->host;
62 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
63 struct gfs2_bufdata *bd = bh->b_private;
64 struct gfs2_glock *gl;
66 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
67 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
68 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
69 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
70 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
77 fs_warn(sdp, "gl = (%u, %llu)\n",
78 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
80 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
81 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
82 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
84 if (gl->gl_ops == &gfs2_inode_glops) {
85 struct gfs2_inode *ip = gl->gl_object;
91 fs_warn(sdp, "ip = %llu %llu\n",
92 (unsigned long long)ip->i_num.no_formal_ino,
93 (unsigned long long)ip->i_num.no_addr);
95 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
96 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
97 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
102 * gfs2_aspace_releasepage - free the metadata associated with a page
103 * @page: the page that's being released
104 * @gfp_mask: passed from Linux VFS, ignored by us
106 * Call try_to_free_buffers() if the buffers in this page can be
112 static int gfs2_aspace_releasepage(struct page *page, gfp_t gfp_mask)
114 struct inode *aspace = page->mapping->host;
115 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
116 struct buffer_head *bh, *head;
117 struct gfs2_bufdata *bd;
120 if (!page_has_buffers(page))
123 head = bh = page_buffers(page);
127 while (atomic_read(&bh->b_count)) {
128 if (atomic_read(&aspace->i_writecount)) {
129 if (time_after_eq(jiffies, t +
130 gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
131 stuck_releasepage(bh);
142 gfs2_assert_warn(sdp, !buffer_pinned(bh));
146 gfs2_assert_warn(sdp, bd->bd_bh == bh);
147 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
148 gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
149 gfs2_assert_warn(sdp, !bd->bd_ail);
150 kmem_cache_free(gfs2_bufdata_cachep, bd);
151 bh->b_private = NULL;
154 bh = bh->b_this_page;
159 return try_to_free_buffers(page);
162 static const struct address_space_operations aspace_aops = {
163 .writepage = gfs2_aspace_writepage,
164 .releasepage = gfs2_aspace_releasepage,
168 * gfs2_aspace_get - Create and initialize a struct inode structure
169 * @sdp: the filesystem the aspace is in
171 * Right now a struct inode is just a struct inode. Maybe Linux
172 * will supply a more lightweight address space construct (that works)
175 * Make sure pages/buffers in this aspace aren't in high memory.
177 * Returns: the aspace
180 struct inode *gfs2_aspace_get(struct gfs2_sbd *sdp)
182 struct inode *aspace;
184 aspace = new_inode(sdp->sd_vfs);
186 mapping_set_gfp_mask(aspace->i_mapping, GFP_KERNEL);
187 aspace->i_mapping->a_ops = &aspace_aops;
188 aspace->i_size = ~0ULL;
189 aspace->u.generic_ip = NULL;
190 insert_inode_hash(aspace);
195 void gfs2_aspace_put(struct inode *aspace)
197 remove_inode_hash(aspace);
202 * gfs2_ail1_start_one - Start I/O on a part of the AIL
203 * @sdp: the filesystem
204 * @tr: the part of the AIL
208 void gfs2_ail1_start_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
210 struct gfs2_bufdata *bd, *s;
211 struct buffer_head *bh;
214 BUG_ON(!spin_is_locked(&sdp->sd_log_lock));
219 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
223 gfs2_assert(sdp, bd->bd_ail == ai);
225 if (!buffer_busy(bh)) {
226 if (!buffer_uptodate(bh)) {
227 gfs2_log_unlock(sdp);
228 gfs2_io_error_bh(sdp, bh);
231 list_move(&bd->bd_ail_st_list,
236 if (!buffer_dirty(bh))
239 list_move(&bd->bd_ail_st_list, &ai->ai_ail1_list);
241 gfs2_log_unlock(sdp);
243 ll_rw_block(WRITE, 1, &bh);
253 * gfs2_ail1_empty_one - Check whether or not a trans in the AIL has been synced
254 * @sdp: the filesystem
259 int gfs2_ail1_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai, int flags)
261 struct gfs2_bufdata *bd, *s;
262 struct buffer_head *bh;
264 list_for_each_entry_safe_reverse(bd, s, &ai->ai_ail1_list,
268 gfs2_assert(sdp, bd->bd_ail == ai);
270 if (buffer_busy(bh)) {
277 if (!buffer_uptodate(bh))
278 gfs2_io_error_bh(sdp, bh);
280 list_move(&bd->bd_ail_st_list, &ai->ai_ail2_list);
283 return list_empty(&ai->ai_ail1_list);
287 * gfs2_ail2_empty_one - Check whether or not a trans in the AIL has been synced
288 * @sdp: the filesystem
293 void gfs2_ail2_empty_one(struct gfs2_sbd *sdp, struct gfs2_ail *ai)
295 struct list_head *head = &ai->ai_ail2_list;
296 struct gfs2_bufdata *bd;
298 while (!list_empty(head)) {
299 bd = list_entry(head->prev, struct gfs2_bufdata,
301 gfs2_assert(sdp, bd->bd_ail == ai);
303 list_del(&bd->bd_ail_st_list);
304 list_del(&bd->bd_ail_gl_list);
305 atomic_dec(&bd->bd_gl->gl_ail_count);
311 * ail_empty_gl - remove all buffers for a given lock from the AIL
314 * None of the buffers should be dirty, locked, or pinned.
