2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2005 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>
21 #include <asm/semaphore.h>
24 #include "lm_interface.h"
36 #define buffer_busy(bh) \
37 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock) | (1ul << BH_Pinned)))
38 #define buffer_in_io(bh) \
39 ((bh)->b_state & ((1ul << BH_Dirty) | (1ul << BH_Lock)))
41 static int aspace_get_block(struct inode
*inode
, sector_t lblock
,
42 struct buffer_head
*bh_result
, int create
)
44 gfs2_assert_warn(inode
->i_sb
->s_fs_info
, 0);
48 static int gfs2_aspace_writepage(struct page
*page
,
49 struct writeback_control
*wbc
)
51 return block_write_full_page(page
, aspace_get_block
, wbc
);
55 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
56 * @bh: the buffer we're stuck on
60 static void stuck_releasepage(struct buffer_head
*bh
)
62 struct inode
*inode
= bh
->b_page
->mapping
->host
;
63 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
64 struct gfs2_bufdata
*bd
= bh
->b_private
;
65 struct gfs2_glock
*gl
;
67 fs_warn(sdp
, "stuck in gfs2_releasepage() %p\n", inode
);
68 fs_warn(sdp
, "blkno = %llu, bh->b_count = %d\n",
69 (uint64_t)bh
->b_blocknr
, atomic_read(&bh
->b_count
));
70 fs_warn(sdp
, "pinned = %u\n", buffer_pinned(bh
));
71 fs_warn(sdp
, "bh->b_private = %s\n", (bd
) ? "!NULL" : "NULL");
78 fs_warn(sdp
, "gl = (%u, %llu)\n",
79 gl
->gl_name
.ln_type
, gl
->gl_name
.ln_number
);
81 fs_warn(sdp
, "bd_list_tr = %s, bd_le.le_list = %s\n",
82 (list_empty(&bd
->bd_list_tr
)) ? "no" : "yes",
83 (list_empty(&bd
->bd_le
.le_list
)) ? "no" : "yes");
85 if (gl
->gl_ops
== &gfs2_inode_glops
) {
86 struct gfs2_inode
*ip
= gl
->gl_object
;
92 fs_warn(sdp
, "ip = %llu %llu\n",
93 ip
->i_num
.no_formal_ino
, ip
->i_num
.no_addr
);
94 fs_warn(sdp
, "ip->i_count = %d, ip->i_vnode = %s\n",
95 atomic_read(&ip
->i_count
),
96 (ip
->i_vnode
) ? "!NULL" : "NULL");
98 for (x
= 0; x
< GFS2_MAX_META_HEIGHT
; x
++)
99 fs_warn(sdp
, "ip->i_cache[%u] = %s\n",
100 x
, (ip
->i_cache
[x
]) ? "!NULL" : "NULL");
105 * gfs2_aspace_releasepage - free the metadata associated with a page
106 * @page: the page that's being released
107 * @gfp_mask: passed from Linux VFS, ignored by us
109 * Call try_to_free_buffers() if the buffers in this page can be
115 static int gfs2_aspace_releasepage(struct page
*page
, gfp_t gfp_mask
)
117 struct inode
*aspace
= page
->mapping
->host
;
118 struct gfs2_sbd
*sdp
= aspace
->i_sb
->s_fs_info
;
119 struct buffer_head
*bh
, *head
;
120 struct gfs2_bufdata
*bd
;
123 if (!page_has_buffers(page
))
126 head
= bh
= page_buffers(page
);
130 while (atomic_read(&bh
->b_count
)) {
131 if (atomic_read(&aspace
->i_writecount
)) {
132 if (time_after_eq(jiffies
, t
+
133 gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
)) {
134 stuck_releasepage(bh
);
145 gfs2_assert_warn(sdp
, !buffer_pinned(bh
));
149 gfs2_assert_warn(sdp
, bd
->bd_bh
== bh
);
150 gfs2_assert_warn(sdp
, list_empty(&bd
->bd_list_tr
));
151 gfs2_assert_warn(sdp
, list_empty(&bd
->bd_le
.le_list
));
152 gfs2_assert_warn(sdp
, !bd
->bd_ail
);
153 kmem_cache_free(gfs2_bufdata_cachep
, bd
);
154 bh
->b_private
= NULL
;
157 bh
= bh
->b_this_page
;
162 return try_to_free_buffers(page
);
165 static struct address_space_operations aspace_aops
= {
166 .writepage
= gfs2_aspace_writepage
,
167 .releasepage
= gfs2_aspace_releasepage
,
171 * gfs2_aspace_get - Create and initialize a struct inode structure
172 * @sdp: the filesystem the aspace is in
174 * Right now a struct inode is just a struct inode. Maybe Linux
175 * will supply a more lightweight address space construct (that works)
178 * Make sure pages/buffers in this aspace aren't in high memory.