317 void gfs2_ail_empty_gl(struct gfs2_glock *gl)
319 struct gfs2_sbd *sdp = gl->gl_sbd;
321 struct list_head *head = &gl->gl_ail_list;
322 struct gfs2_bufdata *bd;
323 struct buffer_head *bh;
327 blocks = atomic_read(&gl->gl_ail_count);
331 error = gfs2_trans_begin(sdp, 0, blocks);
332 if (gfs2_assert_withdraw(sdp, !error))
336 while (!list_empty(head)) {
337 bd = list_entry(head->next, struct gfs2_bufdata,
340 blkno = bh->b_blocknr;
341 gfs2_assert_withdraw(sdp, !buffer_busy(bh));
344 list_del(&bd->bd_ail_st_list);
345 list_del(&bd->bd_ail_gl_list);
346 atomic_dec(&gl->gl_ail_count);
348 gfs2_log_unlock(sdp);
350 gfs2_trans_add_revoke(sdp, blkno);
354 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
355 gfs2_log_unlock(sdp);
358 gfs2_log_flush(sdp, NULL);
362 * gfs2_meta_inval - Invalidate all buffers associated with a glock
367 void gfs2_meta_inval(struct gfs2_glock *gl)
369 struct gfs2_sbd *sdp = gl->gl_sbd;
370 struct inode *aspace = gl->gl_aspace;
371 struct address_space *mapping = gl->gl_aspace->i_mapping;
373 gfs2_assert_withdraw(sdp, !atomic_read(&gl->gl_ail_count));
375 atomic_inc(&aspace->i_writecount);
376 truncate_inode_pages(mapping, 0);
377 atomic_dec(&aspace->i_writecount);
379 gfs2_assert_withdraw(sdp, !mapping->nrpages);
383 * gfs2_meta_sync - Sync all buffers associated with a glock
385 * @flags: DIO_START | DIO_WAIT
389 void gfs2_meta_sync(struct gfs2_glock *gl, int flags)
391 struct address_space *mapping = gl->gl_aspace->i_mapping;
394 if (flags & DIO_START)
395 filemap_fdatawrite(mapping);
396 if (!error && (flags & DIO_WAIT))
397 error = filemap_fdatawait(mapping);
400 gfs2_io_error(gl->gl_sbd);
404 * getbuf - Get a buffer with a given address space
405 * @sdp: the filesystem
406 * @aspace: the address space
407 * @blkno: the block number (filesystem scope)
408 * @create: 1 if the buffer should be created
410 * Returns: the buffer
413 static struct buffer_head *getbuf(struct gfs2_sbd *sdp, struct inode *aspace,
414 uint64_t blkno, int create)
417 struct buffer_head *bh;
422 shift = PAGE_CACHE_SHIFT - sdp->sd_sb.sb_bsize_shift;
423 index = blkno >> shift; /* convert block to page */
424 bufnum = blkno - (index << shift); /* block buf index within page */
428 page = grab_cache_page(aspace->i_mapping, index);
434 page = find_lock_page(aspace->i_mapping, index);
439 if (!page_has_buffers(page))
440 create_empty_buffers(page, sdp->sd_sb.sb_bsize, 0);
442 /* Locate header for our buffer within our page */
443 for (bh = page_buffers(page); bufnum--; bh = bh->b_this_page)
447 if (!buffer_mapped(bh))
448 map_bh(bh, sdp->sd_vfs, blkno);
451 mark_page_accessed(page);
452 page_cache_release(page);
457 static void meta_prep_new(struct buffer_head *bh)
459 struct gfs2_meta_header *mh = (struct gfs2_meta_header *)bh->b_data;
462 clear_buffer_dirty(bh);
463 set_buffer_uptodate(bh);
466 mh->mh_magic = cpu_to_be32(GFS2_MAGIC);
470 * gfs2_meta_new - Get a block
471 * @gl: The glock associated with this block
472 * @blkno: The block number
474 * Returns: The buffer
477 struct buffer_head *gfs2_meta_new(struct gfs2_glock *gl, uint64_t blkno)
479 struct buffer_head *bh;
480 bh = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
486 * gfs2_meta_read - Read a block from disk
487 * @gl: The glock covering the block
488 * @blkno: The block number
489 * @flags: flags to gfs2_dreread()
490 * @bhp: the place where the buffer is returned (NULL on failure)
495 int gfs2_meta_read(struct gfs2_glock *gl, uint64_t blkno, int flags,
496 struct buffer_head **bhp)
500 *bhp = getbuf(gl->gl_sbd, gl->gl_aspace, blkno, CREATE);
501 error = gfs2_meta_reread(gl->gl_sbd, *bhp, flags);
509 * gfs2_meta_reread - Reread a block from disk
510 * @sdp: the filesystem
511 * @bh: The block to read
512 * @flags: Flags that control the read