180 * Returns: the aspace
183 struct inode
*gfs2_aspace_get(struct gfs2_sbd
*sdp
)
185 struct inode
*aspace
;
187 aspace
= new_inode(sdp
->sd_vfs
);
189 mapping_set_gfp_mask(aspace
->i_mapping
, GFP_KERNEL
);
190 aspace
->i_mapping
->a_ops
= &aspace_aops
;
191 aspace
->i_size
= ~0ULL;
192 aspace
->u
.generic_ip
= NULL
;
193 insert_inode_hash(aspace
);
198 void gfs2_aspace_put(struct inode
*aspace
)
200 remove_inode_hash(aspace
);
205 * gfs2_ail1_start_one - Start I/O on a part of the AIL
206 * @sdp: the filesystem
207 * @tr: the part of the AIL
211 void gfs2_ail1_start_one(struct gfs2_sbd
*sdp
, struct gfs2_ail
*ai
)
213 struct gfs2_bufdata
*bd
, *s
;
214 struct buffer_head
*bh
;
217 BUG_ON(!spin_is_locked(&sdp
->sd_log_lock
));
222 list_for_each_entry_safe_reverse(bd
, s
, &ai
->ai_ail1_list
,
226 gfs2_assert(sdp
, bd
->bd_ail
== ai
);
228 if (!buffer_busy(bh
)) {
229 if (!buffer_uptodate(bh
))
230 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
));
571 INIT_LIST_HEAD(&bd
->bd_list_tr
);
573 lops_init_le(&bd
->bd_le
, &gfs2_buf_lops
);
575 lops_init_le(&bd
->bd_le
, &gfs2_databuf_lops
);
581 unlock_page(bh
->b_page
);
585 * gfs2_pin - Pin a buffer in memory
586 * @sdp: the filesystem the buffer belongs to
587 * @bh: The buffer to be pinned
591 void gfs2_pin(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
)
593 struct gfs2_bufdata
*bd
= bh
->b_private
;
595 gfs2_assert_withdraw(sdp
, test_bit(SDF_JOURNAL_LIVE
, &sdp
->sd_flags
));
597 if (test_set_buffer_pinned(bh
))
598 gfs2_assert_withdraw(sdp
, 0);
602 /* If this buffer is in the AIL and it has already been written
603 to in-place disk block, remove it from the AIL. */
606 if (bd
->bd_ail
&& !buffer_in_io(bh
))
607 list_move(&bd
->bd_ail_st_list
, &bd
->bd_ail
->ai_ail2_list
);
608 gfs2_log_unlock(sdp
);
610 clear_buffer_dirty(bh
);
613 if (!buffer_uptodate(bh
))
614 gfs2_io_error_bh(sdp
, bh
);
620 * gfs2_unpin - Unpin a buffer
621 * @sdp: the filesystem the buffer belongs to
622 * @bh: The buffer to unpin
627 void gfs2_unpin(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
,
630 struct gfs2_bufdata
*bd
= bh
->b_private
;
632 gfs2_assert_withdraw(sdp
, buffer_uptodate(bh
));
634 if (!buffer_pinned(bh
))
635 gfs2_assert_withdraw(sdp
, 0);
637 mark_buffer_dirty(bh
);
638 clear_buffer_pinned(bh
);
642 list_del(&bd
->bd_ail_st_list
);
645 struct gfs2_glock
*gl
= bd
->bd_gl
;
646 list_add(&bd
->bd_ail_gl_list
, &gl
->gl_ail_list
);
647 atomic_inc(&gl
->gl_ail_count
);
650 list_add(&bd
->bd_ail_st_list
, &ai
->ai_ail1_list
);
651 gfs2_log_unlock(sdp
);
655 * gfs2_meta_wipe - make inode's buffers so they aren't dirty/pinned anymore
656 * @ip: the inode who owns the buffers
657 * @bstart: the first buffer in the run
658 * @blen: the number of buffers in the run
662 void gfs2_meta_wipe(struct gfs2_inode
*ip
, uint64_t bstart
, uint32_t blen
)
664 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
665 struct inode
*aspace
= ip
->i_gl
->gl_aspace
;
666 struct buffer_head
*bh
;
669 bh
= getbuf(sdp
, aspace
, bstart
, NO_CREATE
);
671 struct gfs2_bufdata
*bd
= bh
->b_private
;
673 if (test_clear_buffer_pinned(bh
)) {
674 struct gfs2_trans
*tr
= current
->journal_info
;
676 list_del_init(&bd
->bd_le
.le_list
);
677 gfs2_assert_warn(sdp
, sdp
->sd_log_num_buf
);
678 sdp
->sd_log_num_buf
--;
679 gfs2_log_unlock(sdp
);
686 uint64_t blkno
= bh
->b_blocknr
;
688 list_del(&bd
->bd_ail_st_list
);
689 list_del(&bd
->bd_ail_gl_list
);
690 atomic_dec(&bd
->bd_gl
->gl_ail_count
);
692 gfs2_log_unlock(sdp
);
693 gfs2_trans_add_revoke(sdp
, blkno
);
695 gfs2_log_unlock(sdp
);
699 clear_buffer_dirty(bh
);
700 clear_buffer_uptodate(bh
);
712 * gfs2_meta_cache_flush - get rid of any references on buffers for this inode
713 * @ip: The GFS2 inode
715 * This releases buffers that are in the most-recently-used array of
716 * blocks used for indirect block addressing for this inode.
719 void gfs2_meta_cache_flush(struct gfs2_inode
*ip
)
721 struct buffer_head
**bh_slot
;
724 spin_lock(&ip
->i_spin
);
726 for (x
= 0; x
< GFS2_MAX_META_HEIGHT
; x
++) {
727 bh_slot
= &ip
->i_cache
[x
];
734 spin_unlock(&ip
->i_spin
);
738 * gfs2_meta_indirect_buffer - Get a metadata buffer
739 * @ip: The GFS2 inode
740 * @height: The level of this buf in the metadata (indir addr) tree (if any)
741 * @num: The block number (device relative) of the buffer
742 * @new: Non-zero if we may create a new buffer
743 * @bhp: the buffer is returned here
745 * Try to use the gfs2_inode's MRU metadata tree cache.
750 int gfs2_meta_indirect_buffer(struct gfs2_inode
*ip
, int height
, uint64_t num
,
751 int new, struct buffer_head
**bhp
)
753 struct buffer_head
*bh
, **bh_slot
= ip
->i_cache
+ height
;
756 spin_lock(&ip
->i_spin
);
759 if (bh
->b_blocknr
== num
)
764 spin_unlock(&ip
->i_spin
);
770 error
= gfs2_meta_reread(ip
->i_sbd
, bh
,
771 DIO_START
| DIO_WAIT
);
779 bh
= gfs2_meta_new(ip
->i_gl
, num
);
781 error
= gfs2_meta_read(ip
->i_gl
, num
,
782 DIO_START
| DIO_WAIT
, &bh
);
787 spin_lock(&ip
->i_spin
);
788 if (*bh_slot
!= bh
) {
793 spin_unlock(&ip
->i_spin
);
797 if (gfs2_assert_warn(ip
->i_sbd
, height
)) {
801 gfs2_trans_add_bh(ip
->i_gl
, bh
, 1);
802 gfs2_metatype_set(bh
, GFS2_METATYPE_IN
, GFS2_FORMAT_IN
);
803 gfs2_buffer_clear_tail(bh
, sizeof(struct gfs2_meta_header
));
805 } else if (gfs2_metatype_check(ip
->i_sbd
, bh
,
806 (height
) ? GFS2_METATYPE_IN
: GFS2_METATYPE_DI
)) {
817 * gfs2_meta_ra - start readahead on an extent of a file
818 * @gl: the glock the blocks belong to
819 * @dblock: the starting disk block
820 * @extlen: the number of blocks in the extent
824 void gfs2_meta_ra(struct gfs2_glock
*gl
, uint64_t dblock
, uint32_t extlen
)
826 struct gfs2_sbd
*sdp
= gl
->gl_sbd
;
827 struct inode
*aspace
= gl
->gl_aspace
;
828 struct buffer_head
*first_bh
, *bh
;
829 uint32_t max_ra
= gfs2_tune_get(sdp
, gt_max_readahead
) >>
830 sdp
->sd_sb
.sb_bsize_shift
;
833 if (!extlen
|| !max_ra
)
838 first_bh
= getbuf(sdp
, aspace
, dblock
, CREATE
);
840 if (buffer_uptodate(first_bh
))
842 if (!buffer_locked(first_bh
)) {
843 error
= gfs2_meta_reread(sdp
, first_bh
, DIO_START
);
852 bh
= getbuf(sdp
, aspace
, dblock
, CREATE
);
854 if (!buffer_uptodate(bh
) && !buffer_locked(bh
)) {
855 error
= gfs2_meta_reread(sdp
, bh
, DIO_START
);
865 if (buffer_uptodate(first_bh
))
874 * gfs2_meta_syncfs - sync all the buffers in a filesystem
875 * @sdp: the filesystem
879 void gfs2_meta_syncfs(struct gfs2_sbd
*sdp
)
881 gfs2_log_flush(sdp
, NULL
);
883 gfs2_ail1_start(sdp
, DIO_ALL
);
884 if (gfs2_ail1_empty(sdp
, DIO_ALL
))