517 int gfs2_meta_reread(struct gfs2_sbd *sdp, struct buffer_head *bh, int flags)
519 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
522 if (flags & DIO_FORCE)
523 clear_buffer_uptodate(bh);
525 if ((flags & DIO_START) && !buffer_uptodate(bh))
526 ll_rw_block(READ, 1, &bh);
528 if (flags & DIO_WAIT) {
531 if (!buffer_uptodate(bh)) {
532 struct gfs2_trans *tr = current->journal_info;
533 if (tr && tr->tr_touched)
534 gfs2_io_error_bh(sdp, bh);
537 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
545 * gfs2_attach_bufdata - attach a struct gfs2_bufdata structure to a buffer
546 * @gl: the glock the buffer belongs to
547 * @bh: The buffer to be attached to
548 * @meta: Flag to indicate whether its metadata or not
551 void gfs2_attach_bufdata(struct gfs2_glock *gl, struct buffer_head *bh,
554 struct gfs2_bufdata *bd;
557 lock_page(bh->b_page);
561 unlock_page(bh->b_page);
565 bd = kmem_cache_alloc(gfs2_bufdata_cachep, GFP_NOFS | __GFP_NOFAIL),
566 memset(bd, 0, sizeof(struct gfs2_bufdata));
570 INIT_LIST_HEAD(&bd->bd_list_tr);
572 lops_init_le(&bd->bd_le, &gfs2_buf_lops);
574 lops_init_le(&bd->bd_le, &gfs2_databuf_lops);
580 unlock_page(bh->b_page);
584 * gfs2_pin - Pin a buffer in memory
585 * @sdp: the filesystem the buffer belongs to
586 * @bh: The buffer to be pinned
590 void gfs2_pin(struct gfs2_sbd *sdp, struct buffer_head *bh)
592 struct gfs2_bufdata *bd = bh->b_private;
594 gfs2_assert_withdraw(sdp, test_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags));
596 if (test_set_buffer_pinned(bh))
597 gfs2_assert_withdraw(sdp, 0);
601 /* If this buffer is in the AIL and it has already been written
602 to in-place disk block, remove it from the AIL. */
605 if (bd->bd_ail && !buffer_in_io(bh))
606 list_move(&bd->bd_ail_st_list, &bd->bd_ail->ai_ail2_list);
607 gfs2_log_unlock(sdp);
609 clear_buffer_dirty(bh);
612 if (!buffer_uptodate(bh))
613 gfs2_io_error_bh(sdp, bh);
619 * gfs2_unpin - Unpin a buffer
620 * @sdp: the filesystem the buffer belongs to
621 * @bh: The buffer to unpin
626 void gfs2_unpin(struct gfs2_sbd *sdp, struct buffer_head *bh,
629 struct gfs2_bufdata *bd = bh->b_private;
631 gfs2_assert_withdraw(sdp, buffer_uptodate(bh));
633 if (!buffer_pinned(bh))
634 gfs2_assert_withdraw(sdp, 0);
636 mark_buffer_dirty(bh);
637 clear_buffer_pinned(bh);
641 list_del(&bd->bd_ail_st_list);
644 struct gfs2_glock *gl = bd->bd_gl;
645 list_add(&bd->bd_ail_gl_list, &gl->gl_ail_list);
646 atomic_inc(&gl->gl_ail_count);
649 list_add(&bd->bd_ail_st_list, &ai->ai_ail1_list);
650 gfs2_log_unlock(sdp);
654 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
655 * @ip: the inode who owns the buffers
656 * @bstart: the first buffer in the run
657 * @blen: the number of buffers in the run
661 void gfs2_meta_wipe(struct gfs2_inode *ip, uint64_t bstart, uint32_t blen)
663 struct gfs2_sbd *sdp = GFS2_SB(&ip->i_inode);
664 struct inode *aspace = ip->i_gl->gl_aspace;
665 struct buffer_head *bh;
668 bh = getbuf(sdp, aspace, bstart, NO_CREATE);
670 struct gfs2_bufdata *bd = bh->b_private;
672 if (test_clear_buffer_pinned(bh)) {
673 struct gfs2_trans *tr = current->journal_info;
675 list_del_init(&bd->bd_le.le_list);
676 gfs2_assert_warn(sdp, sdp->sd_log_num_buf);
677 sdp->sd_log_num_buf--;
678 gfs2_log_unlock(sdp);
685 uint64_t blkno = bh->b_blocknr;
687 list_del(&bd->bd_ail_st_list);
688 list_del(&bd->bd_ail_gl_list);
689 atomic_dec(&bd->bd_gl->gl_ail_count);
691 gfs2_log_unlock(sdp);
692 gfs2_trans_add_revoke(sdp, blkno);
694 gfs2_log_unlock(sdp);
698 clear_buffer_dirty(bh);
699 clear_buffer_uptodate(bh);
711 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
712 * @ip: The GFS2 inode
714 * This releases buffers that are in the most-recently-used array of
715 * blocks used for indirect block addressing for this inode.
718 void gfs2_meta_cache_flush(struct gfs2_inode *ip)
720 struct buffer_head **bh_slot;
723 spin_lock(&ip->i_spin);
725 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++) {
726 bh_slot = &ip->i_cache[x];
733 spin_unlock(&ip->i_spin);
737 * gfs2_meta_indirect_buffer - Get a metadata buffer
738 * @ip: The GFS2 inode
739 * @height: The level of this buf in the metadata (indir addr) tree (if any)
740 * @num: The block number (device relative) of the buffer
741 * @new: Non-zero if we may create a new buffer
742 * @bhp: the buffer is returned here
744 * Try to use the gfs2_inode's MRU metadata tree cache.
749 int gfs2_meta_indirect_buffer(struct gfs2_inode *ip, int height, uint64_t num,
750 int new, struct buffer_head **bhp)
752 struct buffer_head *bh, **bh_slot = ip->i_cache + height;
755 spin_lock(&ip->i_spin);
758 if (bh->b_blocknr == num)
763 spin_unlock(&ip->i_spin);
769 error = gfs2_meta_reread(GFS2_SB(&ip->i_inode), bh,
770 DIO_START | DIO_WAIT);
778 bh = gfs2_meta_new(ip->i_gl, num);
780 error = gfs2_meta_read(ip->i_gl, num,
781 DIO_START | DIO_WAIT, &bh);
786 spin_lock(&ip->i_spin);
787 if (*bh_slot != bh) {
792 spin_unlock(&ip->i_spin);
796 if (gfs2_assert_warn(GFS2_SB(&ip->i_inode), height)) {
800 gfs2_trans_add_bh(ip->i_gl, bh, 1);
801 gfs2_metatype_set(bh, GFS2_METATYPE_IN, GFS2_FORMAT_IN);
802 gfs2_buffer_clear_tail(bh, sizeof(struct gfs2_meta_header));
804 } else if (gfs2_metatype_check(GFS2_SB(&ip->i_inode), bh,
805 (height) ? GFS2_METATYPE_IN : GFS2_METATYPE_DI)) {
816 * gfs2_meta_ra - start readahead on an extent of a file
817 * @gl: the glock the blocks belong to
818 * @dblock: the starting disk block
819 * @extlen: the number of blocks in the extent
823 void gfs2_meta_ra(struct gfs2_glock *gl, uint64_t dblock, uint32_t extlen)
825 struct gfs2_sbd *sdp = gl->gl_sbd;
826 struct inode *aspace = gl->gl_aspace;
827 struct buffer_head *first_bh, *bh;
828 uint32_t max_ra = gfs2_tune_get(sdp, gt_max_readahead) >>
829 sdp->sd_sb.sb_bsize_shift;
832 if (!extlen || !max_ra)
837 first_bh = getbuf(sdp, aspace, dblock, CREATE);
839 if (buffer_uptodate(first_bh))
841 if (!buffer_locked(first_bh)) {
842 error = gfs2_meta_reread(sdp, first_bh, DIO_START);
851 bh = getbuf(sdp, aspace, dblock, CREATE);
853 if (!buffer_uptodate(bh) && !buffer_locked(bh)) {
854 error = gfs2_meta_reread(sdp, bh, DIO_START);
864 if (buffer_uptodate(first_bh))
873 * gfs2_meta_syncfs - sync all the buffers in a filesystem
874 * @sdp: the filesystem
878 void gfs2_meta_syncfs(struct gfs2_sbd *sdp)
880 gfs2_log_flush(sdp, NULL);
882 gfs2_ail1_start(sdp, DIO_ALL);
883 if (gfs2_ail1_empty(sdp, DIO_ALL))